L19: Introduction To Digital Communication: (P&S Pp. 7-12, 45-49, 192-194, 290-298, 302-305 C&C Pp. 263-6, 544-54)

L19: Introduction to Digital Communication
(P&S pp. 7-12, 45-49, 192-194, 290-298, 302-305;

C&C pp. 263-6, 544-54)
In analog comms, parameters of carrier

(in-phase & quadrature amplitudes, inst. freq., phase etc)
are varied linearly/LTI’ly with message.
Typically, message is usually analog.
Design objective is to make demodulated signal “look like”

original message waveform in the presence of noise &
channel distortion.
L19 1
The Digital Approach
For computer data etc, this can be too stringent an aim.

E.g., if we know that M (t)  5 V, a received value of 1V will
very likely correspond to M (t)  5V.
I.e., because of the discrete-valued message, the receiver
can in principle tolerate more noise & distortion.
If the message is analog, converting it to digital form will

yield the same robustness to channel noise & distortion.
Q: What’s the price…?
L19 2
Digital Communication System (Fig 1.1, Proakis)
L19 3
Other Advantages of Digital Comms
 In long-distance comms, repeaters can be used to

regenerate the attenuated signal at regular distances
without passing on amplified noise.
 Can compress data to remove unwanted redundancy &
improve bandwidth & power utilisation (source coding)
 Can introduce redundancy in a carefully controlled way
to counteract channel noise (channel coding)
 Voice, video & data can be transmitted over a common
digital system
 Low cost of digital circuits
 Encryption to protect privacy
L19 4
Disadvantages…?
 Receiver’s clock needs to be synchronised, typically regenerated

from received data stream
 Bandwidth requirements are usually high
Most common format is pulse amplitude modulation:
s(t )   n  d n p(t  nT )


{d n } : discrete-valued amplitudes, each representing a data symbol

T : sampling period
p(.) : signalling waveform (pulse )
When the d n ’s arise from sampling & quantising (A/D conversion) an

analog signal, overall scheme is called pulse-code modulation.
L19 5
Ideal Samplers
A sampler is a device that measures an input analog

waveform x(t) once every Ts s & outputs (the equivalent of)
a discrete-time sequence. { x (nTs )} n
In continuous time, it turns out to be mathematically

convenient to represent the output of an ideal sampler as
 
x (t ) : x (t )   (t  nTs )   x (nT ) (t  nT )
s s
n  n 
I.e. a modulated sequence of impulses.
L19 6
Impulse Train in Fourier Series Form

As i(t) :   (t  mTs ) is periodic, it has a Fourier
m 
series i (t )   c e j 2nt / T
n   n s
where n  0,1,2,,
1 Ts / 2 1  Ts / 2
cn   i (t )e  j 2nt / Ts
dt     (t  mTs )e j 2nt / Ts dt
Ts Ts / 2 Ts m   Ts / 2
1 Ts / 2 1 1
   (t  0  Ts )e j 2nt / Ts dt  e j 2n.0 / Ts  .
Ts Ts / 2 Ts Ts
L19 7
Spectrum of Impulse-Sampled Signal

     f  n 

 I ( f )   n   cn Fe
 j 2nt / Ts 1
Ts n    Ts 
As x (t)  x(t)i(t),
1   n
X  ( f )  X ( f ) * I ( f )   X ( f ) *   f  
Ts n    Ts 
1    n
   X ( f  u )  u  du
Ts n      Ts 
1   n
  X  f  
Ts n    Ts 
i.e., superposition of translated original spectra
L19 8
The Sampling Theorem
If original signal x(t) is strictly bandlimited to ( W ,W ) &

we choose f s : 1/Ts  2W (Nyquist sampling
rate) then translated versions of spectra will not overlap.
 Can apply ideal LPF to recover original analog signal

I.e. we don’t lose information by sampling a bandlimited
continuous-time signal above its Nyquist rate.
If signal is not bandlimited or is undersampled, then

aliasing distortion occurs.
L19 9
Quantisation
Sampling = quantising or discretising the time axis.
However, the sampled signal may still be continuous-valued.
Before transmission in a digital system, it needs to be

converted into discrete values: quantisation.
The discretised samples are then encoded into bits.
Inevitably, quantisers cause loss of resolution in data

However, if appropriately designed, then the quantisation
noise so added is outweighed by gains in output SNR.
L19 10
Quantiser Definition
Mathematically, a quantiser maps a real input x to some

finite-valued xˆ : Q ( x)  {xˆ1 ,., xˆ1}  R Typically,
 x̂1 if x  a1
 x̂ if a1  x  a2
 2
Q(x)  

 x̂N if x  aN  1
N is usually a power of 2 so that the index of the selected

point can be encoded into bits.
Note that quantization is not invertible i.e. information is lost.
L19 11
 x̂1 if x  a1
 x̂ if a1  x  a2
 2
Q(x)  

 x̂N if x  aN  1
L19
(Proakis & Salehi) 12
Quantisation Noise
When x is a realisation of an rv, quantiser quality is measured by

the error power, or mean square error (MSE)
D  E[( X  Q ( X )) 2 ]
If this is minimised, then X & X-Q(x) are uncorrelated.

