3.

Principles of Digital Transmission, Line

Coding and Modulation - Christie Alwis
Nyquist sampling theorem
The sampling frequency should be at least twice the highest frequency contained
in the signal.
To achieve sampling theorem the following aspects should be taken into
consideration.
I. The signal should be band limited.
II. The sampling frequency should be equal or greater to twice the
maximum frequency of the sound wave.
III. After sampling, the sampled values should be transmitted through a
transmission medium to a receiving end to be reproduced.
IV. The original transmitted signal has to be reproduced at the end by
low pass filtering.
Telephone signals are band limited signals of 0-4kHz. Therefore to achieve the
second condition the sampling rate should be equal or greater than 8kHz.
Frequency = 1 / Time Period
Sampling time period for telephone = 1 / 8000
= 125 s
Hence telephone signals are sampled every 125 s.
A problem arose when using the sampling theorem practically, concerning the third
aspect stated above. When some kind of information is received first it is
necessary to verify whether that certain information can be true or false.
Suppose that there are two robots operating two exchanges as can be seen on
next page. When R1 sends some sampled signals to R2, R2 have to verify the
incoming samples. In the transmission medium noise is introduced and the
received sample can be different from the sent one. It is very difficult to verify the
values it receives, as this can be any value between -127V to +127V. Therefore it
was necessary to find a solution to make verification easier.

3.2 Quantizing
As a solution Quantizing was introduced. That is, instead of sending the actual
sample, first the sampled signal was put into a known number of levels, which is
informed to the receiver. Suppose instead of sending a whole range of voltages,
Robot 1 informs Robot 2 that it is going to send only 4 voltage levels, say 0-3V. For
example if the sample is 2.7V, first, Robot 1 will convert it into a 3V sample. Then it
will be sent through the transmission medium. Robot 2 at the receiving end gets a
sample of 3.3V. Then immediately he knows that this is not an agreed level, hence
the sent value has been changed. Robot 2 converts 3.3V sample back into a 3V.
There are two types of Quantizing :I. Linear Quantizing
II. Non-linear Quantizing

In the Linear quantizing graph each voltage level consists of ( 3072 / 128 = ) 24mV.
Therefore any value that has to be quantized is put into the nearest multiple of 24.
3

Take the example of 22mV. It has to be quantized as a 24mV, as 24mV is the

nearest multiple.
Take 3022mV.
Calculating the level = 3022 / 24
= 125.9
= level 126
Quantized value = 126 x 24
= 3024 mV
Therefore 3022mV is sampled as a 3024mV.
The quality of voice transmission is measured by signal to noise ratio. That is the
division of the original signal value by the change made when quantizing.
The following S/N ratios were calculated using the above values.

Linear quantizing is not used. Why.?

It is noticeable that even though the noise is the same for both these signals the
S/N ratio is highly different. It seems that Linear Quantizing gives high S/N ratios
for high signals and low S/N ratios for low signals.
There is another disadvantage in using Linear Quantizing. That is, according to
CCITT experiments 90% of voice signals lie between Vmax/2. In order to give high
S/N ratios for the range of Vmax/2, different ranges of voltages should be given
to each level. In the Linear Quantizing graph, the voltage range is 24mV between
any two successive levels. Therefore the small signals are not supported with.
Hence linear quantizing does not serve the purpose.

Non-Linear Quantizing
The main aim for using the non-linear graph is to have a good S/N ratio. In order to
do so the gradient of the curve between Vmax/2 should be changed. Then the
4

voltage ranges will not be the same like in the linear graph. The non-linear graph is
shown on next page.
It is divided into 8 parts. These are named as SEGMENTS. As 90% of the voice
signals lie in the range of Vmax/2, only one segment is given for voltages beyond
Vmax/2. The sample levels, which is the Y axis is divided into 8 parts each
carrying (128/8=) 16 levels. These 8 parts correspond with the 8 segments, which
mean each segment holds 16 levels.
The gradient of upper segments 0, 1 and lower segments 0, 1 are the same.
Therefore they are considered as one segment. Hence, the non-linear graph is
named as 13 segments graph.
Summarized information on the graph :-

For example lets try to find the quantized value of 22mV.

