IE3012 Communication Principles

Contents

1 Week 1 3
1.1 Linear Modulation . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Review of Mathematical Tools . . . . . . . . . . . . . . . . . . . 4

2 Week 2 7
2.1 DSBSC-AM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2 Demodulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.3 Frequency Translation/Shifting (Mixing) . . . . . . . . . . . . . . 8

3 Week 3 9
3.1 Frequency Division Multiplexing . . . . . . . . . . . . . . . . . . 9
3.2 FDM Demultiplexing . . . . . . . . . . . . . . . . . . . . . . . . . 9
3.3 Quadrature Amplitude Modulation . . . . . . . . . . . . . . . . . 10
3.4 QAM Demodulation . . . . . . . . . . . . . . . . . . . . . . . . . 10

4 Week 4 11
4.1 Noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
4.2 White Noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
4.3 Bandpass Noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
4.4 Demodulation in Noisy Channel . . . . . . . . . . . . . . . . . . . 12

5 Week 5 14
5.1 Phase Modulation . . . . . . . . . . . . . . . . . . . . . . . . . . 14
5.2 Frequency Modulation . . . . . . . . . . . . . . . . . . . . . . . . 15
5.3 Single Tone Modulation . . . . . . . . . . . . . . . . . . . . . . . 15
5.4 Narrow-Band FM . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
5.5 Wide-Band FM . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

6 Week 6 18
6.1 Bandwidth of FM signals . . . . . . . . . . . . . . . . . . . . . . 18
6.2 Generation of FM signals . . . . . . . . . . . . . . . . . . . . . . 19

7 Week 7 20
7.1 Demodulation of FM Signals . . . . . . . . . . . . . . . . . . . . 20
7.2 Summary (What to Memorise?) . . . . . . . . . . . . . . . . . . . 23

1
8 Week 8 25
8.1 Review of Fourier Series and Transforms . . . . . . . . . . . . . . 25
8.2 Linear Pulse Coded Modulation (PCM) . . . . . . . . . . . . . . 26

9 Week 9 28
9.1 Review of Expected Value . . . . . . . . . . . . . . . . . . . . . . 28
9.2 Average Quantisation Error . . . . . . . . . . . . . . . . . . . . . 28
9.3 Nonlinear PCM . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
9.4 Delta Modulation . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

2
Chapter 1

Week 1

1.1 Linear Modulation

Modulation is a process in which a modulator converts a carrier signal c(t) into
a modulated signal x(t) by modifying certain parameters of c(t) in accordance
to the modulating/message signal s(t). In the case of a mobile phone, the
modulating signal is the human speech while both carrier and modulated signals
are radio waves. The carrier signal c(t) is typically a cosine wave defined by:

c(t) = Ac cos (2πfc t + ϕc )

Types of modulation includes amplitude, frequency, and phase modulated signal
(AM, FM, and PM respectively). Which means that the modulated signal
carries information regarding the modulating signal s(t) in either its amplitude
(1), frequency (2) or phase shift (3).

x(t) = Ac (t) cos(2πfc t + ϕc ) (1.1)

x(t) = Ac cos(2πfc (t)t + ϕc ) (1.2)
x(t) = Ac cos(2πfc t + ϕc (t)) (1.3)
The purpose of modulation: 1. Enables the communication signal to travel
further, easiest example is communicating using our voices. Barriers such as a
wall can block out our voices from reaching the other end, but when converting
into a radio signal, it can easily travel distances and penetrate through such
obstacles. 2. Enables us to reduce the size of radio frequency antennas by
transmitting at a higher frequency fc . 3. Enables us transmit multiple signals
on one communication channel (multiplexing). 4. Enables us to comply with
radio frequency spectrum regulations.

3
1.2 Review of Mathematical Tools
Fourier Transform
An operation which gives us the frequency component of a signal. What this
means is given a waveform v(t), we can get the same function v(t) as a function
of frequency by performing Fourier transform on v(t).
Z ∞
F {v(t)} = v(t) exp (−j2πf t)dt
−∞

Fourier Spectrum
Plotting the result of a Fourier transform on a waveform is called the Fourier
Spectrum, a plot of Fourier transform on the positive and negative frequency
axes. An example is the Fourier transform of the cosine signal:
1
F {cos (2πfc t)} = [δ(f − fc ) + δ(f + fc )]
2
Here are the plots of the cosine signal

and its resulting Fourier transform signal.

