Unit 3.

Multi-Path Fading
ECE-GY 6023. WIRELESS COMMUNICATIONS
PROF. SUNDEEP RANGAN

1
Fading
From the Introduction of a classic text:
There are two fundamental aspects of wireless
communication that make the problem challenging
and interesting.
…First is the phenomenon of fading …

…Second …there is significant interference …

2
Learning Objectives
Describe up and down-conversion in time- and frequency-domain
Describe the steps in the DAC and ADC including the filtering
Compute a discrete-time and continuous-time base equivalent channels from the passband
Simulate fractional delays and gains in the sampled data
Describe and simulate a deterministic multi-path wireless channel
Compute the time-varying frequency response given the path parameters
Describe a statistical model for multi-path fading
Approximately compute the coherence time and bandwidth given a channel

3
Outline
Review of Up- and Downconversion
Review of TX and RX Sampling
Doppler and Multi-Path Fading
Statistical Descriptions of Fading

4
Up- and Downconversion
Complex baseband upconversion Real passband

f f
𝑊𝑊 𝑊𝑊 downconversion
-
2 0 2 −𝑓𝑓𝑐𝑐 0 𝑓𝑓𝑐𝑐
RF communication systems:
◦ Information occurs and is processed in complex baseband
◦ Transmitted and received in real passband

Up and down-conversion: Shift center frequency of signals

Also called mixing

5
 Up and Down-Conversion in Time Domain
upconversion
Complex baseband: Real passband: 𝑢𝑢𝑝𝑝 𝑡𝑡
◦ Two real signals, 𝑢𝑢𝐼𝐼 𝑡𝑡 , 𝑢𝑢𝑄𝑄 𝑡𝑡
𝑢𝑢𝑝𝑝 𝑡𝑡 = 𝑅𝑅𝑅𝑅(𝑢𝑢 𝑡𝑡 𝑒𝑒 𝑗𝑗𝜔𝜔𝑐𝑐𝑡𝑡 )
◦ Or one complex signal:

𝑢𝑢 𝑡𝑡 = 𝑢𝑢𝐼𝐼 𝑡𝑡 + 𝑗𝑗𝑢𝑢𝑄𝑄 𝑡𝑡
downconversion

Note: downconversion needs:

𝑣𝑣 𝑡𝑡 = 2𝑢𝑢𝑝𝑝 𝑡𝑡 𝑒𝑒 −𝑗𝑗𝜔𝜔𝑐𝑐𝑡𝑡 • Multiplication by 2
𝑢𝑢 𝑡𝑡 = ℎ𝐿𝐿𝐿𝐿𝐿𝐿 𝑡𝑡 ∗ 𝑣𝑣(𝑡𝑡) • Low pass filtering

6
Mixing in Frequency Domain A
Baseband signals
𝑈𝑈(𝑓𝑓)
◦ Centered around 𝑓𝑓 = 0, complex
𝑊𝑊
◦ = single sided bandwidth
2
◦ 𝑊𝑊 = two sided bandwidth 0
𝑊𝑊
𝑊𝑊 𝑊𝑊
◦ Band-limited to 𝑓𝑓 ≤ −
2 2 2
A/2

Passband signals 𝑈𝑈(𝑓𝑓)

◦ Centered around 𝑓𝑓 = 𝑓𝑓𝑐𝑐 , real
◦ 𝑊𝑊 = bandwidth (per side or image)
𝑊𝑊 −𝑓𝑓𝑐𝑐
◦ Band-limited to 𝑓𝑓 − 𝑓𝑓𝑐𝑐 ≤ 0 𝑓𝑓𝑐𝑐
2
𝑊𝑊

7
 Discrete IQ Mixer
LO = “local oscillator” = square or sine wave at 𝑓𝑓𝑐𝑐
I1, I2 = I and Q inputs.
◦ Generally, lowpass

RF = passband output centered at 𝑓𝑓𝑐𝑐

http://www.markimicrowave.com/Mixers/IQ_Quadrature-
IF_Double-Balanced/IQ-0318.aspx

Datashe RF [GHz] LO IF [MHz] Conversi Image Amplitud Phase Isolations Isolations

et [GHz] on Loss Rejectio e Deviation L-R [dB] L-I [dB]
[dB] n [dB] Deviatio [Degrees]
n [dB]

IQ-0318 3 to 18 3 to 18 DC to 500 7 22 0.75 10 40 20

8
Baseband Equivalent Channel
𝑓𝑓𝑐𝑐 𝑓𝑓𝑐𝑐

𝑠𝑠(𝑡𝑡) 𝑠𝑠𝑝𝑝 (𝑡𝑡)

𝐻𝐻𝑝𝑝 (𝑓𝑓)
𝑦𝑦𝑝𝑝 (𝑡𝑡) 𝑦𝑦(𝑡𝑡)
= 𝑠𝑠(𝑡𝑡)
𝐻𝐻𝑏𝑏 (𝑓𝑓)
𝑦𝑦(𝑡𝑡)

Upconvert Filter Downconvert Filter

Filtering at passband equivalent to complex baseband filter

Assuming downconversion filter is ideal:
𝑊𝑊
𝐻𝐻𝑏𝑏 𝑓𝑓 = 𝐻𝐻𝑝𝑝 𝑓𝑓 + 𝑓𝑓𝑐𝑐 for 𝑓𝑓 ≤
2

◦ Simply shift 𝐻𝐻𝑝𝑝 𝑓𝑓 to the left by 𝑓𝑓𝑐𝑐 .

9
Important Special Case: Delay
Passband channel Baseband equivalent
channel

ℎ𝑝𝑝 𝑡𝑡 = 𝐴𝐴𝐴𝐴 𝑡𝑡 − 𝜏𝜏 ℎ𝑏𝑏 𝑡𝑡 = 𝐴𝐴𝑒𝑒 −2𝜋𝜋𝜋𝜋𝑓𝑓𝑐𝑐 𝜏𝜏 𝛿𝛿(𝑡𝑡 − 𝜏𝜏)

𝐴𝐴 = gain 𝐴𝐴 = gain
𝜏𝜏 = delay 𝜏𝜏 = delay
𝜃𝜃 = 2𝜋𝜋𝑓𝑓𝑐𝑐 𝜏𝜏 = phase rotation

Delay, gain in passband ⇒ delay, gain and phase rotation in baseband

Proof:
◦ Passband frequency response is: 𝐻𝐻𝑝𝑝 𝑓𝑓 = 𝐴𝐴𝐴𝐴 −2𝜋𝜋𝜋𝜋𝜋𝜋𝜋𝜋
◦ Baseband frequency response: 𝐻𝐻𝑏𝑏 𝑓𝑓 = 𝐻𝐻𝑝𝑝 𝑓𝑓 + 𝑓𝑓𝑐𝑐 = 𝐴𝐴𝐴𝐴 −2𝜋𝜋𝜋𝜋 𝑓𝑓𝑐𝑐 +𝑓𝑓 𝜏𝜏

