ECE 5325/6325: Wireless Communication Systems

Lecture Notes, Spring 2010

Prof. Neal Patwari

University of Utah
Department of Electrical and Computer Engineering
c
2010
ECE 5325/6325 Spring 2010 2

Contents
1 Cellular Systems Intro 5
1.1 Generation Zero . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2 Cellular . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.3 Key Terms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.4 The Cellular Radio . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

2 Frequency Reuse 8
2.1 Cellular Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.1.1 Channel Assignment within Group . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2 Handoff . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.3 Co-Channel Interference . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.3.1 Downtilt . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.4 Adjacent Channel Interference . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

3 Trunking 14
3.1 Blocked calls cleared . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.2 Blocked calls delayed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

4 Increasing Capacity and Coverage 16

4.1 Sectoring . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
4.2 Microcells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
4.3 Repeaters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
4.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

5 Free Space Propagation 18

5.1 Received Power Reference . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
5.2 Antennas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
5.3 Power Flux Density . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

6 Large Scale Path Loss Models 22

6.1 Log Distance Path Loss . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
6.2 Multiple Breakpoint Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

7 Reflection and Transmission 24

8 Two-Ray (Ground Reflection) Model 26

8.1 Direct Path . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
8.2 Reflected Path . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
8.3 Total Two-Ray E-Field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

9 Cellular Large Scale Path Loss Models 27

9.1 Okumura-Hata Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

10 Exam 1 29
ECE 5325/6325 Spring 2010 3

11 Link Budgeting 29
11.1 Link Budget Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
11.2 Thermal noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
11.3 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

12 Diffraction 32

13 Rough Surface Scattering 34

14 Multipath Fading 35
14.1 Multipath . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
14.2 Temporal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
14.3 Channel Impulse Response . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
14.4 Received Power . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
14.5 Time Dispersion Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

15 Fade Distribution 38
15.1 Rayleigh Fading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
15.2 Ricean fading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

16 Doppler Fading 41
16.1 One Component . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
16.2 Many Components . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
16.3 System Design Implications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

17 Course Project 43
17.1 Topic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
17.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
17.3 Writeup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
17.4 Presentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

18 Digital Communications: Overview 44

18.1 Orthogonal Waveforms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
18.2 Linear Combinations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
18.3 Reception . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
18.4 How to Choose a Modulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
18.5 Bandwidth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
18.6 Intersymbol Interference . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
18.7 Cost of Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
18.8 Channel Capacity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

19 Modulation 49
19.1 PAM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
19.2 M-ary QAM and PSK . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
19.3 FSK . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
19.4 MSK . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
ECE 5325/6325 Spring 2010 4

20 Fidelity 51
20.1 Link Budgets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
20.2 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

21 Implementation Costs 54
21.1 Power Amplifiers and Constant Envelope . . . . . . . . . . . . . . . . . . . . . . . . 54
21.1.1 Offset QPSK . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
21.1.2 π/4 QPSK . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
21.1.3 Other Modulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
21.2 Sync . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
21.2.1 Energy Detection of FSK . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
21.2.2 Differential PSK . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57

22 Fading Demo 59

23 Multi-carrier Modulation 59
23.1 OFDM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

24 Channel Coding 62
24.1 Block Codes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
24.2 Error Detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
24.3 System Performance and Costs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65

25 Spread Spectrum 65
25.1 FH-SS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
25.2 DS-SS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66

26 Rake Receiver 70

27 Channel Measurement 71

28 Diversity 72

29 Diversity Combining 75
29.1 Selection Combining . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
29.2 Scanning Combining . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
29.3 Equal Gain Combining . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
29.4 Maximal Ratio Combining . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78

30 MIMO 79
30.1 Revisit Maximal Ratio Combining . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
30.2 Alamouti code . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
30.3 MIMO Channel Representation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
30.4 Capacity of MIMO Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
ECE 5325/6325 Spring 2010 5

Lecture 1

Today: (1) Syllabus, (2) Cellular Systems Intro

1 Cellular Systems Intro

Reading: Ch 1, Ch 3.1 from Rappaport.

1.1 Generation Zero

The study of the history of cellular systems can help us understand the need for the system design
concepts we have today.
One of the major developments in WWII was the miniaturization of FM radio components to
a backpack or handheld device (the walkie-talkie), a half-duplex (either transmit or receive, not
both) push-to-talk communication device. After returning from war, veterans had the expectation
that wireless communications should be available in their civilian jobs [19]. But the phone sytem,
the Public Switched Telephone Network (PTSN) was: wired, and manually switched at telephone
exchanges. In 1952, the mobile telephone system (MTS) was designed to serve 25 cities in the US
[6] (including one in Salt Lake City [5]). In each city, an additional telephone exchange office was
created for purpose of connection with the mobile telephones [19]. The MTS and later the improved
mobile telephone system (IMTS), introduced in 1964, were not particularly spectrally efficient.
• They were allocated a total bandwidth of about 2 MHz. Frequency modulation (FM) was the
bet technology, and was narrowband, so they chose to operate as frequency division multiple
access (FDMA), in which each channel was allocated a non-overlapping frequency band within
the 2 MHz.
• The PTSN is full duplex (transmit and receive simultaneously) in IMTS, so it required two
channels for each call, one uplink (to the base station) and one downlink (to the mobile
receiver). Note MTS had been half duplex, i.e., only one party could talk at once.
• The FCC required them to operate over an entire city (25 mile radius). Since the coverage
was city wide, and coverage did not exist outside of the cities, there was no need for handoff.
Control was manual, and the control channel was open for anyone to hear. In fact, users were
required to listen to the control channel. When the switching operator wanted to connect to any
mobile user, they would announce the call on the control channel. If the user responded, they
would tell the user which voice channel to turn to. Any other curious user could listen as well. A
mobile user could also use the control channel to request a call to be connected. The system was
congested, so there was always activity.
The demand was very high, even at the high cost of about $400 per month (in 2009 dollars).
There were a few hundred subscribers in a city [6] but up to 20,000 on the waiting list [19]. The
only way to increase the capacity was to allocate more bandwidth, but satisfying the need would
have required more bandwidth than was available.
The downsides to MTS took a significant amount of technological development to address, and
the business case was not clear (AT&T developed the technologies over 35 years, but then largely
ignored it during the 1980s when it was deployed [6]).
ECE 5325/6325 Spring 2010 6

1.2 Cellular
The cellular concept is to partition a geographical area into “cells”, each covering a small fraction
of a city. Each cell is allocated a “channel group”, i.e., a subset of the total list of channels. A
second cell, distant from a first cell using a particular channel group, can reuse the same channel
group. This is called “frequency reuse”. This is depicted in Figure 3.1 in Rappaport. This assumes
that at a long distance, the signals transmitted in the first cell are too low by the time they reach
the second cell to significantly interfere with the use of those channels in the second cell.
There are dramatic technical implications of the cellular concept. First, rather than one base
station, you need dozens or hundreds, deployed across a city. You need automatic and robust
mobility management (handoff) to allow users to cross cell lines and continue a phone call. Both of
these are actually enabled by semiconductor technology advancement, which made the base stations
and the automated wired PSTN cheaper [19].
What is frequency reuse? What is required for a handoff? These are to be covered in the next
lecture.

1.3 Key Terms

Communication between two parties (a “link”), in general, can be one of the following:

• Simplex : Data/Voice is transferred in only one direction (e.g., paging). Not even an acknowl-
edgement of receipt is returned.

• Half Duplex : Data/Voice is transferred in one direction at a time. One can’t talk and listen
at the same time. One channel is required.

• Full Duplex : Data/Voice can be transferred in both directions between two parties at the
same time. This requires two channels.

In a cellular system, there is full duplex communication, between a base station and a mobile.
The two directions are called either uplink (from mobile to base station) or downlink (from BS
to mobile). The downlink channel is synonymous with “forward channel”; the uplink channel is
synonymous with the “reverse channel”.
Simultaneous communication on the many channels needed for many pairs of radios to commu-
nicate can be accomplished by one of the following:

• Frequency division multiple access (FDMA): Each channel occupies a different band of the
frequency spectrum. Each signal can be upconverted to a frequency band by multiplying it
by a sinusoid at the center frequency of that band, and then filtering out any out-of-band
content, as you have learned in ECE 3500.

• Time division multiple access (TDMA): Every period of time can be divided into short seg-
ments, and each channel can be carried only during its segment. This requires each device to
be synchronized to have the same time clock.

See Figures 9.2 and 9.3, pages 450-453, in the Rappaport book.
Physical “parts” of a cellular system:

1. Public switched telephone network (PSTN): Wired telephone network, connecting homes,
businesses, switching centers.
ECE 5325/6325 Spring 2010 7

2. Mobile switching center (MSC), a.k.a. mobile telephone switching office (MTSO): Controls
connection of wireless phone calls through the base stations to the PSTN. Connected either
by wire or by wireless (microwave relay) to the base stations.

3. Base station (BS): Maintains direct wireless connection to cell phones in its cell. Typically
maintains many connections simultaneously. Has multiple antennas, some for downlink and
some for uplink.

See Figure 1.5, page 14, in the Rappaport book.

In cellular systems, there are actually two types of channels: (1) Control, and (2) Communica-
tion. The control channel is needed to tell the mobile device what to do, or for the mobile to tell
the BS or MSC what to do. The communication channel is the “voice channel” or data channel,
the actual information that the user / system needs to convey in order to operate. Since we also
have a forward and reverse channel (for full duplex comms), we have

1. FCC: Forward control channel

2. FVC: Forward voice channel

3. RCC: Reverse control channel

4. RVC: Reverse voice channel

Information needed for/from the mobile user: (1) mobile identification number (MIN), basically
the telephone number; (2) Electronic serial number (ESN), information about who will get billed
for this call, and needed to find the home location register (HLN), where the information about
this mobile is stored; (3) station class mark, which indicates how much power a mobile is able to
transmit.
We will review “How a call is made from PSTN to mobile” handout from Prof. Cynthia Furse.

Lecture 2

Today: (1) Frequency Reuse, (2) Handoff

1.4 The Cellular Radio

A cell is the area in which a mobile is served by a single BS. What is the power transmitted by the
radios in a cell system? Limits differ by country.

1. Basestation maximum = 100 W maximum Effective Radiated Power (ERP), or up to 500 W

in rural areas [20]

2. Cell phone: typically 0.5 W; but limited by power absorbed by human tissue in test measure-
ments. The measurement is called Specific Absorption Rate (SAR). For TDMA, transmitter
is only on a fraction of the time. For CDMA, transmit power is lowered when close to a BS.

Cell phone exposure limits are typically set to meet both US and European standards. Note that
cellular community uses the term “mobile station” (MS) to describe the cell phone or mobile, even
though it is an oxymoron.
ECE 5325/6325 Spring 2010 8

2 Frequency Reuse
2.1 Cellular Geometry
When the signal from the base station is weak, the mobile will not be able to be served by the BS.
What shape is a cell? See Figure 1. These are in order from best to worst:

1. A random shape dependent on the environment.

2. Circular (theoretical): If path loss was a strictly decreasing function of distance, say, 1/dn ,
where d is the distance from BS to mobile, then the cell will be a perfect circle. This is never
true in reality.

3. An approximation to the theoretical shape: required for a tesselation (non-overlapping repeti-

tive placement of a shape that achieves full coverage. Think floor tiles.) Possible “tile” shapes
include triangles, squares, hexagons. Hexagons are closest to reality.

(a) (b)

Figure 1: Theoretical coverage area, and measured coverage area. In (b), from measurements, with
red, blue, green, and yellow indicating signal strength, in decreasing order. From Newport et. al.
[12].

What is the size of the coverage area of a cell? (From C. Furse)

• Macrocell: Diameter typically greater than 2000 feet, up to 25 miles

• Microcell: Diameter typically 400 to 2000 feet

• Picocell: Diameter typically 100 feet

There are also cell sites in trucks that can be driven to replace downed cell towers after natural
disasters, or to create additional capacity for large gatherings (football games, rock concerts), called
“cell on wheels” (COW) or “cell on light truck sites” (COLTS).
As we mentioned in lecture 1, a cellular system assigns subsets, “channel groups”, of the total
set of channels to each cell. Call the total number of channels S, and the number of channel groups
N . Then there are on average k = S/N channels per cell. (In reality, k may vary between groups.)
Then with N channel groups, how do we assign them? We want cells that reuse group A, for
example, to be as far apart as possible.
ECE 5325/6325 Spring 2010 9

2.1.1 Channel Assignment within Group

See Section 3.3. Which channels should be assigned to a cell? First, it is best to separate channels
in the group in frequency as much as possible to reduce adjacent channel interference (studied
later). But which channels are assigned? Two ways:

1. Fixed assignment: Each basestation has a fixed set of channels to use. Simple, but a busy
cell will run out of channels before a neighboring cell. System performance will be limited by
the most crowded cell.

2. Dynamic allocation: Each basestation can change the channels it uses. Channels in neigh-
boring cells must still be different. This requires more careful control, but increases the
capacity.

For example, a typical city needs more channels in its business districts during the day, and in its
residential areas at night and on weekends.
For general shapes, this can be seen as a graph coloring problem, and is typically covered in a
graph theory course.
For hexagons, we have simple channel group assignment. Consider N = 3, 4, 7, or 12 as seen in
Figure 2. A tesselation of these channel groupings would be a “cut and paste” tiling of the figure.
The tiling of the N = 4 example is shown in Figure 3.

Figure 2: Hexagonal tesselation and channel groupings for N = 3, 4, 7, and 12.

Figure 3: Frequency reuse for N = 4.

Example: Call capacity of N = 4 system

Assume that 50 MHz is available for forward channels, and you will deploy GSM. Each channel is
ECE 5325/6325 Spring 2010 10

200 kHz, but using TDMA, 8 simultaneous calls can be made on each channel. How large is k?
How many forward calls can be made simultaneously for the cellular system depicted in Figure 3?
Solution: There are 50 MHz / 0.2 MHz or 250 total channels. With N = 4, then k = 250/4 = 62.5,
and with (about) 62.5 channels, 8(62.5) = 500 calls can be made simultaneously in each cell. There
are 28 cells on the cell map in Figure 3, so the total forward calls is 28(500) = 14 × 103 calls can
be made simultaneously.

Why wouldn’t you choose N as low as possible? There are interference limits, which will be
discussed in more detail later.
How do you generally “move” from one cell to the co-channel cell (a second cell assigned the same
channel group)? All cellular tiling patterns can be represented using two non-negative integers, i
and j. The integer i is the number of cells to move from one cell in one direction. Then, turn 60
degrees counter-clockwise and move j cells in the new direction. For Figure 3, this is i = 2, j = 0.
In this notation, the number of cells can be shown to be:

N = i2 + ij + j 2

What is the distance between two co-channel cell BSes? If the distance between the BS and a
vertex in its cell is called R, its “radius”, then you can show this co-channel reuse distance D is:
√
D = R 3N
√
The ratio of D/R = 3N is called Q, the co-channel reuse ratio.

2.2 Handoff
See Section 3.4. As a mobile travels beyond the coverage region of its serving BS, it must be trans-
ferred to better BS. If the received power drops too low prior to handoff, the call is “dropped”.
Rappaport denotes this minimum received power, below which a call cannot be received, as
Pr,minimum useable . We want to initiate a handoff much prior to this point, so we set a higher
threshold Pr,handof f at which the MSC initiates the handoff procedure. Note the signal strength
varies quickly due to multipath fading, but we are most interested an short-term averaged received
power.
Because power may go down quickly, particularly at high mobile speeds, this handoff needs to
happen quickly. In GSM, handoff is typically within 1-2 seconds. In AMPS, this was 10 seconds
(higher potential for dropped calls!)
Define handoff margin as ∆

∆ = Pr,handof f − Pr,minimum useable .

How much margin is needed to handle a mobile at driving speeds?

Example: Handoff Margin

Let the speed of a mobile be v = 35 meters/sec. For n = 4, a cell radius of 500 meters (the distance
at which the power is at the threshold), and a 2 second handoff, what ∆ is needed?
Solution: Assume the mobile is driving directly away from the BS, so distance d changes by 70
meters in two seconds. Consider the received power at the two times:

Pr,minimum useable = Π0 − 10n log10 d

Pr,handof f = Π0 − 10n log10 (d − 70)
ECE 5325/6325 Spring 2010 11

Taking the difference of the two equations (the 2nd minus the 1st),

d
∆ = 10n log 10 d − 10n log10 (d − 50) = 10n log10
d − 70

Plugging in that the call is dropped at d = 500 meters, we have ∆ = 40 log1 0 500
430 = 2.6 dB.

Note that in this simple example, the propagation equation used is for “large scale path loss”
only, which changes slowly. Typically, shadowing (caused by large geographical features and build-
ings blocking the signal) will play a more important role in quick changes in received power.
Mobile handoff strategies:

1. MSC controlled: all BSes measure all RVC, using a spare “locator receiver”. The MSC decides
when to handoff.

2. Mobile-assisted hand-off (MAHO): mobile measures the FCC from neighboring BSes, and
reports them to the MSC. This ends up leading to faster handoffs. MAHO is used in 2G
systems.

This assumes that there is a channel in the new BS to offer the entering mobile! But there may
not be, and the call may be dropped for this reason. Users complain about dropped calls. So BSes
may reserve “guard channels” purely for handoff purposes, which then are not offered to mobiles
making new calls.
CDMA (Verizon, e.g.) phones do not require handoff as we’ve described above (here called
“hard handoff”). In CDMA, a user does not need to switch “channel”, so handoff changes only
which BS is receiving the signal and sending the replies. In fact, if multiple BSes receive the signal
from the same mobile, the MSC can combine / choose the best among the three.
Discussion: What are some of the problems with handoff vs. what the Rappaport book has
presented?

2.3 Co-Channel Interference

What is the ratio of signal power to interference power? This is the critical question regarding the
limits on how low we may set N . This ratio is abbreviated S/I. Signal power is the desired signal,
from the base station which is serving the mobile. The interference is the sum of the signals sent
by co-channel base stations, which is not intended to be heard by mobiles in this cell. The S/I
ratio is defined as:
S S
= Pi0
I i=1 Ii
where Ii is the power received by the mobile from a co-channel BS, of which there are i0 , and S is
the power received by the mobile from the serving BS.
As a first order, before we get more complicated, we model the received power as inversely
proportional to distance to the n power, for some constant path loss exponent n:

S = cd−n

for some real value constant c.

We typically look at the worst case, when the S/I is the lowest. This happens when the mobile
is at the vertex of the hexagonal cell, i.e., at the radius R from the serving BS. So we know
ECE 5325/6325 Spring 2010 12

Figure 4: Desired, and interfering signal for a mobile (M) from a serving and co-channel base
station.

S = cR−n . What are the distances to the neighboring cells from the mobile at the vertex? This
requires some trigonometry work. The easiest approximation is (1) that only the first “tier” of
co-channel BSes matter; (2) all mobile-to-co-channel-BS distances are approximately equal to D,
the distance between the two co-channel BSes. In this case,

S S cR−n (D/R)n (3N )n/2

= Pi0 ≈ = = (1)
I i=1 Ii
i0 (cD−n ) i0 i0

where i0 is the number of co-channel cells in the first tier. For all N , we typically have i0 = 6
(try it out!); this will change when using sector antennas, so it can be useful to leave i0 in the
denominator. It is useful to report the S/I in dB, because S/I requirements are typically given in
dB.

Example: AMPS design

Assume that 18 dB of S/I is required for acceptable system operation. What minimum N is
required? Test for n = 3 and n = 4.
n/2
Solution: 18 dB is 1018/10 = 101.8 = 63.1. Using (1), we need (3N6) ≥ 63.1, so
1
N ≥ [6(63.1)]2/n
3
For n = 3, N = 17.4; for n = 4, N = 6.5. Clearly, a high path loss exponent is important for
frequency reuse.

2.3.1 Downtilt
The Rappaport does not cover antenna downtilt, but it is an important practical concept. Compare
the elevation angles from the BS to mobile (Q1 in Figure 4) and co-channel BS to the mobile (Q2
in Figure 4). Note Q2 is lower (closer to the horizon) than from the serving BS. The great thing is,
we can provide less gain at angle Q2 than at Q1, by pointing the antenna main lobe downwards.
This is called downtilt. For example, if the gain at Q1 is 5 dB more than the gain at Q2, then the
we have added 5 dB to the S/I ratio. Having a narrow beam in the vertical plane is also useful to
reduce the delay spread and thus inter-symbol interference (ISI) [2], which we will introduce in the
2nd part of this course. This narrow vertical beam is pointed downwards, typically in the range of
5-10 degrees. The effect is to decrease received power more quickly as distance increases; effectively
increasing n. This is shown in Figure 5. How do you calculate the elevation angle from a BS to a
mobile? This angle is the inverse tangent of the ratio between BS height ht and horizontal distance
from the mobile to BS, d. But, at very low ratios, we can approximate tan−1 (x) ≈ x. So the angle
is ht /d.
ECE 5325/6325 Spring 2010 13

Figure 5: A diagram of a BS antenna employing downtilt to effectively increase the path loss at
large distances. From [9].

