Course 4: Baseband data

transmission

Data is a lot like humans: It is

born. Matures. Gets married to
other data, divorced. Gets old. One
thing that it doesn't do is die. It has
to be killed.
Arthur Miller,
American playwright and
essayist

Agenda
General model of the baseband data transmission
systems
ISI-less data transmission
Nyquist filters: raised cosine/squared root raised
cosine
Baseband transmission with controlled ISI
Evaluating the ISI level using the eye diagram
Error probabilities for the baseband transmission

Model for baseband data transmission

{an}

PAM

a(t)

GE()

s(t)

GR()

C()

Legend:

r(t)

Threshold
comparison

{ean}

n(t)

an: sequence of bits to be transmitted

t0+kT

PAM: pulse amplitude modulator

GE(), GR(): emission/reception filters (their transfer function)
s(t): signal which carry data, transmitted in the channel
C(): channel (physical environment) transfer function
n(t): additive white noise
r(t): received signal (based on it, a decision is made)
Threshold comparison: e.g. positive value leads to 1, negative to 0
ean: estimation of the received bits (ideally, identical to an)

Model for the baseband data transmission

General remarks
r (t ) = As (t 0 )
(1)
For an ideal channel:
An ideal channel attenuates the input signal (A) and
introduces a delay (tau)
Both are constant !!! (they DO NOT DEPEND on the frequency)

C () = Ae j0

( 2)

The derivate of the phase of C() is called group delay, and is

constant in this scenario
( 3)
C () = A()e j( )
In practice:
The real channels distort the input signal and they are
frequency-selective (dispersive)

Model for the baseband data transmission

The baseband chain: mathematical approach [1]

{an}

PAM

a(t)

GE()

s(t)

GR()

C()

r(t)

Threshold
comparison

{ean}

n(t)
t0+kT

The PAM gives a first signal model for the data sequence:
N 1

a(t ) = an (t nT ) (4 )
n =0

GE shapes the signal for transmission:

s (t ) = a (t ) * ge (t ) =

g e (t ) =

N 1

an ge (t nT )

(5)

n =0

1
jt
GE ()e d (6 )
2

The pulse amplitude modulator (PAM) is an imaginary block; its output

is a sequence of Dirac pulses, separated in time by the symbol time (T). Thus, for
every bit of 1, a positive pulse is issued, whereas for every bit of 0 a negative
pulse is generated. Although such pulses cannot be generated in practice, the PAM
helps to the mathematical modeling of the transmission system.
ge(t) is a pulse shaping filter (at the transmission side). This filter shapes the signal,
such a way that every digital information symbol is represented by a certain
waveform g(t).

Model for the baseband data transmission

The baseband chain: mathematical approach [2]

{an}

PAM

a(t)

GE()

s(t)

GR()

C()

r(t)

Threshold
comparison

{ean}

n(t)
t0+kT

If g(t) stands for the equivalent impulse response of the chain

emission filter channel reception filter, then:
N 1

r (t ) = an g (t nT ) + n(t ) (7 )
n =0

The received signal is a weighted sum of g(t) waveforms,

affected by additive noise

The received signal (eq. 7) is a weighted sum of g(t) shapes; g(t) is the equivalent
impulse response of the assembly transmission filter physical channel reception
filter. According to eq. 7, the received signal is a weighted sum of g(t) waveforms,
perturbed by the additive noise n(t).

Model for the baseband data transmission

The baseband chain: mathematical approach [3]

In order to identify the k-th symbol, the corresponding k-th
sample is collected by the receiver:
N 1

rk = r( t0 + kT ) = an g( t0 + ( k n )T ) + n( t0 + kT ) ( 8 )
n =0

Re-written:
N 1

rk = r (t0 + kT ) = an g k n + nk
n =0

(9 )

The k-th received sample depends on all the N transmitted

symbols and on the sampled values of g(t) and n(t)

The received analog signal r(t) (but which conveys digital data) must be sampled,
and a decision block will make an estimation of the transmitted bits. According to
eq. (9), the kth received sample depend on all the transmitted bits an, although it
would be desirable to depend only on the current (k-th) received sample.