◦ Equivalent impulse response: ℎ𝑏𝑏 𝑡𝑡 = 𝐴𝐴𝑒𝑒 −2𝜋𝜋𝜋𝜋𝑓𝑓𝑐𝑐 𝜏𝜏 𝛿𝛿(𝑡𝑡 − 𝜏𝜏)

10
Synchronization and Delay Errors
𝑦𝑦 𝑡𝑡
𝑟𝑟 𝑡𝑡 = 𝑟𝑟 𝑡𝑡 + 𝜏𝜏̂
Delay = 𝐴𝐴𝑒𝑒 𝑗𝑗𝑗𝑗 𝑠𝑠(𝑡𝑡 − 𝜏𝜏) = 𝐴𝐴𝑒𝑒 𝑗𝑗𝑗𝑗 𝑠𝑠(𝑡𝑡 − Δ𝜏𝜏)
𝑠𝑠(𝑡𝑡) & gain Shift

𝐶𝐶 𝐴𝐴𝐴𝐴𝑒𝑒 𝑗𝑗𝑗𝑗
Delay estimate 𝜏𝜏̂
0 𝜏𝜏
Sync
Time synchronization at the receiver:
◦ Estimate the arrival time of the signal 𝜏𝜏̂
◦ Starts processing remainder of signal starting at 𝜏𝜏̂
◦ Equivalent to shifting received signal ahead in time by 𝜏𝜏:̂ 𝑦𝑦 𝑡𝑡 = 𝑟𝑟 𝑡𝑡 + 𝜏𝜏̂
◦ Remaining time error: Δ𝜏𝜏 = 𝜏𝜏 − 𝜏𝜏̂
Later, we will discuss:
◦ How to estimate 𝜏𝜏 (synchronization) and how to correct for gain and phase error (equalization)

11
Frequency Errors

𝑠𝑠(𝑡𝑡) × × 𝑟𝑟(𝑡𝑡) = 𝑠𝑠(𝑡𝑡)𝑒𝑒 𝑗𝑗((𝜔𝜔0−𝜔𝜔1)𝑡𝑡+(𝜃𝜃0−𝜃𝜃1))

LO 1 𝜔𝜔1 , 𝜃𝜃1 LO 2 𝜔𝜔2 , 𝜃𝜃2

Oscillators at TX and RX always have some mismatch. To analyze, suppose:

◦ Upconversion: 𝑠𝑠𝑝𝑝 𝑡𝑡 = 𝑅𝑅𝑅𝑅 𝑠𝑠 𝑡𝑡 𝑒𝑒 𝑗𝑗𝜔𝜔1𝑡𝑡+𝜃𝜃1 ,
◦ Downcoversion: 𝑟𝑟 𝑡𝑡 = 𝐿𝐿𝐿𝐿𝐿𝐿( 2𝑠𝑠𝑝𝑝 𝑡𝑡 𝑒𝑒 −(𝑗𝑗𝜔𝜔2𝑡𝑡+𝜃𝜃2) )

LO error leads to time-varying gain: 𝑟𝑟 𝑡𝑡 = 𝑔𝑔 𝑡𝑡 𝑠𝑠 𝑡𝑡 , 𝑔𝑔 𝑡𝑡 = 𝑒𝑒 𝑗𝑗((𝜔𝜔0 −𝜔𝜔1 )𝑡𝑡+(𝜃𝜃0 −𝜃𝜃1 ))

◦ Frequency and phase shift

12
In-Class Problems

13
Outline
Review of Up- and Downconversion
Review of TX and RX Sampling
Doppler and Multi-Path Fading
Statistical Descriptions of Fading

14
Typical Digital Communication Path
𝑢𝑢𝛿𝛿 (𝑡𝑡) 𝑢𝑢(𝑡𝑡) 𝑟𝑟(𝑡𝑡) 𝑣𝑣(𝑡𝑡)
𝑠𝑠[𝑛𝑛] 𝑟𝑟[𝑛𝑛]

TX samples Ideal D/A 𝑝𝑝𝑡𝑡𝑡𝑡 (𝑡𝑡) ℎ𝑐𝑐𝑐𝑐𝑐𝑐𝑐 (𝑡𝑡) 𝑝𝑝𝑟𝑟𝑟𝑟 (𝑡𝑡) Ideal A/D RX samples

All modern communication systems TX and RX digital samples

Transmitter: DAC + filtering with 𝑝𝑝𝑡𝑡𝑡𝑡 (𝑡𝑡). Filtering used to:
◦ Suppress out of band emissions

Receiver: Filters with 𝑝𝑝𝑟𝑟𝑟𝑟 (𝑡𝑡) then performs ADC. Filtering plays two roles:
◦ Reduces noise
◦ Remove out-of-band signals before ADC. (i.e. Anti-aliasing)

Filter design discussed in digital communications class

15
Review of DTFT
Given discrete-time sequence 𝑠𝑠[𝑛𝑛]
◦ Real or complex

Discrete-time Fourier Transform: 𝑆𝑆 Ω = ∑𝑛𝑛 𝑠𝑠 𝑛𝑛 𝑒𝑒 −𝑗𝑗Ω𝑛𝑛

1 𝜋𝜋
Inverse DTFT: 𝑠𝑠 𝑛𝑛 = ∫ 𝑆𝑆 Ω 𝑒𝑒 𝑗𝑗Ω𝑛𝑛 𝑑𝑑Ω
2𝜋𝜋 −𝜋𝜋

Note 𝑆𝑆 Ω is always a 2𝜋𝜋 periodic signal

◦ Can take integral for inverse DTFT on any period of 2𝜋𝜋

Ω is the discrete frequency. Units is radians per sample.

