Binary Phase Shift Keying (BPSK)
Lecture Notes 6: Basic Modulation Schemes
The first modulation considered is binary phase shift keying. In this scheme during every bit
In this lecture we examine a number of different simple modulation schemes. We examine the duration, denoted by T , one of two phases of the carrier is transmitted. These two phases are
implementation of the optimum receiver, the error probability and the bandwidth occupancy. 180 degrees apart. This makes these two waveforms antipodal. Any binary modulation where
We would like the simplest possible receiver, with the lowest error probability and smallest the two signals are antipodal gives the minimum error probability (for fixed energy) over any
bandwidth for a given data rate. other set of binary signals. The error probability can only be made smaller (for fixed energy
per bit) by allowing more than two waveforms for transmitting information.




VI-1 VI-2

BPSK Modulator


bt st rt
and height 1.


 



     

  




∞
bt ∑ bl pT t lT bl 1 1






"!





l ∞
2P cos 2π f ct


nt




Modulator
The transmitted signal then is given by
Figure 33: Modulator for BPSK ∞
st 2P ∑ bl cos 2π f ct pT t lT


 
 


 
  
 
l ∞

 
To mathematically described the transmitted signal we define appulse
t function p T t as
T
2P b t cos 2π f ct 2P cos 2π f ct φt

 


1 where φ t is the phase waveform. The signal power is P. The energy of each transmitted bit
1 0 t T



pT t is E PT .


 

0 otherwise.


The phase of a BPSK signal can take on one of two values as shown in Figure VI-3.


t
T

Let b t denote the data waveform consisting of an infinite sequence of pulses of duration T





VI-3 VI-4
 


bt

 
t iT


rt
0 dec bi 1 1


1

 


 



LPF




  



0 dec bi 1 1



t X iT
T 2T 3T 4T 5T


-1 2 T cos 2π f ct



Figure 35: Demodulator for BPSK
φt
The optimum receiver for BPSK in the presence of additive white Gaussian noise is shown in
 

Figure VI-3. The low pass filter (LPF) is a filter “matched” to the baseband signal being
π transmitted. For BPSK this is just a rectangular pulse of duration T . The impulse response is
ht pT t The output of the low pass filter is


" 
0 ∞


t
T 2T 3T 4T 5T X t 2 T cos 2π f c τ h t τ r τ dτ

 


"
∞




Figure 34: Data and Phase waveforms for BPSK




VI-5 VI-6


The sampled version of the output is given by
∞
X iT 2 T cos 2π f c τ pT iT τ r τ dτ
 
 

∞


  
   
 

 

iT
2 T cos 2π f c τ 2P b τ cos 2π f c τ


 "

i 1T

n τ dτ e,-1






iT




P
2 P T bi 1 cos 2π f c τ cos 2π f c τ d τ ηi e,+1








 

i 1T





ηi is Gaussian random variable, mean 0 variance N0 2. Assuming 2π f c T 2πn for some





integer n (or that f c T 1)






X (iT )
X iT PT bi 1 ηi E bi 1 ηi E 0 E






"


Figure 36: Probability Density of Decision Statistic for Binary Phase Shift
Keying




VI-7 VI-8
 Error Probability of BPSK


Bit Error Probability of BPSK
P e,b 1

-1
10
2E 2Eb
Pe b Q Q


N0 N0 10 -2








where 10 -3
∞ 1 u2 2
Qx e du



-4


x 2π 10

-5
For binary signals this is the smallest bit error probability, i.e. BPSK are optimal signals and 10
the receiver shown above is optimum (in additive white Gaussian noise). For binary signals 10 -6

the energy transmitted per information bit Eb is equal to the energy per signal E. For
-7
Pe b 10 5 we need a bit-energy, Eb to noise density N0 ratio of Eb N0 9 6dB. Note: Q x 10




"


is a decreasing function which is 1/2 at x 0. There are efficient algorithms (based on Taylor 10 -8

2
series expansions) to calculate Q x . Since Q x e x 2 2 the error probability can be
 



-9
  10



upper bounded by
1 10 -10
Pe b e Eb N0





2 0 2 4 6 8 10 12 14 16


which decreases exponentially with signal-to-noise ratio.

