EE5407 Tutorial Band-Limited Linear Filter Channels and Equalisation

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 8

EE5407 Tutorial

Band-Limited Linear Filter Channels and Equalisation

February 9, 2010

1. In a binary PAM system, the clock that specifies the sampling of the correlator
output is offset from the optimum sampling time by 10 percent.

Show that the sampled sequence at the output of the receiving filter can be written
as,

X 1 1
ym = In x((m − )T − nT ) + v((m − )T ) (1)
−∞
10 10

Assume that the the signal pulse is rectangular with amplitude A and duration T .

(a) Determine the sampled sequence {ym } at the output of the receiving filter.
Identify the ISI terms.

(b) Determine the loss in SNR due to the mistiming.

2. Prove that the relationship for the autocorrelation of the noise sequence {vk } at the
output of the matched filter is,

1  N x ; if |k − j| ≤ L
∗ 0 k−j
E[vk vj ] = (2)
2  0; otherwise

i.e. the noise is correlated unless xk = 0, k 6= 0, with


Z ∞
xk−j = h∗ (a − jT )h(a − kT )da (3)
−∞

where h(t) represents the channel impulse response.

1
3. Binary PAM is used to transmit information over an unequalised linear filter channel.
Assume that the channel is free of Gaussian noise. When the symbol a = 1 is
transmitted, the output of the demodulator is


 0.3; m = 1




 0.9; m = 0
xm = (4)

 0.3; m = −1




 0; otherwise

(a) Design a 3-tap zero-forcing linear equaliser so that the output is



 1; m = 0
qm = (5)
 0; m = ±1

(b) Calculate qm for m = ±2, ±3, by convolving the impulse response of the
equaliser with the channel response.

4. The transmission of a signal pulse with raised cosine spectrum through a channel
results in the following (noise-free) sampled output from the demodulator:


 −0.5; k = −2





 0.1; k = −1




 1; k = 0
xk = (6)

 −0.2; k = 1





 0.05; k = 2




 0; otherwise

(a) Determine the tap weight coefficients of a 3-tap linear equaliser based on the
zero-forcing criterion.

(b) For the coefficients determined in (a), determine the output of the equaliser
for the case of the isolated pulse. Thus, calculate the residue ISI and its span
in time.

5. Consider a 3-tap zero-forcing equaliser. If the ISI spans 3 symbols and is charac-
terised by the values, 

 0.3; k = −1


xk = 1; k = 0 (7)



 0.2; k = 1

2
(a) Determine the optimum tap weight coefficients of the zero-forcing equaliser.

(b) Determined the residue ISI at the output of the equaliser for the optimum tap
coefficients.

6. Let us solve Problem (3) using the MSE as the criterion for optimising the equaliser
tap weight coefficients. Assume that the noise power spectral density is 0.1 W/Hz.

(a) Show that the optimum coefficients {ck } of the MMSE equaliser satisfy:
K
X
cj Γlj = ξl ; l = −K, .., 0, ..K (8)
j=−K

where

 x + N δ ; if |l − j| ≤ L
l−j 0 ij
Γlj = (9)
 0; otherwise

and

 f ∗ ; if − L ≤ l ≤ 0
−l
ξl = (10)
 0; otherwise

with K = 1, and L = 1 for this particular problem.

(b) Find the channel parameters f0 and f1 for the equivalent discrete-time white
noise filter model.

(c) Determine the optimum coefficients {ck } of the MMSE equaliser.

7. A non-ideal band-limited channel introduces ISI over successive symbols. The


(noise-free) response of the matched filter demodulator sampled at the sampling
time kT is 

 Eb ; k = ±0


Z 

∞  0.9Eb ; k = ±1
s(t)s(t − kT )dt = (11)
−∞ 
 0.1Eb ; k = 0 ± 2




 0; otherwise

(a) Determine the tap coefficients of the 3-tap linear equaliser that equalises the
channel (received signal) response to an equivalent partial-response signal

 E ; k = 0, 1
b
yk = (12)
 0; otherwise

3
8. A time dispersive channel having an impulse response h(t) is used for transmission.
The equivalent discrete-time channel is of 2 taps with f0 = 0.8 and f1 = −0.6. The
{ηk } is a white noise sequence having zero mean and variance σ 2 = N0 .

(a) Draw the equivalent discrete-time channel model. Label the {ηk }, the infor-
mation symbol sequence {Ik }, and the received sequence {yk }.

(b) Find the values of the sampled autocorrelation function sequence {xk } for this
channels, where xk is defined as
Z ∞
xk = h∗ (t)h(t + kT )dt (13)
−∞

9. Consider the problem of equalising an equivalent discrete-time channel, which is of


2 taps with f0 = f1 = √1 . The additive noise sequence {vn } is white and real-
2

valued with variance N0 . The information symbol sequence {In } is binary (±1) and
uncorrelated. The received sequence {yn } is processed by a linear 3-tap equaliser
that is optimised on the basis of the MMSE criterion.

