Transmission of Signals Through Linear Systems

Definition : A system refers to any physical device that produces an output signal in
response to an input signal.

Definition : A system is linear if the principle of superposition applies.

If x1(t) produces output y1(t)

x2(t) produces output y2(t)

then a1x1(t)+ a2x2(t) produces an output a1y1(t)+ a2y2(t)

Also, a zero input should produce a zero output.

Example of linear systems include filters and communication channels.

Definition: A filter refers to a frequency selective device that is used to limit the
spectrum of a signal to some band of frequencies.

Definition: A channel refers to a transmission medium that connects the transmitter

and receivers of a communication system .

Time domain and frequency domain may be used to evaluate system performance.

Time response :

Definition: The impulse response h(t) is defined as the

response of a system to an impulse 𝛿(𝑡) applied to the
input at t=0 .

Definition: A system is time-invariant when the shape of

the impulse response is the same no matter when the
impulse is applied to the system.

𝛿(𝑡) h(t), then 𝛿(t − t d ) h(t - td)

When the input to a linear time-invariant system in a signal

x(t) , then the output is given by
∞
y(t) = ∫−∞ x(λ)h(t − λ) dλ
∞
= ∫−∞ h(λ)x(t − λ) dλ ; convolution integral

1
Definition: A system is said to be causal if it does not respond before the excitation is
applied, i.e.,

h(t) = 0 t<0

The causal system is physically realizable.

Definition: A system is said to be stable if the output signal is bounded for all
bounded input signals.

If | x(t) | ≤ M ; M is the maximum value of the input

∞
then | y(t) | ≤ ∫−∞ | h(τ) ||x(t − τ ) | dτ
∞
= M ∫−∞| h(τ) | dτ

A necessary and sufficient condition for stability (a bounded output) is

∞
∫−∞ |h(t)| dt < ∞ ; h(t) is absolutely integrable.

∴ zero initial conditions assumed .

Frequency Response:

Definition: The transfer function of a linear time invariant system is defined as the
Fourier transform of the impulse response h(t)

H(f) = F{h(t)}

Since y(t) = x(t)*h(t) , then

Y(f) = H(f) X(f)

𝑌(𝑓)
or 𝐻(𝑓) = 𝑋(𝑓)

The transfer function H(f) is a complex function of frequency, which can be obtained
as the ratio of the Fourier transform of the output to that of the input.

𝐻(𝑓) = |𝐻(𝑓)|𝑒 𝑗 𝜃(𝑓)

where,
H(f) : amplitude spectrum
𝜃(f) : phase spectrum.

System Input–Output Energy Spectral Density

Let x(t) be applied to a LTI system , then the Fourier transform of the output is related
to the Fourier transform of the input through the relation

2
 Y(f) = H(f) X(f)

Taking the absolute value and squaring both sides, we get

|Y(f)|2 = |H(f)|2 | X(f)|2

SY(f) = |H(f)|2 SX(f)

SY(f): Output Energy Spectral Density

SX(f): Input Energy Spectral Density.

Output energy spectral density = |H(f)|2 x Input energy spectral density

The total output energy

+∞
𝐸𝑦 = ∫−∞ 𝑆𝑌 (𝑓)𝑑𝑓
+∞
= ∫−∞ |H(f)|2 𝑆𝑋 (𝑓)𝑑𝑓.

The total input energy is

+∞
𝐸𝑥 = ∫−∞ 𝑆𝑥 (𝑓)𝑑𝑓 .

Example: Response of a Filter to a Sinusoidal Input

The signal 𝑥(𝑡) = 𝑐𝑜𝑠 𝑤0 𝑡 is applied to a filter described by the transfer function
1
𝐻(𝑓) = 1+𝑗𝑓/𝐵. Find the filter output 𝑦(𝑡).

Solution:

We will find the output using the frequency domain approach.

𝑌(𝑓) = 𝐻(𝑓)𝑋(𝑓)
1 𝑓 𝑓
𝐻(𝑓) = 𝑒 −𝑗𝜃 ; 𝜃 = tan−1 𝐵 ; 𝜃0 = tan−1 𝐵0
𝑓
√1+( )2
𝐵

1 1
𝑌(𝑓) = 𝐻(𝑓)[2 𝛿(𝑓 − 𝑓0 ) + 2 𝛿(𝑓 + 𝑓0 )]

1 1
𝑌(𝑓) = 2 𝐻(𝑓0 )𝛿(𝑓 − 𝑓0 ) + 2 𝐻(−𝑓0 )𝛿(𝑓 + 𝑓0 )

1 1 1 1
𝑌(𝑓) = 2 𝑒 −𝑗𝜃0 𝛿(𝑓 − 𝑓0 ) + 2 𝑒 𝑗𝜃0 𝛿(𝑓 + 𝑓0 )
𝑓 𝑓
√1+( 0 )2 √1+( 0 )2
𝐵 𝐵

3
Taking the inverse Fourier transform, we get
1 1
𝑦(𝑡) = [𝑒 𝑗(2𝜋𝑓0 𝑡−𝜃0 ) + 𝑒 −𝑗(2𝜋𝑓0 𝑡−𝜃0 )]
𝑓 2
√1+( 0 )2
𝐵

1
𝑦(𝑡) = cos(2𝜋𝑓0 𝑡 − 𝜃0 )
𝑓
√1+( 0 )2
𝐵

Note that in the last step we have made use of the Fourier transform pair

𝑒 𝑗2𝜋𝑓𝑐𝑡 ↔ 𝛿(𝑓 − 𝑓𝑐 )

Remark: Note that the amplitude of the output as well as its phase depend on the
frequency of the input and the bandwidth of the filter.
𝑓
Assume, for instance, that 𝑓0 = 𝐵. Then 𝜃0 = tan−1 𝐵0 = tan−1 1 = 45° and the
output can be written as:
1
𝑦(𝑡) = cos(2𝜋𝑓0 𝑡 − 45°)
√1+1

1
𝑦(𝑡) = cos(2𝜋𝑓0 𝑡 − 45°)
√2

1
Exercise: The signal 𝑥(𝑡) = 𝑐𝑜𝑠 𝑤0 𝑡 − 𝜋 𝑐𝑜𝑠 3𝑤0 𝑡 is applied to a filter described by
1
the transfer function 𝐻(𝑓) = 1+𝑗𝑓/𝐵.

a. Use the result of the previous example to find the filter output 𝑦(𝑡).
b. Is the transmission through this filter distortion-less ?

Exercise: Consider the periodic rectangular signal 𝑔(𝑡) defined over one period 𝑇0 as:

+𝐴, −𝑇0 /4 ≤ 𝑡 ≤ 𝑇0 /4
𝑔(𝑡) = {
0, 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒
1
If 𝑔(𝑡) is applied to a filter described by the transfer function 𝐻(𝑓) = 1+𝑗𝑓/𝐵. Use
the result of the previous example to find the filter output 𝑦(𝑡).

Example:
𝑡 1
The signal 𝑔(𝑡) = 𝐴𝑟𝑒𝑐𝑡(𝑇) is applied to the filter (𝑓) = 1+𝑗𝑓/𝐵 . Find the output
energy spectral density.

Solution:

𝑆𝑌 (𝑓) = |𝐻(𝑓)|2 𝑆𝑋 (𝑓)

1
𝑆𝑌 (𝑓) = 𝑓 (𝐴𝑇 |𝑠𝑖𝑛𝑐 𝑇𝑓|)2
1+( )2
𝐵

4
Example:

The signal 𝑔(𝑡) = 𝛿(𝑡) − 𝛿(𝑡 − 1) is applied to a channel described by the transfer
1
function 𝐻(𝑓) = 1+𝑗𝑓/𝐵 . Find the channel output.

