ELEC1200: A System View of

Communications: from Signals to Packets

Lecture 6
Review and Motivation
Our Communication System
Bit Rate vs. BER
Eye diagram for visualizing the effect of ISI
Equalization
Equivalent Models
Recursive Channel Model
Proof of Equivalence

ELEC1200 1
 Our Communication System

bit add start/ encode at SPB add training

transmit
seq stop bits samples per bit sequence

channel
bit decode find start skip past estimate
receive
seq bits bit training threshold

In the lab, we will examine the effect of reducing

the SPB and will find a tradeoff between Bit Rate
and Bit Error Rate.

ELEC1200 2
 Trade-off between Bit Rate and BER
Bit Rate the number of bits that can be transmitted per second.
We want this to be as high as possible
Bit Error Rate (BER) the fraction of bits that are wrongly
decoded by the receiver.
We want this to be low.
BER decreases as the bit time increases due to the ISI (InterSymbol
Interference).
ISI can be visualized using Eye Diagrams.

Bit Rate

Bit Error Rate

ELEC1200 bit time 3

 Eye Diagrams (Varying bit time)
SPB = 20 SPB = 15
1 1

Channel: Exponential, a = 0.9, noise added

0 0

0 20 40 0 15 30

SPB = 10 SPB = 8
1 1

0 0

0 10 20 0 8 16

SPB = 7 SPB = 2
1 1

0 0

0 7 14 0 2 4
ELEC1200 4
 Motivation: Equalization
The channel introduces intersymbol interference,
which causes the eye to close.
The goal of a channel equalizer is to undo the
effect of the channel.
This will cause the eye to open.
1
0.5
input 0

Discrete Time Channel

x(n) 0 5 10

1 1
0.5 0.5
0 0
0 5 10 channel 0 5 10
equalizer
x(n) y(n)
ELEC1200 5
 Developing the Equalizer
How can we figure out what to put into the
channel equalizer?
We need to describe the effect of the channel using a
model that enables us to predict the channel output for
any input.
actual predicted
channel channel
channel
input model output

Using this model, we can hopefully guess (estimate) what

the input to the channel was from knowledge of the
output
estimated actual
channel ??? channel
input output

ELEC1200 6
 ELEC1200: A System View of
Communications: from Signals to Packets
Lecture 6
Review and Motivation
Our Communication System
Bit Rate vs. BER
Eye diagram for visualizing the effect of ISI
Equalization
Equivalent Models
Recursive Channel Model
Proof of Equivalence

ELEC1200 7
 Model 1
We have seen that the communication channel can
be modeled by
Assuming it is linear and time invariant (LTI)
Assuming it has known step response s(n)

For our communication system, the step response

has the form

k s(n) = k(1-an+1)u(n)
0 < a < 1
0
0 10 20 30 40

ELEC1200 8
 Model 1 (cont.)
This model is useful in predicting the output for a
given input.

Discrete Time Channel

n
x(n) u(n) u(n 5) u(n 10)...

y(n) s(n) s(n 5) s(n 10)...

However, it is not easy to use this model to go the

other way (predict the input from the output).
We seek another equivalent model that will allow
us to do this easily.
ELEC1200 9
 Equivalent Models
A model of the channel is a way of predicting the
channel output given the channel input.
One system may be described by many equivalent
models that are written in different ways.
Given the same input, equivalent models make the
same prediction for the output.

Why do we need more than one model?

The more models we have, the better we understand the
system.
Different models have different uses.

ELEC1200 10
 ELEC1200: A System View of
Communications: from Signals to Packets
Lecture 6
Review and Motivation
Our Communication System
Bit Rate vs. BER
Eye diagram for visualizing the effect of ISI
Equalization
Equivalent Models
Recursive Channel Model
Proof of Equivalence

ELEC1200 11
 Recursive models
Recursive = involving repeated application of a rule

A recursive model for a discrete time waveform

x(n) has two parts
a formula that defines the nth sample in terms of the
past samples, e.g.
x(n) = f(x(n-1))
an initial (starting) condition, e.g.
x(0) = 0

Generating the waveform by recursion:

Given x(0), find x(1) = f(x(0)).
Given x(1), find x(2) = f(x(1)).
and so on
ELEC1200 12
 Examples
Can you think of recursive models for the following
sequences:
x(n) = c (a constant)

x(n) = n (a linear ramp)

x(n) = 0.2n

x(n) = an alternating bit stream (0 1 0 1 0 1)

ELEC1200 13
 Recursive model of IR channel
It turns out that the response of the IR channel y(n) to an
input x(n) can be described by a recursive formula:
y(n) =ay(n-1) + (1-a)kx(n)

k 1-a
x(n) y(n)

a delay
y(n-1)

The parameter a lies between 0 and 1.