The segment number for 22mV
= Segment 0
Increment per level
= 1.5
Calculating the level
= (22 / 1.5)
5

= 14.6
= level 15
Quantized value
= 15 x 1.5
= 22.5 mV
Hence the quantized value of 22mV is 22.5mV.
THE A LAW SIGNAL-COMPRESSION CHARACTERISTICS OF 13 LINEAR SEGMENTS

Study the following calculated values for linear and non-linear methods.

00

It can be seen that in non-linear quantizing the smaller sampled voltages have a
higher S/N ratio than in the linear quantizing method. And the higher sampled
voltage values have been reduced to a certain extent.
In the linear quantizing method the difference between the highest S/N ratio and
the lowest is (1525-9=) 1516. In the non-linear method it is (260.8-34=) 226.8.
7

Therefore there is no major difference in the S/N ratios for higher and lower
sampled voltages.
These S/N ratio values in the non-linear method are sufficient enough to transmit a
normal message.
Hence non-linear quantizing is used.
A vocoder that places most of the quantization steps at lower amplitudes by using
a nonlinear
function, such as a logarithm, is said to compress voice upon encoding, then
expand the PCM samples to re-create an analog voice signal. Such a vocoder is
hence called a compander (from compress and expand).

Encoding
As there are 256 levels, the quantized value is encoded into 8 bits.

The first bit indicates whether the sampled value is negative or positive. The
segment number is coded in the next three bits. Last four bits give the level
number. Using the table we can convert any sampled voltage value into 8 bit
binary format.

For example suppose that the sampled value is +367.

The first bit is 1 as it is positive. 367 is in the range of 192 384. Hence the
segment number is 4 which is 100 in binary.
The level position can be calculated by dividing additional value by the appropriate
increment per level for that particular segment. In this case the additional amount
8

comes to (367 192 =)175. And the increment is 12. Therefore the level is 175 /
12 which is 14.6. As 14.6 is closer to 15, 15 is taken as the level number.
Hence the encoded form of +367 would be as follows.
1 100 1111
Take another example value like -50V.
The first bit should be 0 as this is a negative value. 50 is in the range of 48 96.
Hence the segment number is 2 which is 010 in binary.
The additional amount comes to (50 48 =)2. And the increment is 3. Therefore
the level is 2 / 3 which is 0.67. As 0.67 is closer to 1 it is taken as the level number.
Hence the encoded form of -50 would be,
0 010 0001
Likewise any voltage level is converted into binary and is transmitted.

Time Division Multiplexing

The calculated sampling time period for telephone signals is 125s. This means
that Robot 1 sleeps for 125s before the next sample and in that time it samples
only a signal of a single telephone.

Considering modern technology 125s is too much of a sleeping time. Therefore

instead of letting it sample only one telephone, it was decided that signals of many
channels should be sampled during that time. This is called Time Division
Multiplexing.
There are two systems as at present for TDM.
I. European (CEPT) system
II. American system
In European system 32 channels are used while 24 channels are used for the
American system.
Most of the countries use the CEPT 32 channel system. This means in 125s,
Robot1 samples signals of 32 telephones. This is called a FRAME.
9

The above diagram is a frame consisting 32 telephone signals numbering from 0

31. They are referred as TIME SLOTS. Each time slot takes ( 125 / 32 =)3.9s of the
whole time of its frame. And each time slot contains 8 bits of information. See
figure 6.3.

