Discrete Memoryless Channels
Discrete Memoryless Channels
x0 y0
x y1
1
. .
X X P(y k | x j ) Y Y
. .
. .
xJ 1 y K -1
Definition of DMC
Channel with input X & output Y which is
noisy version of X.
Discrete when both of alphabets X & Y finite sizes.
Memoryless when no dependency between
input symbols.
1
Discrete Memoryless Channel (cont’)
Channel Matrix (Transition Probability Matrix)
p( y0 | x0 ) p(y1 | x0 ) ... p( y K 1 | x0 )
p( y | x ) . . . . . p ( y | x )
0 1 K 1 1
P .
.
p( y0 | x J 1 ) . . . . .. p( y K 1 | xJ 1 )
The size is J by K
K 1
p( yk | x j ) 1 for all j
k 0
a priori prob. is :
P k p ( x j ), j 0 ,1 ,.., J 1
2
Discrete Memoryless Channel (cont’)
1-p
x0 0 y0 0
p
1-p y1 1
x1 1
4
Mutual Information
Conditional Entropy
J 1
1
H ( X | Y yk ) p( x j | yk ) log 2 [ ]
j 0 p ( x j | yk )
The mean value
K 1
H (X |Y )
k 0
H ( X | Y yk ) p( yk )
K 1 J 1
1
k 0 j0
p ( x j , y k ) log 2[
p(x j | yk )
]
5
Properties of Mutual Information
( simple ex. needed for 2 by 2 DMC)
Symmetric : I ( X ; Y ) I (Y ; X )
Non-negative :
I ( X ;Y ) 0
I ( X ; Y ) H (Y ) H (Y | X )
I ( X ; Y ) H ( X ) H (Y ) H ( X , Y ) J 1 K 1
where 1
H ( X , Y ) p ( x j , yk ) log 2 [ ]
j 0 k 0 p ( x j , yk )
H(X,Y)
H(X) H(Y)
6
Channel Capacity
For a dms with input X, output Y, & p ( y k | x j, )
J 1 K 1 p( yk | x j )
I ( X ; Y ) p ( x j , yk ) log 2 [ ]
j 0 k 0 p( yk )
J 1
where p ( x j , y k ) p ( yk | x j ) p ( x j ) , p ( y k ) p ( yk | x j ) p ( x j )
j 0
7
Ex.) for BSC
C max I ( X ; Y ) I ( X ; Y ) | p ( x0 ) 0.5
C 1 p log 2 p (1 p) log 2 (1 p) 1 H ( p )
8
Channel Coding Theorem
For reliable communication , needs channel encoding & decoding.
“any coding scheme which gives the error as small as possible, and
which is efficient enough that code rate is not too small?”
Let dms with alphabet X have entropy H(X) and produce symbols
once every Ts, and dmc have capacity C and be used once every Tc .
Then,
ⅱ) if H (X )
C , it is not possible to transmit with arbitrary
Ts Tc small error.
9
Ex.) for BSC with p0 0.5
The condition for reliable comm. ,
1 C
Ts Tc
Tc
Let Ts
be r , then r C
10
Differential Entropy
Differential Entropy
1
h( X ) f x ( x) log 2 [ ]dx
f x ( x)
where f x (x ) is p.d.f.
11
Maximum Differential Entropy for Specified Variance
12
Maximum Differential Entropy for Specified Variance
Sol. is based on calculus of variation & use of Lagrange multiplier
f X ( x) log 2 f X ( x) 1 f X ( x) 2 ( x ) 2 f X ( x) dx
I
I should be stationary to get maximum entropy.
- The desired form of f X ( x) is
1 ( x u )2
f X ( x) exp 2
2 2
Gaussian p.d.f.
- The maximum entropy
1
h ( x ) log 2 (2 e 2 )
2
Gaussian channel model is so widely utilized.
13
Mutual Information for Continuous r.v
f X ( x | y)
I(X :Y )
f X ,Y ( x, y ) log 2
f X ( x)
dxdy
14
Shannon’s Channel Capacity Theorem
For bandlimited, power limited Gaussian channels
P
C Blog2 1 (bits/s)
N
The capacity of a channel of bandwidth B, perturbed by
additive white gaussian noise of psd N0 /2, and limited in bandwidth to B,
P is the average transmitted power, and N is the noise (NoB)
- It is not possible to transmit at rate higher than C reliability by any means.
- It does not say how to find coding and modulation to achieve maximum capacity,
but it indicates that approaching this limit, the transmitted signal should have statistical
property approximately to Gaussian noise.
15
Bandwidth efficient diagram
Define ideal system as Rb C,
P EbC where Eb is the Tx energy per bit
C/B
C EbC 20
Then, log2 (1 )
B No B 10
Rb = C
Rb > C
Eb 2C / B 1
Rb < C
No C/B
For infinite bandwidth channel -1.6
1
Eb Eb/No(dB)
ln 2 0.693 1.6dB
No B
P Shannon's limit
C lim C log2 e
B No 0.1
-10 0 10 20 30 40 50
16