For finite length signals and finite number of Ω, can be computed via FFT
Review in your signals and systems class

16
Common DTFT Pairs
See Wikipedia

17
Discrete-Time Systems
𝑠𝑠[𝑛𝑛] 𝐻𝐻(Ω) 𝑟𝑟[𝑛𝑛]

Consider discrete-time LTI system

Time-domain: Characterized by impulse response ℎ 𝑛𝑛

𝑟𝑟 𝑛𝑛 = ℎ 𝑛𝑛 ∗ 𝑠𝑠 𝑛𝑛 = � ℎ 𝑘𝑘 𝑠𝑠 𝑛𝑛 − 𝑘𝑘
𝑘𝑘
Frequency-domain: Characterized by frequency response 𝐻𝐻 Ω
𝑅𝑅 Ω = 𝐻𝐻 Ω 𝑆𝑆 Ω

1 𝜋𝜋
◦ 𝑅𝑅 Ω = ∑𝑟𝑟 𝑛𝑛 𝑒𝑒 −𝑗𝑗Ω𝑛𝑛 , 𝑟𝑟 𝑛𝑛 = ∫ 𝑅𝑅 Ω 𝑒𝑒 𝑗𝑗Ω𝑛𝑛 𝑑𝑑Ω
2𝜋𝜋 −𝜋𝜋

18
 DT Equivalent Channel
𝑠𝑠[𝑛𝑛] 𝑟𝑟[𝑛𝑛] 𝑠𝑠[𝑛𝑛] 𝑟𝑟[𝑛𝑛]
DAC 𝑝𝑝𝑡𝑡𝑡𝑡 (𝑡𝑡) ℎ𝑐𝑐𝑐𝑐𝑐𝑐𝑐 (𝑡𝑡) 𝑝𝑝𝑟𝑟𝑟𝑟 (𝑡𝑡) ADC = 𝐻𝐻(Ω)

Discrete-time baseband equivalent channel:

◦ Describes equivalent mapping from 𝑠𝑠 𝑛𝑛 to 𝑟𝑟[𝑛𝑛]
◦ Includes effects of TX and RX filtering and continuous-time baseband channel

Band-limited filters:
1
◦ Suppose one of 𝑃𝑃𝑟𝑟𝑟𝑟 , 𝑃𝑃𝑡𝑡𝑡𝑡 is bandlimited to 𝑓𝑓 < (no out-of-band emissions or aliasing)
2𝑇𝑇
◦ Then, discrete-time equivalent channel reduces to:

1 Ω Ω Ω
𝐻𝐻 Ω = 𝑃𝑃𝑟𝑟𝑟𝑟 𝑃𝑃 𝐻𝐻 for Ω < 𝜋𝜋
𝑇𝑇 2𝜋𝜋𝜋𝜋 𝑡𝑡𝑡𝑡 2𝜋𝜋𝜋𝜋 𝑐𝑐𝑐𝑐𝑐𝑐𝑐 2𝜋𝜋𝜋𝜋

19
 Ideal Filtering
𝑠𝑠[𝑛𝑛] 𝑟𝑟[𝑛𝑛] 𝑠𝑠[𝑛𝑛] 𝑟𝑟[𝑛𝑛]
DAC 𝑝𝑝𝑡𝑡𝑡𝑡 (𝑡𝑡) ℎ𝑐𝑐𝑐𝑐𝑐𝑐𝑐 (𝑡𝑡) 𝑝𝑝𝑟𝑟𝑟𝑟 (𝑡𝑡) ADC = 𝐻𝐻(Ω)

1
Suppose sample rate 𝑓𝑓𝑠𝑠 = 𝑃𝑃𝑡𝑡𝑡𝑡 𝑓𝑓 = 𝑃𝑃𝑟𝑟𝑥𝑥 𝑓𝑓
𝑇𝑇

“Ideal” TX and RX filter : 𝑇𝑇

1 𝑡𝑡
◦ 𝑝𝑝𝑡𝑡𝑡𝑡 𝑡𝑡 = 𝑝𝑝𝑟𝑟𝑥𝑥 𝑡𝑡 = Sinc
𝑇𝑇 𝑇𝑇
◦ In frequency domain: 𝑃𝑃𝑟𝑟𝑟𝑟 𝑓𝑓 = 𝑃𝑃𝑡𝑡𝑡𝑡 𝑓𝑓 = 𝑇𝑇Rect 𝑓𝑓𝑓𝑓
◦ Also called “brick wall” filter 𝑓𝑓

Most practical filters match this well 𝑓𝑓𝑠𝑠 𝑓𝑓𝑠𝑠

◦ Up to gain and delay −
2 2

20
 Ideal Filtering
𝑠𝑠[𝑛𝑛] 𝑟𝑟[𝑛𝑛] 𝑠𝑠[𝑛𝑛] 𝑟𝑟[𝑛𝑛]
DAC 𝑝𝑝𝑡𝑡𝑡𝑡 (𝑡𝑡) ℎ𝑐𝑐𝑐𝑐𝑐𝑐𝑐 (𝑡𝑡) 𝑝𝑝𝑟𝑟𝑟𝑟 (𝑡𝑡) ADC = 𝐻𝐻(Ω)

Assume TX and RX filters are ideal

Theorem: DT equivalent channel is the re-scaled continuous-time channel

Ω
𝐻𝐻 Ω = 𝐻𝐻𝑐𝑐𝑐𝑐𝑐𝑐𝑐
2𝜋𝜋𝜋𝜋

◦ Frequency 𝑓𝑓 mapped to Ω = 2𝜋𝜋𝜋𝜋𝜋𝜋

21
 Special Case: Delay
Passband Continuous-Time Baseband Discrete-Time Baseband
Impulse response ℎ𝑝𝑝 𝑡𝑡 = 𝐴𝐴𝐴𝐴 𝑡𝑡 − 𝜏𝜏 ℎ𝑐𝑐𝑐𝑐𝑐𝑐𝑐 𝑡𝑡 = 𝐴𝐴𝑒𝑒 −𝑗𝑗𝜔𝜔𝑐𝑐𝜏𝜏 𝛿𝛿 𝑡𝑡 − 𝜏𝜏 𝜏𝜏𝑛𝑛
ℎ 𝑛𝑛 = 𝐴𝐴𝑒𝑒 −𝑗𝑗𝜔𝜔𝑐𝑐𝜏𝜏 𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠
𝑇𝑇

Frequency response 𝐻𝐻𝑝𝑝 𝑓𝑓 = 𝐴𝐴𝑒𝑒 −𝑗𝑗𝑗𝑗𝑗𝑗𝑗𝑗𝑗 𝐻𝐻𝑐𝑐𝑐𝑐𝑐𝑐𝑐 𝑓𝑓 = 𝐴𝐴𝑒𝑒 −𝑗𝑗𝜔𝜔𝑐𝑐𝜏𝜏 𝑒𝑒 −𝑗𝑗𝑗𝑗𝑗𝑗𝑗𝑗𝑗 𝐻𝐻 Ω = 𝐴𝐴𝑒𝑒 −𝑗𝑗𝜔𝜔𝑐𝑐𝜏𝜏 𝑒𝑒 −𝑗𝑗Ω𝜏𝜏/𝑇𝑇

Suppose passband has a gain and delay.