At t=0 both robots conduct TS0 which is the synchronising bit. Then after 3.9s R1
passes a sample value of subscriber 1. Then again after another 3.9s R1 goes to
subscriber 2. Likewise it goes on and comes to the last subscriber of the frame
which is the 32ndth sub at t=121.1s. Upon finishing it, R1 goes back into frame
synchronizing bit of the next frame after 125s. This is the procedure of Time
Division Multiplexing.
10

In the transmission media all these frames are sent one after the other. In another
word a whole series of 1s and 0s are received at the receiving exchange. What if
one bit gets misplaced? It definitely would make a mess of every bit after that. The
receiving information will be wrong.
In order to avoid such confusion and to obtain the correct sample value, a frame
synchronizing bit is introduced. The time slot 0 of every frame is taken as the
synchronizing bit. That is why the 0 numbered telephones are crossed out in the
previous diagram.
Hence at the beginning of every frame the receiving exchange synchronizes itself
with the oncoming signal. There are 32 channels in a frame. The speed of a frame
can be calculated as follows.
Time taken for the whole frame
= 125s
Number of bits in a frame
= 32 x 8
= 256 bits
Hence the speed of the frame= 256 bits / 125s
= 2.048 Mb/s
Therefore the speed of one link from one exchange to another is 2.048 Mb/s.
Practically these are referred to as 2Ms.
The speed of one telephone channel can be calculated in the same way. One
telephone means one time slot.
Time allocated for the whole frame = 125s
As one telephone channel occurs only once during this whole time, 125s is equal
to
the time taken for one channel.
Number of bits in a single time slot = 8 bits
Hence the speed of one telephone channel = 8 bits
(125 x 10-6) s
Therefore the speed of any telephone channels is 64 kb/s.

11

Waveforms
There are two types of general uni-polar digital waveforms by which any binary
number can be represented using a Mark (1) or a Zero (0).
I. Non-Return to zero (NRZ) form
II. Return to Zero (RZ) form
Both these types have to be synchronized with a clock pulse. Therefore the clock is
generated at both the transmitting and receiving sides.
Before the end of the clock pulse every mark should come to zero in the Return to
Zero wave type. That means a mark only takes half of the time taken for a full
clock pulse. In this case it should be half of 448ns.
The Non-return to Zero means the mark does not come to zero before the end of
the clock pulse. Hence any mark or zero will have a time period of 448ns. This can
be analyzed using an example.
Take the binary number 11010011. This can be represented both by NRZ and RZ
forms as shown in figure 8.1.

As can be seen clearly these two waveforms are uni-polar. Therefore there will be a
positive D.C.(direct current) average value for both. These coded values have to be
12

transmitted through a transformer. As transformers do not pass D.C. voltages, it

was necessary to keep a D.C. average of 0V.
In order to do so the above waveforms must be changed so that they give an
average of 0V. Hence in order to suit for the transmission medium, Alternate Mark
Inversion (AMI) is introduced.
Using AMI the RZ and NRZ waveforms can be converted into bi-polar waves. AMI
rule is consecutive marks should be of opposing polarity. Accordingly the above
waves can be redrawn in AMI form as follows.

Assuming that 50% of the transmitted waves are Marks and the other half are
Zeros, the D.C. average then comes to 0V. Hence these types of waveforms are
suitable for transmission. But this method is not used. To understand the reason
for not using this method, first let us analyze the function of a repeater.

Repeaters
13

For a certain message to pass between two locations there should be a

transmission medium connecting them. In telecommunication this could be either
copper wires, radio links or optical fiber.
These media carry information for a long distance. Therefore interferences such as
electromagnetism and thunder could produce noise inside transmission media
which results in attenuation and energy losses of the transmitted signal. Hence a
way of converting these weak signals into their original forms was to be
considered.

As its name implies repeaters are used as a solution. There are two main types of
repeaters.
I. Active repeaters
II. Passive repeaters
Active repeaters feed power as well as converting the weak signal into its original
form.
Passive repeaters act as a sort of a reflector only. For example satellites belong to
this
kind. Since voice transmission would require power feeding as well as converting
the
weak signal, active repeaters are used.
What does a repeater do?
Repeaters do three main functions to a weak signal.
I. Re-shape
II. Re-time
III. Power feed
Many repeaters are stationed between exchanges for these purposes.