Solution:

The impulse response of the channel is obtained by taking the inverse Fourier
transform of 𝐻(𝑓), which is

ℎ(𝑡) = 2𝜋𝐵𝑒 −2𝜋𝐵𝑡 𝑢(𝑡)

Using the linearity and time invariance property, the output can be obtained as:

𝑦(𝑡) = ℎ(𝑡)𝑢(𝑡) − ℎ(𝑡 − 1)𝑢(𝑡 − 1)

𝑦(𝑡) = 2𝜋𝐵[𝑒 −2𝜋𝐵𝑡 𝑢(𝑡) − 𝑒 −2𝜋𝐵(𝑡−1) 𝑢(𝑡 − 1)]

Exercise: The signal 𝑔(𝑡) = 𝑢(𝑡) − 𝑢(𝑡 − 1) is applied to a channel described by the
1
transfer function 𝐻(𝑓) = 1+𝑗𝑓/𝐵 . Find the channel output 𝑦(𝑡).

Signal Distortion in Transmission

As we have said before, the objective of a communication system is to deliver to the

receiver almost an exact copy of what the source generates. However, communication
channels are not perfect in the sense that impairments on the channel will cause the
received signal to differ from the transmitted one. During the course of transmission,
the signal undergoes attenuation, phase delay, interference from other transmissions,
Doppler shift in the carrier frequency, and many other effects. In this introductory
discussion we will explain some of the reasons that cause the received signal to be
distorted.

Linear Distortion
A signal transmission is said to be distortion-less if the output signal y(t) is an exact
replica of the input signal x(t) , i.e., y(t) has the same shape as the input, except for a
constant amplification (or attenuation) and a constant time delay.

Condition for a distortion-less transmission in the time domain is:

𝒚(𝒕) = 𝒌𝒙(𝒕 − 𝒕𝒅 ); Condition for a distortion-less transmission

where k: is a constant amplitude scaling

td: is a constant time delay

5
 In the frequency domain, the condition for a distortion-less transmission becomes

Y(f) = k X(f) 𝑒 −𝑗2𝜋𝑓𝑡𝑑

Y(f)
or H(f) = = k 𝑒 −𝑗2𝜋𝑓𝑡𝑑 = 𝑘𝑒 −𝑗𝜃(𝑓)
𝑋(𝑓)

That is, for a distortion-less transmission, the transfer function should satisfy two
conditions:

1. |H(f)| = k ; The amplitude of the transfer function is constant (gain or attenuation)

over the frequency range of interest.
2. 𝜽(𝒇) = −𝟐𝝅𝒇𝒕𝒅 = −(𝟐𝝅𝒕𝒅 )𝒇 ; The phase function is linear in frequency with a
negative slope that passes through the origin (or multiples of π).

When |H(f)| is not constant for all frequencies of interest, amplitude distortion results.

When 𝜃(𝑓) ≠ −2𝜋𝑓𝑡𝑑 ± 1800 , then we have phase distortion (or delay distortion).

The following examples demonstrate the two types of distortion mentioned above.

Example : Amplitude Distortion

1
Consider the signal 𝑥(𝑡) = 𝑐𝑜𝑠 𝑤0 𝑡 − 3 𝑐𝑜𝑠 3𝑤0 𝑡. If this signal passes through a

channel with zero time delay (i.e., td = 0) and amplitude spectrum as shown in the
figure
a. Find y(t)
b. Is this a distortion-less transmission?

6
Solution:
x(t) consists of two frequency components, f0 and 3f0 . Upon passing through the
channel, each one of them will be scaled by a different factor.
1 1
a. 𝑦(𝑡) = 𝑐𝑜𝑠 𝑤0 𝑡 − 2 . 𝑐𝑜𝑠 3𝑤0 𝑡
3

b. Since y(t) ≠ k x(t), this is not a distortion-less transmission .

Example: Phase Distortion
If x(t) in the previous example is passed through a channel whose amplitude spectrum
𝜋
is a constant k. Each component in x(t) suffers a 2 phase shift

a. Find y(t).
b. Is this a distortion-less transmission ?
Solution:
1
𝑥(𝑡) = cos 𝑤𝑜 𝑡 − cos 3𝑤𝑜 𝑡
3
𝜋 1 𝜋
𝑦(𝑡) = k cos(𝑤𝑜 𝑡 − ) − 𝑘 cos (3𝑤𝑜 𝑡 − )
2 3 2
𝜋 1 𝜋
𝑦(𝑡) = k cos 𝑤𝑜 (𝑡 − ) − 𝑘 cos (3𝑤𝑜 (𝑡 − ))
2𝑤𝑜 3 2𝑥3𝑤𝑜
1
𝑦(𝑡) = k cos 𝑤𝑜 (𝑡 − 𝑡𝑑1 ) − 𝑘 cos(3𝑤𝑜 (𝑡 − 𝑡𝑑2 ))
3
Note that td1 ≠ td2 , i.e., each component in 𝑥(𝑡) suffers from a different
time delay. Hence this transmission introduces phase (delay) distortion.

Nonlinear Distortion
When a system contains nonlinear elements, it is not described by a transfer function
H(f), but rather by a transfer characteristic of the form

y(t) = a1 x(t) +a2 x2(t) +a3 x3(t) + … (time domain)

In the frequency domain ,

Y(f)= a1 X(f) +a2 X(f)*X(f) +a3 X(f)*X(f)*X(f) + …

Here, the output contains new frequencies not originally present in the original signal.
The nonlinearity produces undesirable frequency component for |f|≤ W, in which W is
the signal bandwidth.

7
Harmonic Distortion in Nonlinear Systems
Let the input to a nonlinear system be the single tone signal

x(t) = cos2πfot

This signal is applied to a channel with characteristic

y(t) = a1x(t) + a2x(t)2 + a3x(t)3

upon substituting x(t) and arranging terms, we get

1 3 1 1
𝑦(𝑡) = 2 𝑎2 + (𝑎1 + 4 𝑎3 ) cos2πf0 t + 2 𝑎2 cos 4𝜋𝑓0 𝑡 + 4 𝑎3 𝑐𝑜𝑠6𝜋 𝑓0 𝑡

Note that the output contains a component proportional to x(t) which is (𝑎1 +
3
4
𝑎3 ) cos2πf0 t, in addition to a second and a third harmonic terms (terms at twice and
three times the frequency of the input). These new terms are the result of the nonlinear
characteristic and are, therefore, considered as harmonic distortion. The DC term does
not constitute a distortion, for it can be removed using a blocking capacitor.