Ever wonder why base station antennas are tall and narrow? The length of an antenna in any
dimension is inversely proportional to the beamwidth in that dimension. The vertical beamwidth
needs to be low (5-10 degrees), so the antenna height is tall. The horizonal pattern beamwidths
are typically wide (120 degrees or more) so the antenna does not need to be very wide.
For more information, perhaps for a project, please consult [7].
Discussion: What are some of the problems with coverage and frequency reuse vs. what the
Rappaport book has presented?

Lecture 3

Today: (1) Interference (from Lecture 2); (2) Trunking

2.4 Adjacent Channel Interference

Our standard, non-ideal radios do not perfectly filter out any out-of-band signals. Thus any signal
that a mobile sends in another channel (besides its assigned channel) is interference at the BS
w.r.t. the desired signal sent by another mobile in that channel. Each mobile’s receiver also must
filter out out-of-band signals from the BS, which does send signals on all channels. One standard
way of making this problem less difficult is to assign non-adjacent channels within each cell’s channel
group.

Example: Assign S = 50 channels into groups for N = 7.

Solution: We should have about k = 50/7 ≈ 7 channels per group. For group 1, use forward
channels {1, 8, 15, 22, 29, 36, 43, 50}. For group i, i = 2, . . . 7, let the channels for group i consist of
{i, i + 7, i + 14, i + 21, i + 28, i + 35, i + 42}.
Because of the spacing in channels used in one group, the specifications for transmit and receive
filters will be less strict, and can be more effective at reducing adjacent channel interference.
There is still the near-far effect. If a TX near the BS is producing just a little bit of out-of-band
noise, it might swamp out the desired signal transmitted by a TX far away to the same BS.
One solution is power control, i.e., reducing the power level TXed by mobiles close to the BS,
since a high TX power is not necessary in that case. This reduces their out-of-band noise as well.
This means that antennas far away must transmit larger power than those nearby. Compared to
ECE 5325/6325 Spring 2010 14

a mobile transmitting full power all the time, power control extends battery life, and generally
reduces even co-channel interference on the reverse channel. However, power control requires, well,
control. Controlling a mobile means communication from the BS to the mobile to inform it whether
to increase or decrease its power, which then requires data overhead.
Tight power control is particularly required in CDMA systems, in which the “near-far problem”
is even more of a problem.

3 Trunking
Trunking refers to sharing few channels among many users. Let U be the number of users, and C
be the number of channels. Each user requires a channel infrequently, so a dedicated channel for
each user is not required. But, the request for a channel happens at random times, and so for any
C < U , it is possible that there will be more requests than channels.

• Erlang: A “unit” of measure of usage or traffic intensity. One Erlang is the traffic intensity
carried by one channel that is occupied all of the time. 0.1 Erlang is the same channel occupied
only 10% of the time.

• Average holding time: Average call duration, denoted H.

• Call rate: Average number of calls per unit time, denoted λ. Typically taken to be at the
busiest time of day.

• Total offered traffic intensity: The total amount of traffic users request of the system, denoted
A.

• Grade of Service (GOS): The probability an offered call will be blocked (and thus not served,
or carried by the system).

Rappaport presents that an average user will request (offer) this much traffic, Au = λH. For
example, if a user makes on average, two calls per hour, and that call lasts an average of 3 minutes,
2
Au = 60 min 3 min = 0.1 Erlang. (Check your units!)
Then, to compute the total offered traffic intensity, and the total offered traffic intensity per
channel (denoted Ac ),
A = U Au , Ac = A/C
For the above example, assume that there are 1000 users and 200 channels. Then A = 1000(0.1) =
100, and Ac = 100/200 = 0.5.
Note that Ac is a measure of the efficiency of the utilization of the channels.
How should we design our system? Obviously, Ac should be less than one (A < C); or
we’ll never satisfy our call demand. But how should we set U , Au , C to satisfy our customers?
First choice: what do we do when a call is offered (requested) but all channels are full?

• Blocked calls cleared: Ignore it!

• Blocked calls delayed: Postpone it!

ECE 5325/6325 Spring 2010 15

3.1 Blocked calls cleared

1. Call requests are a Poisson process. That is, the times between calls are exponentially dis-
tributed, and memoryless.

2. Call durations are also exponentially distributed.

3. Rather than a finite number U of users each requesting Au traffic, we set the total offered
traffic as a constant A, and then let U → ∞ and Au → 0 in a way that preserves U Au = A.
This is the “infinite number of users” assumption that simplifies things considerably.

These assumptions, along with the blocked calls cleared setup of the system, leads to the Erlang
B formula:
AC /C!
GOS = P [blocking] = PC (2)
k
k=0 A /k!
Since C is very high, it’s typically easier to use Figure 3.6 on page 81. By setting the desired GOS,
we can derive what number of channels we need; or the maximum number of users we can support
(remember A = U Au ); or the maximum Au we can support (and set the number of minutes on our
calling plans accordingly).

3.2 Blocked calls delayed

Instead of clearing a call; put it in a queue (a first-in, first-out line). Have it wait its turn for a
channel. (“Calls will be processed in the order received”). There are now two things to determine

1. The probability a call will be delayed (enter the queue), and

2. The probability that the delay will be longer than t seconds.

The first is no longer the same as in (2); it goes up, because blocked calls aren’t cleared, they “stick
around” and wait for the first open channel.
Here, we clarify the meaning of GOS for a blocked calls delayed system. Here it means the
probability that a call will be forced into the queue AND it will wait longer than t seconds before
being served (for some given t).
We need a couple additional assumptions:

1. The queue is infinitely long. In a computer system, this translates to infinite memory.

2. No one who is queued gives up / hangs up (rather than wait).

With these assumptions, we can derive the Erlang C formula, for the probability that a call will be
delayed:
AC
P [delay > 0] = PC−1 k (3)
AC + C! (1 − A/C) k=0 A /k!
It is typically easiest to find a result from Figure 3.7, on page 82. Once it enters the queue, the
probability that the delay is greater than t (for t > 0) is given as
 
C−A
P [delay > t|delay > 0] = exp − t (4)
H
ECE 5325/6325 Spring 2010 16

The two combined are needed to find the marginal (overall) probability that a call will be delayed
AND experience a delay greater than t, the event that we are quantifying in GOS.

GOS = P [delay > t] = P [delay > t|delay > 0] P [delay > 0]

 
C−A
= P [delay > 0] exp − t (5)
H

Example: N = 7 cell cluster

A 7 cell cluster (with N = 7) has 28 MHz allocated to it for forward channels and each channel is
200 kHz. Assume blocked-called-delayed and a probability of delay of 1%, and each user makes one
10 minute call every 3 hours. (a) What is the number of users that can be supported? (b) What
is P [delay > 10] seconds? (c) What if it was a blocked-calls-cleared system with QOS of 1%?
Solution: 30 MHz / 200 kHz = 150 channels, divided among 7 cells, so about 20 channels per cell
(after 1 control channel per cell). (a) With 20 channels, and probability of delay of 1%, looking
at figure 3.7, we see A is about 11. With 11 = Au U and Au = 10/(3 × 60) = 1/18, we have that
U = 198 ≈ 200. But this is per cell. So there can be 7(200) = 1400 users in the 7 cells. (b) in
each cell, C = 20, A = 11, H = 600 sec. So P [delay > t] = (0.01) exp[−(20 − 11)(10)/600] =
0.01 exp(−0.15) = 0.0086. (c) From Figure 3.6, A ≈ 12, so U = 216, for a total of 1512 total users.

3.3 Discussion
What are the problems or benefits we see from the assumptions we’ve made? Are call requests
“memoryless”? Is the exponential interarrival time assumption accurate? When catastrophic events
occur, or major news breaks, what happens? How should a communications system be designed to
handle these cases?

Lecture 4

Today: (1) Sectoring (2) Cell Splitting

4 Increasing Capacity and Coverage

4.1 Sectoring
In sectoring, we divide each cell into three or six “sectors” which are then served by three or six
separate directional antennas, each with beamwidth of about 120 or 60 degrees.
n/2
We showed in Lecture 2 that the S/I ratio is given by (3Ni0) , where N is the reuse ratio, and
i0 is the number of first-tier co-channel base stations. When we used omnidirectional antennas at
each BS, we saw that i0 = 6 regardless of N . By using sector antennas at the BSes, we will show
that i0 will be lower. By reducing the S/I ratio for a given N , we allow a system to be deployed
for a lower N , and therefore a higher capacity system.
However, each cell’s channel group must be divided into three sub-groups. These new groups
have 1/3 or 1/6 the number of channels, and thus the trunking efficiency will be lower.
ECE 5325/6325 Spring 2010 17

Example: Decrease in trunking efficiency for constant N

Let N = 7, each cell has C = 100 channels, and users who make calls with λ = 0.01 per minute with
average holding time 3 minutes. For blocked-calls-cleared and a GOS of 2%, what is the number
of users which can be supported in this cell? Next, using 120 degree sectoring, and otherwise
identical system parameters, what is the number of users which can be supported in this cell?
What percentage reduction in capacity does sectoring with constant N cause?
Solution: For C = 100 and GOS= 0.02, from Figure 3.6, I read A ≈ 99. Thus with Au = 0.01(3) =
0.03, we could support U = 99/0.03 = 3300. For the sectoring case, C = 33.3 in each sector, and
from Figure 3.6, A = 24. So we could support U = 24/0.03 ≈ 800 per sector, or 2400 total in the
cell. The number of users has reduced by 28%.

Example: Reducing N with sector antennas

For the same system, now assume that with 120 degree sectoring, that N can be reduced from 7
to 4. What number of users can be supported?
Solution: Now, the number of channels in each cell goes up to 100(7/4) = 175. So each sector has
C = 58 channels. With GOS = 2%, from Figure 3.6, A ≈ 48, so U ≈ 1600, for a total of 4800 users
per cell. This is a 45% increase upon the N = 7 non-sectored cell.
Why does i0 reduce? Consider again a mobile at the edge of a cell. We need to determine which
of the first tier BSes contribute significantly to the interference signal. Refer to Figures 3.10, 3.11,
for N = 7, P3.28(b) for N = 3, and to Figure 6 for N = 4.

Figure 6: 120 degree sectoring for cellular system with N = 4. Only two first tier BSes significantly
interfere with the middle BS.
So how much does sectoring improve the S/I ratio for particular values of N ?

Example: Sectoring improvement compared to Omni

Compared to when i0 = 6, how much does S/I improve with sectoring?
n/2
Solution: Recall that S/I = (3Ni0) . In dB terms,
S
(dB) = 5n log10 (3N ) − 10 log10 i0
I
So with i0 = 6, the latter term is 7.8 dB. If i0 = 1, 2, and 3, the same term is 0, 3.0, or 4.8 dB. So,
the improvement is 3, 4.8, or 7.8 dB. The particular value of i0 that can be obtained is a function
of N and whether 60 or 120 degree sectoring is used.
ECE 5325/6325 Spring 2010 18

4.2 Microcells
When we introduced “cells” we said the radius was a variable R. The idea of using microcells is
that for a densely populated area, we cut the size of the cell by half. In this microcell-covered
area, the concept of frequency reuse occurs described earlier, only with smaller R. The smaller R
also has the benefit that transmit powers would be cut by a factor of 2n (see Rappaport 3.7.1 for
details). The other main benefit is that by reducing the area of a cell by a factor of four (forced
by cutting R by two) the capacity in the microcell area is increased by four. For example, consider
Figure 7, which shows an original macrocell grid, next to an “inserted” microcell area.
However, at√ the edges of the microcell area, there is a conflict. Cells that were separated
by distance R 3N for the initial R are no longer separated by that much. Conflicts in channel
assignments at the edges are solved by splitting the channel group into two sub-groups. These
subgroups can have different sizes, e.g., the sub-group used for the microcell might have fewer
channels assigned to it compared to the macrocell.
Another problem is that the number of handoffs is increased, since users travel through micro-
cells more quickly. This can be addressed using umbrella cells (page 66) or microcell zones (Section
3.7.4).

4.3 Repeaters
This is Section 3.7.3 in Rappaport. Repeaters can be used to increase the coverage area, particularly
into buildings, tunnels, and canyons. They are bidirectional (they amplify forward and reverse
channels). However, repeaters don’t add any capacity to the system, they just increase the reach
of a BS or MS into “shadowed” areas.

4.4 Discussion
What are some of the problems with the assumptions made in this analysis?

Lecture 5

Today: (1) Free Space (2) Large Scale Path Loss (Intro)

Path loss models are either (1) empirical or (2) theoretical. We’ll start to discuss both. As
you’ll see, empirical models were developed as modifications to theoretical models.

5 Free Space Propagation

Free space is nothing. So why discuss it?
Section 4.3 describes the electric and magnetic fields produced by a small dipole antenna. This
equation is only valid for a small dipole, and is only useful very close to (the near field of) the
antenna. In the “far field” (distances many wavelengths from the antenna), the received power Pr
in free space at a path length d is given in Section 4.2 as
λ 2
 
Pr = Pt Gt Gr (6)
4πd
where Gt and Gr are the transmitter and receiver antenna gains, respectively; Pt is the transmit
power; and λ is the wavelength. Notes:
ECE 5325/6325 Spring 2010 19

(a)

(b)

Figure 7: (a) 68 macrocells vs. (b) 53 macrocells plus 57 microcells.

ECE 5325/6325 Spring 2010 20

• Wavelength λ = c/f , where c = 3×108 meters/sec is the speed of light, and f is the frequency.
We tend to use the center frequency for f , except for UWB signals, it won’t really matter.

• All terms in (6) are in linear units, not dB.

• The effective isotropic radiated power (EIRP) is Pt Gt .

2
• The path loss is Lp = 4πd λ . This term is called the “free space path loss”.

• The received power equation (6) is called the Friis transmission equation, named after Harald
T. Friis (from Wikipedia).

• Free space is used for space communications systems, or radio astronomy. Not for cellular
telephony.

In dB, the expression from (6) becomes

 
4πd
Pr (dBm) = Pt (dBm) + Gt (dB) + Gr (dB) − Lp (dB), where Lp (dB) = 20 log10 (7)
λ

I like to leave Lp (dB) in terms of d/λ, which is a unitless ratio of how many wavelengths the signal
has traveled. The terms Gt (dB) and Gr (dB) are clearly gains, when they are positive, the received
power increases. And as distance increases, Lp (dB) increases, which because of the negative sign,
reduces the received power. We use “G” to denote gains and “L” to denote losses. But a negative
gain is a loss, and a negative loss is a gain.

5.1 Received Power Reference

Note either (6) or (7) can be converted to “refer to” a reference distance. For example, multiply
the top and bottom of (6) by (d0 /d0 )2 for some reference distance d0 :
 2  2
λ d0
Pr = Pt Gt Gr
4πd0 d
 2
d0
= Pr (d0 ) (8)
d

where Pr (d0 ) is the received power at the reference distance d0 , according to (6). Now, we see that
whatever the received power in free space is at distance d0 , the power at d decays as (d0 /d)2 beyond
that distance. In dB terms,
d
Pr (dBm) = Π0 (dBm) − 20 log 10 (9)
d0

where Π0 (dBm) = 10 log 10 Pr (d0 ). This is actually an easier equation to deal with in practice,
because we don’t necessarily know the antenna gains and mismatches, and transmit power; but we
can measure Π0 (dBm). Of course, not in free space – we don’t exist there!
ECE 5325/6325 Spring 2010 21

5.2 Antennas
Antenna gain is a function of angle. The only exception is the (mythical) isotropic radiator.
Def ’n: Isotropic Radiator
An antenna that radiates equally in all directions. In other words, the antenna gain G is 1 (linear
terms) or 0 dB in all directions.

(From Prof. Furse) An isotropic radiator must be infinitesimally small. Does not exist in
practice, but is a good starting point.
Antenna gains can be referred to other ideal antenna types:

• dBi: Gain compared to isotropic radiator. Same as the dB gain we mentioned above because
the isotropic radiator has a gain of 1 or 0 dB.

• dBd: Gain compared to a half-wave dipole antenna. The 1/2 wave dipole has gain 1.64
(linear) or 2.15 dB, so dBi is 2.15 dB greater than dBd.

Technically, any antenna that is not isotropic is directive. Directivity is measured in the far
field from an antenna as:
Pr (maximum)
D=
Pr (isotropic)
where Pr (maximum) is the maximum received power (at the same distance but max across angle),
and Pr (isotropic) is the power that would have been received at that point if the antenna was an
isotropic radiator.
Antennas also have an efficiency. They lose some power without radiating it as EM waves. Thus
the maximum gain is the directivity times the efficiency.
Commonly, we call an antenna directional if it is has a non-uniform horizontal pattern. A dipole
has a “donut-shaped” pattern, which is a circle in its horizontal pattern (slice).
There are also antenna mismatches. We denote these as Γt and Γr . Both are ≤ 1, and only one
if there is a perfect impedance match and no loss.

5.3 Power Flux Density

There is a concept in propagation of power flux density, the amount of power that travels through
a given area. This is a far-field concept only. Power flux is denoted Pd in Rappaport, and has units
of Watts per square meter, W/m2 . In free space,

|E|2
Pd = W/m2
η

where η is the intrinsic impedance of free space, 120πΩ = 377Ω, and |E|2 is the magnitude squared
of the electric field. The idea is that an antenna “captures” some of this power, according to,
effectively, how large the antenna is. We call this the effective antenna aperture, and denote it Ae ,
with units m2 . In short, physically larger antennas are capable of larger Ae , although there is no
exact proportionality. In this case the definition of the received power is

Pr (d) = Pd Ae
ECE 5325/6325 Spring 2010 22

6 Large Scale Path Loss Models

Let’s transition to the real world, where we exist. There are other effects besides radiation, including
attenuation (transmission), reflection, diffraction, scattering, etc. We will discuss each of these in
upcoming lectures. For now, suffice it to say that many signals arrive at the receiver, but with less
power than would be indicated by the Friis equation. The received power varies strongly (5-25 dB)
even for small changes in antenna position, center frequency, and time. But, there is a large effect
caused when the distance (a.k.a. path length) increases by orders of magnitude. This large effect
we call large scale path loss. It is analogous to Lp , but doesn’t necessarily take the same form. We
will re-write Lp as a function of distance in one of two ways:

1. Exponential decay: Lp will include a term proportional to 10αd/10 , where α is a loss factor,
with units dB per meter. In this case, Lp (dB) = αd, which makes it easier to see that α is
dB loss per meter. Equation (6) is typically re-written as:
 2
λ
Pr = Pt Gt Gr 10−αd/10 (10)
4πd

This works well in some conditions, for example, at 60 GHz, at which oxygen molecules absorb
RF radiation, or due to rain at 30 GHz.

2. Power decay: Lp will be proportional to 1/dn , for some path loss exponent n. In free space, it
was proportional to 1/d2 , so this just lets n adjust to the particular environment. Typically, n
ranges between 1.6 and 6, according to Rappaport. From my experience, I’ve seen n between
1.7 and 5.

6.1 Log Distance Path Loss

This is 4.11.3 in Rappaport. This is synonymous with what I call “power decay” above. Actually,
it is the simplest of the models, and makes a big step towards better representation of actual large-
scale path loss. In the log-distance path loss model, we can simply write the received power as a
modification of (9) as
d
Pr (dBm) = Π0 (dBm) − 10n log10 (11)
d0
where Π0 (dBm) is still given by the Friis equation, but now the Lp (dB) term has changed to include
a factor 10n instead of 20. Typically d0 is taken to be on the edge of near-field and far-field, say 1
meter for indoor propagation, and 10-100m for outdoor propagation.
We mentioned that we can find the parameter Π0 (dBm) from measurements. We can also find
the parameter n from measurements. For example, two measurement campaigns I did in office
areas resulted in the estimates of n = 2.30 and 2.98 as shown in Figures 8 and 9.