Eb/N 0 (dB)

Figure 37: Error Probability of BPSK.




VI-9 VI-10

Bandwidth of BPSK


The power spectral density is a measure of the distribution of power with respect to frequency.
The power spectral density for BPSK has the form
PT
S f sinc2 f fc T sinc2 f fc T


2







where
sin πx
sinc x
 "

πx


Notice that 1 2T f fc 1 2T . The drawbacks are that the signal loses its constant envelope



∞








S f df P property (useful for nonlinear amplifiers) and the sensitivity to timing errors is greatly
 
" 

∞
   

increased. The timing sensitivity problem can be greatly alleviated by filtering to a slightly


The power spectrum has zeros or nulls at f f c i T except for i 0; that is there is a null at
larger bandwidth 1 α 2T 1 α 2T .


f fc




f fc 1 T called the first null; a null at f f c 2 T called the second null; etc. The












bandwidth between the first nulls is called the null-to-null bandwidth. For BPSK the
null-to-null bandwidth is 2 T . Notice that the spectrum falls off as f f c 2 as f moves away


from f c . (The spectrum of MSK falls off as the fourth power, versus the second power for
BPSK).
It is possible to reduce the bandwidth of a BPSK signal by filtering. If the filtering is done
properly the (absolute) bandwidth of the signal can be reduced to 1 T without causing any

intersymbol interference; that is all the power is concentrated in the frequency range




VI-11 VI-12
 0.50 0

S(f) (dB)



S(f)

0.40 -20

0.30
-40

-60
0.20

-80
0.10

-100
0.00 -5 -4 -3 -2 -1 0 1 2 3 4 5
-4 -3 -2 -1 0 1 2 3 4
(f-fc)T (f-f c )T

Figure 38: Spectrum of BPSK Figure 39: Spectrum of BPSK




VI-13 VI-14

0 Example


Given:
Noise power spectral density of N0 2 180 dBm/Hz =10 21 Watts/Hz.



-20

S(f) (dB)

Pr 3 10 13 Watts

 



Desired Pe 10 7 .


-40
Find: The data rate that can be used and the bandwidth that is needed.
Solution: Need Q 2Eb N0 10 7 or Eb N0 11 3dB or Eb N0 13 52. But




 "
"

"
 "

-60 Eb N0 Pr T N0 13 52. Thus the data bit must be at least T 9 0 10 8 seconds long, i.e.




the data rate 1 T must be less than 11 Mbits/second. Clearly we also need a (null-to-null)


bandwidth of 22 MHz.
-80 An alternative view of BPSK is that of two antipodal signals; that is

s0 t Eψ t 0 t T






-100 and
-10 -8 -6 -4 -2 0 2 4 6 8 10 s1 t Eψ t 0 t T
  




(f-fc)T
where ψ t 2 T cos 2π f ct 0 t T is a unit energy waveform. The above describes

 


the signals transmitted only during the interval 0 T . Obviously this is repeated for other





Figure 40: Spectrum of BPSK




VI-15 VI-16
 


Effect of Filtering and Nonlinear Amplification
on a BPSK waveform
intervals. The receiver correlates with ψ t over the interval 0 T and compares with a





threshold (usually 0) to make a decision. The correlation receiver is shown below. In this section we illustrate one main drawback to BPSK. The fact that the signal amplitude
has discontinuities causes the spectrum to have fairly large sidelobes. For a system that has a
constraint on the bandwidth this can be a problem. A possible solution is to filter the signal. A
bandpas filter centered at the carrier frequency which removes the sidbands can be inserted
rt
after mixing to the carrier frequency. Alternatly we can filter the data signal at baseband


T γ dec s0

γ before mixing to the carrier frequency.
  


0 dec s1




Below we simulate this type of system to illustrate the effect of filtering and nonlinear


amplification. The data waveform b t is mixed onto a carrier. This modulated waveform is


denoted by
ψt


This is called the “Correlation Receiver.” Note that synchronization to the symbol timing and
s1 t 2P cos 2π f ct



oscillator phase are required.



The signal s1 t is filtered by a fourth order bandpass Butterworth filter with passband from


fc 4Rb to f c 4Rb The filtered signal is denoted by s2 t . The signal s2 t is then amplified.








