Académique Documents
Professionnel Documents
Culture Documents
Muqaibel
Objective :
Receiver design
Performance evaluation (memory, No memory)
Note: These notes are preliminary and are posted by the request of the students. Please,
report to me all the mistakes that you find in the document. (muqaibel@kfupm.edu.sa).
This material is for the sole purpose of in class usage. Please observe the copyright of the
original authors for any content in these notes.
Optimum Receiver
sn (t ) r(t)
n(t)
r (=
t ) sm (t ) + n(t ) 0t T
{sm (t ), m = 1, 2,..., M }
1
nm ( f ) = N0 W / Hz
2
Objective: Upon observation of r(t), what is the optimum receiver design in terms of
probability of making error.
receiver
demodulator detector
Optimum Demodulator
1. Correlators
2. Matched filters
Correlation Demodulator
Decomposes the received signal-to-noise ratio into N-dimensional vectors.
Linearly weighted orthogonal basis functions {fn(t)}, where {fn(t)} spans the
signal space but not the noise space.
We can show that the noise terms that falls outside the signal space is irrelevant to the detection.
f1 (t ) r1
T
(
0
)dt
f 2 (t ) r2
r (t ) T
(
0
)dt to the detector
f N (t ) rN
T
(
0
)dt
T T
r (t ) f=
0
k (t ) dt [s
0
m (t ) + n(t )] f k (t )dt
rk =+
smk nk k=
1, 2, N
where
T
=smk =
0
s (t ) f (t )dt ,
m k k 1, 2, N
T
=nk =
0
n(t ) f (t )dt ,
k k 1, 2, N
N N N
=
r=
(t )
k 1=k 1
smk f k (t ) + nk f k (t ) + n=
'
(t )
=k 1
r fk k (t ) +n ' (t )
N
n ' (t ) =n(t ) nk f k (t )
k =1
n ' (t ) is zero mean Gaussian noise process. It is the unrepresented part of noise inherent.
n ' (t ) is Gaussian because it is the sampled output of a linear filter excited by a Gaussian input.
T
=E (nn ) =
0
E[n(t )] f (t )dt k 0 for all k
T T
E (nk mk ) = E[n( )n( )] f k (t ) f m ( )dtd
0 0
T T
1
= N
2 0 (t ) f
0 0
k (t ) f m ( )dtd
T
1 1
= = N 0 f k (t ) f m ( )dt N mk
2 0
2 0
mk = 1
{nk } are zero mean uncorrelated random variable with common covariance
1
n2 = N 0
2
r = [r1 r2 rN ]
N
=p (r/ sm )
= p ( r/ s )
k =1
m mk 1, 2,..., M
1 (rk smk ) 2
when
= p (r/ smk ) = exp[ ] , k 1, 2,...N
N0 N0
1 N
(rk smk ) 2
p (r/ sm ) = N / 2 exp[ ], m=
1, 2,..., M
( N 0 ) k =1 N0
As a final step we can show that (r1 , r2 ,..., rN ) are sufficient statistics.
No additional relevant information can be extracted from n ' (t ) .
E[n ' (t )rk ] = 0 uncorrelated proof p 235
Gaussian and uncorrelated implies statistically independent which implies ignore n ' (t )
Example
M-ary PAM
g(t)
t
0 T
T T
g =
2 2
Eg
= t )dt
(= a dt a 2T
0 0
1
T
t
0 T
1 1/ T 0 t T
=f (t ) = g (t )
a 2T 0 otherwise
r sm + n
=
E[n] = 0
T T T T
1 1
[ n(t )n( )dtd ]
T 0 0
= E= 2
n E[n(t )n( )]dtd
T 00
T T
N0 1
=
2T (t )dtd=
0 0
2
N0
1 (r sm ) 2
pdf of the sampled output p (r/ sm ) = exp[ ]
N0 N0
t
= r ( ) f
0
k (T t + )d=
k 1, 2,...N
Matched filter: A filter whose impulse response h(t)=s(T-t) or s(t) where s(t) is assumed
to be confined to the time interval 0tT.
s(t)
y(T)
t
T T 2T
f1 (T t )
r (t )
f 2 (T t )
f N (T t )
Property of Matched filter: If a signal s(t) is corrupted by AWGN, the filter with an impulse
response matched to s(t) maximizes the output signal to noise ratio (SNR).