1/2 1/2

f
−fc 0 fc

Amplitude Spectrum
The Fourier spectrum often has two components – the amplitude and phase
spectrum, since some of its transformations still has a j in its value. In this
course, spectrum refers to amplitude spectrum, unless stated otherwise. An
example, determine the amplitude spectrum expression of

x(t) = sin (2πfc t) + cos (2πfc t)

4
Taking the Fourier transform gives us
1 1
F {x(t)} = [δ(f − fc ) + δ(f + fc )] + [δ(f − fc ) + δ(f + fc )]
2j 2
   
−j 1 j 1
= + δ(f − fc ) + + δ(f + fc )
2 2 2 2
Now we see that the coefficients contain both real and imaginary parts, but
since we are looking for the amplitude spectrum, we are only concerned with
the magnitude (i.e. |F {x(t)}|). Note that the || signs denote the magnitude of
the waveform as opposed to just the absolute value.
The Fourier plot of the resulting Fourier transform is given as below:
1 j 1 j
2 + 2 2 − 2

f
−fc 0 fc

Hence the amplitude spectrum can be found by,

   
−j 1 j 1
|F {x(t)}| = + δ(f − fc ) + + δ(f + fc )
2 2 2 2
1 1
= √ δ(f − fc ) + √ δ(f + fc )
2 2
Note that to find the magnitude of an imaginary number z, it is simply
q
2 2
|z| = Re(z) + Im(z)
Further sidenote for self reference, if z = a + jb, then:
Re(z) = a, Im(z) = b

Filtering
A filter is a device that removes certain frequency component(s) of the input
signal. The mathematical expression of filtering is as follows:

Y (f ) = H(f )X(f )
where Y (f ) is the Fourier transform of the output signal, X(f ) is the Fourier
transform of the input signal, and H(f ) is the filter transfer function. Two types
of filter we will often use in this course are the: 1. Lowpass filter, it allows
all frequencies less than some threshold w to pass, and rejects all frequencies
beyond w to pass. An rectangular signal of amplitude 1 can be used as a low
pass filter. 2. Bandpass filter, it will allow frequency components within an
interval w1 ≤ |f | ≤ w2 to pass, rejects everything else.

5
Bandwidth
Defined as the range of positive-value frequency that is occupied by a spec-
trum.

Signal Power
In the time domain, the power of a signal v(t) can be calculated as
Z T0 /2
1
v 2 (t) = |v(t)|2 dt
T0 −T0 /2

In the frequency domain, the power of a signal v(t) can be calculated as

Z ∞
v 2 (t) = Sv (f )df (1.4)
−∞

where Sv (f ) is the Power Spectral Density (PSD) of v(t).

6
Chapter 2

Week 2

2.1 DSBSC-AM
Conventional AM signals are transmitted as

x(t) = Ac [1 + ms(t)] cos 2πfc t = Ac cos 2πfc t + mAc s(t) cos 2πfc t

where Ac cos 2πfc t is the carrier term and the mAc s(t) cos 2πfc t are the side-
bands. The transmission of the carrier component in conventional AM howver,
is a waste of power. In Double-Sideband Suppressed Carrier (DSBSC) AM, we
suppress the transmission of the carrier component so that only the sideband
component is transmitted. Recall that amplitude modulation means that the
message signal to be transmitted is encoded as the amplitude of the carrier sig-
nal. This means that the final transmitted signal has an amplitude envelope
that follows the message signal waveform.

2.2 Demodulation
The job of a demodulator is to recover the message signal s(t) from the AM
signal x(t). Demodulation is said to be successful if the demodulator output is
= non-zero constant ×s(t). A demodulator that can do this for a suppressed-
carrier AM is the ”coherent demodulator”. The structure of a coherent demod-
ulator consists of a low pass filter and a local oscillator. An input AM signal
s(t) is multiplied with the signal provided by the local oscillator of the form
A′c cos(2πfc t + ϕ). The multiplied result (known as the internal signal) v(t) is
then passed through the low pass filter to obtain some constant ×s(t). Given
the AM signal:

x(t) = Ac s(t) cos(2πfc t)

7
 the internal signal v(t) is

v(t) = A′c cos(2πfc t + ϕ)Ac s(t) cos(2πfc t)

= A′c cos(2πfc t + ϕ) · Ac s(t) cos(2πfc t)

1 ′ 1
= Ac Ac s(t) cos ϕ + A′c Ac cos(4πfc t + ϕ)
2 2
then after low pass filtering, with bandwidth of s(t), v0 (t) = 21 A′c Ac cos (ϕ)s(t)
which is a constant ×s(t), the form we want. But if ϕ = ±π/2, v0 (t) = 0, which
must be avoided. The demodulation output is maximum if ϕ = 0, which means
the signal from the local oscillator must have the same frequency and phase as
the carrier component in the AM signal. A low pass filter works here because
the 12 A′c Ac cos(4πfc t + ϕ) term will have frequency split as sidebands due to the
cos(4πfc t).