14

The circuit diagram of a repeater is as follows.

The receiving signal is a combination of many high frequencies. They have very
high rates of attenuation. The line equalizer amplifies low frequencies with a high
gain and high frequencies with a low gain. Hence the incoming signal is equalized.
The (+) threshold detector takes an average positive value and gives out an output
only when the signal voltage go beyond the selected value. Similarly the ()
threshold detector gives an output when the signal go below the selected negative
voltage.
One output of the equalizer is sent through a rectifier where the output voltage
value gets rectified. The tune amplifier is driven by the rectifier. It regenerates the
clock pulse which is then sent through a differentiator. The differentiator gives out
a short spark at the beginning of each clock pulse. This is fed into both (+) and ()
re-shapers.
The (+) re-shaper checks whether there is a mark at the time when the
differentiator gives a spark. If there is, it gives out a mark of the same time period
as the clock. The same procedure is followed by the () re-shaper.
The outputs of both (+) and () re-shapers produce the original waveform as can
be seen in page 27. This is considered a great advantage of Digital Transmission. If
the transmitted wave was analogue this re-shaping could not take place. Hence
analogue waves are not used for transmission between exchanges.
A problem arose in this method. When two people talk over the phone, only one
speaks at a time. The other listens. During listening, the time slots of the listener
will carry series of zeros. These zeros come across the repeater and would not
give a voltage value for the rectifier to drive the tune amplifier. Therefore the
15

clock regenerating process stops. Because of this reason RZ and NRZ are not used
directly. A new coding was introduced.

Remote power supply configuration for 2.048 Mb/s digital link

16

As can be seen above, power is fed from one

terminal. If the transmission line is too long,
power has to be fed from both terminals. Then
the power feeding arrangements at the repeaters
are the same except in the middle repeater. In
such a situation, this repeater power feeding
arrangement is as in the sub diagram on the
right.

High Density Bi-Polar 3 (HDB3)

As this was introduced to suit for the transmission process, converting AMI codes
to HDB3 is called trans-coding. Nature of voice is that it is 50% efficient. The
repeater clock cannot be regenerated. Hence HDB3 coding technique is used.
There are four rules for HDB3 coding.
I. More than three consecutive zeros are not allowed to be present in
the waveform. For the fourth 0 introduce a Violation bit.
II. Violation bit has to be of the same polarity as the previous mark.
III. Two consecutive violation bits has to be of opposite polarity.
IV. If the number of marks between two consecutive violation bits is
even the format should be B00V where B is a stuffing bit and of
opposite polarity to the previous mark. If the number of marks is an
odd number the format should be 000V.
Some examples :17

The receiving exchange will get 1+0001+1-0001-01+. Suppose that these coded
values are sent from exchange A to exchange B. After coding the 1st mark it keeps
its value inits memory. When B receives a 2nd consecutive positive mark, it
immediately decides that a violation has taken place. Therefore B takes the
second mark as a zero. When two more negative marks come, again B identifies
them as another violation. The latter mark comes after 3 consecutive zeros.
Therefore it cannot be a stuffing bit as only two zeros are followed after the
stuffing bit. Then B decides that the latter mark should be a zero. Hence the final
sample becomes 100001000001.

18

100001000001
Assuming that the previous bit was a positive mark this series of zeros will be
coded according to HDB3 rules as follows.

Ref: Digital Communication, Christie Alwis

Coded mark inversion (CMI)

19

At higher data rates the memory and logic circuits required to implement HDB3
become relatively more expensive. CMI was developed for use at higher data rates
as it uses simpler, and therefore cheaper, circuitry. In comparison with AMI, this
code overcomes the problem of strings of zeros as there is at least one transition
for each symbol. The coding rules are shown in Figure A2.3. Binary 1 is chosen
from one of two full-width pulses. Binary 0 is encoded into two alternate half-width
pulses, as shown. As some data is encoded into two pulses of opposite polarity,
the spectrum of CMI is double that of AMI and HDB3, which means that lines of
larger bandwidth are required. Figure A2.4 illustrates how a certain string of data
may be encoded using CMI. Each successive binary 1, rather like AMI, simply uses
the alternate symbol compared with that used by the immediately preceding
binary 1 symbol, irrespective of any preceding or intervening binary zeros.

