DISCRETE-TIME SYSTEMS

A system is a transformation of signals

A system is an input-output relationship

x  n T   y  n

A SISO system

1
Ex: A delay system 𝑦[𝑛] = 𝑥 [𝑛 − Δ]

2
Ex:
𝑦[𝑛] = 𝑎𝑥 [𝑛] + 𝑏𝑥 [𝑛 − 1]

1 sample
𝑥[𝑛] delay

a b

+ 𝑦[𝑛]

This is a linear, time-invariant, finite impulse response (FIR) system.

In general, an LTI, FIR system

𝑁2

𝑦[𝑛] = ∑ 𝑎𝑘 𝑥 [𝑛 − 𝑘]
𝑁1

3
Ex:

𝑦[𝑛] = 𝑎𝑦[𝑛 − 1] + 𝑥 [𝑛]

This is a recursive expression.

𝑥[𝑛] + 𝑦[𝑛]

1 sample
delay

This is a linear, time-invariant, infinite impulse response (IIR) system.

Equivalently its nonrecursive form is,

𝑦 [𝑛 ] = ∑ 𝑥 [𝑘 ]
𝑘=−∞

Note that not all recursive expressions have their nonrecursive counterparts.

4
CLASSIFICATION OF SYSTEMS

 With memory - memoryless

A Quotation from a Recent Research Paper:

“Null Space Component Analysis for Noisy Blind Source

“The solutions for the BSS problem were investigated under

5
WITH MEMORY - MEMORYLESS

𝑦 [𝑛 ] = 𝑥 [𝑛 ]

𝑦[𝑛] = 3𝑥 [𝑛]

𝑦[𝑛] = 4𝑥[𝑛]

are memoryless

whereas
𝑦 [ 𝑛 ] = 𝑥 [ 𝑛 − 1]

𝑦 [ 𝑛 ] = 𝑥 [ 𝑛 + 1]

𝑦 [ 𝑛 ] = 𝑥 [ 𝑛 ] + 𝑥 [ 𝑛 − 1]

have memory

You have heard or you will hear about “dynamic systems”; they have memory.

6
LINEARITY

A system, 𝑇{∙}, is said to be linear if it satisfies

a) additivity: 𝑇{𝑥1 [𝑛] + 𝑥2 [𝑛]} = 𝑇{𝑥1 [𝑛]} + 𝑇{𝑥2 [𝑛]}

b) homogeneity: 𝑇{𝑎𝑥 [𝑛]} = 𝑎𝑇{𝑥 [𝑛]}

7
Ex:

𝑦[𝑛] = ∑𝑛𝑘=−∞ 𝑥 [𝑘] is linear

𝑦[𝑛] = log(|𝑥 [𝑛]|) is nonlinear

𝑦 [𝑛 ] = 𝑥 [𝑛 ] + 3 is nonlinear

Proof: Exercise

8
TIME-INVARIANCE

Let 𝑦1 [𝑛] = 𝑇{𝑥[𝑛]} and 𝑦2 [𝑛] = 𝑇{𝑥 [𝑛 − 𝑁0 ]} be the outputs of the system to
𝑥 [𝑛] and 𝑥 [𝑛 − 𝑁0 ], respectively.

Then, if 𝑦2 [𝑛] = 𝑦1 [𝑛 − 𝑁0 ] the system is said to be time-invariant.

9
Ex: Compressor (downsampler!)

𝑦[𝑛] = 𝑥 [𝑀𝑛] 𝑀 ∈ 𝑍, 𝑀 > 1

𝑥 [𝑛 ] ↓M 𝑦 [𝑛 ]

Is it time-invariant?

10
 Before answering the question, let’s see what a compressor is.

Compressor

x[n]

3
2
… 1
…
-2 -1 0 1 2 3 4
n

x[2n]

2
… … M=2
-2 -1 0 1 2 3 4
n

11
Is it time-invariant?
Answer:

Following the above definition, let

𝑦1 [𝑛] = 𝑥 [𝑀𝑛]
𝑦2 [𝑛] = 𝑥 [𝑀𝑛 − 𝑁0 ]

Since

𝑦2 [𝑛] ≠ 𝑦1 [𝑛 − 𝑁0 ] = 𝑥 [𝑀𝑛 − 𝑀𝑁0 ]

compressor is time-varying.