6.2 Multiple Breakpoint Model

This is Rappaport Section 4.11.4. Empirically measurement studies have shown that the slope of
the Lp vs. distance curve changes after a certain distance [3]. You can see this effect in Figure 9
for d > 20 meters; the path gains at d = 50 meters are all lower than the model, and one can see
the slope changing to an n higher than 2.98. We will discuss theoretical reasons why this might
ECE 5325/6325 Spring 2010 23

−30

−40

Path Gain (dB)

−50

−60

−70
0 1
10 10
Path Length (m)

Figure 8: Wideband path gain measurements (x) at 2.4 GHz as a function of path length d. Linear
fit (—) is with d0 = 1m, n = 2.30, and σdB = 3.92. From [13].

10
Measured Data
0 Channel Model

−10

−20
pi,j − p0

−30

−40

−50

−60

−70

−80 0 1 2
10 10 10
Path Length (m)

Figure 9: Narrowband measurements of path gain minus Π0 (dBm) (o) at 925 MHz as a function
of path length d. Linear fit (—) is with d0 = 1m, n = 2.98, with standard deviation σdB = 7.38.
From [13].
ECE 5325/6325 Spring 2010 24

happen in later lectures. Regardless, we can model the path loss as experiencing more than one
slope in different segments of the log d axis.
(
Π0 (dBm) − 10n1 log10 dd0 , d ≤ d1
Pr (dBm) = (12)
Π1 (dBm) − 10n2 log10 dd1 , d > d1

where Π0 (dBm) is still the Friis received power at a distance d0 , and Π1 (dBm) is the received power
(given by the first line of the equation) at distance d1 , and d0 < d1 . Typically, the slope of the path
loss increases, i.e., n2 > n1 .

Lecture 6

Today: (1) Reflection (2) Two-ray model (3) Cellular Large Scale Path Loss Models

7 Reflection and Transmission

There are electric and magnetic waves that serve to propagate radio energy. The electric waves
can be represented as a sum of two orthogonal polarization components, for example, vertical and
horizontal, or left-hand and right-hand circular. What happens when these two components of the
electric field hit the boundary between two different dielectric media?
We talk about the plane of incidence, that is, the plane containing the direction of travel of the
waves (incident, reflected, and transmitted), and perpendicular to the surface (plane where the two
media meet). See Figure 4.4 on page 115, reproduced in Fig 10.

Figure 10: Figure 4.4 from Rappaport.

Notes about the notation from Rappaport:

• “Parallel” refers to the E-field having direction parallel to the plane of incidence (as in Figure
4.4(a)); “perpendicular” means perpendicular (normal) to the plane of incidence (as in Figure
4.4(b)).

• Use subscripts i, r, and t to refer to the incident, reflected, and tranmitted field.

• ǫ1 , ǫ2 , is the permittivity of medium 1 and 2. (units Farads/meter) (Note: F = sec / Ω)

• µ1 , µ2 is the permeability of medium 1 and 2. (units Henries/meter) (Note: H = Ω sec )

ECE 5325/6325 Spring 2010 25

• σ1 , σ2 is the conductance of medium 1 and 2 (units Siemens/meter). (Note: S = 1 / Ω)

• The complex dialectric

√ constant is ǫ = ǫ0 ǫr − jǫ′ where ǫ0 = 8.85 × 10−12 F/m is free space
σ
permittivity, j = −1, and ǫ′ = 2πf , and ǫr is the relative permittivity. Don’t get confused
by the subscripts in this equation.

• A material is a good conductor when σk > f ǫk .

p
• The intrinsic impedance of medium k is µk /ǫk .

Then the reflection coefficients are given by

Er η2 sin θt − η1 sin θi
Γk , =
Ei η2 sin θt + η1 sin θi
Er η2 sin θi − η1 sin θt
Γ⊥ , = (13)
Ei η2 sin θt + η1 sin θi
where θt is determined by Snell’s Law :
√ √
µ1 ǫ1 sin(90o − θi ) = µ2 ǫ2 sin(90o − θt ) (14)

Also, the angle of incidence is equal to the angle of reflection:

θi = θr

Finally, the reflected and transmitted field strengths are:

Er = ΓEi
Et = (1 + Γ)Ei

where you chose Γ based on the polarization of the incident E-field, i.e., use either Γk or Γ⊥ .
There is a special case of (13) when the first medium is free space (or approximately, air) and
µ1 = µ2 . These two conditions are the case for most of the things we care about. In this case you
can show (good HW problem!) that
√
−ǫr sin θi + ǫr − cos2 θi
Γk = √
ǫr sin θi + ǫr − cos2 θi
√
sin θi − ǫr − cos2 θi
Γ⊥ = √ (15)
sin θi + ǫr − cos2 θi
See Figure 4.6 on page 118 of Rappaport. At some angle θi , there is no reflection of the parallel
E-field from (15). This angle is called the “Brewster angle”, which is given by
r
ǫ1
sin θB =
ǫ1 + ǫ2

When medium 1 is free space, and ǫ2 = ǫ0 ǫr ,

1
sin θB = √
1 + ǫr
This is the same as Equation 4.28 in Rappaport.
ECE 5325/6325 Spring 2010 26

Note that as θi → 0, Γk → 1 and Γ⊥ → −1.

Also, for perfect conductors (as described in Section 4.5.3), we also have Γk = 1 and Γ⊥ = −1.

Example: Reflection from ground

Find the reflection coefficients for typical ground at an incident angle of 15 degrees at 100 MHz.
Solution: Assume free space is medium 1 and that ‘typical ground’ has ǫr = 15. Note sin 15o =
0.259, and cos 15o = 0.933, so from (15),
√
−15(0.259) + 15 − 0.933
Γk = √ = −0.0176
15(0.259) + 15 − 0.933
√
0.259 − 15 − 0.933
Γ⊥ = √ = 0.871
0.259 + 15 − 0.933

8 Two-Ray (Ground Reflection) Model

Section 4.6 in Rappaport develops a theoretical model for propagation slightly better than the free
space assumption. This model includes not just one path, but also another path that reflects off
of the ground. “The World Is Flat”, if you will. The model isn’t hard to develop, and provides an
important theoretical underpinning to the multiple breakpoint model we covered in lecture 5.
Remember, powers of multipath DON’T add together. Only voltages or field strength of multi-
path actually add together. The voltage on the antenna is proportional to the electric field at the
antenna position. So let’s talk about adding electric fields.
See Figure 4.7 on page 121 of Rappaport.

8.1 Direct Path

Recall that the electric field magnitude decays as 1/d in free space. So, similar to how we wrote the
received power with a reference distance d0 , we write the E-field strength as the E-field strength at
a reference distance, multiplied by d0 /d, for a path (distance of travel for waves) length d. Also,
assume the signal is a simple sinusoid at the carrier frequency, fc . So
  
d0 d
E(d, t) = E0 cos 2πfc t − (16)
d c

For the LOS path, given a distance

p along the ground of L, antenna heights ht and hr at the TX
and RX, respectively, the d = L + (ht − hr )2 . So
2

p !!
d0 L2 + (ht − hr )2
ELOS = E0 p cos 2πfc t − (17)
L2 + (ht − hr )2 c

8.2 Reflected Path

Let’s assume that L is very long compared to the antenna heights. So, the angle of incidence is
approximately 0. In this case the reflection coefficient (assume perpendicular polarization) is -1.
ECE 5325/6325 Spring 2010 27

p
The reflected path travels longer than the direct path, for total length L2 + (ht + hr )2 (one can
use the “method of images” to show this). Then
p !!
d0 L2 + (ht + hr )2
Eg = −E0 p cos 2πfc t− (18)
L2 + (ht + hr )2 c

8.3 Total Two-Ray E-Field

We are interested in the magnitude of the total E-field, ET OT = ELOS + Eg , that is, the quantity
that multiplies the cos(2πfc t) term. Using trig identities, and this approximation:
p p 2ht hr
∆= L2 + (ht + hr )2 − L2 + (ht − hr )2 ≈
L
we can show that  
d0 2πfc 2ht hr
|ET OT | ≈ 2E0 sin
d c L
But at large L, the argument of the sin is approximately 0, and the sin x ≈ x. This is when x < 0.3
radians, which in our equation, means that for
20ht hr fc
L>
c
we can use this approximation (also noting that L ≈ d):
 
d0 2πfc 2ht hr const
|ET OT | ≈ 2E0 =
d c d d2

This means that the power decays as 1/d4 ! See Figure 11.
In summary, when there are two paths, one direct and one ground reflection, the theoretical
models show behavior that has two different path loss exponents, 1/d2 for d less than a threshold,
and 1/d4 for d above the threshold. This matches what we’ve observed from measurements and
presented as the empirical “multiple breakpoint” model.
However, a note: this is just a theoretical model. Typical cellular or indoor channels do not
have just two paths. One of the 6325 assignments is to consider T -ray model, for T > 2. For
example, if you had a ceiling reflection as a 3rd path. Or a ceiling-floor two-bounce path as a 4th
ray. As T goes up, you don’t see the 1/d4 behavior.

9 Cellular Large Scale Path Loss Models

Now, let’s consider some empirical models that are used in cellular systems. These types of models
are better than free space or the log-distance model, and have been designed to better fit measured
data. Note that there are many “modifications” that people continually add to these models to
better fit their environments of interest.
ECE 5325/6325 Spring 2010 28

Figure 11: Received power as a function of log distance in two-ray model, Figure 2.5 from Andrea
Goldsmith, “Wireless Communications”, Cambridge University Press, 2005.

9.1 Okumura-Hata Models

The median (50th percentile) propagation loss at frequency f , distance d, and TX and RX antenna
heights hte and hre , is given in the Okumura model as,
L50 (dB) = LF (f, d) + Amu (f, d) − G(hte ) − G(hre ) − GAREA
where all terms are in dB even though Rappaport does not explicitly denote thenm as such, and
where LF (f, d) is the free space path loss in dB at distance d and frequency f , Amu is the median loss
compared to free space (found from Table 4.23), GAREA is a factor due to the type of environment
(open, quasi open, or suburban, as given in Figure 4.24), and
hte
G(hte ) = 20 log 10 , for 30m < hte < 1000m
200m
10 log 10 3hm
 re
, 0m < hre < 3m
G(hre ) = hre
20 log 10 3m , 3m < hre < 10m

Example: Rappaport Example 4.10, page 153

The Hata model is largely function-fitting, expressing a formula that captures most of the results
of Okumura.
Again, these are based on large sets of measurements. There are many adjustments to this
model, particularly for PCS (the COST-231 model for 1900 MHz), for microcells, etc. In general,
these models have a standard deviation of about 10 or more dB when compared to the actual
measurements. So, it is typically useful to obtain more accurate means of predicting received powers,
in particular, software that predicts based on the geometry of a city or area, and measurements
which give feedback to the model to allow it to (to some extent) correct its errors).

Discussion
What are some of the assumptions made in these models that we have not talked about, or in
particular, that may be incorrect?
ECE 5325/6325 Spring 2010 29

Lecture 7

Today: (1) Exam 1 Prep, (2) Link Budgeting

10 Exam 1
Exam 1 (Tue, Feb 9, 3:40-4:40pm) covers: lectures 1-7; homeworks 1-3; this lecture’s material, even
though it is not included in a homework; Rappaport Chapters 3 and 4 (excluding Sections 4.7,
4.8, and 4.9); the reading from Haykin/Moher. Please do example problems in link budgeting to
prepare for exam 1, you can put these in your portfolio. See http://span.ece.utah.edu/node/93,
for more practice. Note that I cannot test you on everything in one hour, but I’m not going to tell
you what is on the exam. So be knowledgeable about all possible problems that can be solved. DO
NOT look at the four problems I put on the practice exam and think that those are the only types
of problems that will be covered!

11 Link Budgeting
Link budgets are, as the name implies, an accounting of the gains and losses that occur in a radio
channel between a transmitter and receiver. We’ve talked about S/I – you need an acceptable
signal to interference ratio. In addition, you need an acceptable signal to noise, or S/N, ratio.
(a.k.a. SNR, C/N , or Pr /PN ratio, where C stands for carrier power, the same thing we’ve been
calling Pr . Since we’ve already used N in our notation for the cellular reuse factor, we denote noise
power as PN instead.) Noise power is due to thermal noise.
In the second part of this course, we will provide more details on where the requirements for S/N
ratio come from. For now, we assume a requirement is given. For a given required S/N ratio, some
valid questions are: What is the required base station (or mobile) transmit power? What is the
maximum cell radius (i.e., path length)? What is the effect of changing the frequency of operation?
Also, there is a concept of path balance, that is, having connectivity in only one direction doesn’t
help in a cellular system. So using too much power in either BS or mobile to make the maximum
path length longer in one direction is wasteful.
As we’ve said, this is accounting. We need to keep track of each loss and each gain that is
experienced. Also, to find the noise power PN , we need to know the characteristics of the receiver.

11.1 Link Budget Procedure

A universal link budget for received power is:
X X
Pr (dBW) = Pt (dBW) + dB Gains − dB Losses (19)

A universal link budget for S/N is:

X X
S/N = Pr (dBW) − PN (dBW) = Pt (dBW) + dB Gains − dB Losses − PN (dBW)

1. There’s no particular reason I chose dBW instead of dBm for Pr and PN . But they must be
the same, otherwise you’ll have a 30 dB error!
ECE 5325/6325 Spring 2010 30

Figure 12: Relationship among link budget variables.

2. If using EIRP transmit power, it includes Pt (dBW) + Gt (dB), so don’t double count Gt by
also including it in the dB Gains sum.

3. The dB noise figure F (dB) is either included in PN (dBW) or in the dB losses, not both!

4. Gains are typically only the antenna gains, compared to isotropic antennas.

5. There are also coding, a.k.a. processing, gains, achieved by using channel coding to reduce
the errors caused by the channels. DS-SS (e.g., CDMA) is a type of modulation which has
a processing gain. These might be subtracted from the required S/N ratio, or added to the
gains. Do one, but not both.

6. Losses include large scale path loss, or reflection losses (and diffraction, scattering, or shadow-
ing losses, if you know these specifically), losses due to imperfect matching in the transmitter
or receiver antenna, any known small scale fading loss or “margin” (what an engineer decides
needs to be included in case the fading is especially bad), etc.

7. Sometimes the receiver sensitivity is given (for example on a RFIC spec sheet). This is the
PN (dB) plus the required S/N (dB).

11.2 Thermal noise

The thermal noise power in the receiver is PN , and is given as PN = F kT0 B, where

• k is Boltzmann’s constant, k = 1.38 × 10−23 J/K. The units are J/K (Joules/Kelvin) or
W·s/K (1 Joule = 1 Watt × second).

• T0 is the ambient temperature, typically taken to be 290-300 K. If not given, use 294 K, which
is 70 degrees Fahrenheit.
ECE 5325/6325 Spring 2010 31

• B is the bandwidth, in Hz (equivalently, 1/s).

• F is the (unitless) noise figure, which quantifies the gain to the noise produced in the receiver.
The noise figure F ≥ 1.

In dB terms,
PN (dBW) = F (dB) + k(dBWs/K) + T0 (dBK) + B(dBHz)
where k(dBWs/K) = 10 log10 1.38 × 10−23 J/K = −228.6 dBWs/K. We can also find F from what
is called the equivalent temperature Te . This is sometimes given instead of the noise figure directly.
Te
F =1+
T0

11.3 Examples

Example: GSM uplink

Consider the uplink of a GSM system, given GSM requires an S/N of 11 dB [10]. Assume a
maximum mobile transmit power of 1.0 W (30 dBm), 0 dBd antenna gain at the mobile, and 12
dBd gain at the BS. Assume path loss given by the urban area Hata model, with fc = 850 MHz,
BS antenna height of 30 meters, mobile height of 1 meter. Assume F = 3 dB and that the system
is noise-limited. What is the maximum range of the link?
Solution:

• S/N required is 11 dB.

• PN = F kT0 B = 2(1.38 × 10−23 J/K)(294K)(200 × 103 Hz) = 1.62 × 10−15 = −147.9(dBW).

• Pt = 0 dBW.

• Gains: include 0 dBd and 12 dBd (or 2.15 dBi and 14.15 dBi) for a total of 16.3 dB gains.

• Losses: Path loss is via urban area Hata, for d in km,

L(dB) = 69.55 + 26.16 log 10 (850) − 13.82 log 10 (30) + [44.9 − 6.55 log 10 (30)] log 10 d
= 125.8 + 35.22 log 10 d

So

11(dB) = 0(dBW) + 16.3(dB) − (125.8 + 35.22 log 10 d) + 147.9(dBW)

d = 1027.4/35.22 = 6.0(km) (20)

Note 1 km = 0.62 miles, so this is 3.7 miles.

Example: Sensor Network

Assume two wireless sensors 1 foot above ground need to communicate over a range of 30
meters. They operate the 802.15.4 standard (DS-SS at 2.4 GHz). Assume the log-distance model
with reference distance 1m, with path loss at 1 m is Π0 = 40 dB, and path loss exponent 3 beyond
1m. Assume the antenna gains are both 3.0 dBi. The transmitter is the TI CC2520, which has
ECE 5325/6325 Spring 2010 32

max Pt = 1 mW, and its spec sheet gives a receiver sensitivity of -98 dBm. What is the fading
margin at a 30 meter range? (Note: By exam 2 you will be able to specify fading margin given a
desired probability that communication will be lost due to a severe fade. Here we’re just finding
what the margin is for these system parameters.)
Solution: The question asks us to find the difference between Pr at 30 meters and the receiver
sensitivity (PN (dB) plus the required S/N (dB). Rearranging (19),
X X
−98dBm = S/N + PN (dBm) = Pt (dBm) + dB Gains − dB Losses (21)

1. Pt (dBm) = 0 dBm.

2. Gains: Two antennas at 3 dBi (the units are effectively dB), so the total gains are 6 dB.

3. Losses: There is the 40 dB loss to 1 m, then an additional 10(3.0) log 10 (30/1) = 44.3 dB.
Fading Margin is a Loss, so we have 84.3 dB + Fade Margin for the total losses.
So
−98dBm = 0(dBm) + 6(dB) − Fade Margin − 84.3(dB)
Which, solving for Fade Margin, is 19.7 dB.

Example: IS-136
Compare IS-136 and GSM in terms of range. Compared the the GSM uplink example above, an
IS-136 mobile has 0.6 W transmit power, and the required S/N is 15 dB [10], and IS-136 has a
lower bandwidth of 30 kHz.
Solution:
• S/N required is 15 dB.

• PN = F kT0 B = 2(1.38 × 10−23 J/K)(294K)(30 × 103 Hz) = 2.43 × 10−16 W = −156.1(dBW).

• Pt = −2.2 dBW.
So

15(dB) = −2.2(dBW) + 16.3(dB) − (125.8 + 35.22 log 10 d) + 156.1(dBW)

d = 1029.4/35.22 = 6.8(km) (22)

Lecture 8

Today: (1) Diffraction (2) Scattering

12 Diffraction
In EM wave propagation Huygens’ principle says that at each point, the wave field is effectively
re-radiating in all directions. In free space, these secondary reradiators sum and “produce” the
effect of a wave front advancing in the direction away from the source. When objects exist in free
ECE 5325/6325 Spring 2010 33

space that block or attenuate some of the wave field, the reradiation enable EM waves to “bend”
around objects. In order to calculate the field at a point in (or near) the “shadow” of an object,
we can use Huygens’ principle to find accurate numerical results. This is a short version of some
advanced electromagnetics. See [17] Chapter 9 for a more detailed treatment.
See Figure 4.13 in Rappaport. The Fresnel-Kirchoff parameter ν is given by,
s
2(d1 + d2 )
ν=h (23)
λd1 d2

depends on the geometry and the frequency, and is unitless:

• d1 , d2 , distance along line-of-sight path from TX or RX to obstruction

• h, screening height

In short, we have a normalized vertical axis at the knife edge. The top of the knife edge is at
position ν – below ν, there is a perfect conductor, and above ν, there is free space. We assume the
knife edge is infinitely narrow. For our point of interest beyond the knife edge, Huygens’ principle
has us consider (sum) the effect of the secondary reradiators along the vertical axis, above the knife
edge. The summation is actually an integral, and is taken from ν to ∞, and is called the complex
Fresnel integral,
1+j ∞ jπt2
Z  
F (ν) = exp − dt (24)
2 ν 2
Another 6325 assignment has been posted regarding (24). The dB magnitude of the power loss
from the Fresnel integral, 20 log 10 F (ν), which we call the knife-edge diffraction GAIN, is given by
Figure 4.14 in Rappaport, and is copied in Figure 13.

Figure 13: Knife-edge diffraction gain in dB.