nBmT Coding
20

Unipolar signaling

21

Unipolar signalling (also called onoff keying, OOK) refers to a line code in which
onebinarysymbol(denotingadigital0,forexample)isrepresentedbytheabsenceofapuls
e(i.e.a space) and the other binary symbol (denoting a digital 1) is represented by
the presence of a pulse (i.e. a mark). There are two common variations on unipolar
signalling, namely non-return to zero (NRZ) and return to zero (RZ). In the former
case the duration () of the mark pulse is equal to the duration ( To) of the symbol
slot. In the latter case is less than To .
Typically RZ pulses fill only the first half of the timeslot, returning to zero for the
second half. (The mark duty cycle, / T o , would be 50% in this case although other
duty cycles can be, and are, used.) The power spectral densities of both NRZ and
RZ signals have a[( sin x)/ x] 2 shape here x =f . RZ signals (assuming a 50%
mark duty cycle) have the disadvantage of occupying twice the bandwidth of NRZ
signals (see Figure 6.11). They have the advantage, however, of possessing a
spectral line at the symbol rate, f o=1/To Hz(and its odd integer multiples), which
can be recovered for use as a symbol timing clock signal, Table 6.1. Non-linear
processing must be used to recover a clock waveform from an NRZ signal (see
section 6.7).
Both NRZ and RZ unipolar signals have a non-zero average (i.e. DC) level
represented in their spectra by a line at 0 Hz, Figure 6.11. Transmission of these
signals over links with either transformer or capacitor coupled (AC) repeaters
results in the removal of this line and the consequent conversion of the signals to a
polar format. Furthermore, since the continuous part of both the RZ and NRZ signal
spectrum is non-zero at 0 Hz then AC coupling results in distortion of the
transmitted pulse shapes. If the AC coupled lines behave as high pass RC filters
(which is typically the case) then the distortion takes the form of an exponential
decay of the signal amplitude after each transition. This effect, referred to as signal
droop, is illustrated in Figure 6.13 for an NRZ signal. Although the long term DC
component is zero, after AC coupling short term DC levels accumulate with long
strings of ones or zeros. The accumulated DC level is most apparent for the first
few symbols after a string represented by a constant voltage. Neither variety of
unipolar signal is therefore suitable for transmission over AC coupled lines.
Since unipolar voltage levels of 0 or V volts are equivalent (in terms of BER)to polar
levels of V/2 volts (section 6.2.1) then unipolar signalling requires twice the
average, and four times the peak, transmitter power when compared with polar
signalling, Table 6.1.

