EnggTree.

com

DIGITAL SIGNAL PROCESSING

PRE-REQUISITE:

Signal & Systems, Digital System Design, Engineering Mathematics

AIM & OBJECTIVES:

 To learn discrete Fourier transform, properties of DFT and its application to

linear filtering.
 To understand the characteristics of digital filters, design digital FIR filters
and apply these filters to filter undesirable signals in various frequency
bands.
 To design digital IIR filters and apply these filters to filter undesirable
signals in various frequency bands.
 To understand the effects of finite precision representation on digital filters.
 To understand the fundamental concepts of multi-rate signal processing and
its applications.

Downloaded From EnggTree.com

 EnggTree.com

UNIT I DISCRETE FOURIER TRANSFORM

Review of discrete-time signals and systems – Discrete Fourier Transform (DFT) and its
properties, Circular convolution, Linear filtering using DFT, Filtering long data sequences -
overlap-save methods - Overlap-add, Fast Fourier Transform (FFT) algorithms – Fast
computation of DFT –Radix- 2 decimation in time FFT - Decimation in frequency FFT – Linear
filtering using FFT.

UNIT II DESIGN OF FINITE IMPULSE RESPONSE FILTERS

Structures for FIR systems – Transversal and Linear phase structures, Design of FIR filters –
Symmetric and Anti-symmetric FIR filters, Design of linear phase FIR filters using Windows
(Rectangular, Hamming and Hanning windows) and Frequency sampling methods

UNIT III DESIGN OF INFINITE IMPULSE RESPONSE FILTERS

Structures for IIR systems – direct, cascade, parallel forms, Comparison of FIR and IIR, Analog
filters – Butterworth filters - Chebyshev type – I filters (upto 3rd order), Analog transformation
of prototype LPF to BPF/BSF/HPF, Transformation of analog filters into equivalent digital
filters using Impulse invariant method and Bilinear Z-transform method.

UNIT IV FINITE WORD LENGTH EFFECTS

Representation of fixed and floating point numbers, ADC quantization -truncation and rounding
- quantization noise, Coefficient quantization Error – Product quantization error – Overflow error
- Round-off noise power, Limit cycle oscillation due to product round-off error- Limit cycle
oscillation due to overflow in digital filters – Principle of scaling.

UNIT V MULTI-RATE SIGNAL PROCESSING

Introduction to multi-rate signal processing – Decimation – Interpolation- Sampling rate
conversion by a rational factor - Polyphase decomposition of FIR filter – Multistage
implementation of sampling rate conversion – Design of narrow band filters–Applications of
multi-rate signal processing

Downloaded From EnggTree.com

 EnggTree.com
UNIT -1
INTRODUCTION
REVIEW OF DISCRETE TIME SIGNALS AND SYSTEMS

Anything that carries some information can be called as signals. Some examples are ECG,
EEG, ac power, seismic, speech, interest rates of a bank, unemployment rate of a country,
temperature, pressure etc.
A signal is also defined as any physical quantity that varies with one or more independent variables.
A discrete time signal is the one which is not defined at intervals between two successive samples
of a signal. It is represented as graphical, functional, tabular representation and sequence.
Some of the elementary discrete time signals are unit step, unit impulse, unit ramp, exponential
and sinusoidal signals (as you read in signals and systems).

Classification of discrete time signals

Energy and Power signals

If the value of E is finite, then the signal x(n) is called energy signal.

If the value of the P is finite, then the signal x(n) is called Power signal.

Periodic and Non periodic signals

A discrete time signal is said to be periodic if and only if it satisfies the condition X (N+n) =x (n),
otherwise non periodic

Symmetric (even) and Anti-symmetric (odd) signals

The signal is said to be even if x(-n)=x(n)
The signal is said to be odd if x(-n)= - x(n)

Causal and non causal signal

The signal is said to be causal if its value is zero for negative values of ‘n’.

Some of the operations on discrete time signals are shifting, time reversal, time scaling, signal
multiplier, scalar multiplication and signal addition or multiplication.

Discrete time systems

A discrete time signal is a device or algorithm that operates on discrete time signals and produces
another discrete time output.

Classification of discrete time systems

Static and dynamic systems
A system is said to be static if its output at present time depend on the input at present time only.

Causal and non causal systems

A system is said to be causal if the response of the system depends on present and past values of the
input but not on the future inputs.

Downloaded From EnggTree.com

 EnggTree.com
Linear and non linear systems
A system is said to be linear if the response of the system to the weighted sum of inputs should be
equal to the corresponding weighted sum of outputs of the systems. This principle is called
superposition principle.

Time invariant and time variant systems

A system is said to be time invariant if the characteristics of the systems do not change with time.

Stable and unstable systems

A system is said to be stable if bounded input produces bounded output only.

TIME DOMAIN ANALYSIS OF DISCRETE TIME SIGNALS AND SYSTEMS

Representation of an arbitrary sequence
Any signal x(n) can be represented as weighted sum of impulses as given below

The response of the system for unit sample input is called impulse response of the system h(n)

By time invariant property, we have

The above equation is called convolution sum.

Some of the properties of convolution are commutative law, associative law and distributive law.

Correlation of two sequences

It is basically used to compare two signals. It is the measure of similarity between two signals. Some
of the applications are communication systems, radar, sonar etc.
The cross correlation of two sequences x(n) and y(n) is given by

One of the important properties of cross correlation is given by

The auto correlation of the signal x(n) is given by

Downloaded From EnggTree.com

 EnggTree.com

Linear time invariant systems characterized by constant coefficient difference equation

The response of the first order difference equation is given by

The first part contain initial condition y(-1) of the system, the second part contains input x(n) of the
system.
The response of the system when it is in relaxed state at n=0 or
y(-1)=0 is called zero state response of the system or forced response.

The output of the system at zero input condition x(n)=0 is called zero input response of the system
or natural response.

The impulse response of the system is given by zero state response of the system

The total response of the system is equal to sum of natural response and forced responses.

FREQUENCY DOMAIN ANALYSIS OF DISCRETE TIME SIGNALS AND SYSTEMS

A s we have observed from the discussion o f Section 4.1, the Fourier series representation o
f a continuous-time periodic signal can consist of an infinite number of frequency components,
where the frequency spacing between two successive harmonically related frequencies is 1 / T p, and
where Tp is the fundamental period.
Since the frequency range for continuous-time signals extends infinity on both sides it is
possible to have signals that contain an infinite number of frequency components.
In contrast, the frequency range for discrete-time signals is unique over the interval. A
discrete-time signal of fundamental period N can consist of frequency components separated by 2n /
N radians.
Consequently, the Fourier series representation o f the discrete-time periodic signal will
contain at most N frequency components. This is the basic difference between the Fourier series
representations for continuous-time and discrete-time periodic signals.

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

PROPERTIES OF DFT:

LINEAR FILTERING METHODS BASED ON THE DFT

Since the D F T provides a discrete frequency representation o f a finite-duration Sequence in

the frequency domain, it is interesting to exp lore its use as a computational tool for linear system
analysis and, especially, for linear filtering. We have already established that a system with
frequency response H { w ) y w hen excited with an input signal that has a spectrum possesses an
output spectrum.
The output sequence y(n) is determined from its spectrum via the inverse Fourier transform.
Computationally, the problem with this frequency domain approach is that are functions o f the

Downloaded From EnggTree.com

 EnggTree.com
continuous variable. As a consequence, the computations cannot be done on a digital computer, since
the computer can only store and perform computations on quantities at discrete frequencies.
On the other hand, the DFT does lend itself to computation on a digital computer. In the discussion
that follows, we describe how the DFT can be used to perform linear filtering in the frequency
domain. In particular, we present a computational procedure that serves as an alternative to time-
domain convolution.
In fact, the frequency-domain approach based on the DFT, is computationally m ore efficient
than time-domain convolution due to the existence of efficient algorithms for computing the DFT .
These algorithms, which are described in Chapter 6, are collectively called fast Fourier transform
(FFT) algorithms.