Then discrete-time frequency-domain: gain and linear phase rotation over frequency
◦ Rotates 2𝜋𝜋 𝜏𝜏/𝑇𝑇 radians every period

In discrete-time time-domain: gain, constant phase rotation and sinc filter with delay

22
 Sinc Filter with Integer Delays
Suppose we have ideal filtering and passband has delay and gain
𝜏𝜏𝑛𝑛
From previous slide, 𝑟𝑟 𝑛𝑛 = ℎ 𝑛𝑛 ∗ 𝑠𝑠 𝑛𝑛 , ℎ 𝑛𝑛 = 𝐴𝐴𝑒𝑒 −𝑗𝑗𝜔𝜔𝑐𝑐 𝜏𝜏 𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠
𝑇𝑇 𝐴𝐴
Special case 1: No delay 𝜏𝜏 = 0: ℎ[𝑛𝑛]
◦ ℎ 𝑛𝑛 = 𝐴𝐴𝐴𝐴[𝑛𝑛] ⇒ 𝑟𝑟 𝑛𝑛 = 𝐴𝐴𝐴𝐴 𝑛𝑛
◦ Baseband channel introduces only gain
0
Special case 2: Integer delays 𝜏𝜏 = 𝑘𝑘𝑘𝑘:
◦ ℎ 𝑛𝑛 = 𝐴𝐴𝐴𝐴[𝑛𝑛 − 𝑘𝑘] ⇒ 𝑟𝑟 𝑛𝑛 = 𝐴𝐴𝐴𝐴 𝑛𝑛 − 𝑘𝑘 𝐴𝐴
◦ Baseband channel introduces gain and integer shift ℎ[𝑛𝑛]

Ex: Suppose sample rate is 20 MHz and signal is delayed by 400 ns.
◦ Integer delay in discrete-time signal is 20 0.4 = 8 samples 𝑘𝑘

23
 Sinc Pulses with Fractional Delay
𝜏𝜏𝜏𝜏
ℎ 𝑛𝑛 = 𝐴𝐴𝑒𝑒 −𝑗𝑗𝜔𝜔𝑐𝑐 𝜏𝜏 𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠
𝑇𝑇
𝜏𝜏 = 5
Causes blurring over multiple samples
Inter-symbol interference
𝜏𝜏 = 13
Will need equalization to correct
◦ More on this later

𝜏𝜏 = 13.2

𝜏𝜏 = 13.5

24
Simulating Fractional Delays in MATLAB
Code on previous slide was create with DSP toolbox

Creates T x D matrix
Row 𝑖𝑖 is delayed by 𝜏𝜏(𝑖𝑖)

25
In-Class Problem:
Fractional Delays on Constellations

26
Outline
Review of Up- and Downconversion
Review of TX and RX Sampling
Doppler and Multi-Path Fading
Statistical Descriptions of Fading

27
Receiver with Local Motion
𝜏𝜏𝑡𝑡 = (𝑑𝑑0 + 𝑣𝑣𝑣𝑣 cos 𝜃𝜃)⁄𝑐𝑐

Mobile velocity 𝑣𝑣
𝜃𝜃
𝜏𝜏0 = 𝑑𝑑0 ⁄𝑐𝑐

With the RX in motion, the propagation delay changes with time.

In complex baseband signal: 𝑡𝑡𝑣𝑣
𝑗𝑗𝑗𝑗𝑗 − cos 𝜃𝜃−𝑑𝑑0 𝑓𝑓𝑐𝑐
𝑟𝑟 𝑡𝑡 = 𝛼𝛼𝑒𝑒 −𝑗𝑗𝑗𝑗𝑐𝑐𝜏𝜏 𝑡𝑡 𝑢𝑢 𝑡𝑡 − 𝜏𝜏𝑡𝑡 = 𝛼𝛼𝑒𝑒 𝑐𝑐 𝑢𝑢(𝑡𝑡 − 𝜏𝜏𝑡𝑡 )
Local motion assumption: 𝑢𝑢 𝑡𝑡 − 𝜏𝜏𝑡𝑡 ≈ 𝑢𝑢(𝑡𝑡 − 𝜏𝜏0 ) for 𝑡𝑡 small
◦ Effect of the change in propagation delay is only in the complex exponential
𝑡𝑡𝑡𝑡
𝑗𝑗𝑗𝑗𝑗 − cos 𝜃𝜃−𝑑𝑑0 𝑓𝑓𝑐𝑐
Then: 𝑟𝑟 𝑡𝑡 = 𝛼𝛼𝑒𝑒 𝑐𝑐 𝑢𝑢(𝑡𝑡 − 𝜏𝜏0 )

28
Doppler Shift
𝜃𝜃 Mobile velocity 𝑣𝑣

Single path with local motion:

𝑟𝑟 𝑡𝑡 = 𝑔𝑔0 𝑒𝑒 𝑗𝑗𝑗𝑗𝑗𝑓𝑓𝑑𝑑 𝑡𝑡 𝑢𝑢(𝑡𝑡 − 𝜏𝜏0 )
◦
Complex gain: 𝑔𝑔0 = 𝛼𝛼𝑒𝑒 −2𝜋𝜋𝜋𝜋𝜏𝜏0 𝑓𝑓𝑐𝑐
𝑣𝑣𝑓𝑓𝑐𝑐 cos 𝜃𝜃
◦ Doppler shift: 𝑓𝑓𝑑𝑑 = −
𝑐𝑐
◦ Delay: 𝜏𝜏0
For a single path: Local motion causes a phase rotation, but no change in amplitude

29
 Example: Computing Doppler Shift
𝑣𝑣=100 km/h
𝜃𝜃

Suppose: carrier frequency is 𝑓𝑓𝑐𝑐 =2.1GHz Car moves towards a base station at 100 km/h.
What is the Doppler shift?
8
Answer: 𝑣𝑣=100km/h= 27.7 m/s, c= 3 10 m/s, 𝜃𝜃 = 180:

𝑣𝑣𝑓𝑓𝑐𝑐 cos 𝜃𝜃 27.7 2.1 10 9 (−1)

Δ𝑓𝑓 = − =− 8
≈ 194 Hz
𝑐𝑐 3 10
If the angle away from BS at 𝜃𝜃 = 45:
𝑣𝑣𝑓𝑓𝑐𝑐 cos 𝜃𝜃 27.7 2.1 10 9 cos(45)
Δ𝑓𝑓 = = 8
≈ −138 Hz
𝑐𝑐 3 10

30
Multi-Path Models
Most channel consists of many paths
◦ Direct paths
◦ Reflections, transmissions, diffraction, …
◦ LOS and NLOS paths (𝛼𝛼ℓ , 𝜏𝜏ℓ , 𝜙𝜙ℓ )
Wideband time-domain baseband model:
𝐿𝐿

𝑟𝑟 𝑡𝑡 = � 𝑔𝑔ℓ 𝑒𝑒 𝑗𝑗𝜔𝜔ℓ 𝑡𝑡 𝑥𝑥(𝑡𝑡 − 𝜏𝜏ℓ )