VI-17 VI-18


Data waveform
1

0.5
The input-output characteristics of the amplifier are

b(t)
0
s3 t 100 tanh 2s1 t




−0.5
This amplifier is fairly close to a hard limiter in which every input greater than zero is mapped
to 100 and every input less than zero is mapped to -100. −1
0 0.2 0.4 0.6 0.8 1 1.2
Simulation Parameters time −5
x 10
Signal waveform
Sampling Frequency= 50MHz 2
Sampling Time =20nseconds
1
Center Frequency= 12.5MHz
Data Rate=390.125kbps
s(t)

0
Simulation Time= 1.31072 m s
−1

−2
0 0.2 0.4 0.6 0.8 1 1.2
time −5
x 10




VI-19 VI-20
 


Signal spectrum 100
−40
80

−60
60

−80 40

20
−100

Output
S(f)

−120
−20

−140 −40

−60
−160
−80

−180
0 0.5 1 1.5 2 2.5 −100
−2 −1.5 −1 −0.5 0 0.5 1 1.5 2
frequency x 10
7
Input




VI-21 VI-22


Filtered signal spectrum Data waveform
−40 1

−45 0.5

b(t)
−50 0

−55 −0.5

−60
−1
0 0.2 0.4 0.6 0.8 1 1.2
S2(f)

time −5
x 10
−65
Filtered signal waveform
2
−70

−75 1
s2(t)

−80 0

−85 −1

−90 −2
0 0.5 1 1.5 2 2.5 0 0.2 0.4 0.6 0.8 1 1.2
frequency 7 time −5
x 10 x 10




VI-23 VI-24
 


Amplified and filtered signal spectrum Data waveform
−20 1

0.5

−30

b(t)
0

−0.5
−40

−1
0 0.2 0.4 0.6 0.8 1 1.2
time −5
S2(f)

−50 x 10
Amplified and filtered signal waveform
100

−60 50

s3(t)
0
−70
−50

−80 −100
0 0.5 1 1.5 2 2.5 0 0.2 0.4 0.6 0.8 1 1.2
frequency 7 time −5
x 10
x 10




VI-25 VI-26

P cos 2π f ct





Quaternary Phase Shift Keying (QPSK)


bc t
The next modulation technique we consider is QPSK. In this modulation technique one of




 

four phases of the carrier is transmitted in a symbol duration denoted by Ts . Since one of four
waveforms is transmitted there are two bits of information transmitted during each symbol



st


duration. An alternative way of describing QPSK is that of two carriers offset in phase by 90




degrees. Each of these carriers is then modulated using BPSK. These two carriers are called


the inphase and quadrature carriers. Because the carriers are 90 degrees offset, at the output of bs t



the correlation receiver they do not interfer with each other (assuming perfect phase

  


synchronization). The advantage of QPSK over BPSK is that the the data rate is twice as high


for the same bandwidth. Alternatively single-sideband BPSK would have the same rate in bits
per second per hertz but would have a more difficult job of recovering the carrier frequency
and phase.

P sin 2π f ct



Figure 41: Modulator for QPSK




VI-27 VI-28
 


∞
bc t ∑ bc l pTs t lTs bc l 1 1

 

 


 
 



! !
l ∞


∞
bs t ∑ bs l pTs t lTs bs l 1 1



l ∞


bc l bs l φl
st P bc t cos 2π f ct bs t sin 2π f ct
+1 +1 π 4
 
 



 

 




2P cos 2π f ct φt

-1 +1 3π 4

The transmitted power is still P. The symbol duration is Ts seconds. The data rate is -1 -1 5π 4

Rb 2 Ts bits seconds. +1 -1 7π 4


The phase φ t , of the transmitted signal is related to the data waveform as follows.


