Chapter 3 Pulse Modulation

3.1 Introduction
Let g (t ) denote the ideal sampled signal

g (t )   g ( nTs )  (t  nTs ) (3.1)
n 

where Ts : sampling period

f s 1 Ts : sampling rate
From Table A6.3 we have

g(t )   (t  nTs ) 
n 

1 m
G( f ) 
Ts
 ( f
m 

Ts
)

  f sG ( f  mf s )
m 

g ( t )  f s  G( f
m 
 mf s ) (3.2)

or we may apply Fourier Transform on (3.1) to obtain


G ( f )   g ( nTs ) exp(  j 2 nf Ts ) (3.3)
n 

or G ( f )  f sG ( f )  f s  G( f
m 
 mf s ) (3.5)
m 0

If G ( f ) 0 for f W and Ts  1
2W

n j n f
G ( f )   g( ) exp(  ) (3.4)
n  2W W

2
With
1.G ( f ) 0 for f W
2. f s 2W
we find from Equation (3.5) that
1
G( f )  G ( f ) ,  W  f  W (3.6)
2W
Substituting (3.4) into (3.6) we may rewrite G ( f ) as
1 
n jnf
G( f ) 
2W

n 
g(
2W
) exp( 
W
) ,  W  f  W (3.7)

n
g (t ) is uniquely determined by g ( ) for    n  
2W
 n 
or  g ( )  contains all information of g (t )
 2W 
3
  n 
To reconstruct g (t ) from  g ( )  , we may have
 2W 

g (t )  G ( f ) exp( j 2ft )df


W 1 
n j n f

W 2W

n 
g(
2W
) exp( 
W
) exp( j 2 f t )df

n 1 
W n 
  g( )  W exp j 2 f (t  2W ) df (3.8)
n  2W 2W

n sin(2 Wt  n )
  g( )
n  2W 2 Wt  n

n
  g( ) sin c( 2Wt  n ) , -   t   (3.9)
n  2W
(3.9) is an interpolation formula of g (t ) 4
Sampling Theorem for strictly band - limited signals
1.a signal which is limited to  W  f  W , can be completely
 n 
described by  g ( ) .
 2W 
 n 
2.The signal can be completely recovered from  g ( )
 2W 
Nyquist rate 2W
Nyquist interval  1
2W
When the signal is not band - limited (under sampling)
aliasing occurs .To avoid aliasing, we may limit the
signal bandwidth or have higher sampling rate.
5
Figure 3.3 (a) Spectrum of a signal. (b) Spectrum of an undersampled versio
n of the signal exhibiting the aliasing phenomenon.

6
Figure 3.4 (a) Anti-alias filtered spectrum of an information-bearing signal. (b) Spec
trum of instantaneously sampled version of the signal, assuming the use of a samplin
g rate greater than the Nyquist rate. (c) Magnitude response of reconstruction filter. 7
3.3 Pulse-Amplitude Modulation

Let s(t ) denote the sequence of flat - top pulses as


s(t )   m( nTs ) h (t  nTs ) (3.10)
n 

 1, 0 tT
1
h (t )  , t 0, t T (3.11)
2
 0, otherwise

The instantaneously sampled version of m(t ) is

m (t )   m( nTs ) (t  nTs ) (3.12)
n 

m (t )  h (t )  m ( )h (t   )d


 


 m(nT ) (  nT )h(t   )d
n 
s s

 
  m( nTs )   (  nTs )h (t   )d (3.13)

n 

Using the sifting property , we have


m (t )  h (t )   m( nTs )h (t  nTs ) (3.14) 8
n 
The PAM signal s(t ) is
s(t ) m (t )  h(t ) (3.15)
 S ( f )  Mδ ( f ) H ( f ) (3.16)

Recall (3.2) g (t )  fs  G( f  mf )
m 
s (3.2)

M ( f )  f s M( f  k f )
k 
s (3.17)

S( f )  fs  M ( f  k f )H ( f )
k 
s (3.18)

9
 Pulse Amplitude Modulation –
Natural and Flat-Top Sampling

 The circuit of Figure 11-3 is used to illustrate pulse am

plitude modulation (PAM). The FET is the switch used
as a sampling gate.

 When the FET is on, the analog voltage is shorted to gr

ound; when off, the FET is essentially open, so that the
analog signal sample appears at the output.

 Op-amp 1 is a noninverting amplifier that isolates the

analog input channel from the switching function.
Pulse Amplitude Modulation –
Natural and Flat-Top Sampling

Figure 11-3. Pulse amplitude modulator, natural sampling.