ℓ=1

◦ 𝑔𝑔ℓ : Complex path gain

2𝜋𝜋𝜋𝜋𝑓𝑓𝑐𝑐
◦ 𝜔𝜔ℓ = − cos 𝜃𝜃ℓ : Doppler shift of path
𝑐𝑐
◦ 𝜏𝜏ℓ : Delay of the path

31
 Time-Varying Frequency Response
Multipath channel: 𝑦𝑦 𝑡𝑡 = ∑𝐿𝐿ℓ=1 𝑔𝑔ℓ 𝑒𝑒 𝑗𝑗𝜔𝜔ℓ 𝑡𝑡 𝑥𝑥(𝑡𝑡 − 𝜏𝜏ℓ )
Consider exponential input: 𝑥𝑥 𝑡𝑡 = 𝑒𝑒 𝑗𝑗𝑗𝑗𝑗𝑗
Output is: 𝑦𝑦 𝑡𝑡 = 𝐻𝐻 𝑡𝑡, 𝜔𝜔 𝑥𝑥 𝑡𝑡
Time-varying frequency response
𝐿𝐿
𝐻𝐻 𝑡𝑡, 𝜔𝜔 = � 𝑔𝑔ℓ 𝑒𝑒 𝑗𝑗(𝜔𝜔ℓ𝑡𝑡−𝜔𝜔𝜏𝜏ℓ )
ℓ=1

May also write: 𝐻𝐻 𝑡𝑡, 𝑓𝑓 = 𝐻𝐻(𝑡𝑡, 2𝜋𝜋𝜋𝜋)

32
Example with Two Paths
𝜏𝜏1 , 𝜃𝜃1

𝜏𝜏2 , 𝜃𝜃2

To simplify understanding, consider two path model

𝜔𝜔𝑖𝑖 = −2 𝜋𝜋𝑓𝑓𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑 cos 𝜃𝜃𝑖𝑖
𝑗𝑗𝜔𝜔1 𝑡𝑡 𝑗𝑗𝜔𝜔2 𝑡𝑡
𝑟𝑟 𝑡𝑡 = ℎ1 𝑒𝑒 𝑢𝑢 𝑡𝑡 − 𝜏𝜏1 + ℎ2 𝑒𝑒 𝑢𝑢(𝑡𝑡 − 𝜏𝜏2 )
Time-varying response:
𝐻𝐻 𝑡𝑡, 𝜔𝜔 = ℎ1 𝑒𝑒 𝑗𝑗(𝜔𝜔1 𝑡𝑡−𝜔𝜔𝜔𝜔1 ) + ℎ2 𝑒𝑒 𝑗𝑗(𝜔𝜔2 𝑡𝑡−𝜔𝜔𝜔𝜔2 )
Power gain:
2 2
𝑃𝑃 𝑡𝑡, 𝜔𝜔 = 𝐻𝐻 𝑡𝑡, 𝜔𝜔 = ℎ1 𝑒𝑒 𝑗𝑗(𝜔𝜔1 𝑡𝑡−𝜔𝜔𝜔𝜔1 ) + ℎ2 𝑒𝑒 𝑗𝑗(𝜔𝜔2 𝑡𝑡−𝜔𝜔𝜔𝜔2 )

33
Variation in Time 𝑃𝑃 𝑡𝑡, 𝜔𝜔0 = ℎ1 𝑒𝑒 𝑗𝑗(𝜔𝜔1𝑡𝑡+𝜙𝜙1) + ℎ2 𝑒𝑒 𝑗𝑗(𝜔𝜔2𝑡𝑡+𝜙𝜙2)
2

 Fixed frequency 𝜔𝜔0

 Look at time variations 𝑃𝑃(𝑡𝑡, 𝜔𝜔0 )
 Rate of variation depends on Doppler spread:
Δ𝑓𝑓 = 𝑓𝑓𝑚𝑚𝑚𝑚𝑚𝑚 cos 𝜃𝜃1 − cos 𝜃𝜃2
 Size of variation depends on spread of gains:
 Avg: ℎ1 2 + ℎ2 2
 Min: ℎ1 − ℎ2 2 : Destructive interference
2
 Max: ℎ1 + ℎ2 : : Constructive interference
 With equal path gains, there are nulls
Plot shows 𝑓𝑓max =10 Hz,
𝜃𝜃1 = 0, 𝜃𝜃2 = 180,
ℎ2 = 10−0.05Δ ℎ2 , ℎ1 2 + ℎ2 2
=1

34
Variation in Frequency 𝑃𝑃 𝑡𝑡0 , 𝜔𝜔 = ℎ1 𝑒𝑒 𝑗𝑗(𝜔𝜔𝜏𝜏1+𝜙𝜙1) + ℎ2 𝑒𝑒 𝑗𝑗(𝜔𝜔𝜏𝜏2+𝜙𝜙2)
2

 Fixed frequency 𝑡𝑡0

 Look at time variations 𝑃𝑃(𝑡𝑡, 𝜔𝜔0 )
 Period of variation depends on delay spread:
1
Δ𝑓𝑓 =
𝜏𝜏2 − 𝜏𝜏1
 Size of variation depends on spread of gains:
 Avg: ℎ1 2 + ℎ2 2
 Min: ℎ1 − ℎ2 2
2
 Max: ℎ1 + ℎ2
Plot shows
𝜏𝜏1 = 0, 𝜏𝜏2 = 200 ns,
ℎ2 = 10−0.05Δ ℎ2 , ℎ1 2
+ ℎ2 2
=1

35
Fading
 Over time and frequency, paths can either
 Constructively interfere ⇒ Peaks
 Destructively interfere ⇒ Nulls

 Process is called fading

 Intermittent channel quality

 One of the most significant challenges in wireless

 Later, we will discuss how to overcome fading

36
 Narrowband Approximation
𝑔𝑔1
Multi-path channel: 𝑟𝑟 𝑡𝑡 = ∑𝐿𝐿ℓ=1 𝑔𝑔ℓ 𝑒𝑒 𝑗𝑗𝜔𝜔ℓ 𝑡𝑡 𝑥𝑥(𝑡𝑡 − 𝜏𝜏ℓ )
𝑔𝑔0
Define delay spread: 𝛿𝛿 = max |𝜏𝜏ℓ − 𝜏𝜏0 | 𝑔𝑔2
ℓ
◦ Max path difference in seconds
1 𝜏𝜏0 𝜏𝜏1 𝜏𝜏2
Narrowband approximation: 𝑋𝑋(𝑓𝑓) is band-limited to 𝑓𝑓 − 𝑓𝑓 0 ≪ then
2𝛿𝛿
𝛿𝛿
0
𝑟𝑟 𝑡𝑡 ≈ 𝐻𝐻(𝑡𝑡, 𝑓𝑓 )𝑥𝑥 𝑡𝑡 − 𝜏𝜏0
◦ Proof below
1 𝑔𝑔̅ 𝑡𝑡
◦ Coherence bandwidth =
2𝛿𝛿
0 𝜏𝜏
Effective single path gain: 𝑔𝑔̅ 𝑡𝑡 = 𝐻𝐻 𝑡𝑡, 𝑓𝑓 0 = ∑𝐿𝐿ℓ=1 𝑔𝑔ℓ 𝑒𝑒 𝑗𝑗𝜔𝜔ℓ 𝑡𝑡−𝑗𝑗𝜔𝜔 ℓ :
◦ Channel appears as a single path channel with time-varying gain
◦ Channel gain 𝑔𝑔̅ 𝑡𝑡 is band-limited to max Doppler max 𝜔𝜔ℓ 𝜏𝜏0
ℓ