ISI-less data transmission

Reminder:

N 1

rk = r( t0 + kT ) = an g k n + nk
n =0

(9)

Re-writing (9) (t0=0 for simplicity):

rk = ... + ak 1g (T ) + ak g (0) + ak +1 g (T ) + ... + nk

(10)

All terms which do not depend on ak are ISI terms!!!

By correctly choosing g(t), ISI can be eliminated:

1, if n = 0
Nyquist criterion for zero ISI g (nT ) =
0, if n 0

(11)

The k-th received sample, rk, is a weighted sum of the type angk-n,
perturbed by the noise sample nk too. Ideally, rk should equal ak, and in this case
the received sample is identical to the transmitted symbol. Therefore, from eq. (10),
only the middle term is useful, while all other terms are undesired. Thus, all terns
depending on an (excepting the middle one) introduce an influence of the adjacent
transmitted symbols an on the current bit: this influence is called ISI (Inter-Symbol
Interference). Another adverse effect is the white noise, nk.
ISI can be eliminated if the left and the right terms are forced to be 0
(Caution: the noises influence still exists!!!). This rule is met by Nyquists criterion
for ISI-less data transmission (eq. 11).
Lets decode now eq. (11): the transmission is free of ISI if the
waveform g(t) exhibits regular zero crossing at all the sampling point from the
reception side (nT), excepting for the current sample (n=0). Notice that the criterion
only introduces a rule related to the sampling points: there is no constraint imposed
on g(t), excepting those particular points. Theoretically, any waveform g(t) that
meets this rule will generate no ISI, but this statement is true only if the receiver
faithfully respect the ideal sampling points.
Reminder: g(t)=ge(t)*c(t)*gr(t). Although the response c(t) only
depends on the physical characteristics of the transmission environment, the
transmitter and reception filter can be designed such a way that g(t) to meet
Nyquists criterion.

ISI-less data transmission

Nyquist theorem
In a channel which is equivalent with an ideal low-pass filter having the
cutoff frequency F, it is possible to transmit symbols with a modulation
rate equal or less to 2F symbols/sec, without ISI
The characteristics of such a channel are shown below
Transfer function

Impulse response

According to the Nyquist theorem, the best spectral efficiency for an ISI-less data
transmission (that is the highest ratio between R (rate) and W (bandwidth)) can be
achieved if every symbol is shaped as a cardinal sine, which corresponds to an
ideal LPF. In this case only, the transmission can be made at a symbol rate which is
twice the bandwidth, while preserving it free of ISI. Thats why this transmission
rate is sometimes referred to as ideal rate or Nyquist rate.

10

ISI-less data transmission

Graphical view of the Nyquist theorem

Sampling
timespulses
in a Nyquist
transmission
Nyquist
with no
ISI

1.5 1.4
1.2

11

1
0.8
0.6

0.5 0.4
0.2
0

-0.2
-0.4

-0.5

50

50

100

100

150

200

150

250

200

300

250

Symbol time = T0/2=1/2F

Symbol time = the time interval between the transmission in the channel of two
consecutive information symbols.

11

ISI-less data transmission

Ideal case: a closer look

The impulse response of the ideal low-pass filter is:
sin(t )
g (t ) =
=
t

2
t
T0
t

sin

(11)

Such a waveform crosses zero every T0/2 seconds

In frequency, this corresponds to an ideal low-pass
(brick-wall) filter, with the cut-off frequency F=1/T0
If a symbol is issued every T0/2 seconds,
transmission can be made without ISI
The rate in this case will be 2F symbols/s (Nyquist
rate, ideal rate)

Although, unlike in the digital baseband transmission (e.g.:NRZ), a single

information symbol is represented by a waveform with infinite duration (a cardinal
sine), we can transmit the next information symbol at the first zero crossing of the
sinc waveform (i.e.: at T0/2); this will prevent ISI to occur and will allow a
transmission at a rate of 2/T0 symbols/sec. The later statement is identical with
Nyquists theorem, formulated on the previous slide.