Define second harmonic distortion

|𝑎𝑚𝑝𝑙𝑖𝑡𝑢𝑑𝑒 𝑜𝑓 𝑠𝑒𝑐𝑜𝑛𝑑 ℎ𝑎𝑟𝑚𝑜𝑛𝑖𝑐 |

𝐷2 =
|𝑎𝑚𝑝𝑙𝑖𝑡𝑢𝑑𝑒 𝑜𝑓 𝑓𝑢𝑛𝑑𝑎𝑚𝑒𝑛𝑡𝑎𝑙 𝑡𝑒𝑟𝑚|

1
| 2 𝑎2 |
𝐷2 = 𝑥 100%
3
| (𝑎1 + 4 𝑎3 ) |

In a similar way we can define the third harmonic distortion as:

|𝑎𝑚𝑝𝑙𝑖𝑡𝑢𝑑𝑒 𝑜𝑓 𝑡ℎ𝑖𝑟𝑑 ℎ𝑎𝑟𝑚𝑜𝑛𝑖𝑐 |

𝐷2 =
|𝑎𝑚𝑝𝑙𝑖𝑡𝑢𝑑𝑒 𝑜𝑓 𝑓𝑢𝑛𝑑𝑎𝑚𝑒𝑛𝑡𝑎𝑙 𝑡𝑒𝑟𝑚|

Therefore,
1
| 4 𝑎3 |
𝐷3 = 𝑥 100%
3
| (𝑎1 + 4 𝑎3 ) |
Remark: In the solution above we have made use of the following two identities:
1
Cos2x = 2 {1 + 𝑐𝑜𝑠2𝑥}
1
Cos3x=4 {3𝑐𝑜𝑠𝑥 + 𝑐𝑜𝑠3𝑥}.

8
 Filters and Filtering
A filter is a frequency selective device. It allows certain frequencies to pass almost
without attenuation wile it suppresses other frequencies

A. Ideal Filter:
Ideal low pass filter

−𝑗2𝜋𝑓𝑡𝑑
𝐻(𝑓) = {𝑘 𝑒 |𝑓| < 𝐵
0 𝑜. 𝑤

ℎ(𝑡) = 2𝐵𝑘 𝑠𝑖𝑛𝑐 2𝐵(𝑡 − 𝑡𝑑 )

since h(t) is the response to an impulse applied at t=0, and because h(t) has nonzero
values for t<0 , the filter is non-causal (physically non realizable).

Band Pass Filter

−𝑗2𝜋𝑓𝑡𝑑
𝐻(𝑓) = {𝑘 𝑒 𝑓𝑙 < |𝑓| < 𝑓𝑢
0 𝑜. 𝑤

Filer bandwidth B = fu ─ fl; difference between upper and lower positive frequencies
𝑓𝑢 +𝑓𝑙
𝑓𝑐 = ; center frequency of the filter
2
ℎ(𝑡) = 2𝐵𝑘 𝑠𝑖𝑛𝑐 𝐵(𝑡 − 𝑡𝑑 ) 𝑐𝑜𝑠 𝑤𝑐 (𝑡 − 𝑡𝑑 ); impulse response

9
High-pass filter

−𝑗2𝜋𝑓𝑡𝑑
𝐻(𝑓) = {𝑘 𝑒 |𝑓| > 𝐵
0 𝑜. 𝑤

Band Rejection or Notch Filter

𝑘 𝑒 −𝑗2𝜋𝑓𝑡𝑑 𝑜. 𝑤
𝐻(𝑓) = {
0 𝑓1 < |𝑓| < 𝑓2

Real Filter
Here, we only consider a Butterworth low pass filter. The transfer function of a low
pass Butterworth filter is of the form
1
𝐻(𝑓) =
𝑗𝑓
𝑃𝑛 ( 𝐵 )
B is the 3-dB bandwidth of the filter and Pn(jf/B) is a complex polynomial of order n .
The family of Butterworth polynomials is defined by the property
𝑗𝑓 2 𝑓 2𝑛
|𝑃𝑛 ( ) | = 1 + ( )
𝐵 𝐵
So that
1
|𝐻(𝑓)| = 2𝑛
√1+( 𝑓 )
𝐵

The first few polynomials are:

𝑃1 (𝑥) = 1 + 𝑥
𝑃2 (𝑥) = 1 + √2𝑥 + 𝑥 2

10
𝑃3 (𝑥) = (1 + 𝑥)(1 + 𝑥 + 𝑥 2 )

A first order LPF

1
𝑗2𝜋𝑓𝑐 1
𝐻(𝑓) = 1 = 1+𝑗2𝜋𝑓𝑅𝐶
𝑅+
𝑗2𝜋𝑓𝑐
1
Let 𝐵 = 2𝜋𝑅𝐶
1 1 1
𝐻(𝑓) = = =
1 + 𝑗𝑓/𝐵 𝑃1 (𝑗𝑓/𝐵) 𝑃1 (𝑥)

A Second order LPF

1
𝐻(𝑓) = 𝑗𝑤𝐿 2
1+ −(2𝜋√𝐿𝐶𝑓)
𝑅
1
𝐻(𝑓) =
1 + 𝑗√2𝑓/𝐵 − (𝑓/𝐵)2
𝐿 1
𝑤ℎ𝑒𝑟𝑒 𝑅 = √2𝐶, 𝐵 = 2𝜋√𝐿𝐶

1
𝐻(𝑓) =
1 + 𝑗√2𝑓/𝐵 − (𝑓/𝐵)2

1
𝐻(𝑓) =
𝑃2 (𝑗𝑓/𝐵)

11
Hilbert Transform (Details are not required for ENEE 339)

The quadrature filter: is an all pass filter that shifts the phase of positive frequency
by (-90° ) and negative frequency by (+90° ). The transfer function of such a filter is

−𝑗 𝑓>0
H(f) ={
𝑗 𝑓<0

Using the duality property of Fourier transform, the impulse response of the filter is

1
h(t)= 𝜋𝑡

The Hilbert transform is the output of the quadrature filter to the signal g(t)

1 ∞ 𝑔( λ)
𝑔̂(𝑡)= 𝜋𝑡 * g(t) = ∫−∞ 𝜋(𝑡− λ) 𝑑 λ

Note that the Hilbert transform of a signal is a function of time. The Fourier transform
of 𝑔̂(𝑡) is

𝐺̂ (𝑓) = -j sgn(f) G(f)

Hilbert transform can be found using either the time domain approach or the
frequency domain approach depending on the given problem, that is

1
 Direct convolution in the time domain of g(t) and 𝜋𝑡 .
 Find the Fourier transform 𝐺̂ (𝑓), then find the inverse Fourier transform
∞
𝑔̂(𝑡) = ∫ 𝐺̂ (𝑓) 𝑒 𝑗2𝜋𝑓𝑡 𝑑𝑓
−∞

Some properties of the Hilbert transform

1. A signal g(t) and its Hilbert transform 𝑔̂(𝑡) have the same energy spectral
density

2
|𝐺̂ (𝑓)| = |−𝑗 𝑠𝑔𝑛(𝑓)|𝐺(𝑓)||2 = |−𝑗 𝑠𝑔𝑛(𝑓)|2 |𝐺(𝑓)|2
= |𝐺(𝑓)|2

The consequences of this property are:

 If a signal g(t) is bandlimited, then 𝑔̂(𝑡) is bandlimited to the same
bandwidth (note that |𝐺̂ (𝑓)| = |𝐺(𝑓)| )
 𝑔̂(𝑡) and g(t) have the same total energy (or power).
 𝑔̂(𝑡) and g(t) have the same autocorrelation function.

2. A signal g(t) and 𝑔̂(𝑡) are orthogonal, i.e.,

12
 ∞
∫−∞ g(t) 𝑔̂(𝑡)𝑑𝑡 = 0

This property can be verified using the general formula of Rayleigh energy
theorem

∞ ∞ ∞
∫−∞ g(t) 𝑔̂(𝑡)𝑑𝑡 = ∫−∞ G(f) 𝐺̂ ∗ (𝑓)𝑑𝑓 = ∫−∞ G(f) {−𝑗𝑠𝑔𝑛(𝑓) 𝐺(𝑓)}∗ 𝑑𝑓
∞
= ∫−∞ 𝑗𝑠𝑔𝑛(𝑓) |𝐺(𝑓)|2 𝑑𝑓 = 0

The result above follows from the fact that |𝐺(𝑓)|2 is an even function of 𝑓 while
𝑠𝑔𝑛(𝑓) is an odd function of 𝑓. Their product is odd. The integration of an odd
function over a symmetrical interval is zero.