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

FAST FOURIER TRANSFORM

EFFICIENT COMPUTATION OF DFT:
In this section we represent several methods for computing dft efficiently. In view of the
importance of the DFT in various digital signal processing applications such as linear filtering,
correlation analysis and spectrum analysis, its efficient computation is a topic that has received
considerably attention by many mathematicians, engineers and scientists. Basically the computation
is done using the formula method.

Divide-and-Conquer Approach to Computation of the DFT

The development of computationally efficient algorithms for the DFT is made possible if we
adopt a divide-and-conquer approach. This approach is based on the decomposition of an N-point
DFT into successively smaller DFT. This basic approach leads to a family o f computationally
efficient algorithm s know n collectively as FFT algorithms.

Downloaded From EnggTree.com

 EnggTree.com
T o illustrate the basic notions, let us consider the computation of an N point DFT , where N can be
factored as a product of two integers, that is, N = L M

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

An additional factor of 2 savings in storage of twiddle factors can be obtained by introducing

a 90° phase offset at the mid point of each twiddle array , which can be removed if necessary at the
ouput of the SRFFT computation. The incorporation of this improvement into the SRFFT results in
an other algorithm also due to price called the PFFT algorithm.

Implementation of FFT Algorithms

Now that w e has described the basic radix-2 and radix -4 F FT algorithm s, let us consider
some of the implementation issues. Our remarks apply directly to

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

UNIT-2&3
STRUCTURES OF FIR AND IIR SYSTEMS
STRUCTURES FOR THE REALIZATION OF DISCRETE-TIME SYSTEMS
The major factors that influence our choice o f a specific realization are computational complexity,
memory requirements, and finite-word-length effects in the computations.

STRUCTURES FOR FIR SYSTEMS

Direct-Form Structure
The direct form realization follows immediately from the non recursive difference equation given
below

Cascade-Form Structures
The cascaded realization follows naturally system function given by equation. It is simple matter to
factor H(z) into second order FIR system so that

Downloaded From EnggTree.com

 EnggTree.com

Frequency-Sampling Structures
The frequency-sampling realization is an alternative structure for an FIR filter in which the
parameters that characterize the filter are the values o f the desired frequency response instead of the
impulse response h(n). To derive the frequency sampling structure, we specify the desired frequency
response at a set o f equally spaced frequencies, namely

The frequency response of the system is given by

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

Lattice Structure
In this section w e introduce another F IR filter structure, called the lattice filter or
Lattice realization. Lattice filters are used extensively in digital speech processing
And in the implementation of adaptive filters. Let us begin the development by considering a
sequence of FIR filters with system functions

Downloaded From EnggTree.com

 EnggTree.com

The general form of lattice structure for m stage is given by’

Downloaded From EnggTree.com

 EnggTree.com

STRUCTURES FOR IIR SYSTEMS

In this section we consider different IIR system s structures described by the difference equation
given by the system function. Just as in the case o f FIR system s, there are several types o f
structures or realizations, including direct-form structures, cascade-form structures, lattice structures,
and lattice-ladder structures. In addition, IIR systems lend them selves to a parallel form realization.
We begin by describing two direct-form realizations.

DIRECT FORM STRUCTURES:

Downloaded From EnggTree.com

 EnggTree.com

DIRECT FORM II

Signal Flow Graphs and Transposed Structures

A signal flow graph provides an alternative, N but equivalent, graphical representation to a

block diagram structure that we have been using to illustrate various system realizations. T he basic
elements o f a flow graph are branches and nodes. A signal flow graph is basically a set o f directed
branches that connect at nodes. By definition, the signal out of a branch is equal to the branch gain
(system function) times the signal into the branch. Furthermore, the signal at anode o f a flow graph
is equal to the sum o f the signals from all branches connecting to the node.

Downloaded From EnggTree.com

 EnggTree.com

Cascade-Form Structures
Let us consider a high-order IIR system with system function given by equation. Without loss o
f generality we assume that N > M . T h e system can be factored into a cascade o f second-
order subsystem s, such that H (z) can b e expressed as

Downloaded From EnggTree.com

 EnggTree.com

Parallel-Form Structures
A parallel-form realization o f an IIR system can be obtained by performing a partial-fraction
expansion o f H( z) . Without loss o f generality, w e again assume that N > M and that the poles are
distinct. Then, by performing a partial-fraction expansion o f H( z ), we obtain the result

Downloaded From EnggTree.com

 EnggTree.com

The realization of second order form is given by

The general form of parallel form of structure is f\given by

Lattice and Lattice-Ladder Structures for IIR Systems

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

IIR AND FIR FILTERS

The transfer function is obtained by taking Z transform of finite sample impulse response. The filters
designed by using finite samples of impulse response are called FIR filters.
Some of the advantages of FIR filter are linear phase, both recursive and non recursive, stable and
round off noise can be made smaller.
Some of the disadvantages of FIR filters are large amount of processing is required and non integral
delay may lead to problems.
DESIGN OF FIR FILTERS

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

IIR FILTER DESIGN

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

************************************************************************

Downloaded From EnggTree.com

 EnggTree.com
UNIT - IV

FINITE WORD LENGTH EFFECTS IN DIGITAL FILTER

Finite Word length Effects:

 In the design of FIR Filters, The filter coefficients are determined by the system transfer functions. These
filter co-efficient are quantized/truncated while implementing DSP System because of finite length registers.
 Only Finite numbers of bits are used to perform arithmetic operations. Typical word length is 16 bits,
24 bits, 32 bits etc.
 This finite word length introduces an error which can affect the performance of the DSP system.
 The main errors are
1. Input quantization error
2. Co-efficient quantization error
3. Overflow & round off error (Product Quantization error)
 The effect of error introduced by a signal process depend upon number of factors including the
1. Type of arithmetic
2. Quality of input signal
3. Type of algorithm implemented
1. Input quantization error
 The conversion of continuous-time input signal into digital value produces an error which is known as
input quantization error.
 This error arises due to the representation of the input signal by a fixed number of digits in A/D
conversion process.
2. Co-efficient quantization error
 The filter coefficients are compared to infinite precision. If they are quantized the frequency response of
the resulting filter may differ from the desired frequency response.
i.e poles of the desired filter may change leading to instability.
3. Product Quantization error
 It arises at the output of the multiplier
 When a ‘b’ bit data is multiplied with another ‘b’ bit coefficient the product (‘2b’ bits) should be stored
in ‘b’ bits register. The multiplier Output must be rounded or truncated to ‘b’ bits. This known as
overflow and round off error.
*****************************************************************************************
Types of number representation:
There are two common forms that are used to represent the numbers in a digital or any other digital hardware.
1. Fixed point representation
2. Floating point representation
* Explain the various formulas of the fixed point representation of binary numbers.
1. Fixed point representation
 In the fixed point arithmetic, the position of the binary point is fixed. The bit to the right represents the
fractional part of the number and to those to the left represents the integer part.

Downloaded From EnggTree.com

 EnggTree.com
 For example, the binary number 01.1100 has the value 1.75 in decimal.
(0*21) + (1*20) + (1*2-1) + (1*2-2) + (0*2-3) = 1.75
In general, we can represent the fixed point number ‘N’ to any desired accuracy by the series
n2
N   Ci r i
i  ni

Where, r is called as radix.