2.3 Frequency Translation/Shifting (Mixing)

To shift the spectrum of a signal upward and downward in the frequency domain,
a signal goes through an operation called mixing. A mixer consists of multiplying
the input signal s(t) with the local oscillator’s signal, then applying a bandpass
filter onto the multiplied signal v1 (t). Consider a DSBSC-AM signal:

x(t) = s(t) cos(2πfc t)

We then obtain the internal signal to be

v1 (t) = x(t) cos(2πf1 t) = s(t) cos(2πfc t) cos(2πf1 t)

s(t)
= [cos(2π(fc − f1 )t) + cos(2π(fc + f1 )t)]
2
If v1 (t) is then passed through a bandpass filter, with a center frequency of
f0 = fc − f1 , and a bandwidth = BW of x(t), then the mixer output v2 (t) is a
DSBSC-AM signal with a new carrier frequency f0 .
1
v2 (t) = [v1 (t)]BPF = s(t) cos(2πf0 t)
2
Since the input signal x(t) had a frequency fc , and mixing output signal v2 (t)
has a frequency f0 = fc − f1 , we can say that the signal has been shifted from
the fc band to the f0 band.

8
Chapter 3

Week 3

3.1 Frequency Division Multiplexing

Multiplexing is a technique that combines multiple signals for transmission on
a common channel, but must be separable at the receiver. Basically frequency
division multiplexing (FDM) puts different message signals into differnt non-
overlapping ”frequency channels”. As an example, let there be 3 signals s1 (t) =
triangular signal (t), s2 (t) = sinc2 (t), s3 (t) = sinc(t). To multiplex them,
we cannot simply combine them through summation as the simple summation
overlaps the frequencies. Simple summation of s1 (t) + s2 (t) + s3 (t) is viable as
an FDM signal if its ’channels’ are non-overlapping in frequency. Instead, we
can modulate them by using suppressed carrier AM in different carrier signal in
both time and frequency (i.e. xi (t) = si (t) cos(2πfi t)) while ensuring that the
carrier frequencies do not overlap in the frequency domain. Hence, FDM can be
achieved through a summation of AM modulated signals with non-overlapping
fc .

3.2 FDM Demultiplexing

FDM demultiplexing can be achieved through bandpass filtering to separate the
x1 (t) + x2 (t) + x3 (t) into its respective components: x1 (t), x2 (t), x3 (t). Then
each signal goes through AM demodulation to recover sn (t) from the xn (t).
The BPF center frequency should be the respective carrier frequencies of the
signal the BPF is trying to filter for. The BPF bandwidth is the respective AM
bandwidth. The design criteria for FDM is simply

f2 − f1 > W1 + W2

f3 − f2 > W3 + W2
for the 3 aforementioned signals, and can be generalised (where fi is the carrier
frequency of xi (t) and Wi is the bandwidth of si (t)).

9
3.3 Quadrature Amplitude Modulation
QAM enables two DSBSC-AM signals to be transmitted on the same transmis-
sion bandwidth by using the same carrier frequency. This works by having a
π/2 phase difference between the carrier signals. Say 2 message signals s1 (t) and
s2 (t) are to be transmitted, then instead of having 2 difference non-overlapping
frequencies like FDM, we can have just one fc to transmit both signals. The 2
signals we’ll call them the in-phase (I) signal

x1 (t) = Ac s1 (t) cos 2πfc t

and the quadrature-phase (Q) signal

x2 (t) = Ac s2 (t) sin 2πfc t

That way the multiplexing can be done by a simple summation:

x(t) = x1 (t) + x2 (t) = Ac s1 (t) cos 2πfc t + Ac s2 (t) sin 2πfc t

3.4 QAM Demodulation

The objective of this demodulation is to obtain the s1 (t) and s2 (t) message
signals. The idea is to multiply the QAM modulated signal with cos 2πfc t
to obtain the message signal modulated in-phase, and sin 2πfc t to obtain the
message signal modulated quadrature-phase.

v1 (t) = x(t) cos 2πfc t

1 1 1
= Ac s1 (t) + Ac s1 (t) cos 4πfc t + Ac s2 (t) sin 4πfc t
2 2 2
Similarly we also have,
v2 (t) = x(t) sin 2πfc t
1 1 1
= Ac s2 (t) + Ac s2 (t) cos 4πfc t + Ac s1 (t) sin 4πfc t
2 2 2
Using appropriate LPFs, 21 Ac s1 (t) and 21 Ac s2 (t) can be obtained.

10
Chapter 4

Week 4

4.1 Noise
Noise are random in the time domain, hence we typically deal with the average
or frequency domain properties of the noise. Average power of a noise signal in
the time domain is given as:
Z T /2
1
n2 (t) = lim |n(t)|2 dt
T →∞ T −T /2

T is infinitely large here since a noise signal is not periodic. The average power
of a noise signal in the frequency domain is given as:
Z ∞
n2 (t) = Sn (f )df
−∞

Where Sn (f ) is the power spectral density (PSD) of n(t). The signal-to-noise

ratio (SNR) is defined as the ratio of the (useful/clean) signal power S to the
noise power N , higher means better:
" #
S s2 (t) s2 (t)
= = 10 log10 dB
N n2 (t) n2 (t)

If a signal x(t) with PSD Sx (f ) is filtered by a filter with transfer function H(f ),
the filtered signal y(t) will have PSD Sy (f ) given by:

Sy (f ) = |H(f )|2 Sx (f ) (4.1)

11
4.2 White Noise
An idealised form of noise typically used for noise analysis. An example of white
noise is the well known Thermal Noise. White noise has a flat PSD over a wide
range of frequencies (a constant). Hence, white noise has infinite power. A
bandpass filter can be used to filter as much white noise power away as possible,
the result is called the ‘bandpass noise’.