12

ISI-less data transmission

Nyquist: from fiction to reality

If the equivalent channel was an ideal low-pass filter,
transmission can be made at the Nyquist rate,
without ISI
The later statement is valid only if the receiver
faithfully respects the sampling times
In practice, we never deal with ideal filters (which are
non-causal, infinite duration)
More the transfer function gets closer to the
brickwall, higher will be the speed of signal
variation nearby 0
A small error in the sampling time leads to an
important energy of the ISI
we must seek other forms of Nyquist filters!!!

The ideal LPF is unpractical, because its impulse response is infinite as duration
and not-causal. In frequency, this characteristics are the basis for the ideal,
rectangular (brick-wall) form that corresponds to this filter.
Other disadvantages: the waveform evolution nearby the zero-crossing points is
very quick, so a small mistake in the sampling time will significantly impact the
expected value of the sample. In practice, other forms of filters satisfying Nyquist
criterion must be used.

13

ISI-less data transmission

The road to the Nyquist filters [1]

The equivalent response of the channel, at the
sampling moments nT can be expressed as:
( 2 k +1)

T
1
1
jnT
jnT
g n = g (nT ) =
d =
d (12)

G ()e
G ()e
2
2 k = ( 2 k 1)
T

By substituting
gn =

2 k
T

,we get:

1 /T
2 k jnT
) e
d (13)
G (
2 / T k
T

Starting with eq. (12), a mathematical method to find other filters that satisfy Nyquist
criterion for zero ISI. Eq. (12), shows the expression of the n-th sample of g(t),
expressed as the inverse Fourier transform of G(), computed at t=nT.
The term in the square brackets (rel. 13) is directly related to the spectrum of a
shape that would satisfy ISI-less data transmission criterion.

14

ISI-less data transmission

The road to the Nyquist filters [2]

Reminder: the ideal LPF has the transfer function:

1
= T , if
G () = 2 F
T
0, otherwise

(14 )

T is the symbol time,

which equals the zerocrossing rate of g(t),
which is T0/2 !!!

By comparing (13) and (14), we obtain:

2k

), if
G (

= T , if
T
T
2F
Geq () = k
T

0, otherwise
0, if >
T

(15)

Any spectrum that satisfies eq. (15) corresponds to a waveform that meets
Nyquists criterion. The spectra G() that are inside the sum operator are not ideal
LPF, but added up for k=-1,0 and 1 they will lead to an ideal LPF.

15

2k

), if
G (

= T , if
T
T
k
2F
Geq () =
T

0, otherwise
0, if >
T

(15)

The ultimate consequence of (15)

Ideal LPF from (14) can be

approximated piece-by-piece (k by k
from 15)
Any transfer function G which respects
(15) can be used as Nyquist filter

16

ISI-less data transmission

The road to the Nyquist filters [3]

Band for k=-1

Band for k=0

Band for k=1

G() satisfies Nyquists zero ISI criterion if it exhibits symmetry around

/T (A and AI) points
In this case, the side bands (k=-1 and k=1) compensate the
frequency response of the main lobe (k=0), such a way that eq. (15) is
satisfied
A common choice for G is provided by the raised cosine filters

It is straightforward that, in order to satisfy (15), the spectrum G() must exhibit
symmetry around the cut-off frequency =/T, as shown in the figure above. If
G() and the ideal LPF determine equal areas to the left and to the right of , then
G() corresponds to a waveform that meets Nyquist criterion.

17

ISI-less data transmission

The raised-cosine family

The previously formulated criteria is met if:

= T,
(1 )

G () =
1 sin( 2 2 ), (1 ) (1 + ) (16 )
2

0, > (1 + )

The impulse response will be:

g (t ) =

sin(t ) cos(t )

2 t 2
t
)
1 (

(17 )

There is a whole category of filters that meet Nyquists criterion and respect the
symmetry of their spectra around the cut-off frequency. These filters are referred to
as Nyquist pulse-shaping filters, or raised-cosine filters, because the shape of their
spectrum, which looks pretty much like a cosine in the frequency domain. Both their
frequency (16) and their impulse response (17) depend on a parameter , called
roll-off factor or excess bandwidth, which spans from 0 to 1.

18

ISI-less data transmission

Comments on the choice of the roll-off factor

spans from 0 to 1 and is referred to as

excess bandwidth or roll-off factor

For =0 the ideal low-pass filter is obtained,
whereas =1 defines the square cosine filter
Common choices in practice range from 0.1
to 0.5
If we want to achieve a data rate of R,, then
the bandwidth we need will be:
B = (1 + )

R
2

(18 )

What lies behind the excess bandwidth name?