Expressions exist for the multiple knife edge diffraction problem – when multiple obstructions
block a propagating wave. However, these are largely computed via numerical analysis, so we won’t
elaborate on them.
ECE 5325/6325 Spring 2010 34

13 Rough Surface Scattering

When we discussed reflection, we said a wave was impinging on a flat surface. But most surfaces
(e.g., walls, ground) are rough, not flat. When is a surface considered rough? When the maximum
height “protuberance” from the surface, h, is greater than hc ,
λ
hc = (25)
8 sin θi
where θi is, again, the angle of incidence.
Scattering has two effects important to us:

1. Rough surface scattering reduces the power in the reflected wave.

2. Scattering causes additional multipath to be received in directions other than the specular
direction (recall θr = θi ).

For 1., if the surface is rough, then the reflected wave has reflection coefficient multiplied by ρs ,
so that Γrough = ρS Γ. Multiple expressions exist to compute ρS . Two given in the book are:
"   #
πσh sin θi 2
ρS = exp −8
λ
"   # "   #
πσh sin θi 2 πσh sin θi 2
ρS = exp −8 I0 8
λ λ

where σh is the standard deviation of the height of the rough surface. The second expression is
considered to be more accurate.
For 2., scattering is a very useful and impactful phenomena. Scattering is the basis of radar,
weather systems, and passive RFID tags. Engineers have put much effort into understanding the
phenomena of EM wave scattering.
Similar to Huygens’ principal, the wave field at the scatterer is assumed to become a secondary
re-radiator. However, the object (the scatterer) is assumed to absorb the incident power, and then
re-radiate it (The re-radiation is not assumed to be occurring from the free space near the ob-
ject). The object is considered a new point source, where power is “received” and “retransmitted”.
Essentially we have two links, one from the transmitter to the scatterer, and one away from the
scatterer to the end receiver.
Typical airplane, weather radar is monostatic, i.e., the transmitter and receiver are co-located.
But in wireless comm systems (and more robust airplane radar systems) the TX and RX are not
in the same place. The bistatic radar equation describes the received power Pr in a scattered wave
at a receiver that is not at the transmitter location. In linear and dB terms,

Pt Gt Gr σRCS λ2
Pr =
(4π)3 d2t d2r
Pr (dBW) = Pt (dBW) + Gt (dB) + Gr (dB) + σRCS (dBm2 )
+20 log 10 λ − 30 log10 4π − 20 log10 dt − 20 log 10 dr

Note that there are two 1/d2 terms, one corresponding to each “link” described above. The σRCS
term has units of m2 and is the radar cross section of the scatterer. It is an area, like an antenna
ECE 5325/6325 Spring 2010 35

effective aperture, that describes how much power is absorbed by the scatterer to be reradiated.
Also, note that in the dB expression, the dBm2 units cancel because there are two dBm2 terms on
top (σRCS and 20 log 10 λ2 ) and two dBm2 terms on bottom (20 log 10 dt and 20 log10 dr ).

Lecture 10

Today: (1) Multipath Fading

14 Multipath Fading
We’ve talked about physics, that is, how wave propagation and its interaction with the environment
causes reflection, transmission, diffraction, and scattering. Many individual propagating waves
arrive at the receiver, these waves are called multipath components, or collectively, multipath. These
multipath cause fading effects (changes in the received power) grouped together and called multipath
fading. There are many kinds of multipath fading.
The challenges caused by multipath fading in wireless communication systems are one the most
significant challenges addressed by wireless engineers today. Engineers have developed a variety of
modulation and diversity schemes in order to counteract the negative influence of multipath fading.
And, we are developing methods which take advantage of multipath in particular ways as a benefit
for communication systems. All this to say, understanding of the fundamentals of fading is a critical
skill for wireless engineers.
We’re first going to talk about received power when mutiple multipath signals are reiceved at
the receiver. Then, we’ll present the spatial and temporal characteristics of multipath.

14.1 Multipath
We’ve been talking about the EM field. Specifically, we’ve presented expressions for the E-field Eb
when we send a sine at frequency fc through the channel, as

Eb = E0 cos(2πfc t + θ)

where E0 is positive real number. In other words,

h i h i
Eb = R E0 ej2πfc t+jθ = R ej2πfc t E0 ejθ

The above expressions are called bandpass representation. When we want to write a simpler
expression, we write the complex baseband-equivalent representation:

E = E0 ejθ

and we can “translate” any complex baseband-equivalent signal into its bandpass signal by applying
h i
Eb = R ej2πfc t E

The simpler expression E has more easily viewable magnitude (amplitude) E0 and phase (angle) θ.
The voltage received by the antenna is proportional to E and has the same complex baseband
or real-valued baseband representation. For example, V == αEV0 ejθ , for some constant α.
ECE 5325/6325 Spring 2010 36

As we discussed, many such multipath wave components arrive at the receiver. They add
together as voltages. DO NOT add the powers of the multipath together – there is no such physical
antenna that add together the powers of multipath. (If you find a way, patent it quick!)
Let’s say there are M multipath components, numbered 0 through M − 1. (Rappaport uses N
as the number of components, don’t confuse with the frequency reuse factor.) Component i has
amplitude Vi and phase θi . Then the total voltage at the receiver antenna will be:
M
X −1
VT OT = Vi ejθi
i=0

The received power is then proportional to |VT OT |2 (by a resistance).

We did this for the two-ray model, remember? The magnitude |ET OT | was a sum of two cosine
terms, each with a different phase, which we needed to add together, before taking the magnitude
squared. It would be a good exercise to re-derive Section 4.6 using phase angles rather than cosines.

14.2 Temporal
Let’s expand on where these phase angles come from. Recall that Vi ejθi is the representation of
Vi cos(2πfc t + θi ). If Vi cos(2πfc t) is transmitted from the transmitter antenna, how do the phases
of the multipath components behave with respect to each other? Well, each component has its
own path length. It really did travel that length. And EM waves all travel at the same speed –
c = 3 × 108 m/s. So some waves arrive later than others. Let τi denote the time delay of arrival
for multipath i relative to the transmit time. It is di /c, where di is the length of component i.
What happens when a function is delayed by τi ? We replace t with t − τi in the function. So
Vi cos(2πfc (t − τi )) is received. Well, not the full story – reflections and diffractions also cause
phase changes (we discussed specifics for reflection in Section 4.5). Really, Vi cos(2πfc (t − τi ) + φi )
is received, where φi is the sum of all phase changes caused by the physical propagation phenomena.
We’ve been using baseband notation, what is the complex baseband notation? It is Vi ej(−2πfc τi +φi ) .
So what is the total received voltage from all multipath?
M
X −1
VT OT = Vi ej(−2πfc τi +φi ) (26)
i=0

In other words, θi = −2πfc τi + φi . We’ve now written it in terms of its temporal delay, τi . Note
that VT OT has incidentally become a function of frequency fc .

14.3 Channel Impulse Response

What we have in (26) is a frequency response as a function of frequency fc . The equation can
show the frequency response at any frequency. Let f = fc , maybe that makes it clearer. So (26)
is a frequency-domain representation of the total voltage. Let’s convert to the time domain. How?
Using the inverse Fourier transform:
(M −1 ) M −1
X X n o MX−1
j(−2πf τi +φi ) j(−2πf τi +φi )
F Vi e = Vi F e = Vi ejφi δ(τ − τi )
i=0 i=0 i=0

What this says is that in the time delay domain, the arrivals of multipath i occurs at delay τi .
ECE 5325/6325 Spring 2010 37

This leads to how we frame the channel: as an echo-causing filter, with an impulse response
that is a sum of time-delayed impulses. Let s(t) be the transmitted signal and r(t) be the received
signal. Then
1
r(t) = s(t) ⋆ h(τ )
2
where h(τ ) is called the channel impulse response, and is given by
M
X −1
h(τ ) = ai ejφi δ(τ − τi ) (27)
i=0

The ai are proportional to Vi but are unitless – the units are contained in s(t), which has units of
Volts. The amplitude |ai | is the amplitude gain in that path; the squared magnitude |ai |2 is the
power gain in that path. We often plot the squared magnitude of h(τ ) in the dB domain and call
it the power delay profile. This is what Rappaport calls P (τ ). He shows some examples in Figures
5.4 and 5.5.
We’ve also measured many of these in my lab. For example, Figure 14 shows three examples.

0 0
Power (dB from Max)

−5 Power (dB from Max) −5

−10 −10
−15 −15
−20 −20
−25 −25
−30 −30
0 100 200 300 0 100 200 300
(a) Delay (ns) (b) Delay (ns)

0
Power (dB from Max)

−5
−10
−15
−20
−25
−30
0 100 200 300
(c) Delay (ns)

Figure 14: Measured power delay profiles (power gain normalized to the power gain of the maximum
power path) in Salt Lake City (a) in residential area W of U. campus; (b) 4th S commercial district;
(c) Main St., downtown (credit to D. Mass, H. Firooz).

14.4 Received Power

What happens to the transmitted signal s(t) when it is convolved with h(τ )? Well, many copies
show up, at different time delays:
M
X −1
s(t) = ai ejφi s(τ − τi )
i=0
ECE 5325/6325 Spring 2010 38

For example, what if a rectangular

h pulse
i was sent (in Digital Communications, many symbols look
t 1
like a pulse)? Let s(t) = rect Tb b − 2 . In this case,
M −1  
X
jφi t − τi 1
s(t) = ai e rect −
Ts 2
i=0
where Ts is the symbol duration. Essentially, we have versions of the rect pulse piling on top of
each other.
CASE 1: τi << Ts : If the τi s are small compared to Ts ,
M −1
!  
X
jφi t 1
s(t) ≈ ai e rect −
Ts 2
i=0
then we have all of the pulses adding together, in a phasor sum. The sum might be constructive
or destructive. But it acts on the whole pulse.
CASE 2: τi aren’t small compared to Ts : We will have intersymbol interference, and will need
an equalizer in our receiver.
Note Ts is designed by us (the engineer). Why not make Ts big? Because the symbol rate is
1/Ts ! You slow down data rate when increasing Ts .
6325 Only: Study the proof in Section 5.2.1 of Eθ [PW B ].

14.5 Time Dispersion Parameters

There are lots of τi so lets provide a number with more specificity, a single number that summarizes
the size of the time dispersion: the RMS delay spread, στ ,
q
στ = τ¯2 − tau ¯ 2
2
P P
i |ai | τi P (τi )τi
τ̄ = P 2
= Pi
|a i | P (τi )
Pi 2 2 Pi
|a | τ P (τi )τi2
τ¯2 = Pi i 2i = Pi
i |ai | i P (τi )
One good rule of thumb is that the receiver doesn’t need an equalizer if στ ≤ 0.1Ts .

Lecture 10

Today: (1) Fade Distribution, (2) Doppler

15 Fade Distribution
Specular multipath: What we’ve been talking about: individual multipath components, of which
we said there were M total, each with its own amplitude and phase.
Diffuse multipath: multipath which are each infinitely low in power, but there are infinitely many
of them. Typically used to simulate the many many multipath that do arrive due to scattering and
diffraction. It is easier to talk about fading when we can lump lots and lots of multipath into this
“diffuse” camp, rather than specifying each individual multipath and how much power each one
has.
ECE 5325/6325 Spring 2010 39

15.1 Rayleigh Fading

Main idea: all multipath are diffuse. These diffuse multipath have small amplitude and random
phase. Using the central limit theorem, and some probability (ECE 5510), we can come up with a
pdf of the magnitude of the received voltage, r = |VT OT |.
Here, the name of the random variable is R and the value r represents some number of interest.
(  
r r2
σ 2 exp − 2σ 2 , r≥0
fR (r) = (28)
0, o.w.
where σ is a parameter that can be set as needed. For
q example, if you know the mean value of the
signal magnitude is rmean , then you would set σ = π2 rmean .
The most important function for the communication system designer is the cumulative dis-
tribution function (CDF). It tells us the probability that the envelope will fall below some value
r. (  
r2
1 − exp − 2σ 2 , r≥0
FR (r) = P [R ≤ r] = (29)
0, o.w.

Def ’n: Fade Margin

the additional power loss which is included in the link budget in order to keep the received signal
power higher than the minimum, with high probability.

Example: Fade margin design: Rayleigh case

Assume that it is acceptable to have the received power fall below the receiver sensitivity 1% of
the time, and that the channel is Rayleigh. What fade margin (in dB) is required?
Solution: We are given P [R ≤ r] = 0.01. We want to find the threshold in terms of the mean
received signal level, so setting rmean = 1, we have that σ = 0.7979. Then, solving (29) for r,
r2
 
exp − 2 = 0.99
2σ
p
r = −2σ 2 ln 0.99 = 0.1131
(30)
If the envelope falls to 0.1131, then the power falls to (0.1131)2 = 0.0128, which is equal to -18.9
dB. Since the mean value is 1, or 0 dB, the required fade margin is 0 − (−18.9) = 18.9 dB.

15.2 Ricean fading

Main idea: One specular path with amplitude V0 = A plus diffuse power.
The Rice distribution is named after its creator, Stephen O. Rice. I have his 1945 paper in my
office if you want to use it for a project. Note that people just can’t agree on the spelling of the
adjective made out of his name, whether it is “Rician” or “Ricean”. Google gives me 104k results
for “Rician” and 80k results for “Ricean”. I think that the mathematicians use “Ricean” more
while the radio engineers use “Rician” more. Why? Who knows.
The pdf of Ricean fading is
(  2 2
r r +A Ar


σ 2 exp − 2σ 2 I0 σ , r ≥0
2
fR (r) = (31)
0, o.w.
ECE 5325/6325 Spring 2010 40

where A is the peak amplitude of the specular (dominant) signal, and I0 (·) is the modified Bessel
function of the first kind and zeroth order. Note Matlab has a besseli(0, x) function. The
Ricean K-factor is defined as the ratio between the power in the specular components to the power
in the diffuse components,

A2 A2
K= , or K(dB) = 10 log10 . (32)
2σ 2 2σ 2
Note that for K = 0 or K(dB) = −∞, the Ricean pdf is the same as the Rayleigh pdf.
There isn’t an analytical expression for the CDF, analagous to (29), unfortunately. So we need
to use a table or a figure. The Rappaport book only has a CDF plot for K = 6 dB. I am including
in Figure 15 a more complete Figure. Keep this figure with your notes, it will be included in Exam
2. It also include Rayleigh as the K = −∞ line.

r
Figure 15: Ricean CDF for various K-factors. Note “Power Level” is 20 log10 rmedian , and the y-axis
is the probability ×100 (percentage) that the power level (fading gain compared to the median) is
less than the value on the abscissa.

Example: Fade margin design: Ricean case

Assume that it is acceptable to have the received power fall below the receiver sensitivity 1% of
the time, and that the channel is Ricean, with K-factor 3, 9, or 15. What fade margin (in dB) is
required?
ECE 5325/6325 Spring 2010 41

16 Doppler Fading
So far we’ve talked about fading without movement. A static link has a fading loss. But for a link
in motion, (1) the fading loss changes over time, and (2) the frequency shifts. Why is this?
In lecture 9, we came up with the top expression for the complex baseband received voltage:
M
X −1
VT OT = Vi ej(−2πfc τi +φi )
i=0
M
X −1
VT OT = Vi ej(−2πdi /λc +φi )
i=0

The second expression is rewritten with di = cτi , where di is the distance multipath component
i travels. (Is this the straight-line from the TX to RX? Or the length of the line following the
path? Answer: the latter.) This showed us the frequency dependence of the fading channel gain,
which is 20 log10 |VT OT |. Now, let’s talk about what happens when the receiver is moving. Motion
causes the time delays to change, because the distance that the wave must travel is becoming either
shorter or longer. How much shorter or longer?
Let the angle of arrival of component i be θi . (Consider several multipath components and their
geometry.) Let’s assume I move ∆move meters in the direction θmove . We assume that these waves
are effectively plane waves in the local (small) area near the antenna. Thus the only thing that
changes when I move the antenna to a new position is that the wave is shortened or lengthed by a
factor of the distance I moved, multiplied by the cosine of the angle in between θi and my direction
of motion θmove . After my movement of ∆move meters in the direction θmove , my VT OT becomes:
M −1   
X 2π
VT OT = Vi exp j − [di + ∆move cos(θmove − θi )] + φi
λc
i=0

16.1 One Component

First, keep things simple by considering only one arriving multipath.
  
2π
VT OT = V0 exp j − [d0 + ∆move cos(θmove − θ0 )] + φ0
λc
Let’s consider when the antenna is moving at a constant velocity v meters per second. Then, at
time t, we have ∆move = vt. So,
  
2π
VT OT = V0 exp j − [d0 + vt cos(θmove − θ0 )] + φ0
λc
There’s a lot going on here, but we actually have not just a phase, but a complex sinusoid of,
v cos(θmove − θ0 )
fd = − (33)
λc
This is called the Doppler shift. When one sends a frequency fc , the received frequency of the signal
has shifted up (or down, if fd is negative) by this factor.

1. What happens when I move such that θi = θmove ?

ECE 5325/6325 Spring 2010 42

2. What happens when I move such that θi = θmove + 90 degrees?

3. What happens when I move such that θi = θmove + 180 degrees?

The worst case is called the maximum Doppler frequency, fm = |fd | = v/λc .

Example: What are the maximum Doppler frequencies for a mobile in a vehicle on
I-15, at 850 and 1950 MHz?

Solution: Given 1 mile per hour = 0.447 meters per second, let’s say a maximum speed of 80
miles/hr, which gives v = 35.8 m/s. Then fm = 101 Hz, or 232 Hz, for 850 or 1950 MHz,
respectively.

16.2 Many Components

Now, each component contributes a complex sinusoid of frequency fd = − v cos(θmove λc
−θi )
to the sum
VT OT . The effect is frequency spreading. That is, for each frequency that is sent, many frequencies
are received. In the frequency domain, the frequency content of the signal (say, S(ω)) is convolved
with the Doppler spectrum VT OT (ω)).
If there are diffuse multipath coming equally from all directions, then the power spectrum of
VT OT can be determined to be:
1.5
S(f ) = r  2
πfm 1 − ff−f
m
c

which is shown in Figure 5.20 on page 219 in Rappaport.

16.3 System Design Implications

The Doppler spread is said to have an effect when the symbol duration is long compared to 1/fm .
Specifically, Rappaport recommends TC = 0.423 f1m , and if Ts > TC , that is, the symbol rate is
higher than then there you have “fast fading”, and demodulation will be difficult. Instead, one
should design systems with Ts << TC . Typically, this is not a problem.
See Figure 5.15 on page 211 in Rappaport for an example of Rayleigh fading. How often does
the power level drop cross a horizonal “threshold” line? This is the level crossing rate (LCR). For
the Clarke AOA spectrum model, The average number of level crossings per second is
√ 2
NR = 2πfm ρe−ρ

Another useful statistic is how long the signal stays below this threshold when it goes below it.
This is the average fade duration, denoted t̄f ade (or τ̄ in Rappaport). Again, for the Clarke AOA
spectrum model,
2
eρ − 1
t̄f ade = √
ρfm 2π

Example: Fading statistics

Let 1/Ts = 1M Hz, and use the solution for 1950 MHz highway driving above. Find whether we
have fast or slow fading. What is the fade duration, LCR, and t̄f ade ?
ECE 5325/6325 Spring 2010 43

Lecture 11

Today: (1) Intro to Digital Communications

17 Course Project
This course requires an individual project writeup and 10-15 minute presentation, in which you will
further develop a topic of interest in the area of wireless communications systems (details provided
separately). A wireless symposium will be held on Thu. April 22, between 3:40-7:40. You must
present (10-15 minutes) and attend six other presentations during this symposium.

17.1 Topic
Write a book chapter on a topic within wireless communications. Examples: a particular wireless
standard, a type of fading, analysis of a particular type of wireless communication system and its
performance, or another wireless communications or networking topic of interest (please ask if you
have any questions). Emphasis should be given on the lessons learned from the study, not just the
numbers describing the system / protocol. Why are particular choices made? How do you apply
what you’ve learned in class?
You will need to submit your topic to me via WebCT for approval before starting.

17.2 Methods
The topic must include a simulation (or experiment, or theoretical analysis). You will explain
the methods you use in your simulation (or experiment or theoretical analysis) in your ”Methods”
section.

17.3 Writeup
• 7 pages, 1.5 line spacing, 11 point type, 1 inch margins.

• minimum of four references (at least two conference or journal papers). Six references is
A-level work.

• NO copying from any source. Any sentence which is paraphrased must be cited.

• Sections: Introduction, Methods, Results, Conclusion, References

17.4 Presentation
• 10-15 minute slide presentation (powerpoint, beamer, ...)

• Teach your topic to the students in attendence.

