Pulse Code Modulation (PCM)

• PCM -> analog-to-digital (ADC) conversion.

Instantaneous samples of an analog signal are
represented by digital words in a serial bit stream.
• 3 main steps: Sampling (i.e., flat-top PAM), Quantizing
(fixed number of levels is allowed), and Encoding (binary
digital word).

Block Diagram of a PCM Modulator

flat-top PAM quantized binary digital

samples codeword

1-Apr-16 Lecture 11, ELG3175: Introduction to Communication Systems © S. Loyka 1(22)

 PCM Modulator
• Analog signal x(t) is bandlimited to Fmax . If not, LPF is used
(pre-sampling filter).
• The sampling is done at higher than Nyquist rate -> guard
band, f s = 2 Fmax + ∆f . Usually – flat-top PAM.
• Quantizing: the sample level is rounded off to the closest
allowed level (only a fixed finite number of levels are
allowed).
• Encoding: each allowed (quantized) level is represented by
a (unique) binary code word.
• Serial transmission is used for binary digits. Thus, higher
bandwidth is required.

1-Apr-16 Lecture 11, ELG3175: Introduction to Communication Systems © S. Loyka 2(22)

 Quantizing
• The exact sample value x ( nTs ) is replaced by the closest value
allowed. The infinite number of levels is transformed into a finite
number of levels.
• Uniform quantizing: all steps are equal.
• The number of bits required to transmit
each sample: R1 = log 2 N , N – the
number of quantized levels.
• Transmission rate [bit/s]: R = f s log 2 N
• Example: 8-level quantizer. x – input
analog signal, xɵ k - quantized signal.
R1=3 bits, R = 3 f s

• Reversible ? quantization
interval

1-Apr-16 Lecture 11, ELG3175: Introduction to Communication Systems © S. Loyka 3(22)

 Quantizing: Example quantized
signal
quantizer characteristic

quantization error (noise)

output PCM signal

Example: sinusoid + binary quantizing.

1-Apr-16 Lecture 11, ELG3175: Introduction to Communication Systems © S. Loyka 4(22)
 Quantization Noise & SQNR
• The quantization function is noninvertible -> some information is
lost. The effect is described using quantization noise.
• Mean square error (distance):
∞
 2
x − xɵ ) 2ρΧ ( x ) dx, xɵ = Q( x )
D = E ( x − Q( x ) ) = ∫ (

  −∞ ε
1 2
D = ∑ i ∫ x − x ρi ( x ) dx, ρΧ ( x ) - pdf of x
(
ɵ
)
N ∆xi

• Definition of quantization noise (error signal) power:

1 T
∫−T
const D
Pq = lim D (t )dt  →D
T →∞ 2T

• Signal to quantization noise ratio: SQNR = Px / Pq

• Note: for random stationary x(t), the power is the variance:
∞
Px = E  x  = ∫ x 2ρ Χ ( x ) dx
2
−∞
1-Apr-16 Lecture 11, ELG3175: Introduction to Communication Systems © S. Loyka 5(22)
 Uniform PCM
• The input signal range: x ∈ [ − xmax , + xmax ]
• All the quantization intervals are equal: ∆xi = ∆ = 2 xmax / N
• When N is large, ∆ is small and the error ε = x − Q ( x ) is
uniformly distributed within [−∆ / 2, +∆ / 2] for each
quantization interval: ρ ε ( ε ) = 1/ ∆
• Quantization noise power is ∆ 2 2
1 1 ∆ x
Pq = D = ∑ i ∫ ε 2ρε ( ε ) d ε= ∫ 2∆ ε 2 d ε = = max2
N ∆xi ∆ − 12 3 N
2
Px 3N 2 Px
• SQNR is SQNR = = 2 ≤ 3N 2 peak SQNR
Pq xmax
2
xmax 3N 2
• Peak factor β = ≥1 SQNR = ≤ 3N 2
Px β

1-Apr-16 Lecture 11, ELG3175: Introduction to Communication Systems © S. Loyka 6(22)

 Uniform PCM & SQNR
• Log form of the uniform SQNR law (6 dB law):

SQNR dB ≈ −β dB +6v + 4.8

• where N = 2 ν , ν - the number of bits.
• Each extra bit adds 6 dB to SQNR.
• Example: x ∈ [ −1, +1] , uniform PCM with 256 levels.
Find SQNR. 2
1 1 2 1 x
• ν = log N = 8, Px =
∫
2 −1
x dx =
3
, β = max
Px
=3