∞
φt ∑ φl pTs t lTs φl π 4 3π 4 5π 4 7π 4











!
l ∞


The relation between φl and bc l bs l is shown in the following table






VI-29 VI-30

bc t
 


1

t


Ts 2Ts 3Ts 4Ts 5Ts

-1

bs t The constellation of QPSK is shown below. The phase of the overall carrier can be on of four
 

values. Transitions between any of the four values may occur at any symbol transition.
1 Because of this, it is possible that the transition is to the 180 degree opposite phase. When this
happens the amplitude of the signal goes through zero. In theory this is an instantaneous
t


Ts 2Ts 3Ts 4Ts 5Ts transition. In practice, when the signal has been filtered to remove out-of-band components
-1 this transition is slowed down. During this transition the amplitude of the carrier goes throguh
φt zero. This can be undesireable for various reasons. One reason is that nonlinear amplifiers
 

with a non constant envelope signal will regenerate the out-of-band spectral components.
7π 4
Another reason is that at the receiver, certain synchronization circuits need constant envelope

5π 4
to maintain their tracking capability.

3π 4

π 4
t

Ts 2Ts 3Ts 4Ts 5Ts

Figure 42: Timing and Phase of QPSK




VI-31 VI-32
 Quadrature-phase


Channel

The bandwidth of QPSK is given by

(-1,+1) (+1,+1)
S f PTs 2 sinc2 f fc Ts sinc2 f fc Ts







 
PTb sinc2 2 f f c Tb sinc2 2 f f c Tb








 
In-phase
Channel since Ts Tb 2. Thus while the spectrum is compressed by a factor of 2 relative to BPSK




with the same bit rate, the center lobe is also 3dB higher, that is the peak power density is
(-1,-1) (+1,-1) higher for QPSK than BPSK. The null-to-null bandwidth is 2 Ts Rb .




Figure 43: Constellation of QPSK




VI-33 VI-34

1.00 0


S(f) dB
BPSK QPSK
S(f)

-20
0.75

QPSK
-40

0.50

-60

BPSK
0.25
-80

0.00
-4 -3 -2 -1 0 1 2 3 4 -100
-5 -4 -3 -2 -1 0 1 2 3 4 5
(f-f c )T (f-f c )T

Figure 44: Spectrum of QPSK Figure 45: Spectrum of QPSK




VI-35 VI-36
 0


S(f) dB QPSK 2 Ts cos 2π f ct
BPSK



t iTs

  

-20 0 dec bc i 1 1

 
 

LPF


  

 




0 dec bc i 1 1
Xc iTs

-40 rt


t iTs

  
-60
LPF 0 dec bs i 1 1

 
 


     

 

0 dec bs i 1


1
Xs iTs

-80
2 Ts sin 2π f ct




-100
-10 -8 -6 -4 -2 0 2 4 6 8 10
(f-fc)T
Figure 47: QPSK Demodulator
Figure 46: Spectrum of QPSK




VI-37 VI-38


Assuming 2π f c Ts 2πn or 2π f c Ts 1 Example


Xc iTs PTs 2 bc i 1 ηc i E b bc i 1 ηc i Given:

 







Noise power spectral density of N0 2 110 dBm/Hz =10 14 Watts/Hz.





Xs iTs PTs 2 bs i 1 ηs i Eb bs i 1 ηs i 6


Pr 3 10 Watts


 




where Eb PTs 2 is the energy per transmitted bit. Also ηc i and ηs i are Gaussian random Desired Pe 10 7 .




variables, with mean 0 and variance N0 2.

Find: The data rate that can be used and the bandwidth that is needed for QPSK.
Bit Error Probability of QPSK
Solution: Need Q 2Eb N0 10 7 or Eb N0 11 3dB or Eb N0 13 52. But




 


"

"


2Eb
Pe b Q Pr
Ts


N0



2
Eb N0


 
N0

The probability that a symbol error is made is
Pr T N0 13 52

 "
" 
2
Pe2 b

Pe s 1 1 Pe b 2Pe b








since Ts 2T . Thus the data bit must be at least T 9 0 10 8 seconds long, i.e. the data


Thus for the same data rate, transmitted power, and bit error rate (probability of error), QPSK rate 1 T must be less than 11 Mbits/second. Clearly we also need a (null-to-null) bandwidth

has half the (null-to-null) bandwidth of BPSK. of 11 MHz.