The parameter k scales the input.
This is also known as a feedback system, since the output
feeds back as an input to the system to determine the next
output.
ELEC1200 14
 Example
Given the channel model: y(n) a y(n 1) (1 a) k x(n)

Assume that a = 0.5 and k = 1, i.e.,

y(n) 1
2
y(n 1) 1
2
x(n) x(n)

Find the output of the channel if the input is n

n 0 1 2 3 4 5 6 7
x(n) 0 0 1 1 1 0 0 0

y(n) 0

ELEC1200 15
 Effect of Parameter a
y(n) =ay(n-1) + (1-a)kx(n)

The parameter a determines the memory in the

channel
a = 0:
no memory of the past
y(n) = kx(n)
the channel output is just the input multiplied by k
a = 1
infinite memory of the past
y(n) = y(n-1)
Channel output is constant, ignores the channel input

ELEC1200 16
 ELEC1200: A System View of
Communications: from Signals to Packets
Lecture 6
Review and Motivation
Our Communication System
Bit Rate vs. BER
Eye diagram for visualizing the effect of ISI
Equalization
Equivalent Models
Recursive Channel Model
Proof of Equivalence

ELEC1200 17
 Proof of Equivalence
The two channel models are equivalent

Model 1: The channel is an LTI system with step response

s(n) = k(1-an+1)u(n)
where 0 < a < 1

Model 2: The channel has input x(n) and output y(n)

determined by
y(n) =ay(n-1) + (1-a)kx(n)
where 0 < a < 1.

We can establish equivalence by showing that

Model 2 is LTI (linear and time invariant)
Model 2 has a response to a unit step given by s(n)

ELEC1200 18
 Fact 1: Model 2 is LTI
y(n) =ay(n-1) + (1-a)kx(n)

It is linear
Intuitively, the output is a linear function of the input
and past outputs.
Formally, we will prove that it satisfies the two
properties:
Homogeneity
Additivity
It is time invariant
Intuitively, the parameters k and a do not change with n.
Formally, we prove that if we delay the input by d, the
output is the same, just delayed by d.

ELEC1200 19
 Proof of Homogeneity
x(n) system y(n) cx(n) system cy(n)

By the definition of model 2,

y(n) a y(n 1) (1 a) k x(n)

Multiply left and right by c,

c y(n) c a y(n 1) (1 a) k x(n)

Distributive law of multiplication,

c y(n) a c y(n 1) (1 a) k c x(n)

ELEC1200 20
 Proof of Additivity
x1(n) system y1(n)
system
x1(n)+x2(n) y1(n)+y2(n)
x2(n) system y2(n)

By definition of model 2,
y1 (n) a y1 (n 1) (1 a) k x1 (n)
y2 (n) a y2 (n 1) (1 a) k x2 (n)

Adding the two equations above,

y1 (n) y2 (n) a y1 (n 1) y2 (n 1) (1 a) k x1(n) x2 (n)

ELEC1200 21
 Proof of Time Invariance
x(n) system y(n) x(n-d) system y(n-d)

By the definition of model 2,

y(n) a y(n 1) (1 a) k x(n)

Substitute n with n-d,

y(n - d) a y(n - d 1) (1 a) k x(n d)

ELEC1200 22
 Fact 2: Same Step Response (example)
Model 1: Model 2:
s(n) = k(1-an+1)u(n) y(n) =ay(n-1)+(1-a)ku(n)

Assume: y(-1) = 0, a = , k = 1

n Model 1 Model 2
s(n)=1-()n+1 y(n) = y(n-1)+
0

5
ELEC1200 23
 Summary
We have seen that the channel can be described
(modeled) in two equivalent ways
Model 1:
Channel is linear and time invariant
Channel has step response s(n) = k(1-an+1)u(n)
Model 2:
If x(n) is the channel input and y(n) is the output,
y(n) =ay(n-1) + (1-a)kx(n)

In the next lecture, we will see how we can use

Model 2 to undo (equalize) the effect of the
channel, and open the eye.

ELEC1200 24
Rigorous Proof of Equivalence (Optional!)
We use recursion to show that the response of model 2 to the unit
step is the step response of model 1.

Let x(n) = u(n) in model 2, i.e. y(n) =ay(n-1) + (1-a)ku(n)

1. Since u(n) = 0 for all n<0,

y(n) =ay(n-1) for all n < 0
This is satisfied if y(n) = 0 for all n<0.
2. At n = 0,
y(0) = ay(-1) + (1-a)ku(0)
= k(1-a)
3. At n = 1,
y(1) = ay(0) + (1-a)ku(1)
= k(a-a2) +k(1-a) = k(1-a2) = k(1-an+1)
4. Suppose that s(n-1) = k(1-an),
then, s(n) = k(a an+1) + k(1-a) = k(1-an+1)
Thus, by recursion induction, s(n) = k(1-an+1)u(n) for all n.
ELEC1200 25