Pulse Amplitude Modulation –
Natural and Flat-Top Sampling

 Op-amp 2 is a high input-impedance voltage follower capa

ble of driving low-impedance loads (high “fanout”).
 The resistor R is used to limit the output current of op-am
p 1 when the FET is “on” and provides a voltage division
with rd of the FET. (rd, the drain-to-source resistance, is lo
w but not zero)
Pulse Amplitude Modulation –
Natural and Flat-Top Sampling
 The most common technique for sampling voice in
PCM systems is to a sample-and-hold circuit.
 As seen in Figure 11-4, the instantaneous amplitude of
the analog (voice) signal is held as a constant charge on
a capacitor for the duration of the sampling period Ts.
 This technique is useful for holding the sample constant
while other processing is taking place, but it alters the
frequency spectrum and introduces an error, called
aperture error, resulting in an inability to recover
exactly the original analog signal.
Pulse Amplitude Modulation –
Natural and Flat-Top Sampling
 The amount of error depends on how mach the analog
changes during the holding time, called aperture time.

 To estimate the maximum voltage error possible,

determine the maximum slope of the analog signal and
multiply it by the aperture time T in Figure 11-4.
Pulse Amplitude Modulation –
Natural and Flat-Top Sampling

Figure 11-4. Sample-and-hold circuit and flat-top sampling.

Pulse Amplitude Modulation –
Natural and Flat-Top Sampling

Figure 11-5. Flat-top PAM signals.

Recovering the original message signal m(t) from PAM signal

Where the filter bandwidth is W

The filter output is f s M ( f ) H ( f ) . Note that the
Fourier transform of h(t ) is given by
H ( f ) T sinc( f T ) exp( j f T ) (3.19)
amplitude distortion delay T
2
 aparture effect
Let the equalizer response is
1 1 f
  (3.20)
H ( f ) T sinc( f T ) sin( f T )
Ideally the original signal m(t ) can be recovered completely.
10
 3.4 Other Forms of Pulse Modulation
a. Pulse-duration modulation (PDM)
b. Pulse-position modulation (PPM)
PPM has a similar noise performance as FM.

11
Pulse Width and Pulse Position Modulation

 In pulse width modulation (PWM), the width of

each pulse is made directly proportional to the
amplitude of the information signal.
 In pulse position modulation, constant-width
pulses are used, and the position or time of
occurrence of each pulse from some reference
time is made directly proportional to the
amplitude of the information signal.
 PWM and PPM are compared and contrasted
to PAM in Figure 11-11.
Pulse Width and Pulse Position Modulation

Figure 11-11. Analog/pulse modulation signals.

Pulse Width and Pulse Position Modulation

 Figure 11-12 shows a PWM modulator. This circuit

is simply a high-gain comparator that is switched o
n and off by the sawtooth waveform derived from a
very stable-frequency oscillator.
 Notice that the output will go to +Vcc the instant the
analog signal exceeds the sawtooth voltage.
 The output will go to -Vcc the instant the analog sign
al is less than the sawtooth voltage. With this circui
t the average value of both inputs should be nearly t
he same.
 This is easily achieved with equal value resistors to
ground. Also the +V and –V values should not excee
d Vcc.
Pulse Width and Pulse Position Modulation

Figure 11-12. Pulse width modulator.

Pulse Width and Pulse Position Modulation

 A 710-type IC comparator can be used for positive-only o

utput pulses that are also TTL compatible. PWM can also
be produced by modulation of various voltage-controllabl
e multivibrators.

 One example is the popular 555 timer IC. Other (pulse ou

tput) VCOs, like the 566 and that of the 565 phase-locked
loop IC, will produce PWM.
 This points out the similarity of PWM to continuous anal
og FM. Indeed, PWM has the advantages of FM---consta
nt amplitude and good noise immunity---and also its disa
dvantage---large bandwidth.
 Demodulation
 Since the width of each pulse in the PWM signal
shown in Figure 11-13 is directly proportional to the
amplitude of the modulating voltage.
 The signal can be differentiated as shown in Figure
11-13 (to PPM in part a), then the positive pulses are
used to start a ramp, and the negative clock pulses
stop and reset the ramp.
 This produces frequency-to-amplitude conversion (or
equivalently, pulse width-to-amplitude conversion).
 The variable-amplitude ramp pulses are then time-
averaged (integrated) to recover the analog signal.
Pulse Width and Pulse Position Modulation

Figure 11-13. Pulse position modulator.

 Demodulation
 As illustrated in Figure 11-14, a narrow clock pulse
sets an RS flip-flop output high, and the next PPM
pulses resets the output to zero.