• Use graphics to explain

• Show and explain methods and results

ECE 5325/6325 Spring 2010 44

18 Digital Communications: Overview

My six word summary of digital communications: Use linear combinations of orthogonal
waveforms.

18.1 Orthogonal Waveforms

My “engineering” definition of a set of orthogonal waveforms: They are waveforms that can be
separated at the receiver.
Def ’n: Orthogonal
Two real-valued waveforms (finite-energy functions) x(t) and y(t) are orthogonal if
Z ∞
x(t)y(t)dt = 0
−∞

Two complex-valued waveforms x(t) and y(t) are orthogonal if

Z ∞
x(t)y ∗ (t)dt = 0
−∞

where y ∗ (t) is the complex conjugate of y(t).

Def ’n: Orthogonal Set

N waveforms φ1 (t), . . . , φN (t) are mutually orthogonal, or form an orthogonal set, if every pair of
waveforms φi (t), φj (t), for i 6= j, is orthogonal.

Example: Sine and Cosine

Let

cos(2πt), 0 < t ≤ 1
x(t) =
0, o.w.

sin(2πt), 0 < t ≤ 1
y(t) =
0, o.w.
Are x(t) and y(t) orthogonal?
Solution: Using cos2 x = 12 (1 + cos 2x),
s s   1
Z 1
2
1 1  p
kx(t)k = cos (2πt)dt = 1+ sin(2πt) dt = 1/2
0 2 2π 0
p
The solution for ky(t)k also turns out to be 1/2. Using sin 2x = 2 cos x sin x, the inner product
is,
Z 1
hx(t), y(t)i = cos(2πt) sin(2πt)dt
0
Z 1
1
= sin(4πt)dt
0 2
1
−1  −1
= cos(4πt) = (1 − 1) = 0
8π 0 8π
ECE 5325/6325 Spring 2010 45

So, yes, x(t) and y(t) are orthogonal.

Example: Frequency Shift Keying

Assume Ts >> 1/fc , and show that these two are orthogonal.

cos (2πfc t), 0 ≤ t ≤ Ts
φ0 (t) =
0, o.w.
(  h i 
cos 2π fc + T1s t , 0 ≤ t ≤ Ts
φ1 (t) =
0, o.w.

Solution: The integral of the product of the two must be zero. Checking, and using the identity
for the product of two cosines,
Z Ts    
1
cos (2πfc t) cos 2π fc + t dt
0 Ts
1 Ts
Z Z Ts
= cos (2πt/Ts ) dt + cos (4πfc t + 2πfc t/Ts ) dt
2 0 0
 Ts  Ts
1 Ts  1 
= sin (2πt/Ts ) +
 sin (2π(2fc + 1/Ts )t)
2 2π 0 2π(2fc + 1/Ts ) 0

The second term has a 2π(2fc1+1/Ts ) constant out front. Because fc is very high, this term will be
very very low. The sine term is limited to between -1 and +1 so it will not cause the second term
to be large. So we will approximate this second term as zero.
1 Ts
≈ [sin(2π) − sin(0)] = 0
2 2π
Thus the two different frequency waveforms are orthogonal.

18.2 Linear Combinations

What is a linear combination of orthogonal waveforms? Well, consider the orthogonal set φ1 (t), . . . , φN (t).
A linear combination si (t) is
N
X
si (t) = ai,1 φ1 (t) + ai,2 φ2 (t) + · · · + ai,N φN (t) = ai,k φk (t)
i=k

This is a linear combination. We use subscript i to indicate that it’s not the only possible linear
combination. In fact, we will use M different linear combinations, and we will use different si (t) to
represent different bit strings.
We represent si (t) as a vector ai = [ai,1 , ai,2 , . . . , ai,N ]T just for ease of notation. Vectors are
easy to deal with because they can be plotted in vector space, to show graphically what is going
on. We call the plot of all possible ai (t), that is, for i = 1, . . . M , the constellation diagram. Some
examples are shown in Figure 16.
When a symbol is chosen from one of M combinations, we are sending log2 M bits. Thus there
are log2 M bits per symbol. For example, if there are 8 possible combinations, we would label them
000, 001, 011, 010, 110, 111, 101, 100, and thus sending one symbol conveys 3 bits of information.
ECE 5325/6325 Spring 2010 46

QPSK QPSK

(a)
M=8 M=16

(b)
M=64 QAM

(c)

Figure 16: Signal constellations for (a) M = 4 PSK (a.k.a. BPSK), (b) M = 8 and M = 16 PSK,
and (c) 64-QAM.

18.3 Reception
At a receiver, because we’ve used an orthogonal set, we can determine how much of each waveform
was sent. This can be done using a bank of matched filters. In short, we can recover ai , up to
a scaling, and plus noise. We might write x = gai + ni where g is the total gain introduced by
the channel and antennas, and ni is the random additive noise added in at the receiver. Assume
we have receiver gain that multiplies the received filter by 1/g (to cancel out g). If the resulting
âi = x/g is close enough to the actual transmitted signal ai , then the correct bit string is received.
(Draw matched filter receiver)

18.4 How to Choose a Modulation

A digital modulation is the choice of: (1) the linear combinations a1 , . . . , aM and, (2) the orthogonal
waveforms φ1 (t), . . . , φN (t). We will choose a digital modulation as a tradeoff between the following
characteristics:

1. Bandwidth efficiency: How many bits per second (bps) can be sent per Hertz of signal band-
width. Thus the bandwidth efficiency has units of bps/Hz.
ECE 5325/6325 Spring 2010 47

2. Power efficiency: How much energy per bit is required at the receiver in order to achieve a
desired fidelity (low bit error rate). We typically use S/N or Es /N0 or Eb /N0 as our figure
of merit.

3. Cost of implementation: Things like symbol and carrier synchronization, and linear transmit-
ters, require additional device cost, which might be unacceptable in a particular system.

18.5 Bandwidth
To send a symbol, we will send a finite-duration waveform. This waveform requires some bandwidth.
We can find this bandwidth by taking the Fourier transform of the waveform.
For example, what is the bandwidth required to send a rectangular pulse? Let s0 (t) = rect(t/Ts ).
It has
sin(πf Ts )
S0 (f ) = Ts
πf Ts
This has frequency content at all frequencies! And, it doesn’t decay very quickly.
As another example, send s0 (t) = sin(πt/Ts)
πt/Ts . This is a rect function in the frequency domain!
But, the problem is is takes a long time to send in the time domain.
Also, every pulse shape MUST meet the Nyquist zero intersymbol interference (zero ISI) con-
dition.

18.6 Intersymbol Interference

We mentioned orthogonality. Well, two symbols sent at different times must also be orthogonal,
so that we can send information at one time, and then at the next, without interacting with each
other.
For example, the pulses:

1, 0 < t < Ts
p0 (t) =
0, o.w.

1, Ts < t < 2Ts
p1 (t) =
0, o.w.

1, 2Ts < t < 3Ts
p2 (t) =
0, o.w.
(34)

form an orthogonal set. The problem is, these rectangular pulses have infinite bandwidth. Can
we reduce the bandwidth and still have zero intersymbol interference? Yes. If our pulse shape
meets the Nyquist zero intersymbol interference (zero ISI) condition. This condition is that for any
integer n, 
K, n = 0
hef f (nTs ) =
0, n = ±1, ±2, . . .
where hef f (t) is the combined effect of all filters between the transmitter and receiver,

hef f (t) = ht (t) ⋆ hc (t) ⋆ hr (t)

ECE 5325/6325 Spring 2010 48

where ht (t), hc (t), and hr (t) are the transmit, channel, and receiver filter impulse responses. The
Rappaport book correctly says that ht (t) = p(t), the transmitted pulse shape, but doesn’t state
that hr (t) = p(t) as well. Note in the frequency domain, with ht (t) = hr (t) = p(t), we’d have

Hef f (f ) = Ht (f )Hc (f )Hr (f ) = |P (f )|2 Hc (f )

For zero ISI, we need hef f (t) to cross through zero at every integer multiple of Ts , the symbol
rate. What it does in between, we really don’t care. That’s because the sampler in the receiver
only samples at multiples of Ts .
Well actually, we do care, a little. It’s hard to sample exactly at the right time, we will be off
slightly from the correct nTs . So we’d like the slope at nTs to be low as well, so that we don’t
introduce much ISI from being off in our timing synchronization.
A good solution (although not the only possibility) is to use a pulse shape in the “raised cosine”
class. A pulse shape is raised cosine (RC) if it has frequency domain representation Hef f (f ) =
HRC (f ),
1−α

 Ts ,n
 h  io 0 ≤ |f | ≤ 2Ts
Ts
HRC (f ) = 2 1 + cos πTα
s
|f | − 1−α
2Ts , 1−α 1+α
2Ts ≤ |f | ≤ 2Ts
(35)

0, o.w.


But we usually need do design a system with two identical filters Ht (f ) = Hr (f ), one at the
transmitter and one at the receiver, which when multiplied produce a zero-ISI filter. In other
words, we need |P (f )|2 to meet the Nyquist filtering condition. So, we actually transmit a pulse
with the “square root” raised cosine (SRRC) shape,
 √

 Ts , 0 ≤ |f | ≤ 1−α
2Ts
 r n h  io
T πT 1−α
PSRRC (f ) = 2
s
1 + cos αs |f | − 2Ts 1−α
, 2Ts ≤ |f | ≤ 1+α
2Ts
(36)


0, o.w.


You can find p(t) by taking the inverser Fourier transform of the above expression.

18.7 Cost of Implementation

One issue in receivers is the costs of coherent vs. non-coherent reception.
Def ’n: Coherent Reception
A receiver is coherent if it is phase-synchronous, that is, it estimates and then uses the phase of
the incoming signal.

Coherent reception requires a phase-locked loop (PLL) for each different carrier frequency in
the signal. This can be a problem for FSK receivers. Also, accurate phase-locked loops can be
difficult to achieve for mobile radios for cases with high Doppler.
Differential reception is another technique to aid receivers which operate without accurate phase.
When Doppler is present, and the coherent receiver switches by 180 degrees, a standard receiver
will flip every subsequent bit estimate. A differential receiver encodes only the change in phase. In
the same case, when the phase is flipped by 180 degrees, one bit error will be made because of the
flip, but the subsequent bits will be all correct.
ECE 5325/6325 Spring 2010 49

18.8 Channel Capacity

We’ve talked about S/N , the signal power (S) to noise power (N ) ratio. In the link budget section,
we provided a S/N required for a given modulation. Here, we talk about the fundamental limits
for bandwidth efficiency for a given S/N .
Def ’n: Bandwidth efficiency
The bandwidth efficiency of a digital communication system is ratio of ηB = R/B, where R is
the bits per second achieved on the link, and B is the signal bandwidth occupied by the signal.
Bandwidth efficiency has units of bits per second per Hertz.

The limit on bandwidth efficiency is given by Shannon [16],

 
Rmax S
= log2 1 +
B N
where Rmax is the maximum possible bit rate which can be achieved on the channel for the given
signal to noise ratio.

Example: Maximum for various S/N (dB)

What is the maximum bandwidth efficiency of a link with S/N (dB) = 10, 15, 20? What maximum
bit rate can be achieved per 30 kHz of spectrum (one AMPS/USDC channel)? Is this enough for
3G?
Solution: S/N (dB) = 10, 15, 20 translates into S/N = 10, 31.6, 100. Then
Rmax
= log2 (11) = 3.46
B
Rmax
= log2 (32.6) = 5.03
B
Rmax
= log2 (101) = 6.66
B
With 30 kHz, multiplying, we have Rmax = 104, 151, 200 kbps. No, this isn’t enough for 3G cellular,
which says it aims to achieve up to 14 Mbps.
Shannon’s bound is great as engineers, we can come up with a quick answer for what we cannot
do. But it doesn’t necessarily tell us what we can achieve. The problem is that it is very difficult
to get within a couple of dB of Shannon’s bound. So, then we have to resort to the performance
equations for particular modulation types.

Lecture 12

Today: (1) Pulse Shaping (2) Modulation (3) Fidelity

19 Modulation
Last lecture we talked about how a digital transmitter sends one of M linear combinations of a small
number N of orthogonal waveforms. The TX’s choice from the M possible linear combinations lets
it send one of log2 M bits per symbol. This section talks specifically about the choices for the
orthogonal waveforms.
ECE 5325/6325 Spring 2010 50

19.1 PAM
Three “types” of modulations only use one (orthogonal) waveform, φ0 (t) = cos(2πfc t)p(t), where
p(t) is the pulse shape. (It has nothing to be orthogonal to, but...) A linear combination, then, is
just ai,0 p(t), for some constant ai,0 . A few different modulations are so named by their choice of
M different choices of ai,0 , for i = 0, . . . , M − 1.
• On-off-keying (OOK): M = 2, and we choose a0,0 = 0, and a1,0 = 1. When sending a “0” bit,
the transmitter doesn’t send anything. It only actually transmits energy when sending a “1”
bit.

• Binary PAM, a.k.a. binary phase-shift-keying (BPSK): M = 2, and we choose a0,0 = −1,
and a1,0 = 1. Now, the amplitude of the sinusoid being transmitted is switched from a +1 to
a −1, or vice versa, when the bit switches. This is also a phase shift of 180o , which is why it
can be called “phase shift” keying.

• M -ary PAM: Binary PAM is extended in M -ary PAM to include M (for M equal to some
power of 2) equally spaced amplitudes, centered on zero. So, for example, M = 4 PAM would
have a0,0 = −3, a1,0 = −1, a2,0 = +1, a3,0 = +3. In general, for M -ary PAM, ai,0 = 2i−M +1.
Note that differential phase shift keying (DPSK) is an implementation variant of BPSK, which
we will discuss next lecture.

19.2 M-ary QAM and PSK

Many types of modulations use the following two orthogonal waveforms:

φ0 (t) = cos(2πfc t)p(t)

φ1 (t) = sin(2πfc t)p(t)

where p(t), again, is the pulse shape. These symbols ai = [ai,0 , ai,1 ]T can then be plotted on a 2-D
graph, typically with the φ0 (t) amplitude plotted on the horizontal axis, and the φ1 (t) amplitude
plotted on the vertical axis.
There are two main “types” of modulations which use these two orthogonal waveforms:
• Phase-shift keying (PSK): PSK places all ai uniformly on the unit circle. That is, kai k = 1
for all i. For M = 2, this is the same as BPSK. For M = 4, the symbols are spaced every 90
degrees and thus the constellation looks like the four corners of a square, and is also called
quadrature phase shift keying (QPSK). For M = 8, symbols are every 45 degrees apart.

• QAM: Generally, any modulation that uses φ0 (t) and φ1 (t) above can be considered as QAM.
So PSK is a sometimes a subtype of QAM. But QAM is not limited to kai k = 1. For example,
square QAM has M = 22k for some integer k ≥ 4, where symbols are arranged in a square
grid, centered at zero. Note M = 4 square QAM is the same as QPSK. But M = 16 and
M = 64 are also common values of M for square QAM.
We drew five different constellation diagrams of PSK and QAM modulations in the Lecture 11
notes, on page 4.
Note that for QPSK or square M -QAM, you send log2 M bits total per symbol. We can also
look at the modulation as sending two independent signals, one on the in-phase (horizontal axis)
and one on the quadrature (vertical axis), each sending 21 log2 M bits per symbol.
ECE 5325/6325 Spring 2010 51

Note that OQPSK and π/4 QPSK are variations on QPSK, that have equivalent fidelity per-
formance and identical bandwidth as QPSK, but they are “constant envelope” modulations, which
we will discuss next lecture.
Bandwidth: Note that for PAM, PSK, and QAM, assuming SRRC pulse shaping with rolloff
factor α, the null-to-null bandwidth of the signal is B = (1 + α)Rs , where Rs is the symbol rate,
Rs = 1/Ts .

19.3 FSK
Binary FSK modulations use the following orthogonal waveforms:

φ0 (t) = cos (2π [fc − n∆f ] t) p(t)

φ1 (t) = cos (2π [fc + n∆f ] t) p(t)

where ∆f = 4T1 s . We showed that for n = 2, that is, when the frequency difference (“offset”) is
1/Ts , that the two are orthogonal. Standard binary FSK uses n = 2.
Bandwidth: For Binary FSK, when a SRRC pulse is used in p(t), the transmitted signal
bandwidth is given by
B = 2n∆f + (1 + α)Rs
where Rs = Rb = 1/Tb . So, for Binary FSK,
1
B=4 + (1 + α)Rs = (2 + α)Rs
4Ts

19.4 MSK
You can also show that orthogonality is achieved for any integer n in (37). Since putting the
two sinusoids closer together reduces the bandwidth of the signal, this is a good idea for spectral
efficiency. When n = 1 binary FSK is called minimum shift keying (MSK).
But MSK is actually the same as OQPSK with a half-cosine pulse shape. This is not something
we discuss until next lecture.

20 Fidelity
Now, let’s compare the fidelity (probability of bit error and probability of symbol error) across
different modulations. For the same energy per bit, using a different modulation would result in a
different fidelity.
Recall N = F kT0 B. We also define N0 = F kT0 , so that N = N0 B. The units of N0 are Watts
per Hertz, or equivalently, Joules.
The signal power can also be written in terms of the energy per symbol or energy per bit. Since
energy is power times time, the energy used to send a symbol duration Ts is Es = STs , with units
Joules per symbol. To calculate the energy needed to send one bit, we calculate Eb = STs / log2 M
Joules per bit. To shorten this, we define Tb = Ts / log2 M and then Eb = STb or Eb = S/R, where
R = 1/Tb is the bit rate.
The probability of bit error is a function of the ratio Eb /N0 . For example, for BPSK,
r !
2Eb
Pe,BP SK = Q
N0
ECE 5325/6325 Spring 2010 52

Name P [symbol error] P [bit error]

q 
2Eb
BPSK =Q same
q N0 
Eb
OOK =Q same
 N0 
DPSK = 12 exp − NEb0 same
q 
2(M −1) 6 log2 M Eb 1
M-PAM = M Q M 2 −1 N0
≈ log2 M P [symbol error]
q 
2Eb
QPSK =Q N0
q 
M-PSK ≤ 2Q 2(log2 M ) sin2 (π/M ) NEb0 ≈ 1
log2 M P [symbol error]
√ q 
4 ( M−1) 3 log2 M Eb
Square M-QAM ≈ log2 M
√
M
Q M −1 N0
h i
1 Eb
2-non-co-FSK =2 exp − 2N0 h same
PM −1 M −1 (−1)n+1 n log2 M Eb
i
M/2
M-non-co-FSK = n=1 ( n ) n+1 exp − n+1 N0 = M −1 P [symbol error]
q 
Eb
2-co-FSK =Q N0 same
q 
M/2
M-co-FSK ≤ (M − 1)Q log2 M NEb0 = M −1 P [symbol error]

Table 1: Summary of probability of bit and symbol error formulas for several modulations.

where Q (z) is the tail probability of a Gaussian r.v., given by

 
1 z
Q (z) = erfc √
2 2
Put this into Matlab or your calculator, it is a useful function. You might also use the table or
graph on pages 645-646 of the Rappaport book.
Other symbol error rates and bit error rates for a variety of modulations is shown in Table 20.
What are the most important? In my opinion: BPSK, DPSK, QPSK, 16-QAM, 64-QAM,
2-non-co-FSK, 2-co-FSK. For the homework, you will make a table, for each modulation, listing
P [bit error] formula, and bandwidth. You will also plot P [bit error] as a function of NEb0 . Both
will be very useful to have in your portfolio so that you have them available on the exam 2. An
example is shown in Figure 17.

20.1 Link Budgets

Link budgeting was discussed in the first part of this course assuming that the required SNR was
known. But actually, we deal with a required NEb0 . Note

Eb S 1 S B
= =
N0 N 0 Rb N Rb
B
Typically the ratio of bandwidth to bit rate is known. For example, what is Rb for SRRC pulse
shaping?
ECE 5325/6325 Spring 2010 53

0
10

Probability of Bit Error

−2
10

−4
10

BPSK
−6 DBPSK
10
0 2 4 6 8 10 12 14
Eb/N0, dB

Figure 17: Comparison of probability of bit error for BPSK and Differential BPSK.

Figure 18: Link Budgeting including Modulation. This relationship graph shows the dependencies
between important variables in the link budget.
ECE 5325/6325 Spring 2010 54

20.2 Examples

Example: 200 kHz

What bit rate can be achieved on a 200 kHz if the S/N ratio is 20 dB?

Example: 1.25 MHz

What bit rate can be achieved on a 200 kHz if the S/N ratio is 12 dB?