I.e. the quantiser output decomposes into signal X +
uncorrelated noise Q( X )  X .
Can define signal-to-quantisation-noise ratio

E[ X 2 ]
SQNR :
E[( X  Q ( X )) 2 ]
L19 13
Uniform Quantisation
If X ~ U[a,b) , then D min. by partitioning [a,b) into N
intervals of equal length  : b  a & setting x̂ midpt.

N

 X  Q( X ) ~ U[  / 2,  / 2]  E  X  Q( X ) 
2
 
/2 n2
 / 2 
2 E[ X 2 ]
dn  
12 N2
Even if X is nonuniform, for N  1 we still have
X  Q(X ) ~ U[  / 2,  / 2] roughly, and so

E  X  Q( X ) 
2
 2 (b  a ) 2
 
12 12 N 2
provided P[ X  a  X  b] is comparatively small.
L19 14
Non‐Uniform Quantisation
To minimise quant. noise power, we should thus try to

ba
keep  : small. With N fixed, this means reducing b
N
or increasing a. However, this increases the probability
P[ X  a  X  b] of overload…
 Non-uniform quantisation required.
In general, the optimal quantiser has levels closer together

in areas of high prob., & far apart in areas of low prob.
L19 15
Companding Quantisation
Companding is a way of using a uniform quantiser to generate non-

uniform quantisation:
xˆ  g 1 QU  g ( x) 
g(.) = compressor, g  1 (.) =
expander E.g., in voice transmission,
ln(1   | x |)
‐ Law: g ( x)  sgn( x) (Typ.   255)
ln(1   )
 Ax ,
| x | 1/ A,
 1  ln A
A-law: g(x)   (Typ. A = 87.6)
1  ln(A | x |)
sgn(x), | x | 1/ A.
 1  ln A
(NB: typically, x(t) is normalised to have amplitude =1.)
L19 16
(Proakis & Salehi)
L19 17
(Proakis & Salehi)
L19 18

L19: Introduction To Digital Communication: (P&S Pp. 7-12, 45-49, 192-194, 290-298, 302-305 C&C Pp. 263-6, 544-54)

Uploaded by

Copyright:

Available Formats

L19: Introduction To Digital Communication: (P&S Pp. 7-12, 45-49, 192-194, 290-298, 302-305 C&C Pp. 263-6, 544-54)

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

L19: Introduction To Digital Communication: (P&S Pp. 7-12, 45-49, 192-194, 290-298, 302-305 C&C Pp. 263-6, 544-54)

Uploaded by

Copyright:

Available Formats

L19: Introduction to Digital Communication

(P&S pp. 7-12, 45-49, 192-194, 290-298, 302-305;

In analog comms, parameters of carrier

Typically, message is usually analog.

Design objective is to make demodulated signal “look like”

For computer data etc, this can be too stringent an aim.

If the message is analog, converting it to digital form will

 In long-distance comms, repeaters can be used to

 Receiver’s clock needs to be synchronised, typically regenerated

Most common format is pulse amplitude modulation:

s(t )   n  d n p(t  nT )

{d n } : discrete-valued amplitudes, each representing a data symbol

When the d n ’s arise from sampling & quantising (A/D conversion) an

A sampler is a device that measures an input analog

In continuous time, it turns out to be mathematically

I.e. a modulated sequence of impulses.

If original signal x(t) is strictly bandlimited to ( W ,W ) &

 Can apply ideal LPF to recover original analog signal

If signal is not bandlimited or is undersampled, then

Sampling = quantising or discretising the time axis.

However, the sampled signal may still be continuous-valued.

Before transmission in a digital system, it needs to be

Inevitably, quantisers cause loss of resolution in data

Mathematically, a quantiser maps a real input x to some

N is usually a power of 2 so that the index of the selected

When x is a realisation of an rv, quantiser quality is measured by

If this is minimised, then X & X-Q(x) are uncorrelated.

Can define signal-to-quantisation-noise ratio

If X ~ U[a,b) , then D min. by partitioning [a,b) into N

intervals of equal length  : b  a & setting x̂ midpt.

To minimise quant. noise power, we should thus try to

 Non-uniform quantisation required.

In general, the optimal quantiser has levels closer together

Companding is a way of using a uniform quantiser to generate non-

(NB: typically, x(t) is normalised to have amplitude =1.)

You might also like