37
Example: 3GPP Cluster Delay Line Model
3GPP has several deterministic multi-path models
Called Cluster-Delay Line (CDL)
Can be downloaded in MATLAB 5G Toolbox
◦ Gives the gain, delay and angles of each path
◦ This ex: CDL-C with 24 paths
Path delay profile Path AoA azimith profile

38
Computing the Doppler of Each Path
Computing the Doppler shift of each path
◦ Suppose that RX has velocity vector 𝒗𝒗 = (𝑣𝑣𝑥𝑥 , 𝑣𝑣𝑦𝑦 , 𝑣𝑣𝑧𝑧 ) Direction of
𝑓𝑓 𝑓𝑓𝑐𝑐 arrival vector 𝒖𝒖ℓ
◦ Doppler shift of path ℓ is: 𝑓𝑓ℓ = − 𝑐𝑐 𝒗𝒗𝑇𝑇 𝒖𝒖ℓ = − 𝒗𝒗 cos 𝜃𝜃ℓ
𝑐𝑐 𝑐𝑐
𝜃𝜃ℓ
In this simulation: 𝑣𝑣 = 30 m/s in x-axis RX velocity 𝒗𝒗

39
Computing the Narrowband Response
0 𝜏𝜏
Narrowband response 𝐻𝐻 𝑡𝑡, 𝑓𝑓 0 = ∑𝐿𝐿ℓ=1 𝑔𝑔ℓ 𝑒𝑒 𝑗𝑗𝜔𝜔ℓ 𝑡𝑡−𝑗𝑗𝜔𝜔 ℓ

Plot at 𝑓𝑓 0 = 0 and 1 MHz

Max Doppler f𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑 ≈ 200 Hz
1
◦ See fast variation on order = 5 ms
𝑓𝑓𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑

Can see deep fades

40
 Narrowband Approximation Proof
 Want to show: If 𝑋𝑋(𝑓𝑓) band-limited to 𝑓𝑓 − 𝑓𝑓 0 ≪
1 𝑔𝑔1
then 𝑟𝑟 𝑡𝑡 ≈ 𝐻𝐻(𝑡𝑡, 𝑓𝑓 0 )𝑥𝑥 𝑡𝑡 − 𝜏𝜏0
2𝛿𝛿 𝑔𝑔0
Prove this for 𝑓𝑓 0 = 0. Other frequencies are similar. 𝑔𝑔2
1
Thus, 𝑋𝑋(𝑓𝑓) is bandlimited to 𝑓𝑓 ≤ .
2𝛿𝛿
𝜏𝜏0 𝜏𝜏1 𝜏𝜏2
Therefore,𝑥𝑥 𝑠𝑠1 ≈ 𝑥𝑥 𝑠𝑠2 𝑓𝑓𝑓𝑓𝑓𝑓 𝑠𝑠1 − 𝑠𝑠2 ≤ 𝛿𝛿
In particular 𝑥𝑥 𝑡𝑡 − 𝜏𝜏𝑖𝑖 ≈ 𝑥𝑥(𝑡𝑡 − 𝜏𝜏0 ) since 𝜏𝜏 − 𝜏𝜏0 ≤ 𝛿𝛿 𝛿𝛿
Hence:
𝐿𝐿 𝐿𝐿 𝑔𝑔̅ 𝑡𝑡
𝑟𝑟 𝑡𝑡 = � 𝑔𝑔ℓ 𝑒𝑒 𝑗𝑗𝜔𝜔ℓ 𝑡𝑡) 𝑥𝑥(𝑡𝑡 − 𝜏𝜏ℓ ) ≈ � 𝑔𝑔ℓ 𝑒𝑒 𝑗𝑗𝜔𝜔ℓ 𝑡𝑡 𝑥𝑥 𝑡𝑡 − 𝜏𝜏0
ℓ=1 ℓ=1
= 𝐻𝐻(𝑡𝑡, 0)𝑥𝑥 𝑡𝑡 − 𝜏𝜏0
𝜏𝜏0

41
Examples: When is Narrowband Valid?
Bluetooth
◦ Bluetooth hops over channels of bandwidth 𝐵𝐵 =1 MHz each.
◦ Indoor delay spread typically ≪ 50 ns
1 1
◦ Coherence bandwidth = = 20 MHz ≫ B
𝛿𝛿 50
◦ Narrowband approximation valid

LTE outdoor cellular system

◦ A typical channel is 𝐵𝐵 =20 MHz
◦ Outdoor delay spread 𝛿𝛿 ≈ 1 𝜇𝜇𝜇𝜇
1 1
◦ Coherence bandwidth = = 1 MHz ≪ B
𝛿𝛿 1
◦ Narrowband approximation not valid

42
 OFDM Time-Frequency Grid
Subcarrier OFDM modulation: Widely-used method
spacing ◦ 4G and 5G cellular systems
1
Δ𝑓𝑓 =
𝑇𝑇
◦ Many 802.11 standards
Frequency

Divide channel into sub-carriers and OFDM symbols

◦ Resource element: One time-frequency point

Data is transmitted is an array: 𝑋𝑋 𝑛𝑛, 𝑘𝑘

◦ 𝑘𝑘 = OFDM symbol index
◦ 𝑛𝑛 = subcarrier index
◦ One complex value per RE.
OFDM symbol
◦ Called a modulation symbol
𝑇𝑇𝑠𝑠𝑠𝑠𝑠𝑠
See digital communication class
Time ◦ We will also review again when we discuss equalization

43
OFDM Channel with Fading
OFDM channel acts as multiplication:
Under normal operation (delay spread is contained in CP):

𝑌𝑌 𝑘𝑘, 𝑛𝑛 = 𝐻𝐻 𝑘𝑘, 𝑛𝑛 𝑋𝑋[𝑘𝑘, 𝑛𝑛]

RX symbols Channel TX symbols

OFDM channel gains can be computed from the multi-path components
𝐿𝐿

𝐻𝐻 𝑘𝑘, 𝑛𝑛 = � 𝐸𝐸ℓ 𝑒𝑒 −2𝜋𝜋𝑗𝑗 (𝑇𝑇𝑇𝑇𝑓𝑓ℓ +𝑆𝑆𝑆𝑆𝜏𝜏ℓ +𝜙𝜙ℓ )