Lecture 14

Today: (1) Implementation Costs

21 Implementation Costs
21.1 Power Amplifiers and Constant Envelope
A power amplifier at the transmitter to take the transmitted signal and amplify it to the desired
output power (e.g., 500 mW for cellular). Power amplifiers waste some energy, and are rated by
their efficiency. Amplifiers that can deal with any signal you send to it are about 50% efficient
(Class A amplifiers). However, the most efficient (Class C) amplifiers (90% efficient) require a
constant envelope modulation. Designers thus tend to choose constant envelope modulations for
battery-powered transmitters.
A constant envelope signal s(t) has envelope (magnitude) that does not change significantly
over time (symbol periods). We can check the envelope of s(t) by looking at the constellation
diagram. In the next section, we compare QPSK (which is not constant envelope) with O-QPSK
and π/4-QPSK (which are closer to constant envelope).
A metric to quantify the envelope changes is the peak-to-average ratio (PAR). The PAR is
defined as
Maximum Envelope
P AR =
Average Envelope
where “Envelope” is |s(t)| for transmitted signal s(t).

21.1.1 Offset QPSK

For QPSK, we can write the modulated signal as

s(t) = p(t) cos(2πfc t + ∠a(t))

where ∠a(t) is the angle of the symbol chosen to be sent at time t. It is in a discrete set,
 
π 3π 5π 7π
∠a(t) ∈ , , ,
4 4 4 4
and p(t) is the pulse shape. We could have also written s(t) as

s(t) = p(t) [a0 (t) cos(2πfc t) + a1 (t) sin(2πfc t)]

ECE 5325/6325 Spring 2010 55

QPSK without offset: Offset QPSK:

Q-channel Q-channel

I-channel I-channel

(a) (b)
Figure 19: Constellation Diagram for (a) QPSK and (b) O-QPSK.

The problem is that when the phase changes 180 degrees, the signal s(t) will go through zero, which
precludes use of a class C amplifier. See Figure 20(b) and Figure 19 (a) to see this graphically.
For offset QPSK (OQPSK), we delay the quadrature Ts /2 with respect to the in-phase. Rewrit-
ing s(t)       
Ts Ts Ts
s(t) = p(t)a0 (t) cos(2πfc t) + p t − a1 t − sin 2πfc t −
2 2 2
At the receiver, we just need to delay the sampling on the quadrature half of a sample period
with respect to the in-phase signal. The new transmitted signal takes the same bandwidth and
average power, and has the same NEb0 vs. probability of bit error performance. However, the envelope
|s(t)| is largely constant. See Figure 20 for a comparison of QPSK and OQPSK.

21.1.2 π/4 QPSK

Another way to avoid sending the envelope to 0 is to use π/4 QPSK. The idea is simple. It doesn’t
matter if we rotate the QPSK constellation by 45 degrees (π/4 radians). So let’s alternate between
the standard QPSK constellation in Figure 19 (a), and a π/4 radians rotated version of the same
constellation. For example, if kTs is the start of the kth symbol, if k is odd, then use the standard
constellation; if k is even, use the rotated version. Now, the possible transitions are shown in Figure
21. Note that there are no self-loops as in Figure 19(a) because each symbol location changes from
time k to k + 1. The desireable part of π/4 QPSK is that the transitions never cross through the
origin.

21.1.3 Other Modulations

What are other modulations that do or do not have a constant envelope?

1. FSK: FSK modulations DO have a constant envelope.

2. M -QAM: Square M -QAM modulations DO NOT have a constant envelope, and are generally
even worse than QPSK in that they have greater needs for linear amplifiers.

21.2 Sync
There are three main kinds of synchronization which we require in a receiver:
ECE 5325/6325 Spring 2010 56

0.2

0.1 0.15
Real

0
0.1
−0.1
0.05

Imag
0 1 2 3 4
Time t 0

0.1 −0.05
Imag

0
−0.1
−0.1
−0.15
0 1 2 3 4 −0.2 −0.1 0 0.1 0.2
(a) Time t (b) Real

0.2

0.1 0.15
Real

0
0.1
−0.1
0.05
0 1 2 3 4 Imag
Time t 0

0.1 −0.05
Imag

0
−0.1
−0.1
−0.15
0 1 2 3 4 −0.2 −0.1 0 0.1 0.2
(c) Time t (d) Real

Figure 20: Matlab simulation of (a-b) QPSK and (c-d) O-QPSK, showing the (d) largely constant
envelope of OQPSK, compared to (b) that for QPSK.

1. Phase: To multiply by the correct sine or cosine, we need to know the phase of the incoming
signal. Typically, we track this with a phase-locked loop (PLL). PLLs can be difficult to
implement.

2. Symbol Timing: One must synchronize to the incoming signal to know when to sample the
symbol value. Errors in sampling time introduce inter-symbol interference (ISI). Using a lower
α makes a receiver more sensitive to symbol synch errors.

In this section, we present two different ways to simplify phase synchronization.

21.2.1 Energy Detection of FSK

FSK reception can be done via a matched filter receiver, in which there is a multiply and integrate
with each orthogonal waveform (in this case, sinusoids at different frequencies). This is shown in
Figure 22. This is called “coherent reception” of FSK. For binary FSK, there must be 2 PLLs,
which have more difficulty, because the received signal contains each frequency sinusoid only half of
the time. FSK is popular in inexpensive radios; so two PLLs might be significant enough additional
cost to preclude their use.
A non-coherent binary FSK receiver avoids PLLs altogether by simply computing the energy
Efk at each frequency fk , for k = 0, 1. It decides the bit by seeing which energy, Ef0 or Ef1 , is
higher.
ECE 5325/6325 Spring 2010 57

Figure 21: Constellations and transitions for π/4 QPSK.

To calculate the energy, it still must multiply and integrate with a cosine and sine at each
frequency. That’s because if it just multiplies with the cosine, for example, and the phase of the
received signal makes it a sine, then the multiply and integrate will result in zero (sine and cosine
are orthogonal, after all). But if you find the energy in the cosine, xIk , and the energy in the cosine,
xQk , then the total energy at frequency k is

Efk = (xIk )2 + (xQ

k)
2

is calculated for each frequency fk , k = 0, 1.

The downside of non-coherent reception of FSK is that qtheprobability of bit error increases.
Eb
The optimal, coherent FSK receiver has performance Q N0 . The non-coherent binary FSK
h i
Eb
receiver has probability of bit error of 21 exp − 2N 0
. There is about a 1 dB difference between the
two – to achieve the same probability of bit error, the non-coherent receiver must have about 1 dB
more received power.

21.2.2 Differential PSK

In BPSK, the bit is encoded by the sign of the sinusoid. But the sign is just a 180 degree phase
shift in the sinusoid. So when we receive the signal and lock our PLL to the signal, we’ll see the
phase changing 180 degrees, but we won’t know which one is positive and which one is negative.
One thing that we can do is to encode the bit using the change of phase. That is, change the phase
180 degrees to indicate a “1” bit, and keep the phase the same to indicate a “0” bit. This is called
differential encoding. In the RX, we’ll keep track of the phase of the last symbol and compare
it to the phase of the current symbol. When that change is closer to 180 degrees than 0 degrees
we’ll decide that a “1” bit was sent, or alternatively, if that change is closer to 0 degrees than 180
degrees, we’ll decide that a “1” bit was sent. This is called differential phase shift keying (DPSK)
instead of BPSK. The downside of BPSK is that once we’re off by 180 degrees, every bit will be
decided in error. Instead, DPSK never becomes 100% wrong, it degrades gracefully with phase
(and thus frequency) offset.
ECE 5325/6325 Spring 2010 58

Figure 22: Phase-coherent demodulation of M -ary FSK signals, from Proakis & Salehi [15], Figure
7.46.

Figure 23: Demodulation and square-law detection of binary FSK signals, from Proakis & Salehi
[15], Figure 7.49.
ECE 5325/6325 Spring 2010 59

The downside ofq

DPSKis that the probability of bit error increases. The optimal BPSK
h ireceiver
2Eb 1 Eb
has performance Q N0 . The DPSK receiver has probability of bit error of 2 exp − N0 . There
is about a 1 dB difference between the two – to achieve the same probability of bit error, the DPSK
receiver must have about 1 dB more received power.

Lecture 14

Today: (1) OFDM, (2) Fading Demo

22 Fading Demo
To run the demo I am running, you need:

• Two wireless sensors programmed with SPIN, and one programmed with BaseStation, two
TinyOS programs. You may borrow these from me.

• A computer with Python, SciPy, and MatPlotLib

• The java package net.tinyos.tools

A Ubuntu OS makes the installation easy as adding packages. Other OSes, I haven’t tried.
My sensors are 2.4-2.48 GHz radios, using DS-SS, with OQPSK modulation, at a data rate of
250 kbps, with a BW of about 5 MHz. The antennas are inverted F on-board antennas. They
measure received power in dBm. Both directions of the link are measured.
The questions to answer while participating in the demo are:

1. What type of fading is measured while one transceiver is moving?

2. What type of fading causes the measured Pr to differ when the transceiver has changed
positions slightly?

3. What would I see if I could change the center frequency?

4. What is the effect of a person walking near either node, or in the line between them?

5. Can you tell whether a link is in a deep fade or not?

6. What is the dB difference in power between a deep fade and an anti-fade (opposite of a fade)?

23 Multi-carrier Modulation
The basic idea for multi-carrier modulation is to divide the total available bandwidth B into N
different subchannels. In each subchannel, we’ll use some particular modulation, and send all
signals on all subchannels simultaneously. Considering orthogonal waveforms, we’ll use both the
sin and the cosine function at many different frequencies.
The benefit is that with narrower bandwidth subchannels, we can achieve flat fading on each
subchannel. Remember, if the bandwidth is low enough, we don’t need an equalizer, because the
ECE 5325/6325 Spring 2010 60

channel experiences flat fading. (Q: What was the rule given in the Rappaport book given Ts and
στ to determine if fading is flat or frequency selective?)
So we will set N high enough so that the symbol rate in each subchannel (which is B/N wide)
will be high enough to “qualify” as flat fading. How do you find Ts to fit in B/N ? Recall the
bandwidth is given by T1s (1 + α) for rolloff factor α so now Ts = N
B (1 + α).
If we use Ts ≥ 10στ then now we can choose N as
1
N ≥ 10στ B
1+α
Now, bit errors are not an ‘all or nothing’ game. In frequency multiplexing, there are N parallel
bitstreams, each of rate log2 (M )/Ts , where log2 (M )/Ts is the bit rate on each subchannel. As a
first order approximation, a subchannel either experiences a high SNR and makes no errors; or is
in a severe fade, has a very low SNR, and experiences a BER of 0.5 (the worst bit error rate!). If β
is the probability that a subchannel experiences a severe fade, the overall probability of error will
be 0.5β.
Frequency multiplexing is typically combined with channel coding designed to correct a small
percentage of bit errors.
Note that the handout uses TN where we use Ts . They reserve Ts to be TN /N . But the symbol
period, the duration of the symbol, is in fact what we call Ts . We know N symbols are sent
simultaneously, but each one still has duration Ts . Please make a note.

23.1 OFDM
In the previous section we made no restriction on the frequencies of the subcarriers. Well, we know
that to have orthogonal sinusoids at different frequencies, we need a particular condition on ∆f
between the two frequencies.
In FSK, we use a single basis function at each of different frequencies. In QAM, we use two
basis functions at the same frequency. OFDM is the combination:

cos(2πfc t), 0 ≤ t ≤ Ts
φ0,I (t) =
0, o.w.

sin(2πfc t), 0 ≤ t ≤ Ts
φ0,Q (t) =
0, o.w.

cos(2πfc t + 2π∆f t), 0 ≤ t ≤ Ts
φ1,I (t) =
0, o.w.

sin(2πfc t + 2π∆f t), 0 ≤ t ≤ Ts
φ1,Q (t) =
0, o.w.
..
. 
cos(2πfc t + 2π(M − 1)∆f t), 0 ≤ t ≤ Ts
φM −1,I (t) =
0, o.w.

sin(2πfc t + 2π(M − 1)∆f t), 0 ≤ t ≤ Ts
φM −1,Q (t) =
0, o.w.

where ∆f = 2T1 s . These are all orthogonal functions! We can transmit much more information
than possible in M -ary FSK. (Note we have 2M basis functions here!)
ECE 5325/6325 Spring 2010 61

The signal on subchannel k might be represented as:

xk (t) = ak,I (t) cos(2πfk t) + ak,Q (t) sin(2πfk t)

The complex baseband signal of the sum of all K signals might then be represented as
(N )
X
xl (t) = R (ak,I (t) + jak,Q (t))ej2πk∆f t
k=1
(N )
X
j2πk∆f t
xl (t) = R Ak (t)e (37)
k=1

where Ak (t) = ak,I (t) + jak,Q (t). Does this look like an inverse discrete Fourier transform? If yes,
than you can see why it is possible to use an IFFT and FFT to generate the complex baseband
signal.

1. FFT implementation: There is a particular implementation of the transmitter and receiver

that use FFT/IFFT operations. This avoids having N independent transmitter chains and
receiver chains. The FFT implementation (and the speed and ease of implementation of the
FFT in hardware) is why OFDM is popular.

Since the N carriers are orthogonal, the signal is like N -ary FSK. But, rather than transmitting
on one of the K carriers at a given time (like FSK) we transmit information in parallel on all N
channels simultaneously. An example state space diagram for N = 3 and PAM on each channel is
shown in Figure 24.

OFDM, 3 subchannels
of 4-ary PAM

Figure 24: Signal space diagram for N = 3 subchannel OFDM with 4-PAM on each channel.

However, because the FFT implementation assumes that the data is periodic, and it is not, we
must add a cyclic prefix at the start of each block of N symbols. The cyclic prefix is length µ and
is an overhead (doesn’t contain data) and thus reduces our overall data rate. The actual data rate
will be the total bit rate multiplied by NN+µ . We determine µ to be longer than the expected delay
spread στ . So, µ = στ /Ts ,
ECE 5325/6325 Spring 2010 62

Example: 802.11a
IEEE 802.11a uses OFDM with 52 subcarriers. Four of the subcarriers are reserved for pilot tones,
so effectively 48 subcarriers are used for data. Each data subcarrier can be modulated in different
ways. One example is to use 16 square QAM on each subcarrier (which is 4 bits per symbol per
subcarrier). The symbol rate in 802.11a is 250k/sec. Thus the bit rate is

OFDM symbols subcarriers coded bits Mb (coded)

250 × 103 48 4 = 48
sec OFDM symbol subcarrier sec

Lecture 15

Today: (1) Channel Coding

24 Channel Coding
The material for this section comes largely from Jeff Frolik’s MUSE channel coding video, available
at:

• http://www.uvm.edu/~muse/CTA.html

Def ’n: Forward error correction coding or channel coding

Adding redundancy to our data at the transmitter with the purpose of detecting and correcting
errors at the receiver.
The transmitter takes in data bits and puts out coded bits. Our notation is that for each k
data bits input to the FEC operator, the FEC operation will produce n > k coded bits out.

24.1 Block Codes

Def ’n: (k, n) Block Code
A (k, n) block code inputs k-bits which are accumulated (via serial-to-parallel conversion) in a k-
length vector d. Block encoding multiplies d by a k × n generator matrix, G, to output a n-length
bit vector c. Block decoding then multiplies the received vector r by the syndrome matrix S to
determine if any errors occurred and determine which (if any) bits were in error out of the n sent.

The syndrome is just a rearrangement of the transpose of the generator matrix, as shown by
example below.
In contrast, a convolutional code is a “running” code. For encoding, bits are input into what is
effectively a binary filter, the output bits are dependent on the current and past bits.
Compare the advantages and disadvantages:

• Block code: Advantages: Better for data that is not coming in large streams (bursty data
sources, <1000 bits), e.g., wireless sensor networks. Pretty simple computation. Not the best
one can do in terms of improving energy efficiency / removing errors. Block codes are used
in CDs, DVDs, and disk drives.
ECE 5325/6325 Spring 2010 63

• Convolutional codes: Advantages: Best for very large data streams. More energy efficient
than block codes when you have large streams of data. Convolutional codes are used in: deep
space communication (Voyager program), satellite and terrestrial digital video broadcasting.
Disadvantages: Computational complexity increases exponentially in the length of the code.
Andrew Viterbi (founder of Qualcomm) is credited with the optimal decoder, called the
Viterbi algorithm.
Let the input be denoted d, a k-bit vector. Let the output be c, a n-bit vector. Let G be the
generator matrix. Then
c = dG
Thus the G matrix has size k × n. Multiplication by a matrix means, multiply all of the pairs, and
then exclusive or them, that is, sum them mod 2. That is, if the sum of the products is even, the
answer is 0, if the sum is odd, the answer is 1.
Def ’n: Systematic
The first k bits of the n bits output, are the same as the k bits in d.

Example: (6, 3) systematic block code which can correct one bit error
Let G be given by:  
1 0 0 1 0 1
G= 0 1 0 0 1 1 
0 0 1 1 1 0
Encode the data bits d = [1, 0, 1].
Solution: c = [1, 0, 1, 0, 1, 1]

Example: Reception
Given you receive r = [1, 1, 1, 0, 1, 1], what was is the estimate of the transmitted data?
Solution: At the receiver, multiply by the syndrome
1 0 1
 
 0 1 1 
 
 1 1 0 
S=  
 1 0 0 

 0 1 0 
0 0 1

Compute: rS = [0, 1, 1].

Look at all of the rows of the syndrome. Where we find the output, it gives us the bit that
was in error. All zeros indicates no errors. Since the second bit was in error, instead of [1, 1, 1], we
know the correct output was [1, 0, 1].

Example: (7, 4) Block Code

Encode d = [0, 1, 1, 0] with the (7, 4) block code with generator,
 
1 0 0 0 1 1 1
 0 1 0 0 0 1 1 
G=  0

0 1 0 1 0 1 
0 0 0 1 1 1 0
ECE 5325/6325 Spring 2010 64

If r = [0, 1, 1, 0, 1, 1, 1] is received, what would the receiver determine to be the demodulated bits?
Solution: I get c = [0, 1, 1, 0, 1, 1, 0]. Then, multiplying [0, 1, 1, 0, 1, 1, 1]S,

1 1 1
 

 0 1 1 


 1 0 1 

S=
 1 1 0 


 1 0 0 

 0 1 0 
0 0 1

I get [0, 0, 1], which is the same as the 7th row, which says that the last row was incorrectly received,
and so the 7th bit was incorrect. Thus the correct four bits sent were [0, 1, 1, 0].

24.2 Error Detection

Rather than trying to correct the data, can also simply try to detect the error. This lets us know
not to use the data we’ve received. If we request the data again, it tells the transmitter that we
didn’t get the data (perhaps the SNR was too low, or the interference power was too high), which
it might need to know anyway. For error detection, we have the following packet structure:

1. Header: Information about who the data is meant for, what purpose data is being sent, etc.

2. Payload: Data that we need to communicate.

3. Footer: Where the CRC is included.

The cyclic redundancy check (CRC) is a check to make sure that the data in the payload was
correctly received. The CRC is described by a polynomial, for example, c(x) = x + 1. The
data we want to send, d, can also be described as a polynomial. For example d = [0, 1, 0, 1]
corresponds to the polynomial d(x) = x2 + 1. To determine the polynomial, just multiply the (row)
bit vector with the column vector [x3 , x2 , x1 , 1]T , or, in general, for n-length data string, multiply
by [xn−1 , xn , . . . , 1]T .
Then, to find the CRC bit, we divide the two polynomials, d(x) ÷ c(x) and the CRC bit is the
remainder. So in fact the CRC is d(x) mod c(x). In this example, the CRC is zero.
Then, what we send through the channel is the desired bits appended by the CRC bit, in this
example, [0, 1, 0, 1, 0].

Example: CRC for 5 bit payload

Let d = [1, 1, 0, 1, 1]. What CRC bit should be appended for c(x) = x + 1?
Solution: When dividing d(x) = x4 + x3 + x + 1 by c(x) = x + 1, you get a remainder of 0. So 0
should be appended.
Reception: To determine if the data was correctly received for the 1-bit CRC, just check if the
data (including CRC bit) has an even number of 1s.