VI-39 VI-40
 h


P cos 2π f ct





bc t Ts 2


Offset Quaternary Phase Shift Keying (OQPSK)


 





The disadvantages of QPSK can be fixed by offsetting one of the data streams by a fraction st


(usually 1/2) of a symbol duration. By doing this we only allow one data bit to change at a




time. When this is done the possible phase transitions are 90 deg. In this way the transitions


bs t


through the origin are illiminated. Offset QPSK then gives the same performance as QPSK



  

but will have less distorition when there is filtering and nonlinearities.

 
P sin 2π f ct




Figure 48: Modulator for OQPSK




VI-41 VI-42

Quadrature-phase


∞ Channel
bc t ∑ bc l pTs t lTs bc l 1 1










!

l ∞


∞
bs t ∑ bs l pTs t lTs bs l 1 1










!

l ∞
 

(-1,+1) (+1,+1)
st P bc t Ts 2 cos 2π f ct bs t sin 2π f ct










 

st 2P cos 2π f ct φt







The transmitted power is still P. The symbols duration is Ts seconds. The data rate is
In-phase
Rb 2 Ts bits seconds. The bandwidth (null-to-null) is 2 Ts Rb . This modification of
Channel




QPSK removes the possibility of both data bits changing simultaneously. However, one of the
data bits may change every Ts 2 seconds but 180 degree changes are not allowed. The (-1,-1) (+1,-1)

bandwidth of OQPKS is the same as QPSK. OQPSK has advantage over QPSK when passed
through nonlinearities (such as in a satellite) in that the out of band interference generated by
first bandlimiting and then hard limiting is less with OQPSK than QPSK.

Figure 49: Constellation of QPSK




VI-43 VI-44
 bc t Ts 2





1 2 Ts cos 2π f ct



t iTs Ts 2

 



t


Ts 2Ts 3Ts 4Ts 5Ts 0 dec bc i 1 1

 
 

-1 LPF

   







0 dec bc i 1 1
Xc iTs Ts 2

bs t

 
rt


1
t iTs

 
t


Ts 2Ts 3Ts 4Ts 5Ts
LPF 0 dec bs i 1 1
-1

 
 


    



0 dec bs i 1

 



φt Xs iTs 1

 

7π 4

5π 4 2 Ts sin 2π f ct



3π 4

π 4 Figure 51: Demodulator for OQPSK


Ts 2Ts 3Ts 4Ts 5Ts

Figure 50: Data and Phase Waveforms for OQPSK




VI-45 VI-46


Assuming 2π f c Ts 2πn or 2π f c Ts 1 Minimum Shift Keying (MSK)
 

Xc iTs Ts 2 PTs 2 bc i 1 ηc i Eb bc i 1 ηc i

 

 
 

 

Minimum shift keying can be viewed in several different ways and has a number of significant
Xs iTs PTs 2 bs i 1 ηs i Eb bs i 1 ηs i
advantages over the previously considered modulation schemes. MSK can be thought of as a


where Eb PTs 2 is the energy per transmitted bit. Also ηc i and ηs i are Gaussian random variant of OQPSK where the data pulse waveforms are shaped to allow smooth transition


variables, with mean 0 variance N0 2. between phases. It can also be thought of a a form of frequency shift keying where the two

Bit Error Probability of OQPSK frequencies are separated by the minimum amount to maintain orthogonality and have
continuous phase when switching from one frequency to another (hence the name minimum
2Eb shift keying). The advantages of MSK include a better spectral efficiency in most cases. In
Pe b Q


N0



fact the spectrum of MSK falls off at a faster rate than BPSK, QPSK and OQPSK. In addition


The probability that a symbol error is made is there is an easier implementation than OQPSK (called serial MSK) that avoids the problem of
Pe s 1 1 Pe b 2
2Pe b Pe2 b having a precisely controlled time offset between the two data streams. An additional








advantage is that MSK can be demodulator noncoherently as well as coherently. So for

This is the same as QPSK. applications requiring a low cost receiver MSK may be a good choice.




VI-47 VI-48
 ∞


bc t ∑ bc l pTs t lTs bc l 1 1











!
l ∞
P cos 2π f ct


ct Ts 2


∞






bs t ∑ bs l pTs t lTs bs l 1 1











!
l ∞

 



bc t Ts 2

 2 sin πt Ts 2 cos πt Ts


ct ct Ts 2


  

  



   








st



 

bs t st P bc t Ts 2 c t Ts 2 cos 2π f ct bs t c t sin 2π f ct


 











 
 

   

    
st 2P bc t Ts 2 cos πt Ts cos 2π f ct

  




 

 

 !