4.3 Bandpass Noise

To find the power of the bandpass noise n(t), we can use eq. 4.1 where Sx (f )
is the white noise PSD and Sy (f ) is the bandpass noise PSD. Filtering the
bandpass noise will reduce the power from infinite to some finite value. The
bandpass noise can be written as the following:

n(t) = nc (t) cos(2πf0 t) − ns (t) sin(2πf0 t) (4.2)

where nc (t) is the I-phase noise and ns (t) is the Q-phase noise. Proof is ommit-
ted but the power of nc (t) is equal to the power of ns (t) is equal to the power
of n(t).

4.4 Demodulation in Noisy Channel

The idea is if a message signal x(t) is polluted with white noise, passing through
a bandpass wilter will eliminate the white noise replacing it into a bandpass
noise n(t). Then the demodulator will result in αs(t) + n0 (t), where n0 (t) is the
demodulated bandpass noise. A BPF should filter away as much white noise as
possible and should not filter the AM signal x(t). Hence the BPF bandwidth
B should be equal to the AM signal bandwidth. Let the message signal s(t) be
modulated using suppressed AM modulation and demodulated using a coherent
demodulator. The input to the demodulator will be s(t) cos 2πfc t + n(t) where
n(t) is the bandpass noise. The output demodulated signal will be
1
[s(t) cos2 (2πfc t)]LPF = [s(t) (1 + cos 4πfc t)]LPF
2
= s(t)/2
after being multiplied with the local oscillator’s signal and passed through a
LPF. Hence the demodulated signal power [s(t)/2]2 = s(t)2 /4. As for the
bandpass noise component of the demodulator input, the output noise is given
as
n0 (t) = [n(t) cos 2πfc t]LPF

12
We write n(t) as eq. 4.2,

n0 (t) = [nc (t) cos2 2πfc t − ns (t) sin 2πfc t cos 2πfc t]LPF

= [nc (t)(1/2)(1 + cos 4πfc t) − ns (t)(1/2) sin 4πfc t]LPF

= nc (t)/2
The output noise power will be [nc (t)/2]2 = nc (t)2 /4 = ηB/4 which can be
found from white noise PSD (η/2) and AM bandwidth B. Hence, the generic
formula for the output SNR of a suppressed carrier AM signal demodulated by
a coherent demodulator is

= signal power/noise power

= s(t)2 /ηB

13
Chapter 5

Week 5

Phase Modulation (PM) and Frequency Modulation (FM), the phase and fre-
quency of the carrier signal is varied acording to the message signal while the
amplitude is maintained constant. The advantage is that it has better noise and
interference suppression compared to AM. The disadvantage is that it occupies
a wider transmission bandwidth compared to the AM. Recall the relationship
between angular frequency w and phase θ of a sinusoidal signal:

w(t) = 2πf

cos(2πf t) ↔ cos(wt) ↔ cos(θ)

dθ(t)
w(t) =
dt
Z t Z t
θ(t) = w(τ )dτ = 2π f (τ )dτ (5.1)
−∞ −∞

5.1 Phase Modulation

In phase modulation, the instantaneous phase is varied linearly with the message
signal m(t), i.e.
ϕ(t) = ϕ0 + kp m(t)
where ϕ0 is the initial phase with default value 0, kp is the phase sensitivity,
and kp m(t) is the instantaneous phase deviation ∆ϕ(t). So if we have a carrier
waveform
c(t) = Ac cos[2πfc t + ϕ] (5.2)
the PM signal will be

fP M (t) = Ac cos[2πfc t + kp m(t)]

14
5.2 Frequency Modulation
In frequency modulation, the instantaneous frequency fi (t) is varied linearly
with the message signal m(t), i.e.
fi (t) = fc + kf m(t)
The FM signal is written as
fF M (t) = Ac cos[2π(fc + kf m(t))t + 0]
Using eq. 5.1,
Z t  Z t 
ϕi (t) = 2π fi (τ )dτ = 2π fc t + kf m(τ )dτ
−∞ −∞

Then rearranging the FM signal equation, we see that

Z t
fF M (t) = Ac cos[2πfc t + 2πkf m(τ )dτ ]
−∞

With a varying phase and a constant carrier frequency, this may look like a PM
signal. But note that the difference in phase in the FM signal is not directly
proportional to message signal m(t), as it is proportional to the integral of m(t).

5.3 Single Tone Modulation

Consider a single frequency (single tone) message signal, m(t) = Am cos 2πfm t.