SQNR = 3N 2 / β ≈ 6.6 ⋅ 104 ≈ 48 dB

• Homework: do the same for xmax=2 and x ∈ [ −1, +1] .
Compare with the result above and make conclusions.
1-Apr-16 Lecture 11, ELG3175: Introduction to Communication Systems © S. Loyka 7(22)
 Bandwidth of PCM
• If using rectangular pulses, absolute bandwidth is infinite. Power
bandwidth is finite.
• Non-rectangular (“rounded”) pulses may be used to transmit digital
codewords (110100..), which are bandlimited.
• Fundamental limit is obtained using the sampling theorem. The
minimum number of samples for a perfect reconstruction of a
bandlimited signal is fs /second. If N quantization levels are used,
then
Nyquist
R = f s log 2 N R = 2 Fmax log 2 N [bit/s]

• The minimum bandwidth to transmit R bits/s using binary mod. is

R/2 (sampling theorem again! See Lec. 12 for more details). Hence,

1 Nyquist
∆f min = f s log 2 N ∆f min = Fmax log 2 N
2
1-Apr-16 Lecture 11, ELG3175: Introduction to Communication Systems © S. Loyka 8(22)
 Example: PCM for Telephone System
• Telephone spectrum: [300 Hz, 3400 Hz]
• Min. sampling frequency: f s ,min = 2 Fmax = 6.8 kHz (or [sam./s])
• Some guard band is required: f s = 2 Fmax + ∆f g = 8 kHz
• 8-bit codewords are used -> N=256.
• The transmission rate: R = f s ν = 64 kbit/s
• Minimum absolute bandwidth: ∆f min = R / 2 = 32 kHz
• Peak SQNR:
SQNR = 3N 2 ≈ 2 ⋅ 105 ≈ 53 dB
• Another example: CD player (see the text by Proakis and Salehi
(2nd ed.), section 6.8).

1-Apr-16 Lecture 11, ELG3175: Introduction to Communication Systems © S. Loyka 9(22)

 Nonuniform PCM
• Uniform PCM is good for uniform signal distributions, but not
efficient for nonuniform ones.
• Example: speech signal has large probability of small values and
small prob. of large ones.
• Solution: allocate more levels for small amplitudes and less for
large. Total quantizing noise is greatly reduced (see equations
above).
• Typical solution for nonuniform PCM modulator: compress signal
first, then apply uniform PCM. Rx end: demodulate uniform PCM
and expand it. The technique is called companding.

1-Apr-16 Lecture 11, ELG3175: Introduction to Communication Systems © S. Loyka 10(22)

 µ-Law Companding (Speech)
• Logarithmic function is used, x ≤ 1 , log (1 + µ x )
where µ controls the amount of g ( x) = sgn ( x )
log (1 + µ )
compression.
Input-output characteristic
• Used in US & Canada
( µ = 255 ), + a uniform
128 levels (7 bits)
quantizer.
compressed
• The compander
improves SQNR by
approx. 24 dB.
expanded

1-Apr-16 Lecture 11, ELG3175: Introduction to Communication Systems © S. Loyka 11(22)

 Lecture 11
Differential PCM
• Samples of a bandlimited signal are correlated -> previous
sample gives information about the next one. Example: if
previous samples are small, the next one will be small with
high probability.
• This can be used to improve PCM performance: to
decrease the number of bits used (and, hence, the
bandwidth) or to increase SQNR for a given bandwidth.
• Main idea: quantize and transmit the difference between
two adjacent samples rather than sample values.
• Since two adjacent samples are correlated (bandlimited
signal!), their difference is small and requires less bits to
transmit.

1-Apr-16 Lecture 11, ELG3175: Introduction to Communication Systems © S. Loyka 12(22)

 Simple DPCM System
Modulator
∆n
x (t ) xn yn
+ ..01101..
LPF Sampler Σ Quantizer Encoder
- xn −1
Delay
T=1 ∆ n = xn − xn −1 , yn = Q ∆ n  = ∆ n + ε n

n n n
xn = ∑ yk = ∑ ∆ k + ∑ ε k
⌢

Demodulator k = 0 ⌢ k =0 k⌢= 0 ⌢
= yn + xn −1 → yn = xn − xn −1
..01101..
yn ⌢
xn ⌢
x (t )
Decoder Σ⌢ LPF Very good quantization: yn ≈ ∆ n
xn −1 ⌢
Delay
Hence, n and xn satisfy the
x
T=1 same difference equation -> must
be the same!
Problem: quantization noise accumulation.
1-Apr-16 Lecture 11, ELG3175: Introduction to Communication Systems © S. Loyka 13(22)
 Quantization Noise Accumulation
Demodulator

..01101..
yn ⌢
xn ⌢
x (t ) n x
Decoder Σ⌢ LPF

n n n
xn −1
xn = ∑ y k = ∑ ∆ k + ∑ ε k
⌢
Delay
k =0 k =0 k =0
T=1

n
∑ εk
⌢
∆xn = xn − xn =
Includes past Q. noise contributions!
k =0

2
P {∆xn } = ∆xn
n n
Noise power is always added, never
2 subtracted! (assuming independence)
= ∑ εk = ∑ P {εk }
k =0 k =0

1-Apr-16 Lecture 11, ELG3175: Introduction to Communication Systems © S. Loyka 14(22)

 Improved DPCM System
No quantization noise accumulation.