 If r=10, the representation is known as decimal representation having numbers from 0 to 9. In this
representation the number
12
30.285   Ci 10i
i  3

= (3*101 )+ (0*100)+ (2*10-1)+(8*10-2)+(5*10-3)


If r=2, the representation is known as binary representation with two numbers 0 to 1.
 For example, the binary number
110.010 = (1*22 ) + (1*21) + (0*20) + (0*2-1) + (1*2-2) + (0*2-3) = 6.25
Examples:
Convert the decimal number 30.275 to binary form
0.275 * 2 0.55 0
0.55 * 2 1.10 1
0.10 * 2 0.20 0
0.20 * 2 0.40 0
0.40 * 2 0.80 0
0.80 * 2 1.60 1
0.60 * 2 1.20 1
0.20 * 2 0.40 0
(30.275)10 = (11110.01000110)2

In fixed point arithmetic =, the negative numbers are represented by 3 forms.

1. Sign-magnitude form
2. One’s complement form
3. Two’s complement form
1.1 Sign-magnitude form:
 Here an additional bit called sign bit is added as MSB.
o If this bit is zero  It is a positive number
o If this bit is one  It is a positive number
 For example
o 1.75 is represented as 01.110000.
o -1.75 is represented as 11.110000
1.2 One’s complement form:
 Here the positive number is represented same as that in sign magnitude form.
 But the negative number is obtained by complementing all the bits of the positive number
 For eg: the decimal number -0.875 can be represented as
o (0.875)10=(0.111000)2
o (-0.875)10=(1.000111)2 0.111000
↓ ↓↓↓↓↓↓ (Complement each bit)
1.000111

Downloaded From EnggTree.com

 EnggTree.com
1.3 Two’s complement form:
 Here the positive numbers are represented as same in sign magnitude and one’s complement form.
 The negative numbers are obtained by complementing all the bits of the positive number and adding
one to the least significant bit
(0.875)10=(0.111000)2
↓ ↓↓↓↓↓↓ (Complement each bit)
1.000111
+ 1
1.001000
(-0.875)10=(1.001000)2
Examples:
 Find the sign magnitude, 1’s complement, 2’s complement for the given numbers.
7
1. 
32
7
2. 
8
7
3. 
8
7
1. 
32
0.21875 * 2 0.43750 0
0.43750 * 2 0.87500 0
0.87500* 2  1.750000 1
0.75* 2 1.50 1
0.50* 2 1.00 1
7
 =(-0.21875)10 =(1.00111)2
32
Sign magnitude form = 1.00111
1’s complement form = 1.11000
2’s complement form = 1.11001
7
2. 
8
0.875 * 2 1.75 1
0.750 * 2 1.500 1
0.500 * 2 1.000 1
7
 =(-0.875)10 =(0.111)2
8
Sign magnitude form = 0.111
1’s complement form = 1.000
2’s complement form = 1.001
7
3. 
8
Sign magnitude form = 0.111
1’s complement form = 0.111
2’s complement form = 0.111
Addition of two fixed point numbers:
 Add (0.5)10 + (0.125)10

Downloaded From EnggTree.com

 EnggTree.com
(0.5)10 = (0.100)2
(0.125)10 = (0.001)2
(0.101)2 = (0.625)10
 Addition of two fixed point numbers causes an overflow.
For example
(0.100)2
(0.101)2
(1.001)2 = (-0.125)10 in sign magnitude form
Subtraction of two fixed point numbers:
 Subtraction of two numbers can be easily performed easily by using two’s complement representation.
 Subtract 0.25 from 0.5
0.25 * 2 0.50 0 Sign magnitude form = (0.010)2
0.50 * 2 1.00 1 1’s complement form = (1.101)2
0.00 * 2 0.00 0 2’s complement form = (1.110)2
(0.5)10 = (0.100)2
-(0.25)10 = (1.110)2 Two’s complement of -0.25
(10.010)2
Here the carry is generated after the addition. Neglect the carry bit to get the result in decimal.
(0.010)2 = (0.25)10
 Subtract 0.5 from 0.25
0.5 * 2 1.00 1 Sign magnitude form = (0.100)2
0.00 * 2 0.00 0 1’s complement form = (1.011)2
0.00 * 2 0.00 0 2’s complement form = (1.100)2
(0.25)10 = (0.010)2
-(0.5)10 = (1.100)2
(1.110)2
Here the carry is not generated after the addition. So the result is negative.
Multiplication in fixed point arithmetic:
 Here the sign magnitude components are separated.
 The magnitudes of the numbers are multiplied. Then the sign of the product is determined and applied to
the result.
 In the fixed point arithmetic, multiplication of two fractions results in a fraction.
 For multiplications with fractions, overflow can never occur.
 Eg:
(0.1001)2 * (0.0011)2 = (0.00011011)2
2. Floating point representation
 Here, a number ‘x’ is represented by
X=M.re
Where, M  Mantissa which requires a sign bit for representing positive number and negative
numbers.
R  base (or) radix
e  exponent which require an additional and it may be either positive or negative.
o For eg, 278 can be represented in floating point representation.
278 X 1000
278 = = 0.278*103
1000
0.278  Mantissa (M)
10  base (or) radix (r)
3  exponents (e)
Downloaded From EnggTree.com
 EnggTree.com
 Similarly, to represent a binary floating point number
X=M.2e in which the fractional part of a number should fall (or) lie in the range of 1/2 to 1.
5 X 8
5 = 0.625 X23
8
Mantissa (M) = 0.625
Base (or) radix (r) = 2
Exponent (e) = 3
 Some decimal numbers and their floating point representations are given below:
4.5  0.5625 X 23 =0.1001 X 2011
1.5  0.75 X 21 =0.1100 X 2001
6.5  0.8125 X 23 =0.1100 X 2011
0.625  0.625 X 20 =0.1010 X 2000
 Negative floating point numbers are generally represented by considering the mantissa as a fixed point
number. The sign of the floating point number is obtained from the first bit of mantissa.
 To represent floating point in multiplication
Consider X 1  M 1 r e1
X 2  M 2 r e2
X 1 X 2  M 1 * M 2 r   
e1 e2

Example
Given X 1  3.5 * 10 12 , X 2  4.75 * 10 6 . Find the product X 1 X 2
X=(3.5 X 4.75) 10(-12+6)
= (16.625)10-6  in decimal
In binary: (1.5)10 X (1.25)10 = (210.75) X (210.625)
= 2001 X 0.1100 X 2001 X 0.1010
= 2010 X 0.01111
Addition and subtraction:
 Here the exponent of a smaller number is adjusted until it matches the exponent of a larger number.
 Then, the mantissa are added or subtracted
 The resulting representation is rescaled so that its mantissa lies in the range 0.5 to 1.
 Eg: Add (3.0)10 & (0.125)10
(3.0)10 =2010 X 0.1100 = r e1 X M 1
(0.125)10 =2000 X 0.0010 = r e2 X M 2
Now adjust e2 Such that e1=e2
(0.125)10 =2010 X 0.0000100
Addition 2010 (0.110000 + 0.0000100) 2010 X 0.110010
Subraction 2010 X 1.001101
Compare floating point with fixed point arithmetic.
Sl.No Fixed point arithmetic Floating point arithmetic
1 Fast operation Slow operation
2 Relatively economical More expensive because of costlier hardware
3 Small dynamic range Increased Dynamic range
4 Round off errors occurs only for Round off errors can occur with addition and
addition multiplication
5 Overflow occur in addition Overflow does not arise
6 Used in small computers Used in large general purpose computers.

