ELE635 Communication Systems

Chapter 3 : Angle (Exponential) Modulation

Part II

1
Section 3.5: FM Bandwidth Estimation with Arbitrary m(t)

2
 Bandwidth of FM signal

So far all what we have achieved was to determine the spectrum of a single-tone
modulated signal, m(t) = Am cos(2πfm t). As we stated at the onset “computing the
spectrum for the general case is not an easy task!”

Let us consider a more general case with an arbitrary m(t) with

∫ t
a(t) = m(λ)dλ
0

and express the FM signal in a form using the simplified notation:

( ∫ t )
φF M (t) = Ac cos ωc t + kf m(λ)dλ
0
= Ac cos (ωc t + kf a(t)) (1)

Let’s apply power series:

x2 nx
n
e jx
= 1 + jx − + ··· + j + ···
2! n!

3
We now use power series to conduct signal bandwidth analysis. The FM waveform
(1) can be expressed as
{ }
j[ωc t+kf a(t)]
φF M (t) = Re Ac e
{ }
j(ωc t) jkf a(t)
= Re Ac e e
{ [ ]}
2
kf 2
= Ac · Re (cos(ωc t) + j sin(ωc t)) · 1 + jkf a(t) − a (t) + · · ·
2!
{ [ ]}
2
kf 2
= Ac · Re cos(ωc t) · 1 + jkf a(t) − a (t) + · · · (2)
2!
{ [ ]}
2
kf 2
+Ac · Re j sin(ωc t) · 1 + jkf a(t) − a (t) + · · · (3)
2!

Discard those imaginary terms we have,

[ ]
kf2 kf3
φF M (t) = Ac cos ωc t − kf a(t) sin ωc t − a2 (t) cos ωc t + a3 (t) sin ωc t + · · · (4)
2! 3!

4
Note that if m(t) is bandlimited to B (equivalently used as Bm ) Hz,
• a(t) is bandlimited to B Hz, since integration is a linear operation;
• a2 (t) is bandlimited to 2B Hz;
• an (t) is bandlimited to nB Hz.
Then the bandwidth for each term in Eqn. (4) will be illustrated as:

φF M (t) is not band-limited. It has an infinite bandwidth. However, most of the

modulated signal power resides in a finite bandwidth. Higher order terms in (4)
decays fast.

5
• Narrow-band Angle Modulation (NBFM): when kf Am is very small (that is,
|kf a(t)| << 1), then all but the first two terms in (4) are negligible, and we have
φF M (t) ≈ Ac [cos ωc t − kf a(t) sin ωc t] (5)
The bandwidth of φF M (t) is 2Bm .
• NBPM case is similarly given by
φP M (t) ≈ Ac [cos ωc t − kp m(t) sin ωc t] (6)

6
 Bandwidth Estimation for Wideband FM

If |kf a(t)| ≪ 1 is not satisfied, we cannot ignore higher order terms in Eqn. (4), and
the preceding analysis becomes too complicated to lead to a fruitful solution.
We shall use following intuitively simply reasoning to estimate FM bandwidth. We
Carson Rule
will obtain the well-known to calculate the bandwidth for generalized
FM.
Let m(t) be band-limited to Bm Hz, and its peak value,

mp = max[m(t)] = − min[m(t)]

1
We use following staircase to approximate m(t). Each cell has a width 2Bm
(Nyquisite rate: fsample = 2Bm ).

7
 x(t)

τ τ
−
2 2
! "
t
x(t) = rect
τ

8
9
 kf
fk+1 = fc + mp
2π

10
11
 kf
fk+8 = fc − mp
2π

12
 ∆f ∆f
fmax

fmin

13
• Significant frequency component range (mp = maxt [m(t)]):
kf
fmin = fc − mp − 2Bm = fc − ∆f − 2Bm (7)
2π
kf
fmax = fc + mp + 2Bm = fc + ∆f + 2Bm (8)
2π
BF M ≈ fmax − fmin = 2(∆f + 2Bm ) (9)
kf m p
where ∆f = 2π is the maximum frequency deviation.
• The above bandwidth estimate is somewhat higher than the actual value, since it
is obtained based on the staircase approximation of m(t), not the actual m(t),
which is considerably smoother. Hence, the actual bandwidth should smaller
than this value.
• A better estimation is between

[2∆f, 2∆f + 4Bm ]

Therefore, we should adjust our bandwidth estimation.