3. If 𝑔̂(𝑡) is a Hilbert transform of g(t) , then the Hilbert transform of 𝑔̂(𝑡) is

−𝑔(𝑡).

Example on Hilbert Transform

Find the Hilbert transform of the impulse function 𝑔(𝑡) = 𝛿(𝑡)

Solution:
Here, we use the convolution in the time domain

1
𝑔̂(𝑡)= 𝜋𝑡 ∗ 𝛿(𝑡)

As we know, the convolution of the delta function with a continuous function is the
function itself. Therefore,

1
𝑔̂(𝑡)= 𝜋𝑡

13
Example on Hilbert Transform
sin 𝑡
Find the Hilbert transform of 𝑔(𝑡) = 𝑡

Solution
Here, we will first find the Fourier transform of 𝑔(𝑡), find 𝐺̂ (𝑓), and then find 𝑔̂(𝑡)

𝑡 𝑡𝑟𝑎𝑛𝑠𝑓𝑜𝑟𝑚 1
𝐴 𝑟𝑒𝑐𝑡 ( ) ↔ 𝐴𝜏 𝑠𝑖𝑛𝑐 𝑓𝜏 ; 𝑤ℎ𝑒𝑛 𝜏 =
𝜏 𝜋
𝑡 𝑡𝑟𝑎𝑛𝑠𝑓𝑜𝑟𝑚 1 sin 𝜋𝑓𝜏 1 sin 𝑓
𝐴 rect ( ) ↔ 𝐴 =
1/𝜋 𝜋 𝜋𝑓𝜏 𝜋 𝑓
𝑡 𝑡𝑟𝑎𝑛𝑠𝑓𝑜𝑟𝑚 sin 𝑓
𝜋 𝑟𝑒𝑐𝑡 ( ) ↔
1/𝜋 𝑓
So by the duality property, we get the pair
𝑓 𝑡𝑟𝑎𝑛𝑠𝑓𝑜𝑟𝑚 sin 𝑡
𝜋 𝑟𝑒𝑐𝑡 ( ) ↔
1/𝜋 𝑡
𝑓
i.e. , 𝐺(𝑓) = 𝜋 𝑟𝑒𝑐𝑡(1/𝜋) , (See the figure below)

−𝑗𝜋 0 < 𝑓 < 1/2𝜋

𝐺̂ (𝑓) = −𝑗𝑠𝑔𝑛(𝑓)𝐺(𝑓) = {
𝑗𝜋 − 1/2𝜋 < 𝑓 < 0
∞
𝑔̂(𝑡) = ∫−∞ 𝐺̂ (𝑓) 𝑒 𝑗2𝜋𝑓𝑡 𝑑𝑓
0 1/2𝜋
= ∫−1/2𝜋 𝑗𝜋 𝑒 𝑗2𝜋𝑓𝑡 𝑑𝑓 − ∫0 𝑗𝜋 𝑒 𝑗2𝜋𝑓𝑡 𝑑𝑓
1 1
= (1 − 𝑒 −𝑗𝑡 ) − 2𝑡 (𝑒 𝑗𝑡 − 1)
2𝑡
1 1 (𝑒 𝑗𝑡 +𝑒 −𝑗𝑡 )
= −𝑡
𝑡 2

1−cos 𝑡
= 𝑡

14
Correlation and Spectral Density (Details are not required for ENEE 339)

Here, we consider the relationship between the autocorrelation function and the power
spectral density. In this discussion we restrict our attention to real signals. First, we
consider power signals and then energy signals.

Definition: The autocorrelation function of a signal g(t) is a measure of similarity

between g(t) and a delayed version of g(t).

a. Autocorrelation function of a power signal

The autocorrelation function of a power signal g(t) is defined as:

𝑅𝑔 (𝜏) = 〈𝑔(𝑡)𝑔(𝑡 − 𝜏)〉; 〈(. )〉: represents time average.

1 𝑇
𝑅𝑔 (𝜏) = lim 2𝑇 ∫−𝑇 𝑔(𝑡)𝑔(𝑡 − 𝜏)𝑑𝑡
𝑇→∞

Exercise: Show that for a periodic signal with period T0, the above definition
becomes
1 𝑇
𝑅𝑔 (𝜏) = 𝑇 ∫0 0 𝑔(𝑡)𝑔(𝑡 − 𝜏)𝑑𝑡
0

Remark: We can take this as a definition for the autocorrelation function of a

periodic signal.

Exercise: Show that if 𝑔(𝑡) is periodic with period 𝑇0 , then 𝑅𝑔 (𝜏) is also periodic
with the same period 𝑇0 .

Hint: Expand 𝑔(𝑡) in a complex Fourier series 𝑔(𝑡) = ∑∞ 𝑛=−∞ 𝐶𝑛 𝑒

𝑗𝑛𝜔0 𝜏
. Form the
delayed signal 𝑔(𝑡 − 𝜏), and then perform the integration over a complete period 𝑇0 .
You should get the following result:

𝑅𝑔 (𝜏) = ∑∞
𝑛=−∞ 𝐷𝑛 𝑒
𝑗𝑛𝜔0 𝜏
; 𝐷𝑛 = |𝐶𝑛 |2

This formula bears two results

a. 𝑅𝑔 (𝜏) is periodic with period 𝑇0 .

b. The Complex Fourier coefficients 𝐷𝑛 of 𝑅𝑔 (𝜏) are related to the complex
Fourier coefficients 𝐶𝑛 of 𝑔(𝑡) by the relation 𝐷𝑛 = |𝐶𝑛 |2 .

Properties of R(τ)
1 𝑇
 𝑅𝑔 (0) = 𝑇 ∫0 0 𝑔(𝑡)2 𝑑𝑡; is the total average signal power.
0

 𝑅𝑔 (𝜏) is an even function of 𝜏, i.e., 𝑅𝑔 (𝜏) = 𝑅𝑔 (−𝜏).

 𝑅𝑔 (𝜏) has a maximum (positive) magnitude at τ = 0 , i.e. |𝑅𝑔 (𝜏)| ≤ 𝑅𝑔 (0).
 If g(t) is periodic with period T0, then 𝑅𝑔 (𝜏) is also periodic with the same
period T0.

15
  The autocorrelation function of a periodic signal and its power spectral density
(represented by a discrete set of impulse functions) are Fourier transform pairs

𝑆𝑔 (𝑓) = F{𝑅𝑔 (𝜏)}

𝑆𝑔 (𝑓) = ∑∞ 2
𝑛=−∞|𝐶𝑛 | 𝛿(𝑓 − 𝑛𝑓0 )

Cross Correlation Function

The cross correlation function of two periodic signals 𝑔1 (𝑡) and 𝑔2 (𝑡) with the same
period 𝑇0 is defined as:

1 𝑇
𝑅1,2 (𝜏) = 𝑇 ∫0 0 𝑔1 (𝑡)𝑔2 (𝑡 − 𝜏)𝑑𝑡
0

b- Autocorrelation function of an energy signal

When g(t) is an energy signal, 𝑅𝑔 (𝜏) is defined as:

∞
𝑅𝑔 (𝜏) = ∫−∞ 𝑔(𝑡)𝑔(𝑡 − 𝜏)𝑑𝑡

Properties of R(τ)
∞
 𝑅𝑔 (0) = ∫−∞ 𝑔(𝑡)2 𝑑𝑡; is the total signal energy.
 𝑅𝑔 (𝜏) is an even function of 𝜏, i.e., 𝑅𝑔 (𝜏) = 𝑅𝑔 (−𝜏).
 𝑅𝑔 (𝜏) has a maximum ( positive ) magnitude at τ = 0 , i.e. |𝑅𝑔 (𝜏)| ≤ 𝑅𝑔 (0).
 The autocorrelation function of an energy signal and its energy spectral
density (a continuous function of frequency) are Fourier transform pairs, i.e.,

𝑆𝑔 (𝑓) = F{𝑅𝑔 (𝜏)}

∞
𝑆𝑔 (𝑓) = ∫−∞ 𝑅𝑔 (𝜏)𝑒 −𝑗2𝜋𝑓𝜏 𝑑𝜏

∞
𝑅𝑔 (𝜏) = ∫−∞ 𝑆𝑔 (𝑓)𝑒 𝑗2𝜋𝑓𝜏 𝑑𝑓.