Downloaded From EnggTree.com

 EnggTree.com
Quantization:
*Discuss the various methods of quantization.
*Derive the expression for rounding and truncation errors
* Discuss in detail about Quantization error that occurs due to finite word length of registers.
The common methods of quantization are
1. Truncation
2. Rounding
1. Truncation
 The abrupt termination of given number having a large string of bits (or)
 Truncation is a process of discarding all bits less significant than the LSB that is retained.
 Suppose if we truncate the following binary number from 8 bits to 4 bits, we obtain
 0.00110011 to 0.0011
(8 bits) (4 bits)
 1.01001001 to 1.0100
(8 bits) (4 bits)
 When we truncate the number, the signal value is approximated by the highest quantization level that is
not greater than the signal.
2. Rounding (or) Round off
 Rounding is the process of reducing the size of a binary number to finite word size of ‘b’ bits such that
the rounded b-bit number is closest to the original unquantised number.
Error Due to truncation and rounding:
 While storing (or) computation on a number we face registers length problems. Hence given number is
quantized to truncation (or) round off.
i.e. Number of bits in the original number is reduced register length.
Truncation error in sign magnitude form:
 Consider a 5 bit number which has value of
0.110012  (0.7815)10
 This 5 bit number is truncated to a 4 bit number
0.11002  (0.75)10
i.e. 5 bit number 0.11001 has ‘l’ bits
4 bit number 0.1100 has ‘b’ bits
 Truncation error, et = 0.1100 – 0.11001
= -0.00001  (-0.03125)10
 Here original length is ‘l’ bits. (l=5). The truncated length is ‘b’ bits.
 The truncation error, et = 2-b-2-l
= -(2-l-2-b)
et = -(2-5-2-4) = -2-1
 The truncation error for a positive number is
 
 2  b  2  l  et  0  Non causal
 The truncation error for a negative number is
 
0  et  2  b  2  l  Causal
Truncation error in two’s complement:
 For a positive number, the truncation results in a smaller number and hence remains same as in the case
of sign magnitude form.
 For a negative number, the truncation produces negative error in two’s complement
   
 2  b  2  l  et  2  b  2  l

Downloaded From EnggTree.com

 EnggTree.com
Round off error (Error due to rounding):
 Let us consider a number with original length as ‘5’ bits and round off length as ‘4’ bits.
0.11001 Round
 offto
 0.1101
2 b  2 l
 Now error due to rounding er =
2
Where bNumber of bits to the right of binary point after rounding
LNumber of bits to the right of binary point before rounding
 Rounding off error for positive Number:
2 b  2 l
  er  0
2
 Rounding off error for negative Number:
2 b  2 l
0  er 
2
 For two’s complement
2 b  2 l 2 b  2 l
  er 
2 2
***************************************************************************************
Quantization Noise:
*Derive the expression for signal to quantization noise ratio
*What is called Quantization Noise? Derive the expression for quantization noise power.

 The analog signal is converted into digital signal by ADC

 At first, the signal x(t) is sampled at regular intervals t=nT, where n=0,1,2… to create sequence x(n).
This is done by a sampler.
 Then the numeric equivalent of each sample x(n) is expressed by a finite number of bits giving the
sequence xq(n)
 The difference signal e(n)= xq(n)- x(n) is called quantization noise (or) A/D conversion noise.
 Let us assume a sinusoidal signal varying between +1 & -1 having a dynamic range 2
 ADC employs (b+1) bits including sign bit. In this case, the number of levels available for quantizing
x(n) is 2b+1.
 The interval between the successive levels is
2
q  b 1  2 b
2
Where q  quantization step size
If b=3 bits, then q=2-3=0.125
***************************************************************************************
Quantization Noise power:
Input Quantization error:
*Derive the equation for quantization noise power (or) Steady state Input Noise Power.

Downloaded From EnggTree.com

 EnggTree.com

Probability density function for round off error in A/D conversion is

 en   , then the error lies between

q q
If rounding is used for quantization, which is bounded by 
2 2
q q
 to with equal probability, where q quantization step size.
2 2
Properties of analog to digital conversion error, e(n):
1. The error sequence e(n) is a sample sequence of a stationary random process.
2. The error sequence is uncorrelated with x(n) and other signals in the system.
3. The error is a white noise process with uniform amplitude probability distribution over the range of
quantization error.
The variance of e(n) is given by
 
 e2  E e 2 n   E 2 en  ---------------------------->(1)
 
Where E e 2 n  Average of e2(n)
E en  Mean value of e(n).
q q
For rounding, e(n) lies between  and with equal probability
2 2

  e
E e 2 n   2
n  pe de ---------------------------->(2)


pe  ,   en   ---------------------------->(3)

1 q q
q 2 2
Substituting (3) in (2)
q

 
2
E e 2 n    e 2 n  de
1
q q

2
q

  1 2
E e 2 n    e 2 n de ------------------------------->(4)
q q

2

E en   0
E 2 en   0 ------------------------------------------->(5)
Substituting (4) and (5) in (1)
q
2
e n de  0
1 2
q q
 e2 

2

Downloaded From EnggTree.com

 EnggTree.com
q
1  e3  2
  
q  3  q
2

1  q   q  
3 3

       
3q  2   2  
1  q 3   q 3 
      
3q  8   8 

1  q 3   q 3 
     
3q  8   8 
1  2q 3 
  
3q  8 
q2
 e2  ------------------------------------------------->(6)
12
1
In general,  2 b  q -------------------------------------------->(7)
2b

 2

2 b 2

e
12
2 b
2
 e2  ----------------------------------------------->(8)
12
Equation (8) is known as the steady state noise power due to input quantization.
R
q b  in two’s complement representation.
2
R
q b  in sign magnitude (or) one’s complement representation.
2 1
R  Range of analog signal to be quantized.
Steady state Output Noise power:

After quantization, we have noise power  e2 as input noise power. Therefore, Output noise power of system is
given by
 
 eo2   e2  h 2 n  ------------------------------------>(9)
 n 0 

where h(n)  impulse response of the system.

Let error E(n) be output noise power due to quantization
Error E n   en   hn 

  hn en  k 
k 0

Downloaded From EnggTree.com

 EnggTree.com
The variance of error E(n) is called output noise power,  e2 .
By using Parseval’s theorem,

 eo2   e2  h 2 n 
n 0

1
2j  H Z H Z 1
dZ
z
  e2  
Where the closed contour integration is evaluated using the method of residue by taking only the poles that
lie inside the unit circle.

Z transform of h(n), H Z    hn z  n
n 0
 
Z transform of h2(n) = Z[h2(n)]   h 2 n z  n   hn hn z  n -------------------->(10)
n 0 n 0

hn   H Z Z n 1 dZ -------------------------------->(11)
1
2j 
By Inverse Z transform,

Substituting (11) in (10)

 

 h 2 n z n   H Z Z n 1 dZ hn z  n
1
n 0 n 0 2 j 
 

1
 H Z   hn Z 1  dZ
2j  n 0 

  dZ
 h 2 n    hn Z 1  n
1
n 0 2j  H Z  
 n 0 Z
 

1
2j 
H Z  hn  Z
n
  1
Z 1 dZ 
 n 0 

 h n   2j  H Z H Z  Z ----------------------------------->(12)

2 1 dZ 1

n 0

Substituting (12) in (9)

 1 
 eo2   e2   H Z H Z 1 Z 1 dZ   
 2j 
Problem:
The output signal of an A/D converter is passed through a first order low pass filter, with transfer
function given by
H ( z) 
1  a z for 0  a  1. Find the steady state output noise power due to quantization at the
za
output of the digital filter. [Nov/Dec-2015]
Solution:
1
 H (z)H (z
1
e  e ) z  1 dz
2 2

2 j c

H(z) 
1  a  z 1
)
1  a  z 1
Given
(z  a)
H(z
z 1
 a
1
Substituti ng H(z) and H(z ) in equation (1), we have
e
2
1  a  z 1  a  z 1 e
2
1  a 2 dz
e   z  1 dz   z 1
2