! 
 

bs t sin πt Ts sin 2π f ct


ct P sin 2π f ct





2P cos 2π f ct φt


Figure 52: Modulator for MSK




where

cos φ t bc t Ts 2 cos πt Ts














VI-49 VI-50


bc t Ts 2 bs t φt





 
πt
+1 +1 π Ts
πt
+1 -1 π Ts
πt
-1 +1 Ts
πt
sin φ t bs t sin πt Ts -1 -1


Ts






In the above table, because of the delay of the bit stream corresponding to the cosine branch,
only one bit is allowed to change at a time. During each time interval of duration Ts 2 during
bs t sin πt Ts

φt tan 1 which the data bits remain constant there is a phase shift of π 2. Because the phase changes


  
 
 


bc t Ts 2 cos πt Ts


linearly with time MSK can also be viewed as frequency shift keying. The two different




frequencies are f c 2T1 s and f c 2T1 s . The change in frequency is ∆ f T1s 1

2T where



b
Tb 1 2 Ts is the data bit rate. The transmitted power is still P. The symbols duration is Ts



seconds. The data rate is Rb 2 Ts bits seconds. The signal has constant envelope which is



useful for nonlinear amplifiers. The bandwidth is different because of the pulse shaping
waveforms.




VI-51 VI-52
 bc t Ts 2

 


4


1 bc 0 bc 3
3

7Ts t φ(t)/π


Ts 2Ts 3Ts 4Ts 5Ts 6Ts
-1 bc 1 bc 2 bc 4 2

bs t

 1  bs 1 bs 3
1

7Ts t


Ts 2Ts 3Ts 4Ts 5Ts 6Ts
-1b bs 2 bs 4 −1
s0

φt
−2
π
 

π 2

−3
0
7Ts t


Ts 2Ts 3Ts 4Ts 5Ts 6Ts
π 2

 

π −4

2 4 6 8 10 12 14

time/Tb

Figure 53: Data and phase waveforms for MSK Figure 54: Phase of MSK signals




VI-53 VI-54

Quadrature-phase


Channel

The spectrum of MSK is given by

(-1,+1) (+1,+1)
8PTb cos2 2πTb f fc cos2 2πTb f fc


S f



 
 
 
π2 1 4Tb f fc 2 2 1 4Tb f fc 2 2



 
 




 
 





The nulls in the spectrum are at f f c Tb = 0.75, 1.25, 1.75,.... Because we force the signal
to be continuous in phase MSK has significantly faster decay of the power spectrum as the
In-phase
frequency from the carrier becomes larger. MSK decays as 1 f 4 while QPSK, OQPSK, and
Channel

BPSK decay as 1 f 2 as the frequency differs more and more from the center frequency.

(-1,-1) (+1,-1)

Figure 55: Constellation of MSK




VI-55 VI-56
 


Spectrum of MSK Spectrum of MSK
1.00 0

S(f) (dB)
B PSK Q PSK
-10
QPSK M SK M SK
-20
0.75
-30

-40

0.50 -50

-60

-70
0.25
-80
BPSK
-90

0.00 -100
-4 -3 -2 -1 0 1 2 3 4 -5 -4 -3 -2 -1 0 1 2 3 4 5
(f-f c )T (f-f c )T

Figure 56: Spectrum of MSK Figure 57: Spectrum of MSK




VI-57 VI-58


0
S(f)

ct Ts 2 2 Ts cos 2π fc t



t iTs Ts 2




-25


0 dec bc i 1 1



 
 

LPF

   


0 dec bc i 1


1
Xc iTs Ts 2

-50 rt

t iTs

 
-75 LPF 0 dec bs i 1 1

 
 


    






0 dec bs i 1 1
Xs iTs

-100 ct 2 Ts sin 2π fc t




-8 -6 -4 -2 0 2 4 6 8
(f-fc)T
Figure 59: Coherent Demodulator for MSK
Figure 58: Spectrum of MSK