Phase Modulation
The phase deviation is
∆ϕ(t) = kp m(t) = kp Am cos 2πfm t = βp cos 2πfm t
where βp = kp Am is the peak phase deviation (modulation index for PM).
The PM signal is written as:
fP M (t) = Ac cos[2πfc t + βp cos 2πfm t]

Frequency Modulation
The frequency deviation is
∆f (t) = kf m(t) = kf Am cos 2πfm t = ∆f cos 2πfm t
Where ∆f = kf Am is the peak frequency deviation. We can find the phase
function of the FM wave using the following,
Z t
∆f
∆ϕ(t) = 2π ∆f (τ )dτ = sin 2πfm t = β sin 2πfm t
0 fm

15
 k A
where β = ffmm is the peak phase deviation, the modulation index of FM.
The FM signal can be written as

fF M (t) = Ac cos[2πfc t + β sin 2πfm t] (5.3)

Note: Small β creates narrow-band FM resulting in less noise suppression. A

large β creates a wide-band FM which results in better noise suppression.

5.4 Narrow-Band FM
Narrow-band FM or PM signal means that β ≤ 0.2, it is the same β in eq. 5.3.
Using the cos(a ± b) trigonometric identity, we can manipulate the eq. 5.3. into
the following

fF M (t) = Ac [cos wc t cos(β sin wm t) − sin wc t sin(β sin wm t)] (5.4)

Since β ≤ 0.2, we can approximate β sin wm t ≪ 1, and using small angle ap-
proximation for trigonometric functions we get,

cos(β sin wm t) ≈ 1, sin(β sin wm t) ≈ β sin wm t

and so approximating the narrow band pass FM signal gives us

fN BF M (t) ≈ Ac [cos wc t − β sin wm t sin wc t] (5.5)

expanding it using the sin(x) sin(y) identity gives us

βAc βAc
= Ac cos wc t + cos(wc + wm )t − cos(wc − wm )t (5.6)
2 2
The math above shows that the bandwidth of the NBFM signal is 2fm centered
at fc , same as the AM bandwidth.

5.5 Wide-Band FM
If β is not less than 0.2 rad, we can express the wide band FM spectrum as
∞
X
fF M (t) = Ac Jn (β) cos[(wc + nwm )t] (5.7)
n=−∞

(proof ommitted). The Jn is called the Bessel function where the function takes
in a parameter β and is different for different values of n. Note that the Bessel
function form of fF M (t) is valid for any value of β meaning that it works for
both WBFM and NBFM. It is also useful for plotting the FM spectrum.

16
In the exam a table will be provided to obtain the values of the Bessel function.
Only non-negative n will be provided, hence the following useful properties

Jn (β) = J−n (β) for even n

Jn (β) = −J−n (β) for odd n

∞
X
Jn2 (β) = 1 (5.8)
n=−∞

Also note that the power of the signal defined in e.q. 5.7 (simplified using e.q.
5.8) can be written as

A2c X 2 A2
PF M = Jn (β) = c
2 n 2

17
Chapter 6

Week 6

6.1 Bandwidth of FM signals

An FM signal has an infinite number of sidebands, hence the BW of and FM
signal is theoretically ∞. In practice, a large portion of the FM power is in the
’significant’ sidebands which lie within some finite BW, and the ’insignificant’
portion can be discarded without serious distortion of the FM signal. Common
approximation rules of the FM transmission BW is the 1% rule, or Carson’s
rule.

1% rule
For a single tone FM/PM, if the FM/PM signal has n′ significant sideband
frequency pairs with |Jn′ (β)| ≥ 0.01, its 1% rule BW is 2n′ fm . Basically means
you only take the ones where |Jn′ (β)| ≥ 0.01.

Carson’s Rule
A very convenient way to approximate the BW without the use of a Bassel
function table is the following formula,

BW = 2(β + 1)fm = 2(∆f + fm )

Here for β ≪ 1, BW ≈ 2fm which agrees with the narrow-band FM analysis,

and for β ≫ 1, BW ≈ 2βfm .

18
6.2 Generation of FM signals
Indirect Method (Armstrong Method)
The message signal m(t) is first used to generate a NBFM signal with β ≤ 0.2 to
minimise distortion. The signal then goes through a frequency multipler to
widen the BW, then using a frequency converter (mixer) to get the carrier
frequency right. Note, the effect of a frequency multiplier is multiplying both
carrier frequency fc and ∆f by n times. The frequency converter (mixer) is the
same mixer explained in the AM portion (see 2.3).

19
Chapter 7

Week 7

7.1 Demodulation of FM Signals

A FM demodulator should produce an output signal whose amplitude is lin-
early proportional to the instantaneous frequency fi (t) of the FM signal, since
the fi (t) is proportional to the message signal m(t). There are two common
approaches, the Frequency Discriminator (direct approach) – A device that
has a linear frequency to voltage transfer function, containing a Time Dif-
ferentiator – and the Phase-Lock Loop (indirect approach) – A frequency
modulator in the return branch of a feedback system.