Example: 16 bit CRC

Let c(x) = x16 + x12 + x5 + 1. Now the footer will be 16 bits. Now, we can detect a burst error of
up to 16 bits anywhere in the packet. A burst error is a few sequential bit errors.
ECE 5325/6325 Spring 2010 65

24.3 System Performance and Costs

Using a (7,4) block code, we can correct a single error in the 7 bits. But we need to increase the
number of bits to send, which then requires more energy. So when using channel coding, we reduce
the transmit power when comparing uncoded and coded transmissions. Still, the probability of bit
error goes down for equal NEb0 . Equivalently, we can achieve the same bit error rate at 1 dB lower
power. This value, 1 dB, is the coding gain. In our link budgets, coding goes in the Gains column,
added in with the antenna gains.
However, coding requires sending additional bits. So, in a coded system, there is always
a ratio of data bits to coded bits, r, called the code rate. In the (7,4) block code it is r =
4 data bits /7 coded bits. For a fixed bandwidth, this reduces the achievable data rate by r. For a
fixed data rate, it increases the bandwidth by a factor of 1/r.
Using a CRC, we add a few additional bits to the packet, much fewer than a FEC code would
add. This allows us to detect an error, but not correct it. However, it is often important to ensure
that the data received did not have any errors, beyond any reasonable doubt. For example, for
cell phone audio, it may be okay to make a bit error, and have it cause some temporary noise in
the audio signal; but when transferring a file, it may be critical not to have the file show up with
incorrect data, and not know that it was corrupted.
In many systems, we will use both FEC and CRC, to increase the energy efficiency (with the
FEC) and to make sure that received data is correct.

Lecture 16

Today: (1) Spread Spectrum

25 Spread Spectrum
“Spread spectrum” is the use of a much wider bandwidth than necessary in a radio communications
system, in order to achieve other objectives.
For example, frequency modulation (FM) historically was considered to be widely spread in
spectrum (broadcast FM radio uses 200 kHz to send audio, which is inherently a 3-10 kHz bandwidth
signal). However, FM was used because it could mitigate the effects of multipath fading, and could
achieve better signal quality at low SNR.
The two major types of spread spectrum modulation are:

1. Frequency Hopping Spread Spectrum (FH-SS)

2. Direct-Sequence Spread Spectrum (DS-SS)

Both types of SS use pseudo-random codes, periodic sequences of zeros and ones. These are not
actually random – a computer generates them using simple binary feedback logic. But, an average
person would look at the sequence and judge them to be random. So they are called pseudo-random.

25.1 FH-SS
FH-SS pseudo-randomly changes center frequency each “hopping period”, Th . Bluetooth is a FH-SS
system, which achieves a (coded) bit rate of 1 Mbps (potentially up to 3 Mbps), but uses 80 MHz
ECE 5325/6325 Spring 2010 66

of spectrum, in 79 different center frequencies, with a hopping period Th = 1/1600 s/hop. While at
each center frequency, the modulation is similar to things we have been talking about, e.g., FSK,
DPSK, QPSK, etc.
Three benefits of FH-SS are:

1. Interference avoidance: There may be significant interference at a few of the center frequen-
cies. But even if we totally lose all bits during those hops, we will be able to recover using
the bits received during successful (non-interfered) hops. We also avoid being an interferer to
someone else’s signal for too long.

2. Multiple Access: Two devices can occupy the same spectrum and operate without coordinat-
ing medium access at all. Their transmissions will “collide” some small fraction of the time,
but not often enough to cause failure.

3. Stealth: There is an advantage to switching randomly among frequencies when an eavesdrop-

per doesn’t know your hopping pattern – they will not be able to easily follow your signal.
This was the original reason for the discovery and use of FH-SS (by actor and inventor Hedy
Lamarr, in 1940 [11]).

25.2 DS-SS
We will see that DS-SS has the same three advantages as FH-SS, but for different reasons.
Direct-sequence spread spectrum simply uses a pulse shape that has a wide bandwidth. This
pulse shape is known as a pseudo-noise signal, which is, essentially, itself a BPSK-modulated signal,
but which “pseudo-random” data that is known to both transmitter and receiver.

Figure 25: Pseudo-noise (“Barker”) code used in 802.11b DS-SS modulation.

Figure 25 shows the PN signal used in 802.11b. Can you draw on top of Figure 25 what the
pulse shape would be if it were not DS-SS, and instead, simply modulated BPSK, with a SRRC
pulse shape? You can see that the bandwidth of the DS-SS signal will be very large compared to
that of the BPSK signal.
We can describe the modulated signal, sent by user k, as (from Rappaport):

sk (t) = ak (t)pk (t) cos(2πfc t)

where ak (t) is the bit signal, +1 or -1, which indicates the data bits to be sent. For example, the
Barker code uses bits [+1, 1, +1, +1, 1, +1, +1, +1, 1, 1, 1]. As before, we have a pulse shape pk (t),
ECE 5325/6325 Spring 2010 67

however, in this case, pk (t) is not simply a SRRC pulse. It is a high bandwidth pseudo-noise signal.
Essentially, pk (t) is a BPSK-modulated signal itself. The “bits” of the PN signal are called “chips”
to distinguish them from bits. We denote the number of chips in pk (t) as P G, it is the number of
chips per bit. It is called PG because it is also called the processing gain. The period of each chip
is denoted Tc , so
Ts Rc
PG = =
Tc Rs
Where Rc = 1/Tc is the chip rate. For example, 802.11b has a chip rate of 11 M (chips per second)
and a symbol rate of 1 M (symbols per second). As another example, in IS-95, the “short code”
has P G = 215+1 = 32768.
The bandwidth of a DS-SS symbol, when chips have the SRRC shape, is

B = (1 + α)Rc

Which is then P G times the bandwidth of the BPSK signal would have been as a narrowband
signal.
Recall that the SNR required to demodulate a signal is given by:
Eb Rb
SN R =
N0 B
So with DS-SS, the SNR is lowered by a factor of P G
Eb Rb Eb
SN R = =
N0 (1 + α)Rc N0 (1 + α)P G
However, If you thought this signal was just a plain-old BPSK signal, i.e., didn’t know the PN
signal, you’d need the regular SNR NEb0 1+α , which is P G times higher. This makes us understand
advantage #3 of DS-SS: Stealth. Knowing the PN signal allows one to demodulate the signal with
P G times less SNR than an eavesdropper could. If the P G is high enough, the signal would be
extremely difficult to detect at all, but could still be used by the intended receiver.
Advantage #1: Reception of DS-SS uses the same principle as discussed earlier – the received
signal is correlated with the known PN signal. What if a narrowband interference signal was also in
the received signal? Well, this narrowband signal would effectively be spread, when it is multiplied
by the PN signal in the receiver. In contrast, the desired signal is de-spread (becomes narrowband
again) due to correlation with the PN signal. The spread interference can then be partially filtered
out using a narrowband filter. See Figure 6.50 in Rappaport (page 333).
Advantage #2: Further, DS-SS can be designed for some particular benefits for multiple access.
These relate to the near-orthogonality of the particular PN codes used in DS-SS. In short, some sets
of PN signals are nearly orthogonal or completely orthogonal to each other ; and some PN signals
have the property that the PN signal is orthogonal to itself at a different time delay.
First, consider sets of PN signals orthogonal to each other. One example is the set of Gold
codes (used in the GPS network). Gold code signals are nearly orthogonal with each other.
Another example is the Walsh-Hadamard sequence set, used in IS-95 (CDMA). The WH-64
sequences, shown in Figure 26(c), are used in IS-95. The 64 WH signals are exactly orthogonal to
each other. These signals provide a means on the downlink to send 64 simultaneous user signals,
and yet have each mobile perform a correlation operation that completely zeros out the signal sent
from the BS to all 63 other mobiles. When one mobile correlates with its signal, pk (t), it has zero
correlation with the other 63 WH signals.
ECE 5325/6325 Spring 2010 68

(a) (b)

(c)

Figure 26: Walsh Hadamard sequences are in the rows of the above images. The signal is +1
during black squares, and -1 during red squares: (a) two WH-2 sequences in two rows, (b) four
WH-4 sequences in four rows, and (c) sixty-four WH-64 sequences in 64 rows.

Note that WH sequences are not used on the uplink, because they are only orthogonal if time-
synchronized. Mobiles aren’t able to time-synchronize with each other very well.
Second, consider the autocorrelation of a PN signal (the correlation of a signal with itself at
different time delays). The autocorrelation is defined as
Z Ts
Rp (τ ) = ak (t)pk (t)pk (t − τ )dt
0

First assume that the data signal ak (t) = 1. If τ = 0, the value of Rp (τ ) is simply the energy in the
PN signal pk (t) over a duration Ts . For τ a multiple of Ts , that is, the period of pk (t), we get the
same value, i.e., Rp (nTs ) = Rp (0). For in between values of τ , (say, Tc < τ < Ts − Tc , PN signals
have a nearly (but not quite) zero autocorrelation. Generally, for these τ , Rp (τ ) ≈ −Rp (0)/P G.
Figure 28(top plot) shows an example for the Barker code used in 802.11b.
Now, if the data was modulated, the exact −Rp (0)/P G is gone, but we still do have |Rp (τ )| ≤
Rp (0)/P G. Notice that the autocorrelation becomes strongly negative when the bit sent was a -1.

Example: PN code generation

Consider Figure 28(a). Let each state of s1 , . . . , s4 to be either zero or one (but not all zeros to
start out). At each stage, one bit is output, and there is mod 2 addition performed to fill s1 . Let’s
calculate the output.

Lecture 18

Today: (1) Rake Receiver, (2) Channel Measurement

ECE 5325/6325 Spring 2010 69

0.5

0 T 2T 3T 4T

−1
0 T 2T 3T 4T
Time Delay

Figure 27: Autocorrelation of “Barker” code used in 802.11b DS-SS modulation, (a) with all symbol
values = +1, that is, ak (t) = 1, and (b) with data modulation.

(a)

(b)
2

1.5

0.5

−0.5

−1
(c) 5 10 15 20 25 30

Figure 28: Linear feedback shift register (LFSR) generation of an (a) 4-stage, or (b) 5-stage,
maximal-length PN code; and (c) the generated code sequence for the 4-stage LFSR.
ECE 5325/6325 Spring 2010 70

26 Rake Receiver
We have presented in Lecture 16 the DS-SS signal, which we represent as
sk (t) = ak (t)pk (t) cos(2πfc t)
where ak (t) is the bit signal, +1 or -1, which indicates the symbol value being sent, and pk (t) is
the “pulse shape”, which for DS-SS is a PN signal, as shown in Figure 29.

Figure 29: A DS-SS transmitter multiplies symbol values by the PN signal and cos(2πfc t) to
produce the transmitted signal.

Here, we discuss the receiver for a DS-SS signal. We saw at the end of Lecture 16 that the PN
signal was nearly orthogonal to itself at different time delays. That is, after the correlator with the
PN signal in the receiver, the output has a spike when the PN signal lines up with itself, and is
nearly zero at other times. First, assume that there are no multipath. In this case, the received
signal would look like what is shown in Figure 30(a). There is one peak corresponding at 0, Ts ,
2Ts , and so on. The value of that peak indicates which symbol value was sent.

(a)

(b)

Figure 30: (a) A simplified DS-SS receiver (which essentially assumes no multipath exist) correlates
the downconverted signal with the PN signal and samples the output each symbol to determine
which symbol value was sent. (b) A “rake” DS-SS receiver samples the correlator output at each
multipath “peak” in order to benefit from each arriving multipath signal’s power.

But in real world channels, multipath will arrive at different time delays. In the DS-SS receiver,
this can be measured by looking at the correlator output, as shown in Figure 30(b). Instead of one
ECE 5325/6325 Spring 2010 71

peak, there are multiple peaks, one at each multipath time delay. (Actually, they may begin to
merge together to form fewer wider peaks, rather than individual peaks.) In any case, any of these
peaks could be used to determine the symbol value. So instead of looking at just one, we should
look at all of them.
The use of the values of the correlator output at multiple time delays is called a rake receiver.
In some books, “rake” is written as “RAKE”, but it is not an acronym, it is an analogy to a garden
rake. A rake has several “fingers” that each can pick up things simultaneously.
The use of a rake receiver is called “multipath diversity”. Effectively, “diversity” is not putting
your eggs all in one basket. In this case, any one multipath may be blocked or may fade. If we use
a rake receiver, we don’t completely fail when one multipath component is received with less power.
The rake effectively increases the SNR by combining the power contained in different peaks.

27 Channel Measurement
Wideband channel measurement (“small scale multipath measurements” in Rappaport) is the de-
termination of the channel impulse response (CIR) or channel frequency response for a given link.
Wideband channel measurement is useful to get inputs required for system design such as corre-
lation bandwidth or RMS delay spread, or to determine how well a diversity method will perform
(how well will a rake receiver work?). But, different frequency bands (e.g., 2.4 GHz vs. 24 GHz),
and different environments (e.g., indoor WLANs vs. plane-to-plane communications) will tend to
have different characteristics. The only way to really know what a channel is like in a particular
environment at a particular frequency is to measure it.
We discussed that the radio channel is a filter in Lecture 9. The channel impulse response is
given by
M
X −1
h(τ ) = ai ejφi δ(τ − τi ) (38)
i=0
where ai is the voltage gain, φi is the phase change, and τi is the time delay, of multipath component
i. Note that name “power delay profile” (PDP) is the name given to a measurement of |h(τ )|2 .
The channel frequency response is
M
X −1
H(f ) = ai ej(−2πf τi +φi ) (39)
i=0

Rake reception is a good introduction to channel measurement because the rake receiver pro-
duces as a by-product information about the amplitude, phase, and time delays of the multipath
in the channel, in the output of the correlator. The impulse responses printed in Lecture 9 were
generated using the correlator output of a 802.11b receiver that we built in our lab using a software
radio [4].
The Rappaport book relates three other types of measurement systems:
(1) Direct RF Pulse System: We can’t send δ(τ ), a true impulse, through the air – it would
be interference in every frequency band. But, we can send a rectangular pulse with a narrow pulse
width, as an approximation. When the pulse is received, it will be the convolution of h(τ ) in (38)
and the original pulse shape. This still does require producing a lot of interference, so it is not a
popular method.
(2) DS-SS Sliding Correlator: The DS-SS sliding correlator sends a DS-SS signal from the
transmitter with no data (ak (t) = 1 for all t). The receiver is not a rake receiver. The problem
ECE 5325/6325 Spring 2010 72

Figure 31: A measured power delay profile using a DS-SS sliding correlator method with a PN
signal clock rate of 100 MHz, at fc = 1.8 GHz, in an outdoor environment on the Virginia Tech
campus [14].

with a rake receiver is that it is computationally complex. It requires doing many multiplies
and adds each sample. If the measured bandwidth is high (say, a few hundred MHz), then the
required computational power may be too high for standard hardware. The DS-SS sliding correlator
dramatically slows down the computational requirements. It uses a slower chip rate β in the PN
signal generator in the receiver than in the transmitter (denote the transmit chip rate as α > β).
This effectively slows down the correlation to the difference between the PN signal generator chip
rates, α − β. For example, the transmit chip rate might be α = 100 MHz, but the receiver might
have chip rate β = 99.99 MHz, for a difference α − β = 10 kHz, 104 times slower than α. Now, the
correlator output traces out the measured channel impulse response each 1/(α − β). An example is
shown in Figure 31. Although this figure shows time in µs, the plot was actually recorded in time
1 µs times 104 , or 0.01 second.
(3) Frequency Domain: The frequency sweep method simply transmits a CW signal that
slowly increases frequency from fmin to fmax . Assuming that the transmitted signal had constant
amplitude, the amplitude of the received signal thus measures |H(f )|. A good way to do this is
with a vector network analyzer (VNA). By setting the VNA to measure the s-parameters of the
channel, we obtain the gain in the channel, which is s21 . By having the VNA sweep across input
frequencies, we can measure amplitude and phase of H(f ) across the band we’re interested in. Note
if we really want h(τ ), it is just the Fourier transform of H(f ).

Lecture 21

Today: (1) MIMO

28 Diversity
Diversity is the use of multiple channels to increase the signal to noise ratio in the presence of
random fading losses. The idea of diversity is “don’t put all of your eggs in one basket”.
For fading channels, we know that there is a finite probability that a signal power will fall
below any given fade margin. For a Rayleigh channel, we showed that to have the signal above the
ECE 5325/6325 Spring 2010 73

required SNR 99% of the time, we needed to include a fade margin of 18.9 dB. This is a big “loss”
in our link budget. For example, if we didn’t need to include the fade margin in the link budget, we
could multiply the path length by a factor of 1018.9/20 ≈ 10 (in free space); or increase the number
of bits per symbol in the modulation significantly higher so that we can achieve higher bit rate for
the same bandwidth.
There are several physical means to achieve multiple channels, and to get those channels to be
nearly independent. Each has its advantages and disadvantages.

Space Diversity Space diversity at a receiver is the use of multiple antennas across space. Be-
cause multipath fading changes quickly over space (see lecture notes on fading rate, and Doppler
fading), the signal amplitude received on the different antennas can have a low correlation coeffi-
cient. The low correlation typically comes at separation distances of more than λ/2, where λ is
the carrier wavelength. The Jakes model (equal power from all angles) says that the correlation
coefficient at λ/2 is exactly zero; however, in reality, this is not true. The actual angular power
profile (multipath power vs. angle) determines the actual correlation coefficient. In general, we ei-
ther accept that the correlation coefficient is not perfectly zero, or we separate the antennas further
than λ/2. What is λ/2 at typical carrier frequencies?
The problems with space diversity are most importantly that for consumer radios, we want
them to be small; and multiple antennas means that the device will be larger. This is fine when
space is not a big concern – for base stations, or for laptops and access points. Another problem is,
in general, a receiver with multiple antennas must have one RF chain (downconverter, LNA, filter)
per antenna. An exception is that a receiver can use a scanning combiner, which has an RF switch
that scans between antennas, and switches when the SNR goes low.
The benefits of space diversity are that no additional signal needs to be transmitted, and no
additional bandwidth is required.
Space diversity could be used at a transmitter, by changing the transmit antenna until the
receiver SNR is high enough. However, this requires some closed loop control, and so is less
common.
MIMO is a kind of space diversity and multipath diversity, that is more beneficial than simple
diversity method we cover in this lecture. We will cover it in our final lecture.
The multiple antennas don’t need to have the same gain pattern. Another method of diversity
is gain pattern diversity, although it is not mentioned in the book.

Polarization Diversity Polarization diversity is the use of two antennas with different polar-
izations. We know that reflection coefficients are different for horizontal and vertically polarized
components of the signal. Scattered and diffracted signal amplitudes and phases also are different
for opposite polarizations. Thus we can consider one polarized signal, which is the sum of the am-
plitudes and phases of many reflected, scattered, and diffracted signals, to be nearly uncorrelated
with the other polarized signal.
The advantages of polarization diversity is that the two antennas don’t need to be spaced λ/2
apart, so polarization diversity can possibly be done on a mobile device. It may be combined with
space diversity so to further reduce the correlation coefficient between the signal received at two
antennas. Polarization diversity, like space diversity, doesn’t require any additional bandwidth or
signal transmission from the transmitter.
The disadvantages are simply that there can be only two channels – vertical and horizontal
(or equivalently, right-hand and left-hand circular) polarizations. It may require two receiver RF
ECE 5325/6325 Spring 2010 74

chains (again, unless a scanning combiner is used).

Frequency Diversity Frequency diversity uses multiple transmissions on different center fre-
quencies. This doesn’t typically mean transmitting exactly the same thing on multiple different
bands (which would require multiple times more bandwidth!). Frequency division multiplexing
(FDM) or orthogonal FDM (OFDM) are the typical examples, which divide the data into N differ-
ent bands. Error correction coding is used so that some percent of errors can be corrected, so if a
certain percent of the bands experience deep fades, and all of that data is lost, the data can still be
recovered during decoding. Frequency bands in FDM or OFDM are typically somewhat correlated
– each band needs to be in frequency flat fading so that equalization does not need to be used –
but this means that bands right next to each other still have some positive fading correlation.
FH-SS is another frequency diversity example. FH-SS may experience deep fades (and inter-
ference) on some center frequencies among its hopping set, but it is unlikely to lose more than a
percentage of its data. It also uses error correction coding.
Frequency diversity methods can also be set to control which frequency bands/ channels the
transmitter uses, to remove the bands that are in deep fades. Again, this requires closed loop
control.
Advantages of frequency diversity are that only one antenna, and one RF chain, is needed. A
disadvantage is that, because some of the transmit power is used to send data in bands that are in
deep fades, the power efficiency is less compared to space diversity, in which the transmitter sends
all of its power in one channel.