VI-59 VI-60
 


Bit Error Probability of MSK with Coherent Demodulation
Assuming 2π f c Ts 2πn or 2π f c Ts 1
 

Since the signals are still antipodal
Xc iTs Ts 2   PTs 2 bc i 1 ηc i Eb bc i 1 ηc i

 
 

 
2Eb

Xs iTs PTs 2 bs i 1 ηs i Eb bs i 1 ηs i Pe b Q


N0






where Eb PTs 2 is the energy per transmitted bit. Also ηc i and ηs i are Gaussian random The probability that a symbol error is made is


variables, with mean 0 variance N0 2. 2

Pe s 1 1 Pe b 2Pe b Pe2 b












VI-61 VI-62

2 6


1.5
4

2
0.5

phi(t)
y(t)

0 0

-0.5
-2

-1

-4
-1.5

-2 -6
0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20
time/Tb time/Tb

Figure 60: Waveform for Minimum Shift Keying Figure 61: Phase Waveform for Minimum Shift Keying




VI-63 VI-64
 1 60


0.5 40

0 20

-0.5 0

-1
0 5 10 15 20 25 30 -20

-40
1

-60
0.5

0 -80

-0.5 -100

-1 -120
0 5 10 15 20 25 30 0 1 2 3 4 5 6 7 8 9 10

Figure 62: Quadrature Waveforms for Minimum Shift Keying Figure 63: Spectrum for Minimum Shift Keying




VI-65 VI-66


Noncoherent Demodulation of MSK
Because MSK can be viewed as a form of Frequency Shift Keying it can also be demodulated
noncoherently. For the same sequence of data bits the frequency is f c 1 2Ts if


bc t Ts 2 bs t and is f c 1 2Ts if bc t Ts 2 bs t . For the example phase waveform shown previously we have that


  








Consider determining bs i 1 at time i 1 2 Ts . Assume we have already determined bc i 2 at






time i 1 Ts . If we estimate which of two frequencies is sent over the interval Time Interval 0 Ts 2 Ts 2 Ts Ts 3Ts 2 3Ts 2 2Ts 2Ts 5Ts 2


   

  
   












   
    

i 1 Ts i 1 2 Ts the decision rule is to decide that bs i 1 bc i 2 if the frequency Frequency




 



 

detected is f c 1 2Ts and to decide that bs i 1 bc i 2 if the frequency detected is Previous Data bc 1 bs 0 1 bc 0 1 bs 1 1 bc 1 1


 
 


 

fc 1 2Ts .
Detected Data bs 0 1 bc 0 1 bs 1 1 bc 1 1 bs 2 1

Consider determining bc i 1 at time iTs . Assume we have already determined bs i 1 at time

  

So detecting the frequency can also be used to detect the data.

i 1 2 Ts . If we estimate which of two frequencies is sent over the interval i 1 2 Ts iTs


 

the decision rule is to decide that bc i 1 bs i 1 if the frequency detected is f c 1 2Ts and




to decide that bc i 1 bs i 1 if the frequency detected is f c 1 2Ts .







The method to detect which of the two frequencies is transmitted is identical to that of
Frequency Shift Keying which will be considered later.




VI-67 VI-68
 Serial Modulation and Demodulation


The implementation of MSK as parallel branches suffers from significant sensitivity to precise
timing of the data (exact shift by T for the inphase component) and the exact balance between
the inphase and quadriphase carrier signals. An alternative implementation of MSK that is
less complex and does not have these draw backs is known as serial MSK. Serial MSK does
have an additional restriction that f c 2n 1 4T which may be important when f c is about





the same as 1 T but for f c 1 T it is not important. The block diagram for serial MSK
2 sin 2π f 1t pT t

modulator and demodulator is shown below. gt





1 1
where f 1 fc 4T and f 2 fc 4T . (For serial MSK we require f c 2n 1 4T for some








bt st integer n. Otherwise the implementation does not give constant envelope).