12
Compressor is time-varying:

x[n]
3
2
… 1
…
-2 -1 0 1 2 3 4
n

x[3n]
3
2
… 1
… M=3
”
-2 -1 0 1 2 3 4
n

x[3n-1]
3
2
… 1
… The response to 𝑥[𝑛 − 1]
” when M = 3
-2 -1 0 1 2 3 4
n

x[3(n-1)]
3

2
… 1
… 𝑥[3𝑛] delayed by 1 sample
”
n
-2 -1 0 1 2 3 4

13
clear all
close all

k = 0:10;
a = (-1).^k
aa = upsample(a,2)
x = [0 aa 0] %input
n = 0:length(x)-1;
stem(n,x)
xlabel('n, sample index')
ylabel('x[n]')
title('input')
x_1 = circshift(x,[0,1]) % input delayed by one sample
figure
n = 0:length(x_1)-1;
stem(n,x_1)
xlabel('n, sample index')
ylabel('x[n-1]')
title('input delayed by one sample')
y = downsample(x,2)
figure
n = 0:length(y)-1;
stem(n,y)
xlabel('n, sample index')
ylabel('y[n]')
title('response to x[n]')
yy = downsample(x_1,2)
figure
n = 0:length(yy)-1;
stem(n,yy)
xlabel('n, sample index')
ylabel('z[n]\neq y[n-1]')
title('response to x[n-1]')

14
Show that it is linear! (exercise)

15
Ex: Expander (upsampler!)

𝑛
𝑥[ ] 𝑛 = 𝑘𝐿
𝑦 [𝑛 ] = { 𝐿 𝐿 ∈ 𝑍, 𝐿>1
0 𝑛 ≠ 𝑘𝐿

𝑥 [𝑛 ] ↑L 𝑦 [𝑛 ]

16
Is it time-invariant?

𝑛
𝑥[ ] 𝑛 = 𝑘𝐿
𝑦1 [𝑛] = { 𝐿
0 𝑛 ≠ 𝑘𝐿
𝑛
𝑥 [ − 𝑁0 ] 𝑛 = 𝑘𝐿
𝑦2 [𝑛] = { 𝐿
0 𝑛 ≠ 𝑘𝐿

Since

𝑛 − 𝑁0
𝑥[ ] 𝑛 − 𝑁0 = 𝑘𝐿
𝑦2 [𝑛] ≠ 𝑦1 [𝑛 − 𝑁0 ] = { 𝐿
0 𝑛 − 𝑁0 ≠ 𝑘𝐿

expander is time-varying.

Show that it is linear! (exercise)

17
CAUSALITY
A system is said to be causal if the two output signals 𝑦1 [𝑛] and 𝑦2 [𝑛] (due to
two input signals 𝑥1 [𝑛] and 𝑥2 [𝑛]) satisfy

𝑦1 [𝑛] = 𝑦2 [𝑛] 𝑛 ≤ 𝑛0

whenever

𝑥1 [𝑛] = 𝑥2 [𝑛] 𝑛 ≤ 𝑛0

18
Ex:

𝑦 [ 𝑛 ] = 𝑥 [ 𝑛 + 1] − 𝑥 [ 𝑛 ] is noncausal

𝑦 [ 𝑛 ] = 𝑥 [ 𝑛 ] + 𝑥 [ 𝑛 − 1] is causal

𝑦 [𝑛 ] = 𝑥 [𝑛 ] + 5 is causal

Proof: Exercise

19
STABILITY (BIBO)

A system is said to be BIBO stable if “bounded inputs yield bounded outputs.”,

|𝑥 [𝑛]| ≤ 𝑏𝑥 < ∞ ⇒ |𝑦[𝑛]| ≤ 𝑏𝑦 < ∞

for arbitrary finite 𝑏𝑥 and 𝑏𝑦 .

20
Ex:

𝑦 [𝑛 ] = ∑ 𝑥 [𝑘 ]
𝑘=−∞

= 𝑦 [ 𝑛 − 1] + 𝑥 [ 𝑛 ]

UNSTABLE

𝑛+1 𝑛≥0
For example, for 𝑥 [𝑛] = 𝑢[𝑛] the output is 𝑦[𝑛] = {
0 𝑛<0

Therefore bounded input does not yield bounded output.

21
Computation of
𝑦[𝑛] = 𝑎𝑦[𝑛 − 1] + 𝑥 [𝑛]
in MATLAB.

clear all
close all
N = 1000; % signal length
a = 0.97; % system parameter
b = 1; % system parameter
k = 1:N; % discrete-time index vector
w = pi/2; % frequency of input sinuoid
% x = 2*(rand(1,N)-0.5); % random input
x = sin(w*k); % sinusoidal input
x = [x zeros(1,N)]; % zero padding (why?)
y = zeros(1,2*N); % output signal

for n = 2:2*N
y(n) = a*y(n-1)+b*x(n);
end
plot(x)
hold on
plot(y,'r')

22