14
 Carson’s Rule

• Our earlier result shows for narrowband FM, BF M ≈ 2Bm . Therefore, a better
bandwidth estimation is
( )
kf mp
BF M = 2 (∆f + Bm ) = 2 + Bm (10)
2π
This is precisely the result obtained by Carson who investigated this problem
rigorously for tone modulation. This formula goes under his name as “Carson’s
rule” in the literature.
∆f
• Define β = B m
as deviation ratio, which controls the amount of modulation. For
tone-modulated FM, β is called the modulation index (β = ∆f /fm ). From Eqn.
(10), we have
BF M = 2Bm (β + 1) (11)

• Note that Eqn. (11) is consistent with our spectrum analysis for Tone modulated
FM in Section 3.4.

15
 Example 3.5-1: Using Carson’s Rule for BW Estimation

(Follow Example 3.2-2) A modulating signal m(t), is shown below. Let kf = 2π × 105
and kp = 5π.

2 × 10−4
m(t)
1 t
-1

(a) Estimate BF M and BP M ;

(b) Repeat the problem if the amplitude of m(t) is doubled.

Solution: (a) The peak amplitude of m(t) is unity. Hence, mp =1. We now determine
the essential bandwidth Bm of m(t).

16
We can use Fourier Series analysis: let m2 (t) = m(t)+1
2 . Then m2 (t) has the same
spectrum shape as that of m(t) except at f = 0. The period of m2 (t) is T0 with
fundamental frequency
1 1
f0 = = = 5 kHz
T0 2 × 10−4
The Fourier Transform of a full cycle of m2 (t) is:
( ) ( )
t T0 πf T 0
∆ → X(f ) = sinc2
T0 2 2
Then its Fourier series is
( ) ( )
1 1 T0 π · nf0 · T0 1 2 nπ
Dn = X(nf0 ) = · sinc2 = sinc (12)
T0 T0 2 2 2 2
therefore m2 (t) can be written as
∞
∑
m(t) = C0 + Cn cos(2π · nf0 · t), (13)
n=1

17
where 

 0 n=0

Cn = 2Dn = 4
n odd (14)


π 2 n2

0 n even

Since Cn decreases rapidly with n. The third harmonic is only 11% of the
fundamental, and the fifth is only 4% of the fundamental (powers are 1.21% and
0.16% respectively of the fundamental component power), estimated as below:
C3 1 Pn=3 C32 /2
= = 0.11 = 2 = (0.11)2 = 1.21%
C1 9 Pn=1 C1 /2
We can safely assume the essential bandwidth of m(t) as the frequency of the third
harmonic, thus
Bm = 3 × 5 kHz = 15 kHz

18
FM

• Method I
1 1
∆f = kf mp = (2π × 105 )(1) = 100 kHz −→ BF M = 2(∆f + Bm ) = 230 kHz
2π 2π
• Method II
( )
∆f 100 100
β= = −→ BF M = 2Bm (β + 1) = 2 · 15 +1 = 230 kHz
Bm 15 15

PM
2
m′p = ± = ±20, 000 = ±20 kHz
1 × 10−4
• Method I
1 5π
∆f = kp m′p = ·20 = 50 kHz −→ BP M = 2(∆f +Bm ) = 2(50+15) = 130 kHz
2π 2π
• Method II
( )
∆f 50 50
β= = −→ BP M = 2Bm (β + 1) = 2 · 15 +1 = 130 kHz
B 15 15

19
(b) Doubling m(t) doubles its peak value. The spectrum has no change therefore
keeping the same Bm .
4
Bm = 15 kHz, mp = 2, m′p = ± = ±40 kHz
1 × 10−4

FM

• Method I
1 1
∆f = kf mp = (2π × 105 )(2) = 200 kHz −→ BF M = 2(∆f + Bm ) = 430 kHz
2π 2π
• Method II
( )
∆f 200 200
β= = −→ BF M = 2Bm (β + 1) = 2 · 15 +1 = 430 kHz
B 15 15

20
PM

• Method I
1 5π
∆f = kp m′p = ·40 = 100 kHz −→ BP M = 2(∆f +Bm ) = 2(100+15) = 230 kHz
2π 2π
• Method II
( )
∆f 100 100
β= = −→ BP M = 2Bm (β + 1) = 30 +1 = 230 kHz
Bm 15 15

21
Example 3.5-2: An angle-modulated signal is described by the equation
φEM (t) = 10 cos(θ(t)) = 10 cos(2πfc t + 5 sin 3, 000t + 10 sin 2, 000πt)
1. Find the power of the modulated signal: P = 102 /2 = 50.
2. Find the frequency deviation ∆f
d
ωi (t) = θ(t) = ωc + 15, 000 cos 3, 000t + 20, 000π cos 2, 000πt
dt
The maximal value of frequency deviation is ∆ω = 15, 000 + 20, 000π. hence,
∆ω
∆f = = 12, 387.32 Hz
2π
3. Find the deviation ratio β: since the signal BW is the highest frequency in m(t),
we have Bm = 2000π
2π = 1, 000 Hz. Therefore,

∆f 12, 387.32
β= = = 12.387
Bm 1, 000
4. Find the phase deviation ∆ϕ: The angle deviation is (5 sin 3000t + 10 sin 2000πt).
∆ϕ = 15 rad.
5. Estimate the bandwidth of φEM (t): BEM = 2(∆f + Bm ) = 26, 774.65 Hz.