ℓ=1

◦ 𝑇𝑇 = OFDM symbol time, 𝑆𝑆 = sub-carrier spacing

◦ For each path: 𝑓𝑓ℓ =Doppler shift, 𝜏𝜏ℓ =Delay, 𝜙𝜙ℓ = phase of path, 𝐸𝐸ℓ = path received energy

44
Summary
Single path with no motion:
◦ Delay and constant phase shift

Local motion in single path causes Doppler

◦ A time-varying phase rotation
◦ But channel gain is constant

Multiple paths causes fading

◦ Constructive and destructive interference of paths
◦ Variation in gain over time

Described by a time-varying frequency response 𝐻𝐻 𝑡𝑡, 𝑓𝑓

◦ Variations is time due to Doppler spread
◦ Variations in frequency due to delay spread

45
In-Class Exercise: OFDM Channel Response

46
Outline
Review of Up- and Downconversion
Review of TX and RX Sampling
Doppler and Multi-Path Fading
Statistical Descriptions of Fading

47
Statistical Model
Fading depends on the multipath distribution
Multipath is site-specific
Statistical model:
◦ Describes a probability distribution of channels
◦ Trained on an ensemble of channels in some environment

Used in evaluation of communication system

Example questions:
◦ How well does a system do on average?
◦ What is the probability that I will obtain sufficient coverage?

48
Random Path Statistical Model
RX signal has many random, independent paths
𝑔𝑔ℓ , 𝜏𝜏ℓ , 𝜃𝜃ℓ
Time-varying frequency response:
1 𝐿𝐿
ℎ 𝑡𝑡, 𝑓𝑓 = � 𝑔𝑔ℓ 𝑒𝑒 2𝜋𝜋𝜋𝜋(𝑡𝑡𝑓𝑓𝑑𝑑 cos 𝜃𝜃ℓ +𝑓𝑓𝜏𝜏ℓ ) Direction of motion
𝐿𝐿 ℓ=1
◦ Assume (𝑔𝑔ℓ , 𝜏𝜏ℓ , 𝜃𝜃ℓ ) i.i.d.
2
◦ Path gains: 𝑔𝑔ℓ are zero mean and E 𝑔𝑔ℓ = 𝐺𝐺0

As 𝐿𝐿 → ∞ , ℎ(𝑡𝑡, 𝑓𝑓) is a complex Gaussian, ℎ 𝑡𝑡, 𝑓𝑓 ~𝐶𝐶𝐶𝐶(0, 𝐺𝐺0 )

◦ Follows from Central Limit Theorem
◦ Independent real and imaginary components
◦ 𝐺𝐺0 : Average power gain
◦ Variance 𝐺𝐺0 ⁄2 for real and imaginary components

49
Rayleigh Distribution
Rayleigh fading: Channel response is ℎ 𝑡𝑡, 𝑓𝑓 ~𝐶𝐶𝐶𝐶 0, 𝐺𝐺0
Probability distribution
Let 𝑅𝑅 = |ℎ| magnitude
◦ Represents amplitude gain

Has Rayleigh distribution:

2𝑟𝑟 −𝑟𝑟 2 ⁄𝑃𝑃
◦ PDF: 𝑝𝑝 𝑟𝑟 = 𝑒𝑒
𝑃𝑃
2 ⁄𝑃𝑃
◦ CDF: P 𝑅𝑅 ≤ 𝑟𝑟 = 1 − 𝑒𝑒 −𝑟𝑟
𝐺𝐺0 𝜋𝜋
◦ Mean: 𝐸𝐸 𝑅𝑅 =
2
◦ Second moment: 𝐸𝐸𝑅𝑅 = 𝐺𝐺0 2 https://en.wikipedia.org/wiki/Rayleigh_distribution

50
Exponential Distribution
Consider Rayleigh fading complex gain ℎ~𝐶𝐶𝐶𝐶(0, 𝐺𝐺0 )
Magnitude 𝑅𝑅 = |ℎ| is Rayleigh
2 ⁄𝐺𝐺
𝑃𝑃 𝑅𝑅 ≥ 𝑟𝑟 = 𝑒𝑒 −𝑟𝑟 0

2
Instantaneous gain 𝐺𝐺 = ℎ has exponential distribution

𝑃𝑃 𝐺𝐺 ≥ 𝑔𝑔 = 𝑃𝑃 𝑅𝑅 ≥ 𝑔𝑔 = 𝑒𝑒 −𝑔𝑔⁄𝐺𝐺0

2
◦ Average gain is 𝐸𝐸 𝐺𝐺 = 𝐸𝐸 ℎ = 𝐺𝐺0
1
PDF for 𝜆𝜆 =
For channel, 𝐺𝐺 represent power gain (in linear scale) 𝐸𝐸(𝐺𝐺)

51
Example Calculation
Suppose the channel experiences Rayleigh fading.
What is probability gain will be 15 dB below the average?
◦ Called a 15 dB fade.
0
10

Answer:
◦ Gain is 15 dB below average when 𝐺𝐺 ≤ 10−0.1 15 𝐺𝐺0 10
-1

◦ From exponential distribution:

Probability
𝑃𝑃 𝐺𝐺 ≤ 𝛽𝛽𝐺𝐺0 = 1 − 𝑒𝑒 −𝛽𝛽𝐺𝐺0⁄𝐺𝐺0 = 1 − 𝑒𝑒 −𝛽𝛽 -2
10

◦ For small 𝛽𝛽, 𝑃𝑃 𝐺𝐺 ≤ 𝛽𝛽𝐺𝐺0 ≈ 𝛽𝛽

◦ For 15 dB fade, 𝛽𝛽 = 10−0.1(15) ≈ 0.032. -3
10
-10 0 10 20 30
Fade [dB]

52
 Doppler Spectra
Consider statistical model: 𝐿𝐿
1
ℎ 𝑡𝑡, 𝑓𝑓 = � 𝑔𝑔ℓ 𝑒𝑒 2𝜋𝜋𝜋𝜋(𝑡𝑡𝑓𝑓𝑑𝑑 cos 𝜃𝜃ℓ +𝑓𝑓𝜏𝜏ℓ )
𝐿𝐿 ℓ=1
2
◦ Paths are i.i.d. and 𝑔𝑔ℓ are zero mean, 𝐸𝐸 𝑔𝑔ℓ = 𝐺𝐺0
◦ Assume 𝐿𝐿 is large
For a given (𝑡𝑡, 𝑓𝑓), complex gain ℎ 𝑡𝑡, 𝑓𝑓 ~𝐶𝐶𝐶𝐶(0, 𝐺𝐺0 )
As varies (𝑡𝑡, 𝑓𝑓), ℎ 𝑡𝑡, 𝑓𝑓 is a Gaussian random process
Auto-correlation:
𝑅𝑅 𝛿𝛿𝑡𝑡, 𝛿𝛿𝛿𝛿 = 𝐸𝐸 ℎ 𝑡𝑡, 𝑓𝑓 ℎ∗ 𝑡𝑡 + 𝛿𝛿𝛿𝛿, 𝑓𝑓 + 𝛿𝛿𝛿𝛿
= 𝐺𝐺0 𝐸𝐸 𝑒𝑒 2𝜋𝜋𝜋𝜋(𝛿𝛿𝛿𝛿𝑓𝑓𝑑𝑑 cos 𝜃𝜃ℓ +𝛿𝛿𝛿𝛿𝜏𝜏ℓ )
◦ Describes how correlated the process is over time and frequency
◦ Depends on the distribution of angles 𝜃𝜃ℓ and delays 𝜏𝜏ℓ