2 j c
(z  a) z 1
 a 2 j c
( z  a ) z  a 
1

Downloaded From EnggTree.com

 EnggTree.com

2 1
  e  residue of H(z) H(z 1 ) z 1at z  a  residue of H(z) H(z 1 ) z 1at z  
 a
2
  e  z  a 
1  a 2 z 1  0
 z  a z 1  a  
2  1  a   2  1  a  
2
  e  1   e 
 z  a 

 1  a 
2 2 b
Where,  e 
2

12
**************************************************************************************
Find the steady state variance of the noise in the output due to quantization of input for the first order
filter.
y (n )  a y (n  1)  x (n )
Solution:
The impulse response for the above filter is given by h(n)  an u(n)

 2   e2  h 2 ( n )
k 0

  e2  a 2 n
k 0

  1  a 2  a 4  .... 
2
e

1
  e2
1  a2
2 2 b  1 
 ( or )
12  1  a 2 

Taking Z-transform on both sides we have

Y ( z )  az  1Y ( z )  X ( z )
Y (z) 1 z
H (z)   
X ( z ) 1  az  1 za
z 1
H ( z 1 )  1
z a
We know
1
c H ( z ) H ( z )z dz
1 1
 2   e2
2 j 
Substituting H ( z ) and H ( z 1 ) value s in the above equation we get
1 z z 1
c z  a z 1  a z dz
1
 2   e2
2 j 
1 z 1
 2   e2 c ( z  a)( z 1  a)dz
2 j 
 z 1 
 residue of 1
at z  a 
2  ( z  a)( z  a) 
 e
 z 1 
 residue of 1
at z  1 / a 
 ( z  a)( z  a) 
 z 1 
  e2  ( z  a ) 
 ( z  a )( z 1  a ) 
z a 

Downloaded From EnggTree.com

 EnggTree.com
1
a 1
  e2 1
  e2
a a 1  a2

***************************************************************************************
The output of the A/D converter is applied to a digital filter with the system function
H Z  
0.45Z
Z  0.72
Find the output noise power of the digital filter, when the input signal is quantized to 7 bits.
Given:
H Z  
0.45Z
Z  0.72
Solution:
0.45Z 1
 
H Z H Z 1 Z 1 
0.45Z
 1
Z  0.72 Z  0.72
 Z 1

0.45 2 Z 1

Z  0.72 1  0.72 
Z 
0.2025Z 1

Z  0.72 1  0.72Z 
 Z 
0.2025Z 1 Z

Z  0.72 Z  1 
 0.72 
 0.28125

Z  0.72Z  1.3889
Now the poles of H(Z)H(Z-1)Z-1 are p1=0.72 , p2=1.3889
Output noise power due to input quantization
 1 
 eo2   e2    
H Z H Z 1 Z 1 dZ 
 2j 

   
N
  e2  Re s H Z H Z 1 Z 1
i 1 z  pi

   
N
  e2  Re s H Z H Z 1 Z 1
i 1 z  pi
-1 -1
Where p1,p2,…..pn are the poles of H(Z)H(Z ) Z that lies inside the unit circle in z-plane.

 0.28125
 eo2   e2  Z  0.72 
Z  0.72Z  1.3889 Z 0.72
 0.28125
  e2 
0.72  1.3889
 0.4205 e2
***************************************************************************************
1 1
Consider the transfer function H ( z )  H 1 ( z ) H 2 ( z ) where H 1 ( z )  1
and H 2 ( z ) 
1  a1 z 1  a2 z 1

Downloaded From EnggTree.com

 EnggTree.com
Find the output round off noise power. Assume  1  0.5 and  2  0.6 and find output round off noise
power.
Solution:
The round off noise model for H ( z )  H 1 ( z ) H 2 ( z ) is given by,
From the realization we can find that the noise transfer function seen by noise source e1(n) is H(z), where,
1
H ( z) 
1  a1 z 1  a2 z 1                         (1)
1

Whereas, the noise transfer function seen by e2 ( n ) is,

1
H 2 ( z) 
1  a2 z 1                             (2)
The total steady state noise variance can be obtained, we have
 0   01   02                             ( 3)
2 2 2

1
 01   H ( z ) H ( z 1 ) z 1dz
2

2j c
1 1 1 1 1
 e  z 1dz
2
1 1
2j c 1  a1 z 1  a 2 z 1  a1 z 1  a 2 z
2 1 1
  e  of residue of H ( z ) H ( z 1 ) z 1 at po1es z  a1 , z  a 2 , z  and z  
 a1 a2 
If a1 and a2 are less than the poles z=1/a1 and z=1/a2 lies outside of the circle z  1. So, the residue of H(z) H(z-1) z-
1
at z=1/a1 and z=1/a2 are zero. Consequently we have,
 01 
2
 of residue of H(z)H(z 1
) z 1 at po1es z  a1 , z  a 2 
 z 1 z 1 
  z  a1     
 1  a1 z 1  a2 z 1  a1 z 1  a2 z  za1
1 1
z a 2
1  a1 z 1  a2 z 1  a1 z 1  a2 z  za2 
1 1

 
 
 e  
1 1

2

 
a2 
2
  a2 
 1   1  a 2 1  a1a 2  1  1  a1a 2  1  a 2
2
 

 a1   a1  
2 a1 1 1 a2 1 1 
 01   e                 ( 4)
2
. . . .
 a1  a 2 1  a1 1  a1a 2 a 2  a1 1  a 2 1  a1a 2 
2 2

In the same way,

e
2

2j c
 02  H 2 ( z ) H 2 ( z 1 ) z 1dz
2

e
2
1 1
  1
2j c 1  a2 z 1  a2 z
z 1dz

 z 1 
  e   z  a2 
1  a2 z 1 1  a2 z  za2 
2


 z 1 
  e z  a 2 z 1 
1  a2 z 1  a2 z  za2 
2
1


2 1 
 e  2
                                     (5)
 1  a 2 

Downloaded From EnggTree.com

 EnggTree.com
2 1 a1 1 1 a2 1 1 
0  e   
2
. . . . 
1  a 2
2
a1  a 2 1  a1 1  a1a 2 a 2  a1 1  a 2 1  a1a 2 
2 2

2
 e 
1


a1 1  a 2  a 2 1  a1
2 2 2
   
 1  a2
2
   
1  a1 1  a 2 1  a1a 2 a1  a 2 
2 2

2
 e 
1

a1  a2 1  a1a2  
1  a 2
2
   
1  a1 1  a 2 1  a1a 2 a1  a 2 
2 2


 1
2 2 b

1  a1a2  

1  a 2
12 2
  
1  a1 1  a 2 1  a1a 2 
2 2

The steady state noise power for a1  0.5, a 2  0.6 is given by

2 2 b  1 1  0.50.6 
 
12

1  0.6
2
  
1  0.5 1  0.6 1  0.60.5
2 2

2 2 b
 5.4315
12
*************************************************************************************
1
Draw the quantization noise model for a second order system H ( z )  and find the
1  2r cos z 1  r 2 z 2
steady state output noise variance.
Solution:
Given:
1
H ( z) 
1  2r cos z 1  r 2 z 2
The quantization noise model is,
we know,  0   01   02
2 2 2

Both noise sources see the same transfer function

1
H ( z) 
1  2r cos z 1  r 2 z 2
The impulse response of the transfer function is given by
sin(n  1)
h(n )  r n u(n )
sin 
Now the steady state output noise variance is,
 0   01   02
2 2 2