The answer is given by relation (18). Basically, excess bandwidth means the extra
bandwidth we need, when compared to the bandwidth we would need to achieve a
fixed data rate of R when using cardinal sine shapes for the data pulses.
Or, otherwise formulated: the data rate R can be reached with a consumption of
bandwidth of B=R/2 when the pulses are cardinal sine pulses, and with a
consumption of bandwidth B=(1+)(R/2) when we use other types of Nyquist
waveforms than the ideal cardinal sine.

19

ISI-less data transmission

The need for Nyquist filters

These filters are usually referred to as pulse-shaping filters
A certain shape of the data pulses will lead to a certain level of
spectral efficiency
Interpreting Nyquist: Take care how you shape the data
pulses, and you will achieve the efficiency and the robustness
you desire
Keeping low, good spectral efficiency is obtained (), but
high sensitivity to the sampling time accuracy too ( )
In noisy channels square root raised cosine filters must be used
to the transmitter and to the receiver

20

ISI-less data transmission

Nyquist filters: when Time meets Frequency

Keeping fixed the bandwidth !!!

Keeping fixed the data rate !!!

Small roll-off factor means higher transmission rate at the same occupied bandwidth
BUT! The energy of the side-lobes in time is higher, which causes higher sensitivity
in case of synchronization problems
*Pictures downloaded from www.complextoreal.com

The above figures highlight the need for a good compromise when choosing .
Thus, the left figure shows various impulse responses of some raised cosine filters.
A small value for will permit a higher data rate (the zero crossings of the
waveforms appear at smaller time intervals); nevertheless, the side-lobes carry, in
this case, an important amount of energy. The sensitivity to the synchronization
errors will be higher than for high values of (close to 1), when the side-lobes have
smaller amplitudes.
The right figure shows the frequency responses of several filters from the raised
cosine family; when plotting these responses, we consider a fixed transmission rate.
The lowest bandwidth consumption, as expected, is given by the brick-wall and
higher will be the value of , higher the extra bandwidth required, compared to the
brick-wall.

21

ISI-less data transmission

Squared cosine case

S h a p i n g t h e p u ls e s w i t h t h e s q u a r e c o s i n e
1
0 .8
0 .6
0 .4
0 .2
0
-0 . 2
-0 . 4
0

50

100

150

200

250

Very low-energy sidelobes

Symbol time doubled compared to the sinc case

In the case of =1, we get the squared cosine filter. As shown in the above figure,
its side-lobes are almost negligible, which is a good property from the ISI
robustness point of view. The dashed line has the meaning of a comparison
threshold, always needed in practice, in order to make accurate decisions on the
transmitted symbols. E.g.: sample>0.5 means 1, sample<0.5 means 0.

22

ISI-less data transmission

Practical implementation
of the raised cosine filter
clock

The impulse
response is
approximated by
samples

Every sample
represents an
amplification factor

Shift
register

Input
bits
b-4

b-3 b-2

b-1

b0

b1

b2

b3

Analog Addition circuit

Amplifiers
b4
Shaped output
s(t)

Although the raised cosine filters, as defined by equations (16, 17) are
closer to the practical implementation than the ideal LPF, they still have some
properties that makes their exact implementation impossible: their impulse response
is infinite, and they are not causal. In practice, we need to apply a time window of
finite duration on the impulse response, and to shift this windowed version to the
right, such as to obtain a causal system. The two effects are shown on the above
figure, where the impulse response has finite duration and is 0 for negative values
on the time axis.
In practice, these filters are implemented using digital circuits. That is,
instead of having an analog waveform like in the figure above, we will get some
samples of this waveform. The most common implementation relies on FIR (Finite
Impulse Response) filters, that have a finite number of coefficients. Their values are
given by the amplitude of the samples taken from the original waveform. Such a
filter can be implemented using a transversal structure, as shown in the upper figure
from this slide. This implementation uses a shift register with N cells (N=9 in our
case). The result is that we get a FIR filter with 9 taps. Most of the digital
transmission devices (and especially in wireless transmission) incorporate such a
filter.

