L10: 

 FM Bandwidth, Modulation & Demodulation

(PS pp. 107-12; CC sec 5.2 & 5.3)

FM spectrum for single tone message Am cos(2Wt)highly

dependent on modulation index

  k f Am /W .

Shape of spectrum depends critically on message because FM is

non-linear.

Crucial issue - how does FM bandwidth (i.e. range of significant

+ve frequencies) vary with modulation index?

L10 1
Single-Tone FM Spectra (Carlson et al, Fig 5.1-7) ( f m  W )

L10 2
 Approximating the FM Spectrum

Suppose we wish to approximate single tone FM signal by f c & n

frequencies on each side of it, with n large enough that the
neglected side-frequencies are insignificant.

As spacing between impulses is W Hz, define BW B : 2nW .

Question is, how can we select n?

If too large, then BW B will be conservative.

If too small, then signal will be distorted.

One approach: find n so that power contained in higher order

side-frequencies <2% of total FM power, PFM  0.5A 2c .

L10 3
 Carson’s Rule

Parseval’s Theorem  Pdiscarded terms (n,  )  0.5A k n1 J k ( ) .
2 2
c

Pdiscarded terms (n,  ) 

  0.02
  k n1 k
2
We want J ( )
PFM
From Bessel fn. tables, we need n  1   .
 Transmission BW = 2(1   )W  2(W  f ) : BCarson

Note BCarson  with freq. deviation & message BW.

Good estimate for   1 and   1, but commercial FM

operates with 2    10 usually.
In this case, Carlson’s rule underestimates required BW.

L10 4
 Direct FM Generation

Two ways to generate FM.

Direct FM – FM in one step.

At high f c , can use voltage-controlled oscillator with output

frequency that varies according to input voltage.

At lower freq, can use || resonant circuit. Basic idea is to

modulate capacitance of circuit

C(t)  C 0  km(t).

L10 5
 Analysis of Parallel Resonant Circuit

Output is sinusoid at resonant freq.

 0 .5
1 1 1  km(t ) 
f i (t )    1   ,
2 LC (t ) 2 LC 0  km(t )  2 LC 0  C0 
1  0.5km(t )  km(t )
 1    f 0  k f m(t ), , if  1,
2 LC 0  C0  C0

f 0 = unmodulated resonant freq. of circuit. k f  0.5 f 0 k / C 0 .

Typically k | m(t) | / C0  0.005 is acceptable, i.e. f  0.005 f 0 .

For large carrier freq., can be very big – advantage of direct FM.

L10 6
 Indirect FM Signal Generation

Major disadvantage of direct FM – carrier frequency drifts.

In FDM systems such as broadcast radio, this could cause
problems, so additional circuitry is needed to stabilise f c .

Before such circuits became feasible, FM was - & still is –

generated indirectly, i.e. in 2 stages.

Stage 1: Generate narrow-band FM with stable freq., using

standard linear modulation techniques.

Stage 2: Frequency multiplication to generate wide-band FM,

using a non-linear device.

L10 7
 Generating NBFM

Stage 1: Suppose we have at stable oscillator at a relatively low

frequency f1 Hz. Let 1 (t) : 0 m(t)dt

If | 2k11 (t) | 1, can use NBFM approx.

s1 (t)  A1 cos(2f1t)  2k11 (t) A1 sin(2f1t),

 A1cos2k11 (t) cos(2f1t)  sin 2k11 (t) sin(2f1t)
 A1 cos2f1t  2k11 (t) .

L10 8
 Stage 2: Frequency Multiplication

Pass NBFM signal through a memoryless, non-linear element

with a power-law input-output characteristic.

E.g. with a quadratic law device

vo (t )  avi (t )  bvi (t ) 2 ,
 aA1 cos2f1t  2k11 (t )   bA12 cos2f1t  2k11 (t )  ,
2

 0.5bA12  aA1 cos2f1t  2k11 (t )   bA12 cos4f1t  4k11 (t ) .

    
suppressed by BPF passed by BPF

L10 9
 Stage 3: BPF.
If device has p -th power characteristic, then BPF is designed
with centre frequency pf1, so output is
v2 (t)  constant  A1 cos2pf1t  2pk11 (t) ,
p

= FM signal at carrier frequency pf1 and with k f  pk1.

By repeating several times, NBFM  WBFM

If carrier freq. too large, down-convert before transmission.

Disadvantage of indirect FM: power lost in nonlinear elements, so

amplification needed.

L10 10
 Frequency Demodulation

Have seen that

 FM spectrum depends on deviation ratio

D  f /W   if single tone .
 As D  , FM signal BW  .

So far, have stressed that channel BW is a scarce resource.

However, in FM high D , & hence BW, are typical.
Reason – strength of demodulated signal  with deviation ratio.

I.e. unlike AM & linear modulation, FM can trade-off bandwidth &

received signal strength by changing D .

L10 11
 Direct methods: FM-AM Conversion
Frequency demodulation falls into 2 categories, direct & indirect.

Idea behind FM-AM conversion:

  m( )d ,
t
s FM (t )  Ac cos 2f c t  2k f
0

m(t ) sin 2f t  2k  m( )d .

ds FM (t )
 Ac 2f c  2k f
t
 c f
dt 0

I.e. if k f | m(t) | f c , envelope of derivative is scaled & off-set

message
 Can use envelope detector!

L10 12
 A Problem with Simple AM‐FM

Observe that amplitude of AM carrier  f c  MHz. So

differentiated signal may be too strong for equipment.

In freq domain,

ds FM (t) 
F    j2fS FM ( f ) .
 dt 

Differentiator amplitude response at carrier freq. will be huge.