22
NBFM Summary

23
WBFM Summary: Using Carson’s Rule

24
 Section 3.6: Feature of FM – Immunity to Nonlinearities

Nonlinearity for FM
Suppose a second-order nonlinear device with input-output
relation:
y(t) = b1 x(t) + b2 x2 (t) (15)
∫
Let a(t) = kf m(α)dα, and x(t) = cos[ωc t + a(t)]. Then x(t) is a FM signal. Let it
pass through a nonlinear system described in Eqn. (15), the output is

y(t) = b1 cos[ωc t + a(t)] + b2 cos2 [ωc t + a(t)] (16)

b2 b2
= + b1 cos[ωc t + a(t)] + cos[2ωc t + 2a(t)] (17)
2 2
[ ∫ ] [ ∫ ]
b2 b2
= + b1 cos ωc t + kf m(α)dα + cos 2ωc t + 2kf m(α)dα (18)
2 2
The DC term can be filtered out. The output contains the original FM signal plus an
additional FM signal, whose carrier and frequency deviation are multiplied by 2. The
information m(t) is intact in both terms. Thus, the nonlinearity has not distorted
the information in any way.

25
Suppose an nth-order nonlinear device,

y(t) = b0 + b1 x(t) + b2 x2 (t) + b3 x3 (t) + · · · + bn xn (t) (19)

∫
If the input is an FM signal x(t) = cos[ωc t + kf m(α)dα], the output y(t) is given as
[ ∫ ] [ ∫ ]
y(t) = c0 + c1 cos ωc t + kf m(α)dα + c2 cos 2ωc t + 2kf m(α)dα
[ ∫ ]
+ · · · + cn cos nωc t + nkf m(α)dα (20)

Hence, the output will have spectra at fc , 2fc , · · · , nfc , with frequency deviation
∆f, 2∆f, · · · , n∆f , respectively.
∫
The desired signal x(t) = cos[ωc t + kf m(α)dα], is intact in y(t).

26
Nonlinearlty for AM
Let us see how the nonlinearlty distorts the AM signal. Let a
DSB-SC signal
x(t) = m(t) cos(ωc t)
goes through a nonlinear device with input-output relation as

y(t) = b1 x(t) + b3 x3 (t)

then

y(t) = b1 · m(t) cos ωc t + b3 · m3 (t) cos3 ωc t (21)

Let’s simplify cos3 α as below

( )
1 1 1 1
cos3 α = cos α · cos2 α = cos α + cos 2α = cos α + cos α cos 2α
2 2 2 2
1 1 3 1
= cos α + (cos α + cos 3α) = cos α + cos 3α
2 4 4 4

27
Therefore, Eqn. (21) can be written as
( )
3 1
y(t) = b1 · m(t) cos ωc t + b3 · m3 (t) cos ωc t + cos 3ωc t
4 4
[ ]
3b3 3 b3
= b1 · m(t) + m (t) cos ωc t + m3 (t) cos 3ωc t (22)
4 4
In order to suppress out of band component, passing y(t) in (22) through a BPF
centred at ωc yields
3b3 3
y ′ (t) = b1 · m(t) cos ωc t + m (t) cos ωc t (23)
4
Obviously, the second term distorts the desired signal m(t) cos ωc t.

28
 Section 3.7: Generation of NBFM and WBFM Waves

Generation of NBFM
( ∫ t )
φF M (t) = Ac cos ωc t + kf m(λ)dλ = Ac cos [ωc t + kf a(t)]
0
= Ac cos(ωc t) cos[(kf a(t)] − Ac sin[kf a(t)] sin(ωc t) (24)
when kf is small,
φF M (t) = Ac cos(ωc t) − Ac kf a(t) sin(ωc t) (25)

29
 3.7-2 Wideband FM Generation

In order to generate WBFM:

m(t) → φN BF M (t) → φW BF M (t)

Frequency Multiplier
.
A frequency multiplier can be realized by a nonlinear device followed by a bandpass
filter. Let x(t) denote an FM signal as
[ ∫ t ]
x(t) = Ac cos ωc t + kf m(α)dα (26)
−∞

Let x(t) goes through a nonlinear device whose input-output relation is

y(t) = bx2 (t)

30
Then y(t) will be derived as
[ ∫ t ]
y(t) = bA2c cos2 ωc t + kf m(α)dα
−∞
[ ∫ t ]
A2c A2c
= b +b cos 2ωc t + 2kf m(α)dα
2 2 −∞

Thus, a bandpass filter centered at 2ωc would recover an FM signal with twice the
carrier frequency (2ωc ) and twice the frequency deviation (2∆f ).