But  01   02   e
2 2 2
h
n  
2
(n ), which gives us

Downloaded From EnggTree.com

 EnggTree.com
2 2 b 
sin 2 ( n  1)
 0  2.
2

12
r
n 0
2n

sin 2 
2 b 

 r 1  cos 2n  1 

2 1
 2. 2n
 cos2  1  2sin 2
12 2 sin 2  n 0
2 b
2 1   

 2  
6 2 sin   n 0
r 2n
 
n 0
r 2 n cos 2( n  1) 

2 2 b 1  1 1   2 n j 2 ( n 1)  2 n  j 2 ( n 1) 

6 2 sin 2  1  r 2  2   r e  r e 
  n 0 n 0 
2 2 b 1  1 1  e j 2 e  j 2 
     
6 2 sin 2  1  r
2
2  1  r 2 e 2 j 1  r 2 e 2 j 
2 2 b 1  1 cos 2  r 2 
 
1  r 2 1  2r 2 cos 2  r 4 
6 2 sin 2   


2 2 b 1  1  r  1  cos 2 
2

6 2 sin   1  r 1  2r cos 2  r 
2  2 2 4 


2 2 b 1  r  2

6 1  r 1  2r 2 cos 2  r 4 
2

******************************************************************************
Co-efficient quantization error
 We know that the IIR Filter is characterized by the system function
M

b z k
k

H Z   k 0
N
1   ak z k
k 1

 After quantizing ,
M

 b  k q z k
H Z q  k 0
N
1   a k q z  k
k 1

Where ak q  a k  a k
bk q  bk  bk
 The quantization of filter coefficients alters the positions of the poles and zeros in z-plane.
1.
If the poles of desired filter lie close to the unit circle, then the quantized filter poles may lie outside
the unit circle leading into instability of filter.
2. Deviation in poles and zeros also lead to deviation in frequency response.
***************************************************************************************
Consider a second order IIR filter with H ( z )  1 .0 find the effect on quantization
(1  0 .5 z  1 )(1  0 .4 5 z  1 )
on pole locations of the given system function in direct form and in cascade form. Take b=3bits.
[Apr/May-10] [Nov/Dec-11]
Solution:
Given that,
1 .0
H (z) 
(1  0 .5 z  1 )(1  0 .4 5 z  1 )

Downloaded From EnggTree.com

 EnggTree.com
1
H (z) 
z ( z  0.5 z ) z  1 ( z  0.5)
1 1

z2

( z  0.5)( z  0.45)
The roots of the denominator of H(z) are the original poles of H(z). let the original poles of H(z) be p 1 and
p2 .
Here p1=0.5 and p2=0.45
Direct form I:
1 .0
H (z) 
(1  0 .5 z )(1  0 .4 5 z  1 )
1

1
H ( z) 
1  0.5 z  0.45 z 1  0.225 z  2
1

1

1  0.95 z  0.225 z 2
1

Let us quantize the coefficients by truncation.

Convert to Truncate to Convert to
.9510 .11112 .1112 .87510
Binary 3-bits decimal
Convert to Truncate to Convert to
.22510 .00112 .0012 .12510
Binary 3-bits decimal
Let H ( z ) be the transfer function of the IIR system after quantizing the coefficients.
1
H ( z) 
1  0.875 z  0.125 z 2
1

Y ( z) 1
let H ( z )  
X ( z ) 1  0.875 z  0.125 z 2
1

On cross multiplying the above equation we get,

Y ( z )  0.875 z 1Y ( z )  0.125 z 2Y ( z )  X ( z )
Y ( z )  X ( z )  0.875 z 1Y ( z )  0.125 z 2Y ( z )
Cascade form:
Given that
1.0
H ( z) 
(1  0.5 z )(1  0.45 z 1 )
1

In cascade realization the system can be realized as cascade of first order sections.
H(z)=H1(z)+H2(z)
Where,
1 1
H 1 (z)  1
and H 2 (z) 
1  0.5 z 1  0.45 z  1
Let us quantize the coefficients of H1(z) and H2(z) by truncation.
Convert to Truncate to Convert to
.510 .10002 .1002 .510
Binary 3-bits decimal
Convert to Convert to Convert to
.4510 .01112 .0112 .37510
Binary 3-bits decimal
let , H1 (z) and H 2 (z) be the transfer function of the first-order sections after quantizing the coefficients.

Downloaded From EnggTree.com

 EnggTree.com
1
H1 (z) 
1  0.5 z 1
1
H 2 (z) 
1  0.375 z 1
Y (z) 1
let , H1 (z)  1 
X (z) 1  0.5 z 1
Y1 (z)  0.5z 1 Y1 (z)  X(z)
Y1 (z)  X (z)  0.5z 1 Y1 (z)
Y (z) 1
let, H 2 (z)  
Y1 (z) 1  0.375 z 1
on cross multiplying the above equation we get ,
Y (z)  0.375z 1 Y (z)  Y1 (z)
Y (z)  Y1 (z)  0.375z 1Y(z)
***************************************************************************************
Round off effects and overflow in digital filter:
*Explain in detail about round off effects in digital filters.
 The presence of one or more quantizer in the realization of a digital filter results in a non-linear device.
i.e. recursive digital filter may exhibit undesirable oscillations in its output
 In the finite arithmetic operations, some registers may overflow if the input signal level becomes large.
 These overflow represents non-linear distortion leading to limit cycle oscillations
 There are two types of limit cycle oscillations which includes
1. Zero input limit cycle oscillations (Low amplitude compared to overflow limit cycle oscillations)
2. Over flow limit cycle oscillations.
Zero input limit cycle oscillations
 The arithmetic operations produces oscillations even when the input is zero or some non zero constant
values. Such oscillations are called zero input limit cycle oscillations.
Overflow limit cycle oscillations
 The limit cycle occurs due to the overflow of adder is known as overflow limit cycle oscillations.
Dead Band:
The limit cycle occurs as a result of quantization effect in multiplication. The amplitude of the
output during a limit cycle is confined to a range of values called the dead band of the filter.
2 b
y(n  1)  2
1  a 
Consider a first order filter
y n   ay n  1  x n ; n >0
After rounding the product
y q n   Qa  y n  1  x n ;
The round off error
2 b 2 b
  er 
2 2
where, er difference between the quantized value and the actual value.
2 b
Qay n  1  ay n  1 
2
The dead band of the filter for the limit cycle oscillations are
 yn  1 a  0
Qayn  1  
 yn  1 a  0

Downloaded From EnggTree.com

 EnggTree.com
2b
y(n  1)  a y(n  1) 
2
2b
y(n  1) 1  a  
2
2 b
Dead band of the filter, y(n  1)  2
1  a 
**************************************************************************************
Problem: Consider a 1st order FIR system equation y ( n)  x ( n)  ay ( n  1) with
0.875 , n  0
x ( n)  
0 , otherwise
Find the limit cycle effect and the dead band. Assume b=4 and a=0.95. (Nov/Dec-12)(Nov/Dec-15)
[May/June-2016]
Solution:
Given:
0.875 , n  0
x(n)  
0 , otherwise
2 b 2 4
Dead band    0.625
21  a  21  0.95 
y ( n )  x ( n )  0.95 y ( n  1)
Qay ( n  1)
n x(n) y ( n  1) ay ( n  1) y ( n )  x ( n )  Q ay ( n  1)
(round off to 4-bits)
0 0.875 0 0 0.0000 y(0)=0.875
0.875 * 0.95
 0.11012
1 0 0.875  0.8312510 y(1)=0.8125
 0.8125
 0.11010 2
0.8125 * 0.95
 0.1100 2
2 0 0.8125  0.7718710 y ( 2)  0.75
 0.75
 0.1100012
0.75 * 0.95
 0.10112
3 0 0.75  0.712510 y ( 3)  0.6875
 0.6875
 0.10110112
0.6875 * 0.95
 0.1010 2
4 0 0.6875  0.65312510 y ( 4)  0.625
 0.625
 0.1010012
0.625 * 0.95
 0.1010 2
5 0 0.625  0.5937510 y (5)  0.625
 0.625
 0.100112
0.625 * 0.95
 0.1010 2
6 0 0.625  0.5937510 y ( 6)  0.625
 0.625
 0.100112
Conclusion:

Downloaded From EnggTree.com

 EnggTree.com
The dead band of the filter is 0.625. When n  5 the output remains constant at 0.625 causing
limit cycle oscillations.
***************************************************************************************
Overflow Limit cycle oscillations:
*What are called overflow oscillations? How it can be prevented?
 We know that the limit cycle oscillation is caused by rounding the result of multiplication.
 The limit cycle occurs due to the overflow of adder is known as overflow limit cycle oscillations.\
 Several types of limit cycle oscillations are caused by addition, which makes the filter output oscilate
between maximum and minimum amplitudes.
 Let us consider 2 positive numbers n1 & n2
n1=0.1117/8
n2=0.1106/8
n1 + n2=1.101-5/8 in sign magnitude form.
The sum is wrongly interpreted as a negative number.
 The transfer characteristics of an saturation adder is shown in fig below
where n  The input to the adder
f(n)  The corresponding output

Saturation adder transfer characteristics

 From the transfer characteristics, we find that when overflow occurs, the sum of adder is set equal to the
maximum value.
***************************************************************************************
Signal Scaling:
Explain how reduction of round-off errors is achieved in digital filters.

 Saturation arithmetic eliminates limit cycles due to overflow, but it causes undeniable signal distortion
due to the non linearity of the clipper.
 In order to limit the amount of non linear distortion, it is important to scale input signal and unit sample
response between input and any internal summing node in the system to avoid overflow.

Realization of a second order IIR Filter

 Let us consider a second order IIR filter as shown in the above figure. Here a scale factor S0 is
introduced between the input x(n) and the adder 1 to prevent overflow at the output of adder 1.

Downloaded From EnggTree.com

 EnggTree.com
 Now the overall input-output transfer function is

Now the transfer function

b0  b1 z  1  b2 z  2
H ( z)  S0
1  a1 z  1  a 2 z  2
N (z)
 S0
D(z)

From figure
W ( z) S0 S
H ' ( z)   1 2
 0
X ( z ) 1  a1 z  a2 z D( z)
S0 X ( z )
W ( z)   S 0 S( z ) x(Z)
D( z)
1
Where S ( z ) 
D( z)
we have
S
w( n )  0  S (e j ) X (e j )(e jn ) d 
2
S2 2
w(n) 2  0 2  S (e j ) X (e j )(e jn ) d 
2
Using Schwartz inequality

w(n) 2  S 02   S ( e j ) d     X ( e j ) d  
2 2

 2   2 

Applying parsevals theorem


1
w(n) 2  S 02  x 2 ( n ) 
2
2 S ( e j ) d 
n0 2
if z  e j then dz  je j d 
which gives
dz
d 
jz

By substituting all values


1
w(n) 2  S 02  x 2 ( n ) 
2
S ( z ) z  1d z
2 j
c
n0

1
w(n) 2  S 02  x 2 ( n )  S ( z ) S ( z  1 ) z  1d z
2 j
c
n0

w 2 ( n )   x 2 ( n ) when
n0

1
S 02
2 j  c S ( z ) S ( z 1 ) d z  1

Which gives us,

Downloaded From EnggTree.com

 EnggTree.com
1
S02 
1
2 j 
1 1
c S ( z)S ( z ) z d z

1

1 z 1dz
2 j  D( z ) D( z 1 )
c

1
S02 
I
Where I=

Note:
 Because of the process of scaling, the overflow is eliminated. Here so is the scaling factor for the first
stage.
1
 Scaling factor for the second stage = S01 and it is given by S 012  2
S0 I 2
 
1 H 1 Z H 1 Z 1 Z 1
2j c D2 Z D2 Z 1
Where I 2 
  dZ

********************************************************************************************
0.25  0.7 Z 1
For the given transfer function, H Z   , find scaling factor so as to avoid overflow in
1  0.5Z 1
the adder ‘1’ of the filter.

Given:
D(Z) = 1-0.5Z-1
D(Z-1) = 1-0.5 Z
Solution:
1 1 dZ
I  
2j c DZ D Z  
1
Z
1 1 dZ
   
2j c 1  0.5Z 1  0.5Z  Z
1

1 Z 1 dZ
 
2j c Z  0.5 1  0.5Z  Z
Z 1
Residue of 0
Z  0.5 1  0.5Z  Z 0.5
I=1.3333
1
S0 =
I

Downloaded From EnggTree.com

 EnggTree.com
1
S0 =
1.333
= 0.866
***************************************************************************************
Consider the recursive filter shown in fig. The input x(n) has a range of values of ±100V, represented
by 8 bits. Compute the variance of output due to A/D conversion process. (6)

Solution:
Given the range is ±100V
The difference equation of the system is given by y ( n )  0 .8 y ( n  1)  x ( n ) , whose impulse response h(n)
can be obtained as
h ( n )  (0.8) n u ( n )
range of the signal
quantization step size 
No. of quantization levels
200
 8
2
 0.78125
Variance of the error signal
q2
 e2 
12
(0.78125) 2

12
 e  0.05086
2

Variance of output

 2   e2  h 2 ( n )
n0

 (0.05086)  (0.8) 2 n
n0

0.05086
  0.14128
1  (0.8) 2
***************************************************************************************
The input to the system y(n)=0.999y(n-1)+x(n) is applied to an ADC. What is the power produced by
the quantization noise at the output of the filter if the input is quantized to a) 8 bits b)16 bits. May-07
Solution:
y(n)=0.999y(n-1)+x(n)
Taking z-transform on both sides
Y(z)=0.999z-1Y(z)+X(z)

Y (z) 1
H (z)  
X ( z ) 1  0 .9 9 9 z  1

Downloaded From EnggTree.com

 EnggTree.com
z z 1
H ( z ) H ( z 1 ) z 1  ( )( 1 ) z 1
z  0.999 z  0.999
z 1

1
( z  0.999)(  0.999)( z  )
0.999
 0.001

( z  0.999)( z  0.001)

output noise pow er due  2 1

 H (z)H (z
1
  eoi   e
2
) z  1 dz
to input quantization  2  j c
N
 2
e  R e s  H ( z ) H ( z
i 1
1
) z  1 
z  pi
N
  e2   ( z  p i ) H ( z ) H ( z  1 ) z  1 
i 1 z  pi

Where p1,p2,……pN are poles of H(z)H(z-1)z-1, that lies inside the unit circle in z-plane.
0.001
 eoi 2   e 2 ( z  0.999)( )
( z  0.999)( z  0.001) z  0.999
  e 2 500.25

a) b+1=8 bits(Assuming including sign bit)

2 2(7)
 2  (500.25)  2.544  10  3
12

b) b+1=16 bits
2 2 (1 5 )
 2  (5 0 0 .2 5 )  3 .8 8 2  1 0  8
12
***************************************************************************************
Find the effect of coefficient quantization on pole locations of the given second order IIR system, when
it is realized in direct form I and in cascade form. Assume a word length of 4 bits through truncation.
1
H (z) 
1  0.9 z  1  0.2 z  2
Solution:
Direct form I
Let b=4 bits including a sign bit

Downloaded From EnggTree.com

 EnggTree.com
(0.9) 10  (0.111001...) 2
Integer part
0.9  2
1.8
 1 
0.8  2
1.6
 1
0.6  2
1
1.2