23

Transmission with controlled ISI

Motivation:

In practice, the raised cosine filters will not reach the

Nyquist rate
Closer they are to this objective, higher is the transmission
sensitivity to synchronization errors

Solution:

Some degree of ISI can be tolerated, if the ISI is controlled

Higher rates can be obtained
The cosine filter allows to reach the Nyquist rate

Principle: every waveform g(t) will carry two-bits of

information
This is the reason for calling this transmission duobinary

The raised cosine filters can be we well approximated in practice, but they can only
achieve 1/1+ of the Nyquist rate. If we want a transmission robust to ISI, high
values should be chosen for . For example, if =0.5, only 66% of the Nyquist rate
can be obtained. This drawback is the main motivation behind the transmission with
controlled ISI. This transmission, commonly referred to as duo binary transmission,
sacrifices the ISI-less principle, but allows to reach the Nyquist rate. ISI is
controlled, in the sense that, by systems design, the ISI influence is accurately
known and the phenomenon can be counteracted at receiver side.

24

Transmission with controlled ISI

The cosine filter

2

cos
(19)
G ( ) =
2
0, if >
G()

g (t ) =

cos(t )

2t
1

(20)

1.4

1.2

0.8
g(t)

0
0

0.6
0.4

0.2

0.4

0.6

0.8

T=

1.2

1
2F

1.4

1.6

1.8

(21)

The signaling rate is 2F!!!

0.2

-3/2

0
-0.2
-15

-10

3/2
/2

-/2

-5/2
-5

0
t

5/2
10

15

When, instead of a raised cosine, a cosine filter (eq. 19,20) is used for shaping
purposes, the signal so generated will lead to a transmission with controlled ISI.

25

Transmission with controlled ISI

Graphical view
1.4
1.2
1
0.8

g(t)

Three successive bits

of 1 are transmitted
In the sampling
moments, the sample
value depends only on
the current and on the
previous transmitted bit
The ISI is thus
controlled (the amount
of ISI is known)

0.6
0.4
0.2
0
-0.2
-15

-10

-5

10

15

20

25

Sampling times

It can be seen from the figure above how, at the sampling instants the collected
samples depend not on a single transmitted symbol, but on exactly two transmitted
symbols (the current one and the previous one). Thats why the transmission is
called with controlled ISI (ISI exists, but its known and, therefore, it can be
controlled/counteracted).

26

Transmission with controlled ISI

Mathematical approach
The k-th sample can be expressed as:
rk = r (

T
T
T
1
+ kT ) = ak g ( ) + ak 1 g ( ) + a n g[(k n )T ] (24)
2
2
2 n k , k 1
2

In the right-side sum, all the terms are zero (they

match exactly the zero-crossing of g). It follows that:
rk = ak + ak 1

(25)

Theoretically, if ak is bi-polarly encoded rk may take

three values: 0, if a a
k

k 1

(26)
rk = 2, if a k = a k 1 = 1
2, if a = a = 1
k
k 1

The k-th received sample can be described according to (24). If cosine waveforms
are used to shape the bits and the sampling times are faithfully respected, the right
side sum is zero, and rk will only depend on ak and on ak-1.

27

Transmission with controlled ISI

Graphical view

A signal with controlled ISI is show in the figure above. Thus, two successive bits of
1 will issue a collected sample of amplitude 2; when a 1 is followed by a 0 ore
vice-versa, the sample will be 0; finally, two bits of 0 will issue a sample of value 2.
The main disadvantage of a transmission with controlled ISI is the fact that the
decision on the current bit, e.g. ak, depends on the decision of the previous bit, ak-1.
E.g.: if rk=0, we have to know the value of the previous bit, in order to make a
decision about the current one. If ak-1 was, for example, detected as a 0, it follows
that ak is 1.
If the ak-1 is wrongly detected, the error propagates until a new error will turn the
decision to the right path.