22

Figure 6.13 Distortion due to AC coupling of unipolar NRZ signal: (a) input; (b)
output
Polar signalling
In polar signalling systems a binary one is represented by a pulse g 1(t )and a
binary zero by the opposite or antipodal ) pulse g 0(t )= g1(t ), Figure 6.11. Figure
6.12 compares polar and unipolar signals for a typical data stream. The NRZ and
RZ forms of polar signals have identically shaped spectra to the NRZ and RZ forms
of unipolar signals except that, due to the opposite polarity of the one and zero
symbols, neither contain any spectral lines. Polar signals have the same bandwidth
requirements as their equivalent unipolar signals and suffer the same distortion
effects (in particular signal droop) if transmitted over AC coupled lines, Table 6.1.
As pointed out in section 6.4.1 polar signalling has a significant power (or
alternatively BER) advantage over unipolar signalling. Fundamentally, this is
because the pulses in a unipolar scheme are only orthogonal whilst the pulses in a
polar scheme are antipodal. Another way of explaining the difference in
performance is to observe that the average or DC level transmitted with unipolar
signals contains no information and is therefore wasted power.
Polar binary signalling also has the advantage that, providing the symbols are equiprobable, the decision threshold is 0 V. This means that no automatic gain control
(AGC) is required in the receiver
Dipolar signalling
Dipolar signalling is designed to produce a spectral null at 0 Hz. This makes it
especially well suited to AC coupled transmission lines. The symbol interval, T o, is
split into positive and negative pulses each of width T o/2 s, Figure 6.11. This makes
the total area under either pulse type equal to zero which results in the desirable
DC null in the signals spectrum. Both OOK and antipodal forms of dipolar
signalling are possible, the latter being called split phase or Manchester coding
(Figure 6.11). A spectral line at the clock frequency (1/ToHz) is present in the OOK
form but absent in the antipodal form. Manchester coding is widely used for the
distribution of clock signals within very large scale integrated (VLSI) circuits,
for magnetic recording and for Ethernet LANs (see Chapter 21).
Bipolar alternate mark inversion signalling
Bipolar signalling (also called alternate mark inversion, AMI) uses three voltage
levels (+V, 0, V) to represent two binary symbols (0 and 1) and is therefore a
pseudo-ternary line code. Zeros, as in unipolar signalling, are represented by the
absence of a pulse (i.e.0 V) and ones (or marks) are epresented alternately by
voltage levels of + V and V .
Both RZ and NRZ forms of bipolar signalling are possible, Figure 6.11, although the
RZ form is more common. Alternating the mark voltage level ensures that the
bipolar spectrum has a null at DC and that signal droop on AC coupled lines is
avoided. The alternating mark voltage also gives bipolar signalling a single error
23

detection capability and reduces its bandwidth over that required for the
equivalent unipolar or polar format (see Figure 6.11).

24

Figure 6.11 Selection

of commonly used line code symbols (0, 1) and associated spectra

25

Figure 6.12 Various line code waveforms for PCM bit sequence.

Baseband Centre Point Detection

26

The detection of digital signals involves two processes:

1. Reduction of each received voltage pulse (i.e. symbol) to a single numerical
value.
2. Comparison of this value with a reference voltage (or, for multisymbol signalling,
a setof reference voltages) to determine which symbol was transmitted.
In the case of symbols represented by different voltage levels the simplest way of
achieving 1 is to sample the received signal plus noise; 2 is then implemented
using one or more comparators. In the case of equiprobable, binary, symbols (zero
and one) represented by two voltage levels (e.g. 0 V and 3 V) intuition tells us that
a sensible strategy would be to set the reference, V ref (Figure 6.1(b)), mid-way
between the two voltage levels (i.e. at 1.5 V). Decisions would then be made at the
centre of each symbol period on the basis of whether the instantaneous voltage
(signal plus noise) is above or below this reference. Sampling the instantaneous
signal plus noise voltage somewhere near the middle of the symbol period is called
centre point detection , Figure 6.1(a).

The noise present during detection is often either Gaussian or approximately

Gaussian.(This is always the case if thermal noise is dominant, but may also be the
case when other sources of noise dominate owing to operation of the central limit
theorem.) Since noise with a Gaussian probability density function (pdf) is common
and analytically tractable, the bit error rate (BER) of a communications system is
often modelled assuming this type of noise alone.

27

Bite Error Ratio/Rate (BER)

The bit error rate or bit error ratio (BER) is the number of bit errors divided by the
total number of transferred bits during a studied time interval. BER is a unitless
performance measure, often expressed as a percentage.

Example
As an example, assume this transmitted bit sequence:
0 1 1 0 0 0 1 0 1 1,
and the following received bit sequence:
0 0 1 0 1 0 1 0 0 1,
The number of bit errors (the underlined bits) is in this case 3. The BER is 3 incorrect bits divided by 10
transferred bits, resulting in a BER of 0.3 or 30%.