53
 Jakes Model
At a fixed frequency 𝑓𝑓0 : Gain (dB) Angle (rads)
1 𝐿𝐿
ℎ 𝑡𝑡, 𝑓𝑓0 = � 𝑔𝑔ℓ 𝑒𝑒 2𝜋𝜋𝜋𝜋(𝑡𝑡𝑓𝑓𝑑𝑑 cos 𝜃𝜃ℓ +𝑓𝑓0 𝜏𝜏ℓ )
𝐿𝐿 ℓ=1
Jakes
1 𝐿𝐿
= � 𝑔𝑔ℓ′ 𝑒𝑒 2𝜋𝜋𝜋𝜋𝜋𝜋𝑓𝑓𝑑𝑑 cos 𝜃𝜃ℓ Angles unif [0,2𝜋𝜋]
𝐿𝐿 ℓ=1
Complex Gaussian process
Statistics depend on angular distribution
Asym Jakes
Jakes model: cos 𝜃𝜃 ∈ [0.9,1]
◦ Angles uniform from [0,2𝜋𝜋]
Asymmetric Jakes:
◦ 𝜃𝜃 ∈ [𝜃𝜃1 , 𝜃𝜃2 ] uniform Asym Jakes
Angular spread: cos 𝜃𝜃 ∈ [−0.1,0.1]
◦ Arises from diffuse reflection

54
Fading Models in MATLAB
Comm Toolbox:
◦ Efficient, general fading models

Create a comm.RayleighChannel object

Run the channel to get:
◦ Output and gain

55
 Auto-Correlation
Fix a frequency 𝑓𝑓0 and consider auto-correlation over time:

𝑅𝑅(𝛿𝛿𝛿𝛿) = 𝐸𝐸 ℎ 𝑡𝑡, 𝑓𝑓0 ℎ∗ 𝑡𝑡 + 𝛿𝛿𝛿𝛿, 𝑓𝑓0 = 𝐺𝐺0 𝐸𝐸 𝑒𝑒 2𝜋𝜋𝜋𝜋(𝛿𝛿𝛿𝛿𝑓𝑓𝑑𝑑 cos 𝜃𝜃ℓ )

◦ Expectation is over angle 𝜃𝜃ℓ
◦ Depends on distribution of 𝜃𝜃ℓ
Plot: 𝑅𝑅(𝛿𝛿𝛿𝛿) for 𝜃𝜃 uniform in −𝜃𝜃𝑚𝑚 , 𝜃𝜃𝑚𝑚
◦ Computed numerically in MATLAB (see demo)
◦ Plotted vs normalized delay 𝑓𝑓𝑑𝑑 𝛿𝛿𝛿𝛿
◦ For 𝜃𝜃𝑚𝑚 = 180∘ ⇒ Jakes spectra
1
◦ Uncorrelated at 𝛿𝛿𝛿𝛿 ≈
2𝑓𝑓𝑑𝑑

As angular distribution is smaller:

◦ Correlation is higher with delay
◦ Highly directional channel vary slower

56
Computing Auto-Correlation Numerically
Correlation in previous slide

𝑅𝑅(𝛿𝛿𝛿𝛿) = 𝐺𝐺0 𝐸𝐸 𝑒𝑒 2𝜋𝜋𝜋𝜋(𝛿𝛿𝛿𝛿𝑓𝑓𝑑𝑑 cos 𝜃𝜃ℓ )

◦ Expectation is over angle 𝜃𝜃ℓ

Generally, no analytic solution

Compute via numerical integration
◦ See demo

57
Coherence Time and Frequency
Consider time varying frequency response 𝐻𝐻 𝑡𝑡, 𝑓𝑓
Coherence time:
◦ Max interval Δt where 𝐻𝐻 𝑡𝑡, 𝑓𝑓 ≈ 𝐻𝐻 𝑡𝑡 + Δ𝑡𝑡, 𝑓𝑓
◦ How fast channel changes in time
1
◦ Related to Doppler spread≈
𝑓𝑓𝑚𝑚𝑚𝑚𝑚𝑚 −𝑓𝑓𝑚𝑚𝑚𝑚𝑚𝑚

Coherence bandwidth
◦ Max interval Δf where 𝐻𝐻 𝑡𝑡, 𝑓𝑓 ≈ 𝐻𝐻 𝑡𝑡, 𝑓𝑓 + +Δ𝑓𝑓
◦ How fast channel changes in frequency
1
◦ Related to delay spread≈
𝜏𝜏𝑚𝑚𝑚𝑚𝑚𝑚 −𝜏𝜏𝑚𝑚𝑚𝑚𝑚𝑚
Realization of a Jakes
Critical for many procedures: process with 1/fmax = 0.1 sec
◦ Channel estimation, tracking, coding, ARQ, …
◦ More on this later

58
Winner-3GPP-Spatial Cluster Model
Cluster n
N

BS array
Subpath m ∆n ,m ,AoA
Ω MS θv v
From 3GPP SCM-132
∆n ,m ,AoD
θn ,m , AoA
δn ,AoA

N
δn, AoD
Ω BS θMS

θn ,m , AoD MS array broadside MS array

θBS
MS direction
BS array broadside of travel

Paths arrive in clusters.

Clusters have subpaths (also called rays)
Each cluster has:
◦ Center angle and a statistical model for the delay and angular spread

59
Fading at Different Time Scales
Three mechanisms for path loss variations

Distance-based path loss Slower

Shadowing
Small-scale multi-path fading
Faster

From Goldsmith, “Wireless Communications”

60
Fading at Different Scales Models
Source of variation Mathematical model Typical spatial Typical temporal
coherence coherence
Small-scale fading from Rayleigh or Rician ~ 1 wavelength 15 ms
multi-path fading distribution (v=10m/s, fc=2GHz)
Large-scale fading Lognormal distribution 10 to 100 m 1 to 10 sec
from variations in
shadowing
Path loss variations Path loss exponent 100 m or larger 10 sec

Different fading processes and variations occur at much different time / space scales
Methods to combat these are different

61