28

Transmission with controlled ISI

Pre-coding
Disadvantage of controlled ISI: every decision
depends on two successive bits (error propagation)
Solution: pre-coding performed
Instead of ak, another sequence (bk) is transmitted,
computed as:
bk = a k bk 1

(27)

This leads to a one-sample based decision, as

follows:
0, if rk = 2
ak = bk bk 1 =
1, if rk = 0

(28)

The pre-coding, at the transmitter side, removes the dependency of the decision on
the two bits. Somehow, this inter-bit dependency is transferred at the
transmitter side. This can be simply done by means of a pre-coding operation.
The transmission steps are now:
1. The original bits an are randomly generated
2. The precoded bits bn are computed
3. The precoded bits are bipolarly encoded (i.e.1 -> +1V, 0 -> -1V).
4. The signal so obtained is passed through a cosine filter, to generate a
transmission with controlled ISI.
Thus, instead of transmitting directly the original bits an, some new bits bn (the
pre-coded stream) are generated, according to (27). Taking into account this
pre-coding, the receiver has to inverter, in order to extract an, the original bits,
inversion shown by equation (28). The reception steps are:
1. The signal is sampled (rk) values are collected.
2. Decision is made:

if rk is 2 or -2, it follows that bk=bk-1, and ak=0 (see eq. 27)

if rk is 0, it follows that bkbk-1 and ak=1

29

Evaluating ISI level using the eye diagram

Eye diagramat a glance

The eye diagram allows to evaluate the degree of ISI
It is obtained by superposing the time domain representation of the
signal for a large number of transmitted symbols

*Picture downloaded from the paper Analyzing signals using eye diagram author G Breed High Frequency Electronics Journal

The eye diagram is a very useful tool to assess the accuracy of a digital
transmission. It can be applied to rectangular or raised cosine waveforms (those are
the two practical cases that can be met in a digital transmission).
The eye diagram is an oscilloscope representation on a time domain signal, in
which waveforms for every T sec. interval are superposed. E.g. Waveform that
represent a data signal in the interval [0,T] is superposed with the waveform for the
interval [T,2T], with the waveform for the interval [2T,3T] and so on.
In a radio system, the point of measurement for the eye diagram may be prior to the
modulator in a transmitter, or following the demodulator in a receiver, depending on
which portion of the system requires examination. The eye diagram can also be
used to examine signal integrity in a purely digital base-band systemsuch as fiber
optic transmission, network cables or on a circuit board. The figure above shows the
type of information that is given by the eye diagram.

30

Evaluating ISI level using the eye diagram

Ideal low-pass versus square cosine

The square cosine eye is more widely opened
Lower sensitivity to the sampling times due to the reduced-energy sidelobes
BUT!!! Using sinc waveforms (ideal low-pass) we achieve twice the data rate of
the square cosine
Eye Diagram

Eye Diagram

1
0.8

0.6
0.4
Amplitude

Amplitude

0.5

0
-0.2
-0.4

-0.5

Ideal low-pass

Square cosine

-0.6

-1

-0.5

0.2

-0.8

0
Time

0.5

-1
-0.5

0
Time

0.5

In the slide, we compare the eye diagram from two pulse-shaping waveforms (the
two extreme cases of raised cosine).
For the cardinal sine (roll-off =0), the eye is not so widely open, pointing out high
sensitivity to the sampling point accuracy. By the contrary, in the case of square
cosine (roll-off =0), the eye is widely open, the transmission is not sensitive to the
synchronization errors, but barely one half of the Nyquist rate can be achieved.

31

Evaluating ISI level using the eye diagram

Eye diagram for unfiltered rectangular

waveforms

Ideal eye diagram of a

square waveform (no jitter,
no noise)

Misalignment of rise and

fall times (jitter)

At higher rates, jitter has

more impact (even if its
absolute values decreases)

Figures above show some eye diagrams for purely digital systems, in which the
data signal is made of nearly rectangular waveforms. The jitter error is clearly
highlighted in figures 2 and 3, where misalignment of rise and fall times causes a
kind of grid, or spreading, to occur towards the end of the bit intervals.