Baseband binary error rates in Gaussian noise

Figure6.2(a) shows the pdf of abinary information signa lwhich can take on voltage levels V 0 and V 1 only. Figure
6.2(b) shows the pdf of a zero mean Gaussian noise process, v n(t ),with RMS value V 0andV 1. Figure 6.2(c) is
thus the convolution of Figures 6.2(a) and 6.2(b).) For equiprobable symbols the optimum decision level is set at
(V 0+V 1)/2. (This is not the optimum threshold if the symbols are not equiprobable:) Given that the symbol 0 is
transmitted (i.e. a voltage level V 0) then the probability,Pe1,that the signalplus noise will be above the threshold at
the decision instant is given by twice [ p0(vn) and p1(vn), as dened here, each represent a total probability of 0.5.
Strictly speaking they are not,therefore, pdfs although their sum is the total signal plus noise pdf irrespective of
whether a one or zero is transmitted.

28

The pdf of the signal plus noise conditional on a zero being transmitted is 2 p 0(vn).] the shaded area under the
curve p0(vn)in Figure 6.2(c), i.e.:

29

30

31

Signa
l-to-Quantization-Noise Ratio (SQNR)
The quantisation process in itself, however, actually degrades the quality of the
information signal. This is easy to see since the quantised PAM signal no longer
exactly represents the original, continuous
analogue signal but a distorted version of it. Figure 5.18 (which is drawn with PAM
pulse width, , equal to the sampling period, Ts ) shows that the quantised signal
can be decomposed into the sum of the analogue signal and the difference
between the quantised and the analogue signals. The difference signal is
essentially random and can therefore be thought of as a special type of noise
process. Like any other signal the power or RMS value of this quantisation noise
32

can be calculated or measured. This leads to the concept of a signal to

quantisation noise ratio (SNqR).
Signal-to-Quantization-Noise Ratio (SQNR or SNqR) is widely used quality measure
in analysing digitizing schemes such as PCM (pulse code modulation)
and multimedia codecs. The SQNR reflects the relationship between the maximum
nominal signal strength and the quantization error (also known as quantization
noise) introduced in the analog-to-digital conversion.
To calculate the SNqR of a quantised signal it is convenient to make the following
assumptions:
1. Linear quantisation (i.e. equal increments between quantisation levels).
2. Zero mean signal (i.e. symmetrical pdf about 0 V).
3. Uniform signal pdf (i.e. all signal levels equally likely).
SNqR for linear PCM

Whilst it is true that PCM signals are more tolerant of noise than the equivalent
quantised PAM signals, it is also true that both suffer the same degradation due to
quantisation noise. For a given number of quantisation levels, M, the number of
binary digits required for each PCM codeword is n = log2 M. The PCM peak signal
to quantisation noise ratio, (SNqR)peak , is therefore:

If the ratio of peak to mean signal power,

average SNqR
is:

, is denoted by then the

For a sinusoidal signal = 2 (or 3 dB). For a (clipped) Gaussianly distributed

random signal (with vpeak/g = 4 where g is the signals standard deviation, or
RMS value, as in section 3.2.5) = 16 (or 12 dB), and for speech = 10 dB. The
SNqR for an n-bit PCM voice system can therefore be estimated using the rule of
thumb 6(n 1) dB.

33

Advantages of PCM

Disadvantages of PCM

EXAMPLE 5.3

A digital communications system is to carry a single voice signal using linearly

quantised PCM. What
PCM bit rate will be required if an ideal anti-aliasing filter with a cut-off frequency
of 3.4 kHz is used
at the transmitter and the SNqR is to be kept above 50 dB?
From equation (5.23):
SNqR = 4.8 + 6n dB
For voice signals = 10 dB, i.e.:

34

10 bit/sample are therefore required. The sampling rate required is given by

Nyquists rule, fs 2 fH .
Taking a practical version of the sampling theorem, equation (5.3), gives:
fs = 2.2 3.4 kHz = 7.48 kHz (or k samples/s)
The PCM bit rate (or more strictly binary baud rate) is therefore:
Rb
= fsn
= 7.48 103 10 bit/s
= 74.8 kbit/s

35