Time Differentiator
Rt
Differentiating fF M (t) = Ac cos [wc t + 2πkf 0
m(τ )dτ ], we get

d
f˙F M (t) = fF M (t)
dt
Z t
= −Ac [wc + 2πkf m(t)] sin [wc t + 2πkf m(τ )dτ ]
0
in which its result may look like an ‘FM signal with an AM amplitude’. Its
envelope (amplitude variations) can be extracted by an ’envelope detector’. We
assume that the +ve envelope is extracted. Then a DC blocking can be applied
to remove the DC signal of the envelope, resulting in just 2πAc kf m(t). The
frequency discriminator process is summed up below
d
Modulated FM → → Envelope Detector → DC Block → Demodulated FM
dt

20
Limiter
We have assumed that Ac is constant throughout, but in practice due to channel
noise and signal fading, Ac may be time varying in practice. Then the differ-
d
entiation result will have terms containing dt Ac (t) term and the amplitude will
d
have a Ac (t) term instead of just Ac . Even if the term containing dt Ac (t) is
negligible, the output of the envelope detector will be non-constant, making it
difficule to remove just be using DC Blocking. A limiter can be used before
the frequency discriminator to remove carrier amplitude that varies with time
(i.e., the signal amplitude at the limiter’s output is constant).

FM Receiver for Noisy Channels

The following receiver structure is used to demodulate the FM signal in the
presence of white noise with PSD η/2. Note that the frequency discriminator
here contains a differentiator, envelop detector and DC block. Let the noisy
signal x(t) be an FM signal fF M (t) combined with white noise w(t).

x(t) → BPF → Limiter → Freq. Discriminator → LPF → Output

The BPF is an ideal bandpass filter with center frequency fc and bandwidth
equal to the FM signal’s bandwidth, removing the out-of-band noise. The LPF
is used to further supress noise, by choosing a LPF BW to the message signal’s
BW.

SNR of FM Demodulation
The input SNR before FM demodulation is

Si mean power of FM signal A2 /2

= = c
Ni mean power of noise within FM signal’s BW ηBF M

The output SNR after FM demodulation is (assuming that Ac (t) = 1 after

limiter output)

So mean power of output message signal 4π 2 kf2 m2 (t)

= =
No mean power of noise No

No Calculation
The noise at the FM demodulator output (omitting calculations) is

dρ(t) 1 d
no (t) = = ns (t)
dt Ac dt

21
The PSD of no (t) is
1
Sno (w) = Sn (w)|H(w)|2
A2c s
d
The Fourier Transform of transfer function h(t) = dt is H(w) = jw.

1 1
Sno (w) = Sn (w)|jw|2 = 2 w2 Sns (w)
A2c s Ac

We know from earlier chapters that Sns (w) = η for |f | ≤ BF M /2,

1 2 (2π)2 2
Sno (w) = w η = f η ∀ |f | ≤ BF M /2
A2c A2c

the above shows that if a white noise with a two-sided PSD η/2 is present at
the demodulator input, the output noise PSD will be parabolic. Now since the
BW of the frequency discriminator output is limited by a low pass filter with
bandwidth fm , the power of the output noise is
Z fm
8ηπ 2 fm 2
Z
2
No = no (t) = 2 Sno (f )df = f df
0 A2c 0

8π 2 ηfm3
No = 2
(7.1)
3Ac
For single-tone modulation,

m(t) = Am cos(wm t), ∆f = Am kf

m2 (t) = A2m /2, β = ∆f /fm

Hence,
So 3A2c β 2
= (7.2)
No 4ηfm
By observation, this tells us that the BW increases linearly with β, and so the
output SNR increases with (BW )2 .

SNR of FM v. AM
For DSBSC-AM with s(t) = Ac cos(wm t) and carrier amplitude Ac ,

A2c
 
So
=
No AM 4ηBAM

Meaning that,    
So 2 So
= 3β
No FM No AM
This means that noise suppression in FM can be achieved over AM by tweaking
the β.

22
Threshold Phenomenon (Capture Effect)
Si
At high input SNR (i.e. N i
> 10dB), FM outperforms AM. However at low
Si
Ni (< 10dB), FM demodulation does not work and FM becomes worse than
AM. This is called the Capture Effect or the Threshold Phenomenon.
Si
Ni = 10dB is called the FM threshold.