Proof :
t t t
y (=
t) r ( )h(t )d=
0
s ( )h(t )d + n( )h(t )d
0 0
at =
t T
T T
y (=
T) s( )h(T )d + n( )h(T )d
0 0
= ys (T ) + yn (T )
ys2 (T )
SNR0 =
E[ yn2 (T )]
where E[ yn2 (T )] noise var iance
T T
E[ yn2 (T )]
= E[n( )n(t )]h(T )h(T t )dtd
0 0
T T
1
2 0 0
= N 0 (t )h(T )h(T t )dtd
T
1
= N 0 h 2 (T t )dt
2 0
T T
[ s ( )h(T )d ]2 [ h( ) s (T )d ]2
=SNR0 = 0
T
0
T
1 1
N 0 h 2 (T t )dt N 0 h 2 (T t )dt
2 0 2 0
Cauchy-Schwarz inequality:
[ g1 (t ) g 2 (t )dt ]2 g12 (t )dt g 2 2 (t )dt
T
2 2E
SNR0
= =
N0 0
s 2 (t )dt
N0
Property: The output SNR from the matched filter depends on the energy of the waveform s(t)
but not on the details (shape) of s(t).
2
Y ( f ) = S ( f ) e j 2 ft
2
=Ys (t )
=
Y ( f )e j 2 ft df
S ( f ) e j 2 ft e j 2 ft df
at t = T
2
=
2
Ys (T )
= S=
( f ) df s (t )dt E Pasaval ' s relation
Noise at the output of the matched filter
1 2
( f ) =H ( f ) N 0
2
1 2 1 2 1
( f )df =
Pn = N 0 H ( f ) df =
2
N 0 S ( f ) df =
2 2
EN 0
Ps E2 2E
Ps =Ys2 (T ) =
> SNR0 = = =
Pn 1 EN N0
0
2
Matched filter and correlator at equivalent at t=T but matched filter is immune to time jitters.
Example 5.1.2
M=4 bi-orthogonal signals
f1 (t ) f 2 (t )
2 2
T T
t t
T
T T T
2 2
h=
1 (t ) f1 (T t ) h=
2 (t ) f 2 (T t )
2 2
T T
t t
T T T T
2 2
y1s (t ) y2 s (t )
1 2 1 2
AT AT
2 2
t t
T T
T
2
Note the response to s1(t) is evaluated at T
=r [r1 =
r2 ] [ E + n1 n2 ]
( E )2 2 E
SNR
= =
1 N0
N0
2
Additional Example (Matched Filters)
Consider the signal s(t)
A
2
t
T T
2
A
2
a) Determine the impulse response of a filtered matched to this signal and sketch it.
h(t)=s(T-t)
A
2
t
T
T
2
A
2
T 3T
2 4
t
T
1 2
AT
8
A pair of pulses that are orthogonal to each other over the interval [0,T], are used for two
dimensional matched filter
A
2
A
2
Another method for approximating realization of matched filter is the (RC) low pass filter
[integrator].
R
input output
C
1 1
=H( f ) = , f0
1+ j f 2 RC
f0
The input signal is rectangular of pulse amplitude A and duration T.
Objective: Optimize the selection of 3-dB cutoff frequency f0 of the filter. So that the peak
output SNR is maximized.
Show that f0=0.2/T is the optimum. Compared to matched filter 1dB loss.
s(t)
A[1 exp(2 f 0T )]
N0 df N 0 f 0
=N out =
2 1 + ( f ) 2 2
f0
The corresponding value for the SNR
2 A2
( SNR)=out [1 exp(2 f 0T )]
N 0 f 0
Differentiating with respect to (f0T) and setting the result equal to zero.
The maximum value of (SNR)out is at f0=0.2/T.
r Detector
assume no memory
MAP = ML if { sm } is equi-probable.
where
D(r | sm ) is Euclidean distance and m= 1,2,M
rn 2 2 rn smn + smn 2
D( r, sm ) =
= n 1= n 1 =n 1
2 2
=r 2 r . sm + sm where m = 1,2,3M
2
'
D (r , sm ) =
2 r . sm + s
Let C (r , sm ) = D ' (r , sm )
To get rid of the minus. Now, we maximize C rather than minimize D.