[ ∫t ]
To generalize, Let x(t) = Ac cos ωc t + kf 0
m(α)dα goes through following
nonlinear system,

y(t) = b0 + b1 x(t) + b2 x2 (t) + · · · + bn xn (t) (27)

31
As discussed in Section 3.6, y(t) is of the form

y(t) = c0 +
[ ∫ t ]
c1 cos ωc t + kf m(α)dα +
0
[ ∫ t ]
c2 cos 2ωc t + 2kf m(α)dα + · · · +
0
[ ∫ t ]
cn cos nωc t + nkf m(α)dα
0

Hence, the output will have spectra at ωc , 2ωc , · · · , nωc , with frequency deviations
∆f, 2∆f, · · · , n∆f , respectively. Each one of these components is an FM signal
separated from the others. Thus, a bandpass filter centering at nωc can recover an
FM signal whose instantaneous frequency has been multiplied by a factor of n, and
∆f is also multiplied by a factor of n. These devices, consisting of nonlinearity and
bandpass filters, are known as frequency multipliers.

32
Frequency Multiplier Summary

• The carrier frequency and ∆f are increased by a factor of N .

∆f
• Since β = Bm , β is increased by a factor of N .
• This forms the basis for φN BF M (t) to φW BF M (t).
• Carrier frequency is also increased by a factor of n. This may not be needed.
Then we can apply frequency mixing to shift down the carrier frequency to the
desired value.

33
Indirect Method of Armstrong

34
Check if NBFM condition is satisfied (β < 0.3)

As m(t) is bandlimited to 100 Hz to 15 kHz, we can assume that Bm = 15 kHz.

Then the modulation/deviation index β is:
∆fN B 20
β= =
15000 15000
NBFM condition is indeed satisfied.

35
36
37
 Example 2

Design an Armstrong indirect FM modulator to generate an FM signal with carrier

frequency 97.3 MHz and ∆f4 = 10.24 kHz. A NBFM generator of fc1 = 20 kHz and
∆f1 = 5 Hz is available. Only frequency doubler can be used as multipliers.
Additionally, a local oscillator (LO) with adjustable frequency between 400 and 500
kHz is readily available for frequency mixing.

∆f1 = 5 Hz
∆f4 = 10.24 kHz
fc1 = 20 kHz fc2 fc3
fc4 = 97.3 MHz
Frequency Frequency
m(t) NBFM Multiplier Multiplier ϕFM (t)
Mixer
Modulator X N1 X N2

fo = 400 − 500kHz
~ ~

Solution: the NBFM generator generates:

fc1 = 20, 000 and ∆f1 = 5

38
The final WBFM should have

fc4 = 97.3 × 106 and ∆f4 = 10, 240

The total factor of frequency mulplication needed as

∆f4
N1 · N2 = = 2048 = 211
∆f1
Because only frequency doublers can be used, we have three equations as:

N 1 = 2 n1 N 2 = 2 n2 n1 + n2 = 11

It is also clear that

fc2 = 2n1 fc1 , fc4 = 2n2 fc3

To find fo , there are three possible relationships:

fc3 = fc2 ± fo and fc3 = fo − fc2

Each should be tested to determine the one that will fall in the range of

400, 000 ≤ fo ≤ 500, 000

39
(a) First, we test fc3 = fc2 − fo . This case leads to

97.3 × 106 = 2n2 (2n1 fc1 − fo ) = 2n1 +n2 fc1 − 2n2 fo = 211 20 × 103 − 2n2 fo

Thus we have
fo = 2−n2 (4.096 × 107 − 9.73 × 107 ) ≤ 0

This is outside the local oscillator frequency range.

(b) Next, we test fc3 = fc2 + fo . This leads to

97.3 × 106 = 2n2 (2n1 fc1 + fo ) = 2n1 +n2 fc1 + 2n2 fo = 211 20 × 103 + 2n2 fo

Thus, we have
fo = 2−n2 (5.634 × 107 )

If n2 = 7, then fo = 440 kHz, which is within the realizable range of the local
oscillator.

40
(c) If we choose fc3 = fo − fc2 . Then we have

97.3 × 106 = fo − 2n2 (2n1 fc1 ) = 2n2 fo − 211 (20 × 103 )

Thus, we have
fo = 2−n2 (13.826 × 107 )
No integer n2 will lead to a realizable fo .
Thus, the final design is N1 = 24 = 16, N2 = 27 = 128, and fo = 440 kHz.

41