Hard to implement.

L10 13
 Slope Circuits

Key property: linear freq. response over FM signal band. So

consider

 j2a( f  f c  C) if | f  f c | B / 2,
 H ( f ) * if | f  f c | B / 2,
H( f )  
 0 otherwise.


- Slope circuit. Most filters required to have flat magnitude

response in pass-band, but here a gradient is needed.

Easier to implement than differentiator, e.g. by using tuned RLC

circuits.

L10 14
 Analysis of Slope Circuit
To find output of slope circuit, use base-band equivalent signal &
filter.

Step 1: Find complex envelope of FM:


sFM (t )   Ac e
j 2k f 0t m ( ) d  j2f c t
 sl (t )  Ac e
j 2k f 0t m ( ) d
.

Step 2: Find slope circuit CE spectrum:

2H ( f  f c )  j4a( f  C) if | f | B / 2,
Hl ( f )  
 0 otherwise.

L10 15
 Slope Circuit Analysis, cont.

Step 3: Find CE of output:

Yl ( f )  0.5 H l ( f ) Sl ( f )  j2afSl ( f )  j2aCSl ( f ),

dsl (t )
 yl (t )  a  j2aCsl (t )
dt
j 2k f 0t m ( ) d j2k f 0t m ( ) d
 aAc j2k f m(t )e  j2aCAc e ,

 j2aAc C  k f m(t ) e
j 2k f 0t m ( ) d
.

L10 16
 Slope Circuit Analysis, Step 4

Re-insert carrier in eqns:

j2  f ct  k f 0t m ( ) d  
y (t )   j2aAc C  k f m(t ) e

,
 
 
 2aAc C  k f m(t ) sin 2 f c t  k f 0t m( )d . 
Need C  k f max t | m(t) |: f for env. detection. Common
choice is C  B / 2 .
 Demodulated signal amplitude  k f !
 Demodulated FM SNR can be made much higher than in AM/
linear mod, without increasing transmit power.
Cost: higher bandwidth & complexity.

L10 17
L10 18
 Limiter

AM & linearly modulated signals – highly sensitive to amplitude

distortions due to time-varying / random channel losses.

In FM, message is in phase of carrier, so such distortions do not

affect it:


s ' FM (t )  1   (t )  Ac cos 2f c t  2k f
t
0 m( )d 
However, if an ideal discriminator is used directly on distorted
signal, FM-AM conversion will yield distorted message.

Solution: distort amplitude even more…

L10 19
 Analysis of Limiter

Limiter - highly non-linear circuit s.t.

vo (t)  A'sgn vi (t) .

In terms of the angle  (t) of FM signal,

s' FM (t)  0   (t)    / 2  2n,  / 2  2n .

 vo (t)  A' p (t) /(2 ) ,
where p is a square wave between +1 and -1.

L10 20
 Analysis of Limiter, cont.
As p is periodic, with period 1s, & furthermore even, it has a
Fourier cosine expansion, with fundamental frequency 1 Hz. I.e.

p ( x)  n 1 c n cos(2nx), x  R ,


 
  t

 vo (t )  A' n 1 c n cosn (t )   A' n 1 c n cos 2n f c t  k f 0 m( )d ,
 
 A' c1 cos 2 f c t  k f 0 m( )d

t

FM signal with constant amplitude!

 unwanted terms at harmonics of f c .

So insert a BPF at limiter output.

L10 21
 Zero Crossings Detector

Another direct demodulation method.

Consider any interval t 0 ,t 0  T . If T  1/W , then message 

constant over interval. Hence FM inst. freq. is

f i (t)  f c  k f m(t)  constant, t  t 0 ,t 0  T .

In each such interval, there are roughly 2 f i (t)T  O(1) zero
crossings:
If T  1/ f c also, no. zero-crossings is large. So can ignore
integer round-off &
no. zero crossings in interval
f c  k f m(t)  .
2T

L10 22
 Zero Crossings Detector, cont.
Problem - not that easy to detect zeros of FM signal.

However, observe that limited FM signal has same zeros. As they

also occur at sharp transitions, easier to detect. So,

1) Limit FM signal

2) Next apply to pulse-detector (monostable flip-flop) which

generates short pulse for every +ve or –ve transition.

3) Then continuously integrate over interval T. Integral

proportional to # pulses in T, so we recover message.

L10 23
 Non-Linear Distortion in FM Systems
Have observed that we can inflict severe amplitude distortion on
FM signal by limiting it, but we can still recover signal.

If a non-linearity is inserted intentionally in a comm.s system,

called strong. Other e.g.s: square-law AM modulators, envelope
detectors, frequency multiplier.

If it is unwanted but arises whenever signal levels become too

large, it is called weak. E.g. when amplifiers are operated over full
dynamic range.

Another FM advantage - it is impervious to weak non-linearities.

L10 24
 Amplifier with Distortion
E.g., amplifier with cubic characteristic:
vo (t)  avi (t)  bvi (t) 2  cv i (t) 3 ,
 aAc cos2f c t   (t)   bAc cos2f c t   (t)   cAc cos2f c t   (t)  ,
2 2 3 3

 0.5bAc2  aAc  0.75cAc cos2f c t   (t)   0.5bAc cos4f c t  2 (t) 

3 2

 0.25cAc cos6f c t  3 (t) 

3

A BPF centred at f c Hz yields (scaled) FM signal again.

 FM used in microwave radio links, where amplifiers &

transmitters are operated at large signal levels & are thus highly
non-linear.

L12: Demodulation using phase locked loops (PS pp. 226-228,CCR pp. 278-286)

L10 25