Multipath diversity Multipath diversity is the capturing of multipath signals into independent
channels. In DS-SS, a rake receiver achieves multipath diversity by isolating multipath components
separately from each other based on their differing time delays. If one time delay group fades,
another time delay group may not fade. These “fingers” of the rake receiver do not require different
RF chains (an advantage compared to space diversity) and benefit most when the multipath channel
is the worst, for example, in urban areas, or in mountain canyons. The disadvantage of DS-SS is the
large frequency band required – for example, 20 MHz for 802.11b, or 1.25 MHz for IS-95 (cellular
CDMA). There is also significant computational complexity in the receiver, although standard ICs
now exist to do this computation for these common commercial devices.
The Rappaport book calls this “time diversity”, but I think it is confusing – perhaps “multipath
diversity” or even “multipath time delay diversity” are better names.

Time Diversity Time diversity is the use of a changing channel (due to motion of the TX or
RX) at different times. For example, one might send the same data at multiple different times,
ECE 5325/6325 Spring 2010 75

but this would require multiple times the transmit power, and reduce the data rate possible on
one channel. This incurs additional latency (delay). However, it is used in almost all common
commercial systems in the form of “interleaving”. Interleaving takes an incoming coded bitstream
and spreads the bits across a transmitted packet in a known pattern. An example interleaver
used by a transmitter is shown in Figure 7.17 in the Rappaport book. In the receiver, the inverse
interleaving operation is performed. This way, a burst of (multiple sequential) coded bit errors
caused by the channel are spread across the packet by the interleaver. Error correction codes are
more effective when errors are not grouped together (recall our block coding and decoding – we
assumed at most one error per 6 or 7 received coded bits). In general, coding methods correct a
few out of each group of coded bits received, but not more.
Interleaving’s only disadvantage is additional latency – you need to receive the entire block of
coded bits before they can be put in order and decoded (and then converted into an audio signal,
for example). For different applications, latency requirements are different. Voice communications
are typically the most latency-sensitive, and even cell phone voice data is interleaved.
The disadvantage is that temporal correlation can be very long for most applications, even for
vehicular communications.
Packet retransmissions (e.g., TCP) can be viewed as time diversity.

29 Diversity Combining
In the previous section, we described how we might achieve M different (nearly) independent
channels. In this section, we discuss what to do with those independent signals once we get
them. These are called combining methods. We need them for space, polarization, and multipath
diversity methods. For frequency diversity (FDM and OFDM) combining is done by the FDM or
OFDM receiver using all frequency band signals. For time diversity (interleaving) we described
the combining above. We only want one bitstream, so somehow we need to combine the channels’
signals together. Here are some options, in order of complexity:

1. Scanning Combiner : Scan among the channels, changing when the current SNR goes below
the threshold.

2. Selection Combiner : Select the maximum SNR channel’s signal and use only it.

3. Equal Gain Combiner : Co-phase the signals and then add them together.

4. Maximal Ratio Combiner : The optimal solution in terms of SNR – co-phase and weight
(multiply) each signal by the square root of its signal to noise ratio (SNR), and then add
them together.

“Co-phase the signals” means that we need to multiply signals by ejφi for some constant phase angle
φi on channel i, so that the (otherwise random) phases of the signals on the different channels line
up. If we don’t co-phase the signals before combining them, we end up with the same multipath
fading problem we’ve always had - signals sometimes add together destructively.
You should be prepared to describe any of these combining methods, and discuss its effect on
the fade margin required for a link.
ECE 5325/6325 Spring 2010 76

Figure 32: Rappaport Figure 7.11, the impact of selection combining.

ECE 5325/6325 Spring 2010 77

29.1 Selection Combining

Let’s say that we have M statistically independent channels. This independence means that one
channel’s fading does not influence, or is not correlated in any way with, another channel’s fading.
Let’s assume that each channel is Rayleigh with identical mean SNR Γ. At any given instant,
the SNR on channel i is denoted γi . Based on the Rayleigh assumption for γi , it has a CDF of:

P [γi ≤ γ] = 1 − e−γ/Γ

This means that the probability that the SNR on channel i is less than the threshold γ is given by
1 − e−γ/Γ , where again, Γ is the mean SNR for the channel. In past lectures, we showed that we
can determine a fade margin for a single channel (M = 1) based on this equation. For example,
setting the probability of being less than the threshold to 1%,

0.01 = 1 − e−γ/Γ
0.99 = e−γ/Γ
γ = Γ(− ln 0.99) = Γ(0.0101) = Γ(dB) − 19.98(dB)
(40)

Thus compared to the mean SNR on the link, we need an additional 20 dB of fade margin (this is
slightly less when we use the median SNR).
In contrast, in selection combining, we only fail to achieve the threshold SNR when all channels
are below the threshold SNR. Put in another way, if any of the channels achieve good enough SNR,
we’ll select that one, and then our SNR after the combiner will be good enough. What is the
probability all of the M channels will fail to achieve the threshold SNR γ? All M channels have to
have SNR below γ. The probability is the product of each one:
h i h i h iM
P [γi < γ, ∀i = 1, . . . , M ] = 1 − e−γ/Γ · · · 1 − e−γ/Γ = 1 − e−γ/Γ

Example: What is the required fade margin when assuming Rayleigh fading and M = 2
independent channels, for a 99% probability of being above the receiver threshold?
2
Again, set 0.01 equal, this time, to 1 − e−γ/Γ , so


0.1 = 1 − e−γ/Γ
0.9 = e−γ/Γ
γ = Γ(− ln 0.9) = Γ(0.1054) = Γ(dB) − 9.77(dB)
(41)

So the fade margin has gone down to less than 10 dB, a reduction in fade margin of 10 dB!
As M increases beyond 2, you will see diminishing returns. For example, for M = 3, the required
fade margin improves to 6.15 dB, a reduction of 3.6 dB, which isn’t as great as the reduction in
fade margin due to changing M from 1 to 2.
ECE 5325/6325 Spring 2010 78

Figure 33: Goldsmith Figure 7.5, the impact of maximal ratio combining.

29.2 Scanning Combining

Selection combining assumes we know all signal amplitudes so that we can take the maximum.
Scanning combining is a simplification which says that we only have one receiver, so we can only
know the signal to noise ratio on one channel at a time. But we can switch between them when
one channel’s SNR drops too low. We can often achieve nearly the same results using a scanning
combiner as with selection combining.

29.3 Equal Gain Combining

Here, we simply co-phase the signals and then add them together. The outage probability improves
compared to selection combining. Denoting the SNR of the summed signal as γΣ , an analytical
expression for the outage probability given Rayleigh fading is [6, p. 216],
 p 
P [γΣ < γ] = 1 − e−2γ/Γ − πγ/Γe−γ/Γ 1 − 2Q
p
2γ/Γ

where Q (·) is the tail probability of a zero-mean unit-variance Gaussian random variable, as we’ve
used before to discuss bit error probabilities of modulations.

29.4 Maximal Ratio Combining

For maximal ratio combining, we still co-phase the signals. But then, we weight the signals accord-
ing to their SNR. The intuition is that some channels are more reliable than others, so we should
“listen” to their signal more than others – just like if you hear something from multiple friends, you
probably will not weight each friend equally, because you know who is more reliable than others.
ECE 5325/6325 Spring 2010 79

The outage probability improves compared to equal gain combining. Denoting the SNR of the
summed signal as γΣ , an analytical expression for the outage probability given Rayleigh fading is
[6, p. 214],
M
−γ/Γ
X (γ/Γ)k−1
P [γΣ < γ] = 1 − e .
(k − 1)!
k=1

Lecture 21

Today: (1) MIMO

30 MIMO
Multiple-input multiple output (MIMO) is a particular type of space and/or polarization diversity
in which both the transmitter and receiver may use multiple antennas.

Figure 34: Transmit and receive space diversity schemes: (a) traditional space diversity with
receiver combining, called single input multiple output (SIMO); (b) transmit diversity, which may
use Alamouti’s scheme, called multiple input single output (MISO); (c) 2×2 multiple input multiple
output (MIMO).

30.1 Revisit Maximal Ratio Combining

We’re going to introduce MIMO by comparing it to maximal ratio combining (MRC) using a receiver
with two antennas (space diversity), as shown in Figure 34(a). Let’s assume that the transmitter
sends symbol voltage s. Assume the channel from TX to RX antenna 0 experiences total channel
power gain α20 (or voltage gain α0 ), and the channel from TX to RX antenna 1 experiences total
channel power gain α21 (or voltage gain α1 ).
We said that in MRC, the received signals r0 and r1 are multiplied by the square root of the
SNRs. Multiplying received signals by a constant doesn’t help (it amplifies the noise as much as
the signal) so it really matters to multiply them by different numbers. Here, those numbers turn
out to be α0 and α1 .
rM RC = α0 r0 + α1 r1
If the transmitted symbol voltage
√ was s, then because
√ the channel voltage gains were α0 and
α1 , we must have that r0 = Pt α0 s + n0 and r1 = Pt α1 s + n1 , where n0 , n1 are the additive
ECE 5325/6325 Spring 2010 80

noise introduced by the channel. (There would have been phase shifts as well, but remember that
co-phasing is done prior to MRC.) So,

rM RC = Pt α20 + α21 s + α0 n0 + α1 n1
p  

In the case when we had only one receive antenna, we would have received either r1 or r0 . In
comparison, the noise terms are multiplied by α0 or α1 , but the signal is multiplied by the sum of
α20 + α21 . If one αi fades, we don’t lose the entire signal s.

30.2 Alamouti code

The MIMO started gaining steam in 1998, from two different results, one from Bell Labs, where
they had built an experimental MIMO system they called V-BLAST [21], and a simple transmit
diversity scheme from S. M. Alamouti now called the Alamouti scheme [1]. The Alamouti scheme
is a simple way to achieve a performance similar to MRC using two transmit antennas, and a
single receiver, like the system shown in Figure 34(b). The advantage is that in some cases, the
transmitter is more able to have multiple antennas, while the receiver is more limited in size (for
example, cellular communications on the downlink).
Alamouti presented a simple scheme that sends two symbols simultaneously, but takes two
symbol periods to do so, and over the two transmit antennas. Denote these two symbols s0 and s1 .
The idea is, first transmit s0 out of antenna 0 and s1 out of antenna 1. At the receiver, assuming
the channels are h0 = α0 ejθ0 and h1 α1 ejθ1 , will be

r0 = s0 α0 ejθ0 + s1 α1 ejθ1 (42)

Then, during the subsequent symbol period, send −s∗1 out of antenna 0 and s∗0 out of antenna 1,
where the superscript ∗ is used to denote complex conjugate. During the second symbol period the
receiver will see
r1 = −s∗1 α0 ejθ0 + s∗0 α1 ejθ1 (43)
Note this assumes the channel was the same during the second symbol period as during the first.
The “magic” happens when we combine r0 and r1 in the following way to come up with estimates
of s0 and s1 . We form:

s̃0 = h∗0 r0 + h1 r1∗

s̃1 = h∗1 r0 − h0 r1∗

Plugging in for h0 and h1 ,

s̃0 = α0 e−jθ0 r0 + α1 ejθ1 r1∗

s̃1 = α1 e−jθ1 r0 − α0 ejθ0 r1∗

Plugging in for r0 and r1 as given in (42) and (43), respectively,

s̃0 = α0 e−jθ0 (s0 α0 ejθ0 + s1 α1 ejθ1 ) + α1 ejθ1 (−s1 α0 e−jθ0 + s0 α1 e−jθ1 )

s̃1 = α1 e−jθ1 (s0 α0 ejθ0 + s1 α1 ejθ1 ) − α0 ejθ0 (−s1 α0 e−jθ0 + s0 α1 e−jθ1 )

Simplifying,

s̃0 = α20 s0 + s1 α0 α1 ej(θ1 −θ0 ) − s1 α0 α1 ej(θ1 −θ0 ) + α21 s0

s̃1 = α21 s1 + s0 α0 α1 ej(θ0 −θ1 ) − s0 α0 α1 ej(θ0 −θ1 ) + α20 s1
ECE 5325/6325 Spring 2010 81

The middle terms cancel out in each case, so finally,

s̃0 = (α20 + α21 )s0

s̃1 = (α21 + α20 )s1

In short, in two symbol periods, we’ve managed to convey two symbols of information. Each symbol
arrives with approximately the same signal amplitude that we would have had in the maximal ratio
combining case.
Notes:

1. This is a two-by one code, that is, it works for two transmit antennas and one receive antenna.
This code has been generalized for n×m MIMO systems, and called “space-time block codes”,
by Tarokh et. al. [18]. These can send more symbols in less time – in k symbol periods, you
can send more than k symbols.

2. If you transmit out of two antennas, you would in general need twice as much power as the
receiver diversity case, which had one transmit antenna. So generally we compare the two
when using the same total transmit power, i.e., cut the power in half in the transmitter
diversity case. The performance is thus 3 dB worse than the receiver MRC diversity case.

3. The Alamouti and space-time block codes are not optimal. Space-time coding is the name
of the general area of encoding information the multiple channels. One better-performing
scheme is called space-time trellis coding. But the decoding complexity of space-time trellis
codes increases exponentially as a function of the spectral efficiency [8, p377] [18].

30.3 MIMO Channel Representation

In general for MIMO, we have multiple (Nt ) transmitters and multiple (Nr ) receivers. We refer
to the system as a (Nt , Nr ) or Nt × Nr MIMO system. Figure 34(c) shows the channels for a
(2, 2) MIMO system. For the channel between transmitter k and receiver i, we denote the “channel
voltage gain” as hi,k . This gain is a complex number, with real and imaginary parts. Recall that
the phase of a multipath component changes with distance, frequency, due to reflections, etc. The
channel power gain would be |hi,k |2 . The received voltage signal at i, just from transmitter k, is
sk hi,k , where sk is what was transmitted from antenna k.
To keep all these numbers organized, we use vectors and matrices. The transmitted signal from
antennas 1, . . . , Nt is denoted s,
s = [s1 , . . . , sNt ]T
and the channel gain matrix H is given as
 
h1,1 h2,1 ··· hNt ,1
 h1,2 h2,2 ··· hNt ,2 
H= . .. .. (44)
 
 .. .. 
. . . 
h1,Nr h2,Nr ··· hNt ,Nr

Where there are Nr rows each corresponding to the channels measured at each receiver; and Nt
columns each corresponding to the channels from each transmitter.
ECE 5325/6325 Spring 2010 82

The received signal at receiver i is a linear combination of the sk for k = 1, . . . , Nt terms plus
noise:
Nt
X
xi = hi,k sk + wi
k=1
where wi the additive noise term, and i = 1, . . . , Nr . In matrix form, we can rewrite this as:

x = Hs + w

where x = [x1 , . . . , xNr ]T is the received vector and w = [w1 , . . . , wNr ]T is the noise vector.

30.4 Capacity of MIMO Systems

We said in lecture 11 that there is a theoretical limit to the bps per Hz we can achieve on a channel.
Using multiple antennas at the TX and RX increases this theoretical limit. We said that the limit
on bandwidth efficiency is given as,
Rmax
= log2 (1 + ρ) (45)
B
where Rmax is the maximum possible bit rate which can be achieved on the channel for given signal
to noise ratio ρ and bandwidth B.
In a Nt × Nr MIMO system with channel matrix H as given in (44), with Nt ≥ Nr , the new
Shannon limit on bps per Hz is [8],
   
Rmax 1 †
= E log2 det INr + ρ HH (46)
B Nt

where H † is the complex conjugate of H (I’m copying the notation of the Haykin Moher book),
and ρ is the average signal to noise ratio. Here, we assume that each channel is Rayleigh, that
is each channel voltage gain hi,k is complex Gaussian, and all channel gains are independent from
each other. This is why we need an expected value – the matrix H is filled with random variables.
To get more intuition about the bandwidth efficiency limit, consider that the term HH † is
a Hermitian Nr × Nr matrix with eigendecomposition HH † = U ΛU † where U is the matrix of
eigenvectors of HH † and Λ is a diagonal matrix of eigenvalues λi for i = 1, . . . , Nr . In this case,
we can rewrite (46) as, "N
r  #
Rmax X λi
=E log2 1 + ρ (47)
B Nt
i=1

Compared to (45), Equation (48) is a sum of several Shannon capacities – each with effective SNR
λi
ρN t
. Recall this was the formula for Nt ≥ Nr . For Nr ≥ Nt , the formula is
Nt   
Rmax X λi
= E log2 1 + ρ (48)
B Nr
i=1

These min(Nt , Nr ) “channels” are called the “eigen-mode channels” of a MIMO system.
In summary, we have created min(Nt , Nr ) eigen-mode channels. Results have shown that the to-
tal capacity increases approximately with min(Nt , Nr ). MIMO is so attractive for current and future
communication systems because it multiplies the achievable bit rate by this factor of min(Nt , Nr ),
without requiring additional bandwidth or signal energy.
ECE 5325/6325 Spring 2010 83

References
[1] S. M. Alamouti. A simple transmitter diversity scheme for wireless communications. IEEE J.
Select. Areas Commun., 16(8):1451–1458, 1998.

[2] E. Benner and A. B. Sesay. Effects of antenna height, antenna gain, and pattern downtilting
for cellular mobile radio. IEEE Transactions on Vehicular Technology, 45(2), May 1996.

[3] M. Feuerstein, K. Blackard, T. Rappaport, S. Seidel, and H. Xia. Path loss, delay spread,
and outage models as functions of antenna height for microcellular system design. Vehicular
Technology, IEEE Transactions on, 43(3):487–498, Aug 1994.

[4] M. H. Firooz, D. Maas, J. Zhang, N. Patwari, and S. K. Kasera. Channel sounding for
the masses: Low complexity GNU 802.11b channel impulse response estimation. IEEE
Trans. Wireless Communications. (in review).

[5] G. Fors. The mobile telephone: in Bell System service, 1946-1985. Retrieved Jan. 11, 2010
from http://www.wb6nvh.com/Carphone.htm.

[6] A. Goldsmith. Wireless communications. Cambridge University Press, 2005.

[7] F. Gunnarsson, M. Johansson, A. Furuskär, M. Lundevall, A. Simonsson, C. Tidestav, and

M. Blomgren. Downtilted Base Station Antennas–A Simulation Model Proposal and Impact
on HSPA and LTE Performance. In IEEE 68th Vehicular Technology Conference, 2008. VTC
2008-Fall, pages 1–5, 2008.

[8] S. Haykin and M. Moher. Modern Wireless Communications. Pearson Prentice Hall, 2005.

[9] W. Jianhui and Y. Dongfeng. Antenna downtilt performance in urban environments. In IEEE
Military Communications Conference, 1996. MILCOM’96, Conference Proceedings, volume 3,
1996.

[10] E. Johnson. Link budgets for cellular networks. In Global Wireless Education Consortium,
2001. (presentation slides).

[11] R. Malik. Spread spectrum - secret military technology to 3G. IEEE History of Telecommu-
nications Contest, pages 1–5, 2001.

[12] C. Newport, D. Kotz, Y. Yuan, R. S. Gray, J. Liu, and C. Elliott. Experimental evaluation
of wireless simulation assumptions. SIMULATION: Transactions of The Society for Modeling
and Simulation International, 83(9):643–661, September 2007.

[13] N. Patwari. Location Estimation in Sensor Networks. PhD thesis, University of Michigan,
Ann Arbor, MI, Sept. 2005.

[14] N. Patwari, G. D. Durgin, T. S. Rappaport, and R. J. Boyle. Peer-to-peer low antenna outdoor
radio wave propagation measurements at 1.8 GHz. In IEEE Vehicular Technology Conference,
volume 1, pages 371–375, May 1999.

[15] J. G. Proakis and M. Salehi. Communication System Engineering. Prentice Hall, 2nd edition,
2002.
ECE 5325/6325 Spring 2010 84

[16] C. Shannon. A mathematical theory of communications. Bell System Technical Journal,

27:379–423 and 623–656, 1948.

[17] W. L. Stutzman and G. A. Theile. Antenna Theory and Design. John Wiley & Sons, 1981.

[18] V. Tarokh, H. Jafarkhani, and A. Calderbank. Space-time block codes from orthogonal designs.
IEEE Transactions on Information Theory, 45(5):1456–1467, 1999.

[19] J. Treichler. A panoramic view of personal wireless systems from 1946 through 2016. In
2009 Virginia Tech Symposium on Wireless Personal Communications, June 2009. (keynote
presentation).

[20] US Federal Communications Commission. Radio frequency safety: Information on human

exposure to radiofrequency fields from cellular and PCS radio transmitters. Retrieved Jan. 14,
2010 from http://www.fcc.gov/oet/rfsafety/cellpcs.html.

[21] P. Wolniansky, G. Foschini, G. Golden, and R. Valenzuela. V-BLAST: an architecture for

realizing very high data rates overthe rich-scattering wireless channel. In Signals, Systems,
and Electronics, 1998. ISSSE 98. 1998 URSI International Symposium on, pages 295–300.