7.2 Summary (What to Memorise?)

PM Signal,
fP M (t) = Ac cos[2πfc t + kp m(t)] (7.3)
FM Signal, Z t
fF M (t) = Ac cos[2πfc t + 2πkf m(τ )dτ ] (7.4)
−∞
Rt
kp m(t) and 2πkf −∞ m(τ )dτ → instantaneous phase deviation ∆ϕ(t).
Max value of ∆ϕ(t) is denoted as ∆ϕ or β.
FM Signal,
fF M (t) = Ac cos[2π(fc + kf m(t))t] (7.5)
PM Signal,
kv dm(t)
fP M (t) = Ac cos[2π(fc + )t] (7.6)
2π dt
kv dm(t)
kf m(t) and 2π dt → instantaneous freq deviation.
For a single-tone mesage signal m(t) = Am cos(2πfm t):
PM Signal,

fP M (t) = Ac cos[2πfc t + βp cos(2πfm t)], βp = kp Am (7.7)

FM Signal,
∆f k f Am
fF M (t) = Ac cos[2πfc t + β sin(2πfm t)], β = = (7.8)
fm fm

NBFM: β < 0.2, WBFM: β ≫ 1. To get the FM/PM spectrum, use:

∞
X
Ac cos[2πfc t + β sin(2πfm t)] = Ac Jn (β) cos[2π(fc + nfm )t] (7.9)
n=−∞

23
where J(β) is the Bessel function with the following properties:

Jn (β) = −J−n (β), if n is odd

∞
X
Jn2 (β) = 1
n=−∞

With the Bessel function, we can get the power of the signal in the form of:

A2c X 2
J (β) (7.10)
2 n n

BW Approximation
For a single-tone message signal m(t),
• Carson’s rule: BW = 2(β + 1)fm = 2(∆f + fm )
• 1% rule: BW includes sidebands with Jn (β) ≥ 0.01

WBFM Modulation
• Frequency multiplier: fc and ∆f both multiplied by n
• Frequency converter (mixer): fc → fc ± fshift , ∆f unchanged

FM Demodulation
• Roles of differentiator, envelope detector, DC block
• When input SNR < threshold value (10dB), capture effect occurs, FM
demodulator fails to work.

24
Chapter 8

Week 8

8.1 Review of Fourier Series and Transforms

Signals are vectors, as they have amplitudes and directions. A vector can be
decomposed into several orthogonal components called the basis (e.g. your
standard i, j, and k in the 3-dimensional space). A signal f (t) can also be
decomposed into orthogonal components with sin(2πf t) and cos(2πf t) as the
basis.

Pythagorean Relation and Parseval’s Relation

Pythagorean relation: |r|2 = a2 + b2 where r = ai + bj.
Fourier series: X n
x(t) = xn ej2π T0 t (8.1)
n

A generalization of Pythagorean relation for n-dimension vectors (because the

Fourier series is essentially an n-dimension vector), Parseval’s relation can be
stated as: X
|x(t)|2 = Power of x(t) = |xn |2 (8.2)
n

Properties of Delta Function

For any function f (t) continuous at t0 , we have
Z b (
f (t0 ) a < t0 < b
f (t)δ(t − t0 )dt =
a 0 otherwise

Convolution operation:
f (t) ∗ δ(t) = f (t)
f (t) ∗ δ(t − t0 ) = f (t − t0 )

25
Integral of delta function:
Z t
δ(τ )dτ = u(t) = unit step function
−∞
Z −∞
δ(τ )dτ = 1
−∞

Sampling Theorem
A continuous-time signal v(t) band limited to B Hz can be reconstructed exactly
if sampled at fs ≥ 2B Hz. Note that band limited is not the same as the
bandwidth of a signal. If a signal is band limited by B Hz, that means the
highest frequency covered by this signal is B Hz. Let the sampling period
Ts = 1/fs and denote the periodic impulse train δTs (t) as:
X
δTs (t) = δ(t − nTs )
n

From the table of FT pairs, the FT of δTs (t) is:

1 X
F {δTs (t)} = δ(f − nfs )
Ts n

The sample output signal vs (t) is given by:

X
vs (t) = v(t)δTs (t) = v(t) × δ(t − nTs )
n

The Fourier transform of the sampled output is then given by:

" #
1 X
Vs (f ) = V (f ) ∗ δ(f − nfs )
Ts n

1 X
= V (f − nfs )
Ts n
Note that the ideal sampling Vs (f ) is a repetition of V (f ) in the frequency
domain with a spacing of fs .
• If fs ≥ 2B we can design a LPF to recover V (f ) from Vs (f )
• The minimum sampling rate of fs = 2B is called the Nyquist rate

8.2 Linear Pulse Coded Modulation (PCM)

The conversion of analogue signals to digital signals can be carried out by Lienar
Pulse Coded Modulation (PCM) and Delta Modulation (DM). The steps in
PCM includes Sampling, Quantising, and Encoding.

26
Quantisation
The idea of quantisation is to map analogue amplitudes of a signal to a fixed
set of values, ’digitalising’ the amplitude in a sense.
• Let a message signal m(t) be sampled at fs = 1/Ts Hz and mp be the peak
amplitude of the message signal.
• The whole amplitude range (−mp , mp ) divided into L uniform intervals
where L is a power of 2 with each step size being ∆v = 2mp /L.
• A sample amplitude value is approximated by mid-point of the interval in
which it falls.
• The quantised samples are encoded into binary bits, the two sources of
errors are quantisation error and bit detection error.
• Quantisation error is normally much larger than bit detection error.