Proof:
The autocorrelation function is defined as:
∞
𝑅𝑔 (𝜏) = ∫−∞ 𝑔(𝜆)𝑔(𝜆 − 𝜏)𝑑 𝜆
In this integral we have replaced t by 𝜆 (both are dummy variables of integration).
With this substitution, we can rewrite the integral as

∞
𝑅𝑔 (𝜏) = ∫−∞ 𝑔(𝜆)𝑔(−(𝜏 − 𝜆))𝑑 𝜆
One can realize that 𝑅𝑔 (𝜏) is nothing but the convolution of 𝑔(𝜏) and −𝑔(𝜏). That is,

𝑅𝑔 (𝜏) = 𝑔(𝜏) ∗ 𝑔(−𝜏)

Taking the Fourier transform of both sides, we get

16
 𝐹{𝑅𝑔 (𝜏)} = 𝐺(𝑓)𝐺 ∗ (𝑓)

Therefore, 𝑆𝑔 (𝑓) = F{𝑅𝑔 (𝜏)} = |𝐺(𝑓)|2.

Cross Correlation Function

The cross correlation function of two energy signals 𝑔1 (𝑡) and 𝑔2 (𝑡) is defined as;
∞
𝑅1,2 (𝜏) = ∫−∞ 𝑔1 (𝑡)𝑔2 (𝑡 − 𝜏)𝑑𝑡

Example:

Find the auto-correlation function of the sine signal 𝑔(𝑡) = 𝐴cos(2𝜋𝑓0 𝑡 + 𝜃), where
𝐴 and 𝜃 are constants.

Solution:

As we know, 𝑔(𝑡) is a periodic signal. So, we find 𝑅𝑔 (𝜏) using the definition
1 𝑇
𝑅𝑔 (𝜏) = 𝑇 ∫0 0 𝑔(𝑡)𝑔(𝑡 − 𝜏)𝑑𝑡
0

1 𝑇
𝑅𝑔 (𝜏) = 𝑇 ∫0 0 𝐴cos(2𝜋𝑓0 𝑡 + 𝜃)𝐴cos(2𝜋𝑓0 𝑡 − 2𝜋𝑓0 𝜏 + 𝜃)𝑑𝑡
0

𝐴2 𝑇
𝑅𝑔 (𝜏) = 2𝑇 ∫0 0[cos(4𝜋𝑓0 𝑡 − 2𝜋𝑓0 𝜏 + 2𝜃) + cos(2𝜋𝑓0 𝜏)]𝑑𝑡
0

𝐴2
𝑅𝑔 (𝜏) = 2𝑇 [0 + cos(2𝜋𝑓0 𝜏)𝑇0 ]
0

𝐴2
𝑅𝑔 (𝜏) = cos(2𝜋𝑓0 𝜏); periodic with period 𝑇0 .
2

Example:

Determine the autocorrelation function of the sinc pulse 𝑔(𝑡) = 𝐴𝑠𝑖𝑛𝑐2𝑊𝑡.

Solution:

Using the duality property of the Fourier transform, we can deduce that
𝐴 𝑓
𝐺(𝑓) = 2𝑊 𝑟𝑒𝑐𝑡(2𝑊)

The energy spectral density of 𝑔(𝑡) is

𝐴 𝑓
𝑆𝑔 (𝑓) = |𝐺(𝑓)|2 = (2𝑊)2 𝑟𝑒𝑐𝑡(2𝑊)

Taking the inverse Fourier transform, we get the autocorrelation function

17
 𝐴2
𝑅𝑔 (𝜏) = 2𝑊 𝑠𝑖𝑛𝑐2𝑊𝑡

Exercise:

a. Find and plot the cross correlation function of the two signals

1 0≤𝑡≤2
𝑔1 (𝑡) = {
0, 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒
1 0≤𝑡≤1
𝑔2 (𝑡) = {
−1 1 < 𝑡 ≤ 2

b. Are 𝑔1 (𝑡) and 𝑔2 (𝑡) orthogonal?

Exercise:

Find and plot the autocorrelation function for the periodic saw-tooth signal shown
below:

18
Example:

Find the autocorrelation function of the rectangular pulse g(t).

Solution:

As we saw earlier, this pulse is an energy signal and , therefore, we can find its 𝑅𝑔 (𝜏)
as:
1
𝑅𝑔 (𝜏) = ∫𝜏 (𝐴)(𝐴)𝑑𝑡 = A2 (1-τ) ; 0< τ <1

Using the even symmetry property of the autocorrelation function, we can find 𝑅𝑔 (𝜏)
for – ve values of τ as:

𝑅𝑔 (𝜏) = = A2 (1+τ) ; -1< τ <0

This function is sketched below. Note that that the maximum value occurs at τ = 0 and
that g(t) and g(t-τ) become uncorrelated for τ = 1 sec, which is the duration of the
pulse.

The energy spectral density is Sg(f) =𝐹{Rg(τ)}= A2sinc2f

19
Bandwidth of Signals and Systems: G(f)

Def: A signal g(t) is said to be (absolutely) band-limited to BHz if

G(f) = 0 for |f| >B -B B

Def: A signal x(t) is said to be (absolutely) time-limited if

X(t)
x(t) = 0 for |t| >T

Theorem: An absolutely band-limited waveform cannot be

-T T
absolutely time-limited (theoretically has an infinite time duration) and vice versa.

We have earlier seen examples that support this theorem. For example, the delta
function, which has an almost zero time duration, has a Fourier transform which
extends uniformly over all frequencies (infinite bandwidth). Also, a constant value in
the time domain (a dc) has a Fourier transform, which is an impulse in the frequency
domain. This is repeated here for convenience.

In general, there is an inverse relationship between the signal bandwidth and the time
duration. The bandwidth and the time duration are related through a relation, called
the time bandwidth product, of the form

Bandwidth *Time Duration ≥ constant

The value of the constant depends on the way the bandwidth and the time duration of
1 1
a signal are defined as will be illustrated later (Possible values of the constant = 2 , 4𝜋)

Remarks:

1. The bandwidth of a signal provides a measure of the extent of significant

frequency content of the signal.
2. The bandwidth of a signal is taken to be the width of a positive frequency
band.

20
3. For baseband signals or networks , where the spectrum extends from –B to B,
the bandwidth is taken to be B Hz.
4. For bandpass signals or systems where the spectrum extends between (f1, f2)
and (-f1, -f2), the B.W= f2 – f1.