32

Improving the performance of the baseband

transmission systems
The the k-th received symbol may be expressed as:
N

rk = r (t0 + kt ) = an g k n + nk
n= N

(29 )

The error probability for the k=0-th transmitted symbol is:

Pe = P an g n + n0 > dg0 (30)

n
N
=

n 0

Assuming that there is no useful signal in the channel (only AWGN

noise with unitary variance), the noise power after the receiver filter
is:
1
2
2fn =
GR () d (31)
2
*the fn subscript above emphasizes that we refer to the power of the filtered noise

33

Improving the performance of the baseband transmission systems

SNR maximization
If we assume that there is no noise in the channel, the 0-th
received sample can be computed as:
1
1
jt
r0 = r (t ) |t =0 =
GR () S ()e d |t =0 =
GR () S () d (32)
2
2
| r0 |2
At t=0, we try to choose GR() which maximizes 2
fn
Solution: GR()=S*() (an expression of the matched filter)
For an ideal channel, C()=1 and the received signal form is
determined by the emitters shaping filter: S()=GE()
The choice which maximizes the SNR: GE () = GR () = G ()
Example: for the square root square cosine filter, we have:

GE () = GR () = cos
(33)
4

In the digital transmission systems we deal with two disturbing phenomena: noise
and ISI. While the later can be counteracted using Nyquist filter, we need to focus
our attention on the first effect too. There is a huge amount of detection/estimation
literature, proving that if the noise is of AWGN type, the best solution is to place a
matched filter at the receiver side.
A well balanced choice, which counteracts both effects is to use a pair of filters
(square root raised cosine), one of which is the pulse shaping filter (at transmitter)
and the other one that is a non optimal matched filter (at the receiver). Together,
they will lead to an equivalent response that respects Nyquists criterion for zero ISI,
while the receivers filter will effectively remove the noise.

34

The performance of the baseband

transmission systems
Error probability
Transmission speed
Transmission efficiency

35

The performance of the baseband transmission systems

The error probability in ideal channels [1]

An ideal channel for data transmission does not
modify the signal transmitted through it (C()=1)
Identical transmitter and receiver filters may be used
The average power of the transmitted signal is:
PS =
a2

a2
2

G () d = a 2

(34)

is the average power of the transmitted symbol

36

The performance of the baseband transmission systems

The error probability in ideal channels [2]

We assume M-level signaling, with the levels: d , 3d , 5 d ,
The useful energy is conserved by the shaping filter:
a2 =

Decision thresholds at:

The error probability is:
Pe = (1

d2
( M 2 1) (35)
3
0, 2 d , 4 d ,

1
) P ( n0 > d ) (36)
M

ISI is considered 0 in equation (36)

37

The performance of the baseband transmission systems

Discussion on equation 36
1

Reminder: Pe = (1 ) P ( n0 > d ) (36)

M
For the two
extreme levels,
an error occurs
only if noise
sample has a
certain sign
For the other
levels, the sign
doesnt meter
4

:Signal
:Decision Thresholds

3
2

Amplitude

1
0

-1
-2
-3
-4

50

100

150

200

250

time

38

The performance of the baseband transmission systems

The error probability in ideal channels [3]

Taking into account the pdf of the noise in (36):

Pe = (1

erfc is introduced by:

1
1
)
e
M 2 d

n2

22

dn (37)

erfc( x) =

2
exp( z 2 ) dz (38)

Error probability can be express as (from eq. 37,38)

Pe =

1
1
d
(1 )erfc
(39)
2
M
2

39

The performance of the baseband transmission systems

The error probability in ideal channels [4]

From eq. (35), d can be expressed as:
d=

3
2

M 1

PS

(40)

From (39) and (40):

Pe =

P
1
1
3
S (41)
(1 )erfc
2
2( M 1) PN
M
2

40

The performance of the baseband transmission systems

The error probability in ideal channels [5]

0

10

-1

10

-2

Pe

10

-3

10

-4

10

:M=2
:M=4
:M=8
:M=16

-5

10

-6

10

10

15

20

25

30

SNR [dB]

41

4
:Signal
:Decision Thresholds

3
2

Amplitude

1
0
-1
-2
-3
-4

50

100

150

200

250

time

42

The probability that the noise sample value to exceed the threshold (d=2
in our example) is the delimited area under the Gaussian curve!!!
0.4
0.35
0.3

pdf

0.25
0.2
0.15
0.1
0.05
0
-4

-3

-2

-1

noise sample value

43