2
C (r=
, sm ) 2. r . sm sm
sm can be eliminated if energy is fixed for all m =1,2,M, but cannot if signals have uequal
energy (PAM).
T
C (r , sm ) 2 r (t ) sm (t )dt Em m= 0,1,2,,M
=
0
1
E2
s2 (t ) 2 Select
r (t ) T
( )dt the
0 Largest
1
sN (t ) EN
T 2
(
0
)dt
Summary:
Optimum ML
1) compute D(r , sm ) or D ' (r , sm ) distance metrics and chooses the smallest or
2) compute C (r , sm ) correlator metrics and choose the largest
3) ML= MAP if eqiprobable { sm }, otherwise
PM (r , sm ) = P (r | sm ) P ( sm )
Example:
T
t
T
2
2
C (r=
, sm ) 2. r . sm sm eq. (5.1-44)
T
t
T
For 2
C (r=
, sm ) 2. r . sm sm eq. (5.1-44)
C (=
r , s1 ) 2.(4).(2)
= 4 12
C (r=
, s2 ) 2.(4).(6)
= 36 8
T
t
T
For 2
C (=
r , s1 ) 2.(3).(2)
= 4 8
C (r=
, s2 ) 2.(3).(6) =
36 0
Example:
1
Eb + yn (T ) zero mean and n2 =
r= N0
2
Note: the variance of the sampled noise is N0/2. In general the noise power, P=N0B, According to
Nyquist B=1/(2T), When looking at the energy we E=PT
1 (r Eb ) 2
P(r | s1 )
= exp
2 n 2 n2
1 (r + Eb ) 2
P(r | s2 )
= exp
2 n 2 n2
PM
= (r , s1 ) =
p (r | s1 )
exp
(
r Eb ) 2
2 n 2 n2
1
r + Eb
(1 ) p (r | s2 ) =exp
PM (r , s2 ) =
( ) 2
2 n 2 n2
If PM (r , s1 ) > PM (r , s2 ) then choose s1
s1
PM (r , s1 ) >
1
PM (r , s2 ) <
s2
s1
PM (r , s1 ) (r + Eb ) 2 (r Eb ) 2 >
= exp 1
PM (r , s2 ) 1
2 n
2
<
s2
s1
(r + Eb ) 2 (r Eb ) 2 >
ln
2 n 2
< 1
s 2
s1
> 1 2 1
Eb r n ln = N 0 ln
< 2 1 4 1
s2
Eb Eb
R2 R1
s2 h s1
1) If = , h =0
N0
2) If , knowledge of N0 or is required for optimal detection.
Eb
If the M signals are equi-probable.
M
1
P (c ) = p( s m | r ) p (r )dr MAP
m =1 M Rm
Rm is the region for correct decision. Explain the Union bound Concept of QPSK
P(e)=1-P(c)
The maximum likelihood (ML) sequence detector: searches for the minimum Euclidean
distance path through the trellis that characterizes the memory in the transmitted sequence.
Example NRZI
Viterbi algorithm: is a sequential trellis search algorithm for performing ML sequence detection
Used for decoding convolutional codes
15 Dr. Ali Muqaibel Digital Communications I
[OPTIMUM RECEIVER DESIGN AND EVALUATION FOR AWGN] Dr. Muqaibel
0 / Eb 0 / Eb 0 / Eb
1/ Eb
1/ Eb
1/ Eb
1/ Eb
0 / Eb
t T 2T 3T
At 2T and so on there are two arrows entering the state, we choose the minimum distance
survivor.
D0 (0, 0) = (r1 + Eb ) 2 +(r2 + Eb ) 2
D0 (1,1) = (r1 Eb ) 2 +(r2 + Eb ) 2
Using Viterbi in this example the number of path searched is reduced by a factor of two at
each stage.
The memory length is L (L=1 in the previous example)
You make a decision when the survivor path agree. Variable delay negative ??
Practically at 5L and then make decision at k even then before k-5L is almost identical.
Make a decision.