Some Definitions of Bandwidth:

1- Absolute bandwidth
Here, the Fourier transform of a signal is non zero only within a certain
frequency band. If G(f) = 0 for |f| >B , then g(t) is absolutely band-limited to
BHz. When G(f) ≠ 0 for f 1< |f |< f2 , then the absolute bandwidth is f2 - f 1.

|𝐺(𝑓)|
G(0
2- 3-dB (half power points) bandwidth 1
) 𝐺(0)
The range of frequencies from 0 to some frequency B √2
1
at which |G(f)| drops to √2 of its maximum 0 B f
value (for a low pass signal). G(f)
As for a band pass signal, the B.W = f2 – f1 Gmax
1
𝐺𝑚𝑎𝑥
√2

0 f1 f2 f
3- The 95 % (energy or power) bandwidth.
Here , the B.W is defined as the band of frequencies where the area under the
energy spectral density (or power spectral density) is at least 95% (or 99%) of
the total area .

∞ ∞
Total Energy = ∫−∞ |𝐺(𝑓)|2 𝑑𝑓 = 2∫0 |𝐺(𝑓)|2 𝑑𝑓

𝐵 ∞
∫−𝐵 |𝐺(𝑓)|2 𝑑𝑓 = 0.95 ∫−∞ |𝐺(𝑓)|2 𝑑𝑓

4- Equivalent Rectangular Bandwidth.

It is the width of a fictitious rectangular spectrum such that the power in that
rectangular band is equal to the power associated with the actual spectrum
over positive frequency
Area under rectangle = Area under curve
∞
|G(0)|2*2Beq = ∫−∞ |𝐺(𝑓)|2df
∞
|G(0)|2*2Beq = 2∫0 |𝐺(𝑓)|2 df
1 ∞
Beq = |𝐺(0)|2 ∫0 |𝐺(𝑓)|2df

21
 5- Null – to –null bandwidth:
For baseband signals , B.W is the first null in the envelope of the magnitude
spectrum above zero.

𝑡 𝑠𝑖𝑛𝜋𝑓𝜏
rect(𝜏) → τ sincfτ = τ 𝜋𝑓𝜏

Zero crossing take place when sin(πfτ) = 0

𝑛
πfτ = nπ → f = τ ; n = 1,2,………
1
B.W = smaller τ large bandwidth.
𝜏

For a band pass signal, B.W= f2 – f1

6- Bounded spectrum bandwidth:

Range of frequencies as (0,B) such that outside the band , the power spectral
density must be down by say 50 dB below the maximum value
|𝐺(𝐵)|2
-50 dB = 10 log |𝐺(0)|2

7- RMS Bandwidth:
∞
∫−∞ 𝑓 2 |𝐺(𝑓)|2 df 1\2
Brms =( ∞ )
∫−∞ |𝐺(𝑓)|2 df

The corresponding rms duration of g(t) is

∞
∫−∞ 𝑡 2 |𝑔(𝑡)|2 dt 1\2
Trms = =( ∞ )
∫−∞ |𝑔(𝑡)|2 dt

(here g(t) is assumed to be centered around the origin).

1
Remark: The time bandwidth product is Trms Brms ≥ 4𝜋

Time – Bandwidth Product

To illustrate the time – bandwidth product, consider the equivalent rectangular

bandwidth defined earlier as
∞
∫−∞ |𝐺(𝑓)|2 df
Beq = 2|𝐺(0)|2

Analogous to this definition, we define an equivalent rectangular time duration as :

22
 ∞ 2
(∫−∞ |𝑔(𝑡)|𝑑𝑡)
Teq = ∞
∫−∞ |𝑔(𝑡)|2 dt

The time bandwidth product is

∞ ∞ 2
∫−∞ |𝐺(𝑓)|2 df (∫−∞ |𝑔(𝑡)|𝑑𝑡)
BeqTeq = . ∞
2|𝐺(0)|2 ∫−∞ |𝑔(𝑡)|2 dt

∞ ∞
Note ∫−∞ |𝑔(𝑡)|2 dt = ∫−∞ |𝐺(𝑓)|2 df ; Rayleigh energy theorem. Note also that
∞
𝐺(0) = ∫−∞ 𝑔(𝑡)𝑑𝑡. Using these relations, we get

∞ 2
1 (∫−∞ |𝑔(𝑡)|𝑑𝑡)
BeqTeq = ∞
2 | ∫−∞ 𝑔(𝑡)dt|2

Case 1: When g(t) is positive for all time t, then |g(t)| = g(t) and BeqTeq becomes
1
BeqTeq = 2

Case 2 : For a general g(t) that can take on positive as well as negative values, BeqTeq
satisfies the inequality
1
BeqTeq ≥ 2

Note : For Brms and Trms , the time – bandwidth satisfies the inequality
1
Brms Trms ≥ 4𝜋

Example : Bandwidth of a trapezoidal signal

Find the equivalent rectangular bandwidth, Beq, for the trapezoidal pulse shown.

Solution :
∞ 2 g(t)
(∫−∞ |𝑔(𝑡)|𝑑𝑡)
Teq = ∞ A
∫−∞ |𝑔(𝑡)|2 dt

∫ |𝑔(𝑡)|𝑑𝑡 = 𝐴 ( 𝑡𝑎 + 𝑡𝑏 )
−∞

∞ 2𝐴2 -tb -ta ta tb

∫−∞|𝑔(𝑡)|2 𝑑𝑡 = 3
( 2𝑡𝑎 + 𝑡𝑏 )

3 (𝑡𝑎 +𝑡𝑏 )2
𝑇𝑒𝑞 = 2 (2𝑡𝑎 +𝑡𝑏 )

0.5 2𝑡 +𝑡
Beq = 𝑇 = 3(𝑡 𝑎+𝑡 𝑏)2 .
𝑒𝑞 𝑎 𝑏

23
Remark: Note that using this method we were able to determine the signal bandwidth
without the need to go through the Fourier transform.

Exercise: Use the above method to find the equivalent rectangular bandwidth for the
𝑡
triangular signal 𝑔(𝑡) = 𝑡𝑟𝑖(𝑇).

Example: Bandwidth of a periodic signal:

Find the bandwidth For the periodic square function define over one period as
−𝑇 𝑇
2𝐴, 4 ≤ 𝑡 ≤ 4
𝑔(𝑡) = {
−𝐴, 𝑜. 𝑤

Solution:

The average power, computed using the time average, is

𝑇0
1
𝑃𝑎𝑣 = ∫ |𝑔(𝑡)|2 𝑑𝑡
𝑇0
0

1 2 2
5𝐴2 𝜏 5𝐴2
= [4𝐴 𝜏 + 𝐴 𝜏] = = = 2.5𝐴2
𝑇0 2𝜏 2

Also, by using the Parseval’s theorem, the average power can be computed as:

𝑃𝑎𝑣 = |𝐶0 |2 + 2 ∑∞
𝑛=1|𝐶𝑛 |
2

We recall that the Fourier coefficients for this signal were found in Chapter 1. Using
these values we get

𝐴 2 (3𝐴)2
𝑃𝑎𝑣 = (2 ) + 2 ∑∞
𝑛=1 (𝑛𝜋)2

𝐴2 (3)2
𝑃𝑎𝑣 = + 2𝐴2 ∑∞
𝑛=1 (𝑛𝜋)2
4

Let us take n = 1

9
𝑃1 = 𝐴2 {0.25 + 2 . } = 2.073𝐴2
𝜋2

𝑃1 2.073𝐴2
= = 82.95%
𝑃𝑎𝑣 2.5𝐴2

(This is the percentage of the total power that lies in the dc and the fundamental
frequency ).