 The resulting signal, PWM, has an average voltage

proportional to the time difference between the
PPM pulses and the reference clock pulses.
 Time-averaging (integration) of the output
produces the analog variations.
 PPM has the same disadvantage as continuous
analog phase modulation: a coherent clock
reference signal is necessary for demodulation.
 The reference pulses can be transmitted along with
the PPM signal.
 Demodulation

 This is achieved by full-wave rectifying the PPM pulses

of Figure 11-13a, which has the effect of reversing the p
olarity of the negative (clock-rate) pulses.

 Then an edge-triggered flipflop (J-K or D-type) can be

used to accomplish the same function as the RS flip-flo
p of Figure 11-14, using the clock input.

 The penalty is: more pulses/second will require greater

bandwidth, and the pulse width limit the pulse deviatio
ns for a given pulse period.
Demodulation

Figure 11-14. PPM demodulator.

Pulse Code Modulation (PCM)

 Pulse code modulation (PCM) is produced by analog-t

o-digital conversion process.
 As in the case of other pulse modulation techniques, the
rate at which samples are taken and encoded must conf
orm to the Nyquist sampling rate.

 The sampling rate must be greater than, twice the high

est frequency in the analog signal,
fs > 2fA(max)
 3.6 Quantization Process

Define partition cell

J k : mk  m mk 1 , k 1,2,, L (3.21)
Where mk is the decision level or the decision threshold.
Amplitude quantization : The process of transforming the
sample amplitude m( nTs ) into a discrete amplitude
 ( nTs ) as shown in Fig 3.9
If m(t )  J k then the quantizer output is νk where νk , k 1,2,, L
are the representation or reconstruction levels , mk 1  mk is the step size.
The mapping  g(m ) (3.22)
is called the quantizer characteristic, which is a staircase function.
12
Figure 3.10 Two types of quantization: (a) midtread and (b) midrise.
13
 Quantization Noise

Figure 3.11 Illustration of the quantization process. (Adapted

from Bennett, 1948, with permission of AT&T.) 14
Let the quantization error be denoted by the random
variable Q of sample value q
q m   (3.23)
Q  M  V , ( E [ M ] 0) (3.24)
Assuming a uniform quantizer of the midrise type
2m max
the step - size is   (3.25)
L
 m max  m  m max , L : total number of levels
 1  
,  q
f Q ( q )   2 2 (3.26)
 0, otherwise
 
1
 Q2 E[Q 2 ]  2
 q 2 f Q ( q)dq  2 q 2dq

2  2
2
 (3.28)
12
When the quatized sample is expressed in binary form,
L 2 R (3.29)
where R is the number of bits per sample
R log 2 L (3.30)
2m max
 R (3.31)
2
1 2  2R
 Q  mmax 2
2
(3.32)
3
Let P denote the average power of m(t )
P
 ( SNR ) o  2
Q
3P 2 R
( 2 )2 (3.33)
mmax
(SNR) o increases exponentia lly with increasing R (bandwidth ).
Conditions for Optimality of scalar Quantizers
Let m(t) be a message signal drawn from a stationary process M(t)
-A  m  A

m1= -A
mL+1=A
mk  mk+1 for k=1,2,…., L
The kth partition cell is defined as
Jk: mk< m  mk+1 for k=1,2,…., L
d(m,vk): distortion measure for using vk to represent values inside Jk.
Find the two sets  k k 1 and J k k 1 , that minimize
L L

the average distortion

L
D   d ( m, k ) f M ( m )dm (3.37)
mJ k
k 1

where f M ( m ) is the pdf

The mean - square distortion is used commonly
d ( m,k ) ( m  k ) 2 (3.38)
The optimization is a nonlinear problem which
may not have closed form solution.However the
quantizer consists of two components : an encoder
characteri zed by J k , and a decoder characteri zed by  k
Condition1 . Optimality of the encoder for a given decoder
Given the set  k k 1 , find the set J k k 1 that minimizes D .
L L

That is to find the encoder defined by the nonlinear mapping

g(m )  k , k 1,2, , L (3.40)
such that we have
A L
D  d ( m, g ( m )) f M ( m )dm  
A mJ k
min d (m,k ) f M (m)dm (3.41)
k 1

To attain the lower bound , if

d ( m, k ) d ( m, j ) holds for all j k (3.42)
This is called nearest neighbor condition.
Condition 2 .Optimality of the decoder for a given encoder
Given the set J  , find the set  
L L
k k 1 k k 1 that minimized D .
For mean - square distortion
L 2
D   ( m   k ) f M ( m )dm, f M ( m ) (3.43)
mJ k
k 1

D L 2
 2  ( m   k ) f M ( m )dm 0 (3.44)
 k k 1
mJ k

 k , opt 
mJ k
m f M ( m )dm
(3.45)
 mJ k
f M ( m )dm

Probability Pk (given)
 E M mk  m mk  1 (3.47)
Using iteration, until D reaches a minimum
 Pulse Code Modulation

Figure 3.13 The basic elements of a PCM system.

