Real Time DSP Systems

 Basic signal process performs operation

Sn=K n × S n−1+ Cn
 Where
 Sn is current output value
 Sn−1 is result of previous signal process
 K n and C n are constants that determine characteristics of signal process.

 Effect of sampling is to replicate frequency response of signal process.

 Sampling rates much be higher than Nyquist rate used to avoid aliasing
 Real-time DSP must process samples at the rate at which they arrive
 Must be capable of performing repeated multiply- accumulate operations at a high
speed.

 Pipelines speed up process enabling it to run at required rate.

 Requires CPU that can add & multiply at same time.
 Quantisation sets limit on accuracy of the process (fixed point values)
 In fixed-point arithmetic each addition & multiplication loses resolution.
 Loss of resolution carefully controlled to avoid cumulative disastrous effect.
 Two forms of quantisation errors in DSP system:
 Rounding error – caused by basic level of quantisation.
 Arithmetic error – caused by loss of resolution in cumulative arithmetic operations.
DSP V ASP

ADVANTAGES DISADVANTAGES
Digital Signal Processing Similar architecture for Need for A-D and D-A
wide range of processes – conversion – associated
flexible errors
High accuracy, reliability, Basic hardware more
repeatability expensive
Software programmable – Speed of real-time
possibility of adaptive SPs processes
Low development, Errors associated with
 manufacturing & implementation – drift,
maintenance costs noise, stability.
Suitable for difficult SPs Inflexible
Cheap for simple SPs (e.g. Difficult to implement
gain, level shifting, simple complex processes
filter) Some SPs impossible
Lends itself to single-chip
implementation
Faster real-time processes

Summary
 Simple block diagram of DSP system
 Analogue-to-digital conversion
 Sampling
 Nyquist Sampling Condition
 Aliasing
 Practical approaches – ‘sample & hold’
 Quantisation
 Types: uniform, non-uniform, adaptive
 Errors
 Real-Time systems
 Basic signal process
 Problems associated with real-time
 Errors
 Digital signal processing compared to analogue signal processing
 Advantages and disadvantages of both types of signal processing.

Z – Transforms
 Definition
 Diagrammatic & physical representation
 Properties of z-transform
 Relationship to Laplace transform
 System transform functions
 Z-transform of typical sequences
What is a z-transform
Convenient and powerful theoretical tool for representing, analysing and designing discrete-
time signals and systems.

 Similar role in discrete time systems to laplace transform in continuous time systems –
z-transform easier to understand and apply.
  Relationship between z and laplace transforms allows continuous systems to be
converted into discrete (digital) form and vice-versa.

Definition
 Z-transform of a casual discrete sequence x [n] (i.e. x [0], x [1], x [2]……etc) defined as:
∞
X [ Z ]=∑ x [ n ] Z −n
n=0

 Z is a complex variable
Example
If a discrete signal has values at successive sampling instants of:

-1V, 2V, 0.3V, -1.5V

 Z transform is : −1 z−0+2 z−1+ 0.3 z−2 -1.5 z−3 = −1+2 z−1+ 0.3 z−2 -1.5 z−3
Representation
 Z-transform of a signal found by adding together successive signal samples.
 Samples obtained by putting signal to tapped delay line – delays equal to sampling
period.

Representation
The z−1operator is equivalent to a simple delay element with a time
delay T seconds.

z  ≡ Delay T seconds 
−1

 Z-transform invaluable as means of designing digital circuits

 Easy to make real-life z−1circuit.
 Z-transforms of common signal & system responses available in tables of common z-
transforms.
 Tables useful for finding inverse z-transform
In other words derive discrete time series from its z-transform.
Properties of Z-transform
  Linearity: If sequences x(1) [n] & x(2) [n] have z-transforms x(1) [ z ] & x(2) [ z ] then z-
transform of linear combination is:
ax (1 ) [n] + bx (2) [n ]  aX (1) [ z] + bX (2) [z ]
Delays or shifts: Z-transform of sequence x [n] delayed by m samples is
x [n]X[Z]
x [n−m]  z−mX[Z]
 Differentiation: If X [Z ] is z-transform of x [n], then z-transform of nx [n] is given by:
x [n] X [z ]
dX [z ]
nx [n]−z
dz
 Convolution: Discrete-time linear-time-invariant (LTI) system with input x [n] and
impulse response h [k ] and output system response y [n].

∞
y [ n]= ∑ h[k ] x [ n−k ]
k=−ꝏ
Y [ z ]=H [ z ] X [z ]
 H [ z ] referred to as system transfer function.

Inverse Z-transform

 Recovery of discrete time sequence from z-transform.

 Useful mainly to obtain impulse response of a digital signal process given ots transfer
function.

 Inverse z-transform of G[z] relates output and input sampled data signals in the time
domain
 g[n] is impulse response
g[n]= z−1 {G[ z ]}
 From signal and theory:
y [ n ] =g [ n ]∗x [n]
Y [ z ]=G [ z ] X [z ]

Relationship to Laplace Transform Overview

 Laplace transform for continuous time-signals given by:
L { x ( t ) }=∫ x (t )e dt
st

 Z-transform for discrete-time signals give by:

∞
X [ z ]= ∑ x [n] z−n
k=−ꝏ
  Comparison of two shows that z-transform is discrete-time equivalent of laplace
transform:
z=e =e
sT (σ + jω)T

s=σ + jω
|z|=e σT ;˂ z=ωT
Relationship to Laplace Transform Mapping of s-plane to z-plane.

 Maps left-hand half of s-plane inside unit circle on z-plane.

