You are on page 1of 3

University of Tehran

Department of Electrical and Computer Engineering


DSP- Chapter 3
4 : -

Problem 1:

Consider the following system:

a)

How many multiplications per output sample are needed on average (not counting factors of 1)?

b) Use the noble identity to push the downsampler toward the input of the system as much as possible.
c)

How many multiplications per output sample are needed on average for your new system in part (b) (not
countingfactors of 1)?

Problem 2:
In the system below, w[n]is a real, zero-mean, white, wide-sense stationary random sequence with variance 2 ,
and h[n]is LTI.

(a) Simplify this system as much as possible. As a first step, perform a polyphase decomposition on h[n].
2

(b) Find an expression for the autocorrelation of [] i.e., [, ] = [[][ + ]], in terms of and []
for all values of n. Use your result from part (a) as a first step.

Problem 3:
For each system shown below, design an efficient implementation. Take efficient to mean having a small number
of multiplications per input sample and a small number of delay elements. Make block diagram manipulations in a

step-by-step manner with brief justifications of your steps. Make sure to fully specify any filters, i.e., write out the
transfer functions.

Problem 4:

Consider an arbitrary digital filter with transfer function

() = []
a)

Perform a two-component polyphase decomposition of () by grouping the even-numbered samples


0 [] = [2] and the odd-numbered samples 1 [] = [2 + 1]. Thus show that () can be expressed
as
And determine 0 () and 1 ()

() = 0 ( 2 ) + 1 1 ( 2 )

b) Generalize the result in part (a) by showing that () can be decomposed into an D-component
polyphase filter structure with transfer function
1

c)

Determine ()
For the IIR filter with transfer function

() = ( )
=0

1
1 1
Determine 0 () and 1 () for the two-component decomposition.

Problem 5:

() =

To process sequences on a digital computer, we must quantize the amplitude of the sequence to a set of discrete
levels. This quantization can be expressed in terms of passing the input sequence [] through a quantizer ()
that has an input-output relation as shown in Figure 8-1
If the quantization interval is small compared with changes in the level of the input sequence, we can assume
that the output of the quantizer is of the form
[] = [] + []
where [] = ([]) [] and [] is a stationary random process with a 1st-order probability density uniform
between /2 and /2 , uncorrelated from sample to sample and uncorrelated with [], so that {[][]} =
0 for all m and n.

Let [] be a stationary white-noise process with zero mean and variance 2 .


a) Find the mean, variance, and autocorrelation sequence of []
b) What is the signal-to-quantizing-noise ration
c)

2
2

The quantized signal [] is to be filtered by a digital filter with impulse response [] = [ +


2

() ][]. Determine the variance of the noise produced at the output due to the input quantization
noise, and determine the SNR at the output.

Problem 6:

Let () be a time-limited signal; that is, () = 0 for || > , with Fourier transform () . The function
() is sampled with sampling interval = 1/ .

(a) Show that the function

() = ( )

can be expressed as a Fourier series with coefficients

1
( )

(b) Show that () can be recovered from the samples ( ), < < if 2.

(c) Show that if < 2 , there is time-domain aliasing that prevents exact reconstruction of ().

(d) Show that if 2 , perfect reconstruction of () from the samples ( ) is possible using the
interpolation formula

() = ( )
=

sin[(/)( )]
(/)( )

You might also like