Quantization (nonuniform quantizer)
 - law
log(1   m )
  (3.48)
log(1   )
d m log(1   )
 (1   m ) (3.49)
d 
A - law
 A( m ) 1
 1  log A 0  m 
  A (3.50)
1  log( A m ) 1
  m 1
 1  log A A
 1  log A 1
dm  0 m 
 A A (3.51)
d 1
(1  A) m  m 1
 A
Figure 3.14 Compression laws. (a)  -law. (b) A-law.
Encoding
 Line codes:

1. Unipolar nonreturn-to-zero (NRZ) Signaling

2. Polar nonreturn-to-zero(NRZ) Signaling
3. Unipor nonreturn-to-zero (RZ) Signaling
4. Bipolar nonreturn-to-zero (BRZ) Signaling
5. Split-phase (Manchester code)
Figure 3.15 Line codes for the electrical representations of binary data.
(a) Unipolar NRZ signaling. (b) Polar NRZ signaling.
(c) Unipolar RZ signaling. (d) Bipolar RZ signaling.
(e) Split-phase or Manchester code.
Differential Encoding (encode information in terms of signal
transition; a transition is used to designate Symbol 0)

Regeneration (reamplification, retiming, reshaping )

Two measure factors: bit error rate (BER) and jitter.

Decoding and Filtering
 3.8 Noise consideration in PCM systems
(Channel noise, quantization noise)
(will be discussed in Chapter 4)
 Time-Division Multiplexing

Figure 3.19 Block diagram of TDM system.

Synchronization
Example 2.2 The T1 System
3.10 Digital Multiplexers
3.11 Virtues, Limitations and Modifications of PCM
Advantages of PCM
1. Robustness to noise and interference
2. Efficient regeneration
3. Efficient SNR and bandwidth trade-off
4. Uniform format
5. Ease add and drop
6. Secure
 3.12 Delta Modulation (DM) (Simplicity)

Let mn m(nTs ) , n 0,1,2, 

where Ts is the sampling period and m(nTs ) is a sample of m(t ).
The error signal is
en mn  mq n  1 (3.52)
eq n  sgn(en ) (3.53)
mq n mq n  1  eq n (3.54)
where mq n is the quantizer output , eq n is
the quantized version of en , and  is the step size
Figure 3.23 DM system. (a) Transmitter. (b) Receiver.
The modulator consists of a comparator, a quantizer, and an accumulator
The output of the accumulator is
n
mq n   sgn(ei )
i 1
n
 eq i  (3.55)
i 1

Two types of quantization errors :

Slope overload distortion and granular noise
 Slope Overload Distortion and Granular Noise
Denote the quantization error by qn  ,
mq n  mn   qn  (3.56)
Recall (3.52) , we have
en  mn   mn  1  qn  1 (3.57)
Except for qn  1, the quantizer input is a first
backward difference of the input signal( differentiator )
To avoid slope - overload distortion , we require
 dm(t )
(slope) max (3.58)
Ts dt
On the other hand, granular noise occurs when step size
 is too large relative to the local slope of m(t ).
Delta-Sigma modulation (sigma-delta modulation)
The    modulation which has an integrator can
relieve the draw back of delta modulation (differentiator)
Beneficial effects of using integrator:
1. Pre-emphasize the low-frequency content
2. Increase correlation between adjacent samples
(reduce the variance of the error signal at the quantizer input )
3. Simplify receiver design
Because the transmitter has an integrator , the receiver
consists simply of a low-pass filter.
(The differentiator in the conventional DM receiver is cancelled by t
he integrator )
Figure 3.25 Two equivalent versions of delta-sigma modulation system.
3.13 Linear Prediction (to reduce the sampling rate)
Consider a finite-duration impulse response (FIR)
discrete-time filter which consists of three blocks :
1. Set of p ( p: prediction order) unit-delay elements (z-
1
)
2. Set of multipliers with coefficients w1,w2,…wp
3. Set of adders (  )
The filter output (The linear predition of the input ) is
p
xˆ n   wk x ( n  k ) (3.59)
k 1

The prediction error is

en   xn   xˆ n  (3.60)
Let the index of performance be
 
J  E e 2 n  (mean square error) (3.61)
Find w1 , w2 , , w p to minimize J
From (3.59) (3.60) and (3.61) we have
p
 