Relationship to Laplace Transform Poles & Zeros

 For all practical signals or systems, z- (0r Laplace) transform can be expressed as ratio of
two polynomials in z (or s).
 Roots of numerator are zeros (ς K ) of function and roots of denominator are poles ( pk ) of
function.
−1 −2 −N
a 0+ a1 z + a2 z + … … .+ a N z
G [ z ]= −1 −m
b0 + b1 z +… … … ..+b m z

K ( z−ς 1 )( z −ς 1 ) … … ..(z−ς N )
G [ z ]=
( z− p 1) ( z− p 2) … …..(z −p m)
 Stability condition of linear analogue system is that all poles must lie to the left of
imaginary axis.
 In z-plane (discrete linear system), corresponds to condition:
System is stable if all poles of transfer function lie within unit circle.
System Transfer Function What is it?
 Z-transforms used to represent discrete data signals and response to signals of discrete
data processes.

G[z] is transfer function of system & g[n] is impulse response.

 G[z] can be any (realisable) function chosen to give desired digital signal processor
performance.
 General expression for G[z] as given above is:

−1 −2 −N
a 0+ a1 z + a2 z + … … .+ a N z
G [ z ]= −1 −m
b0 + b1 z +… … ..+b m z
 Where constants a 0, a 1…….a N & b 0, b 1 ,……., b M can be chosen to give DSP system
performance required for specific applications.
 In many DSP systems system response normalised to give maximum gain of 1-
corresponds to setting b 0=1.

System Transfer Function Frequency Response

 Substitute z=e jωT in system transfer function.

 Evaluate |G(e jωT )|- tedious, but straightforward
 Use computer
 Art of DSP design is in choosing values of a’s and b’s to obtain required system
performance (including frequency response).
Summary
 Defined z-transform.
 General properties of z-transform
 Looked at relationship between z and Laplace transforms.
 System transfer functions.
 Now look at typical transforms and some examples of their use (on overheads).
 General DSP
 Transfer Function
 Poles and Zeros
 Analogue vs. Digital Filtering
 Types of Digital Filters
- Finite Impulse Response Filters
- Infinite Impulse Response Filters
 Sources of error
 FIR vs. IIR Filters
Transfer Function
 General transfer function of a digital filter can be represented as:
Y [ z] ∑ ai z
−i
G [ z ]= = −j ; b 0=1
X [z ] ajz
 Problem of filter design: choose a i and b jto obtain desired performance.
Poles and Zeros
Transfer function represented as ratio of two polynomials in z (or z−1):
 Y [ z]
∑ ai z−1
G [ z ]= = i=0
X [z ] 1+∑ b j z − j
j =0

 Or by roots of numerator and denominator:

Y [ z] ∏(z−ς i)
G [ z ]= =
X [z ] ∏(z− p j)
Where ς i are zeros and p j are poles of G [ z ]

Poles and zeros

 Can plot poles and zeros on z-plane pole-zero diagram:
- Example
−1 −2 2 1
1−0.2 z −0.08 z z −0.2 z −0.08 (z−0.4 )(z+ 0.2)
G [ z ]= = =
1+0.25 z
−2 2
z +0.25 ( z−0.5 j)( z +0.5 j)
− Zeros at z = 0.4 & -0.2, and poles at z = j0.5 & z = -j0.5
− Used Matlab to plot diagram:
>>a = [1 -0.2 -0.08];
>>b= [1 0 0.25];
Zplane(a,b)
Poles and Zeros
Z-plane pole-zero diagram: Matlab

Poles and Zeros Rules

 All realisable functions have poles which are either real or exist in conjugate pairs.
 For stable functions, all poles must lie inside unit circle.
Digital vs. Analogue Advantages
 Digital filters have characteristics not possible with analogue filters -such as linear phase
response.
 Performance of digital filter is automatically adjustable if implemented in programmable
processor -used in adaptive filters.
 Several inputs/channels can be filtered by one digital filter -without need to replicate
hardware.
 Both filtered and unfiltered data can be saved for further use.
 Digital filters can be made small in size -due to available technology.
 Precision of digital filters by word length -analogue filters limited to typical stopband
attenuation of 60 to 70dB.
 Performance of digital filters repeatable from unit to unit.
 Digital filters can work at very low frequencies, and across wide frequency range.
 Speed limitation: the maximum bandwidth of signals that digital filters can handle, in
real time, is much lower than for analogue filters.
 Finite word length effects – digital filters subject to ADC noise and to round off noise
during computation.
 Long design and development times – time required for design and development of
digital filters, especially hardware development, can be much longer than for analogue
filters. But designs can be re-used once developed.
Digital Filters: Types
 Two main types:
- Finite Impulse filters (FIR): simplest to construct – all values of b j zero
except b 0 G [ z ]=a0 +a 1 z−1 +a2 z−2+ ...+ an z−n
- Infinite Impulse Response Filter (IIR)- more general form where all
coefficients may be non-zero
−1 −2 −n
a 0+ a1 z a 1+ a2 z +…+ an z
G [ z ]= −1 −2 −n
1+b 1 z +ab 2 z +…+ bn z
The transfer function of FIR filter is:
−1 −2 −n
G [ z ]=a0 +a 1 z +a2 z + ...+ an z
Impulse response given by series:
- a 0 , a 1 , a2 … . an
- Finite series – hence finite impulse response
 Implemented using three components:
- Unit delay elements z−1
- Coefficients multipliers
- An adder (or adders)
FIR Filters-Direct Realisation

FIR Filters
Direct Realisation (alternative)