J.Proakis, M.Salehi, Communications Systems Engineering, Prentice Hall, 2002

• Analysis: very good quantizer

Yn = X n − Yˆn′−1
 Yn = X n − Yˆn′−1 ≈ Yˆn = Yˆn′ − Yˆn′−1 → X n ≈ Yˆn′
 Yˆn′ = Yˆn + Yˆn′−1
Xˆ n = Yˆn + Xˆ n −1 Xˆ n and Yˆn′ satisfy the same difference equation
n
-> must be the same -> Xˆ n = Yˆn′ ≈ X n
Xˆ n = ∑ Yˆi In general, ε n = Yˆn − Yn = Xˆ n − X n
i =0
1-Apr-16 Lecture 11, ELG3175: Introduction to Communication Systems © S. Loyka 15(22)
 Delta Modulation
• This is a simplified version of DPCM. A 1-bit, 2 level -> ±∆
quantizer is used.

J.Proakis, M.Salehi, Communications Systems Engineering, Prentice Hall, 2002

• Since there are only 2 levels, the Yn dynamic range must be

low to keep quantization noise low.
• This, in turn, means that Xn and Xn-1 must be highly correlated
-> sampling frequency must be much higher than the Nyquist
rate.

1-Apr-16 Lecture 11, ELG3175: Introduction to Communication Systems © S. Loyka 16(22)

 Delta Modulation
• Despite of the high sampling frequency, transmission rate is
low (less than for PCM) because there is only 1 bit/sample to
transmit.
• Major advantage -> simple structure.

may be an integrator
or a counter

• Major disadvantage: granular noise and slope-overload

distortion

1-Apr-16 Lecture 11, ELG3175: Introduction to Communication Systems © S. Loyka 17(22)

Granular Noise and Slope-Overload Distortion
large ∆ -> granular noise small ∆ -> slope-overload distortion

original analog signal

demodulated signal
demodulated signal
original analog signal

 Step size is very important.

 Small step size results in slope-overload distortion.
 Large step size results in granular noise.
 Solution: adaptive delta-modulation.

1-Apr-16 Lecture 11, ELG3175: Introduction to Communication Systems © S. Loyka 18(22)

 Example: DM for Sinusoidal Signal
• Maximum slope generated by
DM demodulator is sm = ∆ / Ts = ∆ ⋅ f s
• For a sinusoidal input, the d
sin = x(t ) = Aωin cos ωin t
slope is dt
• The maximum input slope is sin,max = Aωin
• No slope overload distortion if sm ≥ sin ,max → ∆ ≥ 2 πAf in / f s
• SQNR if no overload distortion 3 f s3
(see the text): SQNR = 2 2
8π f in f LPF

• Example: x (t ) = sin 2π103 t , f s = 10 kHz, f LPF = 2 kHz

∆ ≥ 2π103 /104 ≈ 0.6, SQNR ≈ 13 dB

1-Apr-16 Lecture 11, ELG3175: Introduction to Communication Systems © S. Loyka 19(22)

 Granular Noise and Slope-Overload
Distortion

Couch, Digital and Analog Communication Systems, Pearson Education.

1-Apr-16 Lecture 11, ELG3175: Introduction to Communication Systems © S. Loyka 20(22)
 Adaptive Delta-Modulation
• Main idea: change step size according to changes in the
input signal.
• If the input changes rapidly -> large step size. If the input
changes slowly -> small step size.
• How to implement step size
change?
• Simple solution: if two
slow variation
successive outputs have
the same sign -> increase
step size; if they are of
rapid variation
opposite sign -> decrease
step size.

1-Apr-16 Lecture 11, ELG3175: Introduction to Communication Systems © S. Loyka 21(22)

 Lecture 11

Summary
 Pulse code modulation (PCM): Sampling, quantizing and encoding.
 Uniform quantizing. SQNR. Nonuniform quantizing.
• Differential PCM. Block diagrams (simple and improved).
Quantization noise accumulation.
• Delta modulation. Block diagrams.
• Granular noise and slope-overload distortion. Limitation on the step
size.
• Comparison of PCM and delta modulation.
• Adaptive delta-modulation.

Homework: Reading: Couch, 3.1-3.3, 3.7, 3.8. Study

carefully all the examples and make sure you understand
them.

1-Apr-16 Lecture 11, ELG3175: Introduction to Communication Systems © S. Loyka 22(22)