For n = 3

24
 32 32
𝑃3 = 𝐴2 {0.25 + 2 (𝜋2 + 32 𝜋2)} = 2.276𝐴2

𝑃3 2.276𝐴2
= = 91.05%
𝑃𝑎𝑣 2.5𝐴2

(Fraction of power in the dc, fundamental and third harmonic terms)

For n = 5

3 2 3 2 3 2
𝑃5 = 𝐴 {0.25 + 2 (( ) + ( ) + ( ) )} = 2.349𝐴2
2
𝜋 3𝜋 5𝜋

𝑃5 2.349𝐴2
= = 93.97%
𝑃𝑎𝑣 2.5𝐴2

Here , the 93% power band width is 5𝑓0 .

Example: Bandwidth of an energy signal .

If the signal 𝑔(𝑡) = 𝐴𝑒 −∝𝑡 𝑢(𝑡) is passed through an ideal LPF with B.W = B Hz,

find the fraction of the signal energy contained in B.

Solution

The Fourier transform of g(t) is:

𝐴
𝐺(𝑓) = ∝+𝑗2𝜋𝑓

The energy in g(t), using the time domain, is

∞ ∞
𝐴2
𝐸𝑔 = ∫|𝑔(𝑡)|2 𝑑𝑡 = ∫ 𝐴2 𝑒 −2∝𝑡 𝑑𝑡 =
2∝
0 0

Energy contained in the filter output y(t) is

𝐵 𝐵
𝐴2
𝐸𝑦 = ∫|𝐺(𝑓)|2 𝑑𝑓 = ∫ 𝑑𝑓
(∝2 + (2𝜋𝑓)2 )
−𝐵 −𝐵

2𝐴2 2𝜋𝐵
𝐸𝑦 = 𝑡𝑎𝑛−1
2𝜋 ∝ ∝
The ratio of 𝐸𝑦 to the total energy is

𝐸𝑦 2 2𝜋𝐵
= 𝑡𝑎𝑛−1
𝐸𝑔 𝜋 ∝

25
The table below shows this ratio for various values of B .

B (𝑬𝒚 ⁄𝑬𝒈 ) × 𝟏𝟎𝟎

∝ 63.9
4

∝ 80.38
2
∝ 89.95

2∝ 94.94

Thus, the 95% energy bandwidth is 2 ∝.

Exercise: Find the 98% energy bandwidth.

26
 Pulse Response and Risetime
A rectangular pulse contains significant high frequency components. When that
pulse is passed through a LPF, the high frequency components will be attenuated
resulting in signal distortion.

We need to investigate the relationship that should exist between the pulse bandwidth
and the channel bandwidth. This subject is of particular importance, especially, when
we study the transmission of data over band-limited channels. In the simplest form, a
binary digit 1 may be represented by a pulse 𝐴, 0 ≤ 𝑡 ≤ 𝑇𝑏 , while binary digit 0
may be represented by the negative pulse −𝐴, 0 ≤ 𝑡 ≤ 𝑇𝑏 . So, in order to retrieve the
transmitted data, the channel bandwidth must be wide enough to accommodate the
transmitted data.

To convey this idea in a simple form, we first consider the response of a first order
low pass filter to a unit step function and then to a pulse.

Step response of a first order LPF (channel)

Let x(t) = u(t) be applied to a first order RC circuit. This first order filter is a fair
representation of a low pass communication channel.

The system differential equation is

𝑑𝑔(𝑡)
x(t) = Ri(t) + g(t) = RC + g(t)
𝑑𝑡

where g(t) is the channel output.

𝑑𝑔(𝑡)
RC + g(t) = u(t)
𝑑𝑡

The solution to this first order system is

g(t) = (1-𝑒 −𝑡/𝑅𝐶 ) u(t)

The 3-db B.W of the channel is

1
B= (to be derived shortly)
2𝜋𝑅𝐶

g(t) = (1-𝑒 −2𝜋𝐵𝑡 ) u(t)

27
Define the difference between the input and the output as:

e(t) = u(t) - g(t) = 𝑒 −2𝜋𝐵𝑡

Note that e(t) decreases as B increases. Meaning that as the channel bandwidth
increases, the output becomes closer and closer to the input. In the ideal case, when
the channel bandwidth becomes infinity, the output becomes a step function. In
essence, to reproduce a step function (or a rectangular pulse), a channel with infinite
bandwidth is needed.

The Risetime

The Rise time is a measure of the speed of a step response. One common measure is
the 10-90 % rise time defined as the time it takes for the output to rise between 10%
to 90% of the final (steady state) value (1) when a step function is applied to a LIT
system. For the step response g(t) and the first order RC circuit considered above, the
rise time can be easily calculated as:
0.35
tr = t2 - t1 ≈
𝐵

From this result, we conclude that: increasing the bandwidth of the channel will
decrease the rise time (a faster response).
0.35
Exercise: For the system above, verify that the rise time is given as 𝑡𝑟 = 𝐵

Exercise: Find the 10-90% rise time for a second order low pass filter with 3-dB
bandwidth B and transfer function
1
𝐻(𝑓) =
𝑗𝑓
𝑃2 ( 𝐵 )
Where, 𝑃2 (𝑥) = 1 + √2𝑥 + 𝑥 2 .

28
(Hints: You may let B=10, for example, use matlab to find the step response, and then
find the rise time).

Pulse response

It is the response of the circuit to a pulse of duration τ. For the same circuit let us
apply the pulse

x(t) = u(t) - u(t- τ)

Using the linearity and time invariance properties, the output can be obtained from the
step response as:

0 𝑡<0
−𝑡/𝑅𝐶
y(t) ={ 1 − 𝑒 0 < 𝑡 < τ}
τ 𝑡− τ
(1 − 𝑒 −𝑅𝐶 ) . 𝑒 − 𝑅𝐶 t> τ

This is sketched in the figure below.

Bandwidth Considerations

The transfer function of the RC circuit is

1/𝑗2𝜋𝑓𝑐 1
H (f) = =
𝑅+1/𝑗2𝜋𝑓𝑐 1+𝑗2𝜋𝑓𝑅𝑐

1
|H(f)| =
√1+(2𝜋𝑓𝑅𝑐)²

1 1
Let B = ; 3-db bandwidth ; 2𝜋𝑓𝑅𝑐 = 1 ; f =
2𝜋𝑅𝑐 2𝜋𝑅𝑐

1
Then , H(f) =
1+𝑗𝑓/𝐵

1
|H(f)| =
𝑓
√1+( )²
𝐵

29
For the rectangular pulse x(t), we have

𝑋(𝑓) = 𝑠𝑖𝑛𝑐𝑓𝜏
The first null frequency of X(f) is an estimate of the bandwidth Bx of x(t), which is of
1
the order of ≈ .
τ
1
1. When τ is large, such that signal bandwidth Bx = << B (channel B.W)
τ
𝑌(𝑓) = 𝑋(𝑓)𝐻(𝑓) ≈ 𝑋(𝑓)

and the output resembles the input. There is enough time for x(t) to reach the
maximum value .
1
2. When τ is small, such that signal Bx = τ > > B (channel B.W)
𝑌(𝑓) = 𝑋(𝑓)𝐻(𝑓) ≈ 𝐻(𝑓)

The signal suffers a considerable amount of distortion and 𝑌(𝑓) is no longer

proportional to X(f).

30
 Band-pass Signals and Systems
(Details are not required for ENEE 339)

A signal g(t) is called a band pass signal if its Fourier transform G(f) is non-
negligible only in a band of frequencies of total extent 2W centered about fc .