J  E x 2 n   2 wk E  xn xn  k 
k 1
p p
  w j wk E xn  j xn  k  (3.62)
j 1 k 1
Assume X (t ) is stationary process with zero mean ( E[ x[n ]] 0)
 X2 E x 2 n   ( E xn )2

 E x 2 n  
The autocorrelation
RX ( kTs )  RX k   E xn xn  k 
We may simplify J as
p p p
J  X2  2 wk RX k    w j wk RX k  j  (3.63)
k 1 j 1 k 1
p
J
 2 RX k   2 w j RX k  j  0
wk j 1
p

 w R k  j  R k  R  k  , k 1,2, ,p (3.64)
j 1
j X X X

(3.64) are called Wiener - Hopf equations

For convenience,
1
we may rewrite
1
the Wiener-Hopf equations
as , if R X exists w 0 R X rX (3.66)
w 0 w1 , w2 , , w p 
T
where
rX [ R X [1], R X [2],..., R X [ p ]]T
 R X 0 R X 1  R X  p  1
 R 1 R X 0  R X  p  2
RX   X 
    
 
 R X  p  1 R X  p  2  R X 0 
The filter coefficients are uniquely determined by
R X 0 , R X 1 , , R X  p 
Substituting (3.64) into (3.63) yields
p p
J min  2
X  2  wk R X k   wk R X k 
k 1 k 1
p
 2
X  w k R X k 
k 1

 X2  rXT w 0  X2  rXT R X1rX (3.67)

 rXT R X1rX 0,  J min is always less than  X2
 Linear adaptive prediction (If R k  for varying k is not available)
X

The predictor is adaptive in the follow sense

1. Compute wk , k 1,2, , p, starting any initial values
2. Do iteration using the method of steepest descent
Define the gradient vector
J
gk  , k 1,2 , ,p (3.68)
wk
wk n  denotes the value at iteration n . Then update wk n  1
1
wk n  1 wk n   g k , k 1,2 , ,p (3.69)
2
1
where  is a step - size parameter and is for convenience
2
of presentation.
Differentiating (3.63), we have
J P
gk   2 RX k   2 w j RX k  j 
wk j 1
p
 2 E  xn xn  k   2 w j E  xn  j xn  k  , k 1,2, , p (3.70)
j 1

To simplify the computing we use xn xn  k  for E[x[n]x[n - k]]

(ignore the expectation)
p
gˆ k n   2 xn xn  k   2 w j n xn  j xn  k  , k 1,2, , p (3.71)
j 1

Substituting (3.71) into (3.69)

 p

wˆ k n  1 wˆ k n   xn  k  xn    wˆ j n xn  j 
 j 1 
wˆ k n   xn  k en  , k 1,2, , p (3.72)
p
where en   xn    wˆ nxn  j 
j by (3.59)  (3.60) (3.73)
j 1

The above equations are called lease - mean - square algorithm

Figure 3.27
Block diagram illustrating the linear adaptive prediction process.
 3.14 Differential Pulse-Code Modulation (DPCM)
Usually PCM has the sampling rate higher than the Nyquist rate .Th
e encode signal contains redundant information. DPCM can efficientl
y remove this redundancy.

Figure 3.28 DPCM system. (a) Transmitter. (b) Receiver.

Input signal to the quantizer is defined by:
en  mn   mˆ n  (3.74)
mˆ n  is a prediction value.
The quantizer output is
eq n  en   qn  (3.75)
where qn  is quantization error.
The prediction filter input is
mq n  mˆ n   en   qn  (3.77)
From (3.74)
mn 
 mq n  mn   qn  (3.78)
 Processing Gain
The (SNR)o of the DPCM system is
 M2
(SNR)o  2 (3.79)
Q
where  M2 and  Q2 are variances of mn  ( E [m[n ]] 0) and qn 
 M2  E2
(SNR)o ( 2 )( 2 )
 E Q
G p (SNR )Q (3.80)
where  E2 is the variance of the predictions error
and the signal - to - quantization noise ratio is
 E2
(SNR )Q  2 (3.81)
Q
 M2
Processing Gain, G p  2 (3.82)
E
Design a prediction filter to maximize G p (minimize  E2 )
3.15 Adaptive Differential Pulse-Code Modulation (ADPCM)
Need for coding speech at low bit rates , we have two aims in mind:
1. Remove redundancies from the speech signal as far as possible.
2. Assign the available bits in a perceptually efficient manner.

Figure 3.29 Adaptive quantization with backward estimation (AQB).

Figure 3.30 Adaptive prediction with backward estimation (APB).