27
Chapter 9

Week 9

9.1 Review of Expected Value

Average of a discrete random variable X which takes on the possible values
X0 , X1 , ..., Xn−1 with the respective probabilities p(X0 ), p(X1 ), ..., p(Xn−1 ) is
n−1
X
E[X] = X = Xi p(Xi )
i=0

Now if the average of a function of a random variable X, f (X) is

n−1
X
E[f (X)] = f (X) = f (Xi )p(Xi )
i=0

9.2 Average Quantisation Error

The quantisation error q lies in the range of (−mp /L, mp /L) with equal prob-
ability. The probability distribution p(q) forms a rectangular function in the
range above, where
2mp L
x = 1 −→ x =
L 2mp
if x is the height of the rect function. Now to find the average quantisation error
where p(q) is the probability density function of q,
Z ∞
Nq = q =2 q 2 × p(q)dq
−∞

mp /L m2p
Z
L
= q2 × dq =
−mp /L 2mp 3L2

28
Here we find q 2 since finding q will just result in 0 as the -ve and +ve values
cancel out. The signal-to-quantisation-noise ratio (SQNR) is given by

m2 (t) m2 (t)
SQNR = = 3L2 ×
Nq m2p

9.3 Nonlinear PCM

The idea is that for signals where smaller amplitude predominates (e.g. voice
signals), it would be beneficial to have a nonlinear distribution of step sizes.
So have smaller step sizes for small amplitudes since they occur more often,
while having a larger step size for large amplitudes as they occur rarely. This is
achieved in practice by compressing the input signal and then quantising them,
so we ’vertically squeeze’ the larger input values and then set the uniform steps
to quantise.

Compression
Compression is achieved by translating an input signal’s amplitudes into its
compressed form by passing it through a compressing function. There are 2
types of standard of compression signals, i.e., µ-law, A-law.

Expandor
The reverse process of compression operation to recover the original value from
the compressed sample. The output SQNR for the µ - law standard is given by
(proof omitted):
3L2 m2p
SQNR = , µ ≫
ln(1 + µ)2 m2 (t)
Comparison of high (µ = 255) and low (µ = 0) values show that higher (nonlin-
ear) mu is better for lower signal power, but lower µ is more suitable of higher
signal power.

Output SQNR for PCM

To summarize for both the linear and non-linear case:

k = 3×m2 (t) , linear
2 m2p
SQNR = k × L
k = 3
2, nonlinear
ln(1+µ)

= k × 22n , since L = 2n
= k × 22BT /B , since minimum bandwidth BT = nB
= 10 log10 k + 6n (dB)

29
Note that this is only if the signal is sampled using the Nyquist frequency. The
minimum transmission bandwidth BT to transmit a signal of bandwidth B is
nB since:
BT = Rb /2 = nfs /2 = nfN /2 = 2nB/2 = nB
where fN is the Nyquist frequency.

9.4 Delta Modulation

One way to reduce quantisation noise is to redice the peak amplitude of the
transmitted signal mp if the maximum quantisation noise is mp /L and L is
fixed. Instead of transmitting m[k], we can transmit d[k] = m[k]−m[k−1] where
d[k] ≪ m[k] and can be reconstructed at the receiver as m[k] = d[k] + m[k − 1]
can be reconstructed since we have m[k−1]. To improve further, we can transmit
d[k] = m[k] − m̂[k] where m̂[k] is the prediction or estimation of m[k] based on
previous samples m[k − 1], m[k − 2], .... To estimate m[k] we can use time-series
method:
m̂[k] = a1 m[k − 1] + a2 m[k − 2] + ...
and if there is a large correlation between m[k], m[k − 1], then m̂[k] will be
close to m[k]. However note that the receiver does not actually store m but
mq , the quantised values. Hence instead of m̂[k], we estimate m̂q [k] based
on mq [k − 1], mq [k − 2], .... Delta Modulation improves prediction of m̂q [k]
by oversampling (usually 4 times the Nyquist rate) the baseband signal to in-
crease correlation between adjacent samples. And since adjacent values are
close, d[k] = m[k] − m̂q [k] can be represented by one bit, 1 when +ve and 0
when -ve.

Threshold of coding
Variation in m(t) smaller than the step size is lost in delta modulation, this will
cause quantisation noise. To avoid this, we want to make the step size as small
as possible.

Slope overload
m(t) changes too fast such that m̂(t) cannot follow m(t), this will cause slope
overload noise. But to avoid this, we want to make the step size σ large enough to
capture m(t)’s movement. The optimum σ depends on the sampling frequency
fs and the nature of the signal. During the sampling interval Ts = 1/fs , m̂(t) is
capable of changing height of step size σ, hence the maximum slope that m̂(t)
can follow is σfs . So slope overload can be avoided if

dm(t)
< σfs
dt max

30