A signal is called narrowband if 2W is small compared with fc.

A band pass signal g(t) represented in the canonical form:

g(t) = gI(t) cos 𝜔ct - gQ(t) sin𝜔ct.

gI(t) is a low pass signal of B.W = W Hz called the in phase component of g(t) .

gQ(t) is a low pass signal of B.W = W Hz called the quadrature component .

g(t) appears as a modulated signal in which gI(t) and gQ(t) are the low pass signals and
𝑓𝑐 is the carrier frequency. Recall the modulation property of the Fourier transform :
1
x(t) cos 𝜔ct → 2 (X(f- fc)+ X(f+ fc))
1
x(t) sin 𝜔ct →𝑗2 (X(f- fc)- X(f+ fc))

Define the complex envelope of a signal g(t) as:

𝑔̃ (t) = gI(t) + j gQ(t)

𝑔̃ (t) is a low pass signal of B.W =W. The signals g(t) and 𝑔̃ (t) are related by :

g(t) = Re{𝑔̃ (t) 𝑒 𝑗𝜔𝑐𝑡 }

How to get gI(t) and gQ(t) from g(t)

If we multiply g(t) by cos 𝜔ct, we get

g(t) cos𝜔ct = gI(t) cos2 𝜔𝑐 t - gQ(t) sin 𝜔ct cos 𝜔ct

1 1 1
= 2 gI(t) + gI(t) cos 2𝜔ct- 2 gQ(t) sin 2𝜔ct .
2

The first term is the desired low pass signal. The second and third terms are high
frequency components centered about 2 fc.

gI(t) =lowpass{2g(t) cos𝜔ct}

Or, in the frequency domain

𝐺(𝑓 − 𝑓𝑐 ) + 𝐺(𝑓 + 𝑓𝑐 ) −𝑤 ≤𝑓 ≤𝑤
GI(f)= { }
0 𝑜𝑡herwise

31
Now if we multiply g(t) by sin𝜔ct, we get

g(t) sin𝜔ct= gI(t) sin 𝜔Ct cos 𝜔C t - gQ(t) sin2 𝜔ct

1 1 1
=- gQ(t) + 2 gI(t) sin 2𝜔ct+ 2gQ(t) cos 2𝜔ct
2

Again, the first term is a low pass signal, while the second and third are high
frequency terms centered about 2 fc.

gQ(t) = - low pass{2g(t) sin𝜔ct}In the frequency domain, this is equivalent to

𝑗[𝐺(𝑓 − 𝑓𝑐) − 𝐺(𝑓 + 𝑓𝑐) −𝑤 ≤𝑓 ≤𝑤
GQ(t)= { }
0 𝑜𝑡herwise

Band pass systems:

The analysis of band pass systems can be simplified by using the complex envelope
concept. Here, results and techniques from low pass systems can be easily applied to
band pass systems .

The problem to be addressed is :

The input x(t) is a band pass signal

x(t)=xI(t)cos𝜔ct - xQ(t)sin𝜔ct

x(t) is applied to a band pass filter represented as:

h(t) = hI(t)cos𝜔ct - hQ(t)sin𝜔ct

The objective is to find the filter output y(t). The output is, of course, the convolution
of x(t) and h(t) (y(t) = x(t)*h(t)), which can also be expressed as:

y(t) = yI(t)cos𝜔ct - yQ(t)sin𝜔ct

32
Due to the band-pass nature of the problem, carrying out the direct convolution will
be a tedious task due to the presence of the sin and cos functions in all terms. The
complex envelope concept simplifies the problem to a very great extent. The
procedure is summarized as follows:

a. Form the complex envelope for both the input and the channel:
𝑥̃(t)= xI(t) + jxQ(t)

ℎ̃(t)= hI(t) + jhQ(t)

b. Carry out the convolution between 𝑥̃(t) and ℎ̃(t). Note that both signals are low
pass signals and so 𝑦̃(t) is also low pass.

2 𝑦̃(t)= ℎ̃(t) * 𝑥̃(t)

𝑦̃(t) = yI(t) + jyQ(t)

c. The band-pass filter output is obtained from the low pass signal 𝑦̃(t) through
the relation

y(t) =Re{𝑦̃(t) 𝑒 jwc t }

or through the relation

y(t) = yI(t)cos𝜔ct - yQ(t)sin𝜔ct

33
Example :

The rectangular radio frequency (RF) pulse

𝐴 𝑐𝑜𝑠2𝜋𝑓𝑐 𝑡 0≤𝑡≤𝑇
x(t) = { }
0 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒
is applied to a linear filter with impulse response (We will see later that this is a filter
matched to x(t), called the matched filter).

h(t) = x(T - t)
1
Assume that T= nTC; n is an integer, Tc = 𝑓𝑐, determine the response of the filter and
sketch it.

Solution: We follow the three steps outlined above.

h(t) = 𝐴 𝑐𝑜𝑠2𝜋𝑓𝑐 (𝑇 − 𝑡)

= 𝐴 𝑐𝑜𝑠2𝜋𝑓𝑐 𝑇 𝑐𝑜𝑠2𝜋𝑓𝑐 𝑡 + 𝐴 𝑠𝑖𝑛2𝜋𝑓𝑐 𝑇𝑠𝑖𝑛2𝜋𝑓𝑐 t

𝑛𝑇 𝑛𝑇
=𝐴 𝑐𝑜𝑠2𝜋 ( 𝑇 𝑐) 𝑐𝑜𝑠2𝜋𝑓𝑐 𝑡 + 𝐴 𝑠𝑖𝑛2𝜋 ( 𝑇 𝑐) 𝑠𝑖𝑛2𝜋𝑓𝑐 t
𝑐 𝑐

cos2𝑛𝜋≡1 sin2𝑛𝜋≡0

𝐴 𝑐𝑜𝑠2𝜋𝑓𝑐 𝑡 0≤𝑡≤𝑇
Therefore, h(t) = { }
0 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒
The complex envelopes of x(t) and h(t) are (step a)

𝐴 0≤𝑡≤𝑇
𝑥̃ (t) = { }
0 𝑜. 𝑤
𝐴 0≤𝑡≤𝑇
ℎ̃ (𝑡) = { }
0 𝑜. 𝑤

𝑦̃(t) = 𝑥̃(𝑡) ∗ ℎ̃(𝑡) is the triangular signal shown in the Figure (step b).

34
 𝐴²𝑡 0≤𝑡≤𝑇
2𝑦̃(𝑡)={ }
𝐴²(2𝑇 − 𝑡) T ≤ 𝑡 ≤ 2𝑇

The bandpass signal is obtained as (step c)

𝐴²
𝑡 𝑐𝑜𝑠𝑤𝑐 𝑡 0≤𝑡≤𝑇
2
y(t)={𝐴2 }
(2𝑇 − 𝑡) 𝑐𝑜𝑠𝑤𝑐 𝑡 𝑇 ≤ 𝑡 ≤ 2𝑇
2

and is sketched as in the figure below.

Exercise
𝑡
The band-pass signal 𝑥(𝑡) = 𝑒 −𝜏 cos(2𝜋𝑓𝑐 𝑡) 𝑢(𝑡) is applied to a band-pass filter with
impulse response ℎ(𝑡) given as:

𝐴 𝑐𝑜𝑠2𝜋𝑓𝑐 𝑡 0≤𝑡≤𝑇
ℎ(𝑡) = { }
0 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒
Find and sketch the filter output.

35