D1(0,1)=(0.9+1)2+(-0.8-1)2=1.92+1.82=3.61+3.24=6.85
D1(1,0)=(0.9-1)2+(-0.8-1)2=0.12+1.82=0.01+3.24=3.25
Eb Eb
R2 R1
s2 h s1
1 (r E b )2 / N 0
P (r | s1 ) = e
N 0
1 (r + E b )2 / N 0
P (r | s 2 ) = e
N 0
0 0
1 ( r + E b )2 / N 0
P (e | s1 )
= =
P (r | s1 )dr
N 0 e
dr
r Eb2 N
By replacing variables x = , dx = dr =
> dr = 0 dx
N0 / 2 N0 2
x = , = > r =
2 Eb
r ==
0 >x=
N0
2 Eb / N 0
1
2
= e x / 2 dx
2
1
2 Eb
2
= = e x / 2 dx Q
2 2 Eb / N0 N0
1
e dx
2
x /2
Q( x) =
2 x
2 Eb
Eb
s2
0 Eb
s1 = [ Eb 0]
d12 = 2 Eb
s2 = [0 Eb ]
If s1 is transmitted
=r [ Eb + n1 n2 ]
2
C ( r=
, sm ) 2r .sm sm
C (r , s2 ) = 2.[ Eb + n1 n2 ].[0 Eb ] Eb (1)
C (r , s1 ) = 2.[ Eb + n1 n2 ].[ Eb 0] Eb (2)
(1) Can be further simplified to 2n1 Eb Eb and
(2) Can be further simplified to 2E b + 2n1 E b E b = E b + 2n1 E b
Probability of error
C (r , s2 ) > C (r , s1 )
P[e | s1 ] P[C (r , s2 ) > C (r , s1 )]
=
Figure 5.2.4
101
Pb = Q ( )
b
Probability of error Pb
r = 0
Pb = Q ( 2 b )
r = 1
In addition to the above antipodal and orthogonal examples, we can extend the analsys to other
modulation techniques. For example,
Many orthogonal signals
Bi-orthogonal
Simplex
M-ary PAM
2( M 1) d Eg
2
PM = Q
M N0
2( M 1) 6(log 2 M ) Ebav
= Q 2
M ( M 1) N 0
6
Because d 2 Eg = 2 PavT
M 1
M-ary PSK
QAM
Regenerative repeaters
s(t)
r(t)
n(t)
2 Eb
Pb = Q PAM Binary.
N0
k Repeaters assuming single error at a time.
2 Eb
P kQ at repeater.. Why not exact equal sign? (error cancellation)
N 0
Analog: Required Eb/N0 reduced by k
2 Eb
P Q
kN 0
At decision receiver note receiver is connected k= 1+ repeater.
Example
1000 km 10 km repeater k=100 10-5
2 Eb 2 Eb
10-5 = 100 Q => 10-7 = Q => SNR =11.3dB
N N
0 0
2 Eb
10-5 = Q => SNR =29.6 dB 29.6-11.3 = 18.3 dB
N
0
PT GT
GT: antenna transmit gain , GT =1 for isotropic antenna.
4 d 2
PT GT : effective radiated power.
ERP or EIRP compared with isotropic antenna
GR 2
AR: effective area of the antenna, AR = (2)
4
8
c = f , c = 3X10 m/s
Substitute (2) in (1)
PG G
PR = T T R 2
(4 d / )
2
Free space path loss factor Ls =
4 d
Additional losses atmospheric La .
PR = PT GT GR Ls La
Calculation of the antenna gain is antenna specific and depends on (dimensions) diameter D,
Illumination efficiency factor
Effective area AR
Area A
Beam width B
Dish,horn etc
Example
2
D
GR = = 39dB
2
= Ls = 195.6dB
4 d
(PR)dB = 20 + 17 + 39 195.6 = -119.6 dBW
PR = 1.1 X 10-12 W
is this low or high ?
Eb Tb PR 1 PR
Performance is dependent on = =
N0 N0 R N0
PR E
= R b
N0 N 0 Re q
Example for the same previous example.
PR =1.1 X 10-12 W (-119.6 dBW)
N0 = 4.1 X 10-21 W/Hz = N0W = KBT0W
= -203.9 dBW/Hz
PR
= -119.6 + 203.9 = 84.3 dB Hz
N0
Eb
SNR is 10dB
N0
RdB = 84.3-10 = 74.3 dB with respect to 1bit/sec
= 26.9 Mbps
420 PCM (64000 bps)
RdB =
0 0
= ( ) + ( ) + ( ) + ( ) + ( )
0