0.2  2
0.4
 0
0.4  2
0.8
 0
0.8  2
1.6
 1

After truncation we get

(0.111)2=(0.875)10
(0.2)10=(0.00110…)2
0.2  2
(0.2)10 
0.4
 0 
0.4  2
0.8
 0
0.8  2
1.6
 1
0.6  2
1.2
 1
0.2  2
0.4
 0
After truncation we get
(0.001)2=(0.125)10
The system function after coefficient quantization is
1
H (z)  1
1  0.875 z  0.125 z  2
Now the pole locations are given by
z1=0.695
z2=0.178

If we compare the Poles of H(z) and H ( z ) we can observe that the poles of H ( z ) deviate very much from
the original poles.
Cascade form
1
H ( z) 
1  0.5z (1  0.4 z 1 )
1

(0.5)10  (0.1000)2
After truncation we get

Downloaded From EnggTree.com

 EnggTree.com
(0.100)2=(0.5)10
After truncation we get
(0.011)2=(0.375)10
0 .4  2
( 0 .4 ) 1 0 
0 .8
 0 
0 .8  2
1 .6
 0
0 .6  2
1 .2
 1
0 .2  2
0 .4
 1
0 .4  2
0 .8
 0
(0.4)10=(0.01100…)2
The system function after coefficient quantization is
1
H (z) 
(1  0.5 z 1 )(1  0.375 z 1 )

The pole locations are given by

z1=0.5
z2=0.375
on comparing the poles of the cascade system with original poles we can say that one of the poles is same
and other pole is very close to original pole.
***************************************************************************************
A LTI system is characterized by the difference equation y(n)=0.68y(n-1)+0.5x(n).
The input signal x(n) has a range of -5V to +5V, represented by 8-bits. Find the quantization step size,
variance of the error signal and variance of the quantization noise at the output.
Solution:
Given
Range R=-5V to +5V = 5-(-5) =10
Size of binary, B= 8 bits (including sign bit)
Quantization step size,
R 10
q  8  8  0.0390625
2 2
q 2 0.03906252
var iance of error signal ,  e2    1.27116 *10 4
12 12
The difference equation governing the LTI system is
Y (n) =0.68y (n-1) +0.15x (n)
On taking z transform of above equation we get

Downloaded From EnggTree.com

 EnggTree.com
Y ( z )  0.68 z 1Y ( z )  0.15 X ( z )
Y ( z )  0.68 z 1Y ( z )  0.15 X ( z )
Y ( z )[1  0.68 z 1 ]  0.15 X ( z )
Y ( z) 0.15

X ( z ) 1  0.68 z 1
Y ( z) 0.15
H ( z)  
X ( z ) 1  0.68 z 1
0.15 0.15
H ( z ) H ( z 1 ) z 1  1
* * z 1
1  0.68 z 1  0.68 z
0.225 z 1
H ( z ) H ( z 1 ) z 1 
 0.68   1 
1    0.68   z  
 z   0.68 
0.0331z 1 0.0331z 1
H ( z ) H ( z 1 ) z 1  
 z  0.68   z  0.68 z  1.4706 
   z  1.4706 
 z 
Now, poles of H (z) H (z-1) z-1 are p1=0.68, p2=1.4706
Here, p1=0.68 is the only pole that lies inside the unit circle in z-plane
Variance of the input quantization noise at the output.
1
 c H ( z ) H ( z )z dz
1 1
 eoi
2
  e2
2 j 
N
 eoi
2
  e2   Res H ( z ) H ( z 1 ) z 1 
i 1 z  pi
N
 eoi
2
  e2   ( z  pi ) H ( z ) H ( z 1 ) z 1 
i 1 z  pi

0.0331
 eoi
2
  e2 ( z  0.68)*
( z  0.68)( z  1.4706) z 0.68
0.0331
 eoi
2
  e2 *  0.0419 e2
(0.68  1.4706)
 eoi
2
 0.0419*1.2716*104
 eoi
2
 5.328*106

Downloaded From EnggTree.com

 EnggTree.com
Analog to digital conversion:

10. Explain the ADC and DAC in detail.

A/D conversion has three process.

Basic parts of an analog-to digital (A/D) converter

1. Sampling
 Sampling is the conversion of a continuous- tome signal into a discrete-time signal obtained by taking
the samples of continuous-time signal at discrete instants.
 Thus if xa(t) is the input to the sampler, the output is x a(nT)≡x(n), where T is called the sampling
interval.
2. Quantisation
 The process of converting a discrete-time continuous amplitude signal into digital signal is called
quantization.
 The value of each signal sample is represented by a value selected from a finite set of possible values.
 The difference between the unquantised sample x(n) and the quantized output x q(n) is called the
quantization error or quantization noise.
eq(n)= xq(n)-x(n)
 To eliminate the excess bits either discard them by the process of truncation or discard them by rounding
the resulting number by the process of rounding.
 The values allowed in the digital signals are called the quantization levels
 The distance ∆ between two successive quantization levels is called the quantization step size or
resolution.
 The quality of the output of the A/D converter is measured by the signal-to-quantization noise ratio.
3. Coding
 In the coding process, each discrete value xq(n) is represented by a b-bit binary sequence.

Block diagram of basic elements of an A/D Converter

Digital to analog conversion:
 To convert a digital signal into an analog signal, digital to analog converters are used.

Downloaded From EnggTree.com

 EnggTree.com
Basic operations in converting a digital signal into an analog signal
 The D/A converter accepts, at its input, electrical signals that corresponds to a binary word, and
produces an output voltage or current that is proportional to the value of the binary word.
 The task of D/A converter is to interpolate between samples.
 The sampling theorem specifies the optimum interpolation for a band limited signal.
 The simplest D/A converter is the zero order hold which holds constant value of sample until the next
one is received.
 Additional improvement can be obtained by using linear interpolation to connect successive samples
with straight line segment.
 Better interpolation can be achieved y using more sophisticated higher order interpolation techniques.
 Suboptimum interpolation techniques result in passing frequencies above the folding frequency. Such
frequency components are undesirable and are removed by passing the output of the interpolator through
a proper analog filter which is called as post filter or smoothing filter.
 Thus D/A conversion usually involve a suboptimum interpolator followed by a post filter.
***************************************************************************************

Downloaded From EnggTree.com

 EnggTree.com
UNIT 5
MULTIRATE DIGITAL SIGNAL PROCESSING

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

Downloaded From EnggTree.com

 EnggTree.com

****************************************************************************

Downloaded From EnggTree.com

 EnggTree.com

OUTCOMES:

At the end of the course, the student should be able to

 Apply DFT and FFT algorithms for the analysis of digital signals and systems.
 Design FIR filters for various applications.
 Design IIR filters for various applications.
 Characterize the effects of finite precision representation on digital filters.

TEXT BOOKS:

1. John G.Proakis and Dimitris G. Manolakis, “Digital signal processing - Principles, algorithms

and applications”, Pearson education / Prentice hall, Fourth edition, 2007.

REFERENCES:

1. Sanjay K.Mithra, “Digital signal processing - A Computer based approach”, Tata McGraw-Hill, 2007.

2. M.H.Hayes, “Digital signal processing”, Schum’s outlines, Tata McGraw Hill, 2007.

3. A.V.Oppenheim, R. W. Schafer and J. R. Buck, “Discrete-time signal processing”, Pearson, 2004.

4. I.C.Ifeachor and B.W.Jervis, “Digital signal processing – A practical approach”, Pearson 2002

5. L.R. Rabiner and B. Gold, “Theory and application of digital signal processing”, Prentice Hall, 1992.

Downloaded From EnggTree.com