(a) Determine the optimum coefficients of the MMSE equaliser as a function of


N0 .

(b) Determined the the minimum MSE for the 3-tap equaliser as a function of N0 .

10. Consider the equivalent discrete-time channel, which is of 2 taps with f0 = f1 = √1 .


2

The additive white noise sequence {vn } is real-valued with zero mean and variance
N0 . The information symbol sequence is {In }.

Suppose the channel is to be equalised by a DFE having a 2-tap feedforward filter


(c0 , c−1 ), and a 1-tap feedback filter (c1 ). The {cj } are optimised using the MMSE
criterion.

(a) Determine the optimum coefficients of the feedforward/ feedback filters and
their approximated values for N0 ¿ 1.

(b) Assume all previous symbol decisions are correct, determine the exact value
of the minimum MSE and a first-order approximation appropriate to the case
N0 ¿ 1.

4
11. Consider the equivalent discrete-time channel, which is of 2 taps with f0 = 45 , f1 = 35 .
The information symbol sequence is {In }. Assume the AWGN sequence {ηn } is zero.

(a) Suppose the channel is to be equalised by a 3-tap linear equaliser. Determine


the tap coefficients c−1 , c0 , c1 of the 3-tap equaliser.

(b) Determine the equaliser coefficients of a DFE with 2 feedforward taps and 1
feedback tap. Assume all previous symbol decisions are correct.

12. In adaptive MMSE equaliser, an iterative procedure that avoids the matrix inversion
may be used to compute copt . This is achieved by recursively adjusting the coefficient
vector ck (k is the index of iteration) using,

ck+1 = ck − ∆gk ; k = 0, 1, ... (14)

where ∆ is a small positive number to ensure convergence.


Show in details that the gradient vector gk can be written as
1 dJ
gk = = −E[εk vk∗ ]
2 dck (15)
= Γck − ξ

where εk = Ik − Iˆk is the error at the kth iteration; ck = [c−K · · · c0 · · · cK ]T ; vk =


[vk+K · · · vk · · · vk−K ]T is the received signal samples.

13. Consider the random process

x(n) = g v(n) + w(n); n = 0, .., M − 1 (16)

where {vn } is a known sequence, g is a random variable with E(g) = 0, and E(g 2 ) =
G. The process w(n) is a white noise sequence with

γww (m) = σw2 δm (17)

Determine the coefficients of the linear estimator for g, that is,


M
X −1
ĝ(n) = h(n)x(n) (18)
n=0

that minimise the MSE.

5
Figure 1: Figure for Question 3.

14. Consider the adaptive finite impulse response (FIR) filter shown in Figure 1. The
system C(z) is characterised by the system function

1
C(z) = (19)
1 − 0.9z −1

Determine the optimum coefficients of the adaptive transversal (FIR) filter B(z) =
b0 +b1 z −1 that minimise the MSE. The additive noise is white with variance σw2 = 0.1.

Figure 2: Figure for Question 4.

15. Consider a model of the data communication system shown in Figure 2. The input
data sequence {a(n)} is a Bernoulli sequence {±1}, with Pr(1) = Pr(−1) = 0.5.
This data sequence is assumed to be white with zero mean. The channel (including

6
the receiving and whitening filter) is modelled as
 h i
 0.5 1 + cos 2π(n−2) ; n = 1, 2, 3.
U
h(n) = (20)
 0; otherwise

where U controls the amount of amplitude distortion introduced by the channel.

The input to the equaliser is,

x(n) = s(n) + v(n) (21)

where s(n) represents the distorted pulse sequence. The channel noise v(n) is mod-
elled as white Gaussian noise with zero mean and variance σv2 .

In practical modem implementations, the equaliser is initialy designed using a train-


ing sequence that is known to the receiver. It is shown in Figure 2 as the sequence
y(n). The delay, D, in the training sequence is to account for delays introduced in
the channel and in the equaliser; that is y(n) = a(n − D) during the training mode.
The error sequence {e(n)} is further used to design the equaliser {c(n)}.

(a) Show that the channel has a symmetric impulse response.

(b) Let rh (l) = h(l)∗h(−l) be the autocorrelation of h(n). Determine the autocor-
relation rx (l) of x(n) in (21). (hint: ra (l) = δl for {a(n)}, which is white).

(c) Evaluate the cross-correlation between x(n) and y(n) in terms of D.

(d) Determine the cross-correlation between x(n) and y(n) for D = 0, 1, 2, ... etc.
Comment on the results.

16. Consider the performance index

J = h2 + 40h + 28 (22)

Suppose we search for the minimum J by using the steepest-decent algorithm,


1
h(n + 1) = h(n) − ∆g(n) (23)
2

where g(n) is the gradient.

(a) Determine the range of values of ∆ that ensures the convergence of the adjust-
ment process (i.e. that provides an over-damped system).

7
(b) Plot the expression for J as a function of n for a value ∆ = 12 .

– END –

You might also like