G f


  






2P cos 2π f 1t




The filter G f is given by filter

 

jπ f f1 T jπ f f1 T
G f T sinc f f1 T e T sinc f f1 T e







 






VI-69 VI-70

Demodulator


t iT
 

rt


H f LPF The low pass filter (LPF) removes double frequency components. Serial MSK is can also be

  






viewed as a filtered form of BPSK where the BPSK signal center frequency is f 1 but the filter


X iT
is not symmetric with respect to f 1 . The receiver is a filter matched to the transmitted signal


(and hence optimal). The output is then mixed down to baseband where it is filtered (to
2 T cos 2π f 1t
remove the double frequency terms) and sampled.




The filter H f is given by



4T cos 2π f f1 T 0 25 j2π f f1 T
 
  "

H f e
"




π 1 16 f





f1 T 0 25 2








VI-71 VI-72
 


Continuous Phase Modulation
MSK is a special case of a more general form of modulation known as continuous phase
modulation where the phase is continuous. The general form of CPM is given by For example if CPM has h 1 2 and




st 2P cos 2π f ct φt


0 t 0





 



  
where the phase waveform has the form qt t 2 0 t T




 
t k 1 2 t T
φt ∑ bi g τ iT d τ φ0 kT

" 


2πh t k 1T
 


 
 



0 i 0 then the modulation is the same as MSK. Continuous Phase Modulation Techniques have

 
k constant envelope which make them useful for systems involving nonlinear amplifiers which
2πh ∑ bi q t iT φ0 kT t k 1T



also must have very narrow spectral widths.


i 0

The function g is the (instantaneous) frequency function, h is called the modulation index


0 g τ d τ is the phase waveform. The function

t
and bi is the data. The function q t



dg t
gt is the frequence waveform.


dt





VI-73 VI-74


Example

Given:

Noise power spectral density of N0 2 110 dBm/Hz =10 14 Watts/Hz.




Gaussian Minimum Shift Keying

Pr 3 10 6 Watts
 


Desired Pe 10 7 . Gaussian minimum shift keying is a special case of continuous phase modulation discussed in


the previous section. For GMSK the pulse waveforms are given by
Bandwidth available=26MHz (at the 902-928MHz band). The peak power outside must
be 20dB below the peak power inside the band. t T t

 
gt Q Q
σ σ




Find: The data rate that can be used for MSK.
Solution: Need Q 2Eb N0 10 7 or Eb N0 11 3dB or Eb N0 13 52. But



 "
"

"
 "

Eb N0 Pr T N0 13 52. Thus the data bit must be at least T 9 0 10 8 seconds long, i.e.




the data rate 1 T must be less than 11 Mbits/second.




VI-75 VI-76
 


10 2

8 1.8

6 1.6

4 1.4

2 1.2
phi(t)

h(t)
0 1

−2 0.8

−4 0.6

−6 0.4

−8 0.2

−10 0
2 4 6 8 10 12 14 0 2 4 6 8 10 12 14 16 18 20
time/Tb time/Tb

Figure 64: Phase Waveform for Gaussian Minimum Shift Keying (BT=0.3) Figure 65: Data Waveform for Gaussian Minimum Shift Keying (BT=0.3)




VI-77 VI-78


0 1.5

-10
1

-20

0.5
-30
|H(f)|

b(t)
-40 0

-50
-0.5

-60

-1
-70

-80 -1.5
0 1 2 3 4 5 6 7 8 9 10 0 2 4 6 8 10 12 14 16 18 20
f time/Tb

Figure 66: Waveform for Gaussian Minimum Shift Keying (BT=0.3) Figure 67: Waveform for Gaussian Minimum Shift Keying (BT=0.3)




VI-79 VI-80
 


20 2

1.5
10

0
0.5
|X(f)|

y(t)
-10 0

-0.5
-20

-1

-30
-1.5

-40 -2
0 1 2 3 4 5 6 7 8 9 10 0 2 4 6 8 10 12 14 16 18 20
f time/Tb

Figure 68: Waveform for Gaussian Minimum Shift Keying (BT=0.3) Figure 69: Waveform for Gaussian Minimum Shift Keying (BT=0.3)




VI-81 VI-82


6 4

3
4

2
1
phi(t)

phi(t)
0 0

-1
-2

-2

-4
-3

-6 -4
0 2 4 6 8 10 12 14 16 18 20 0 5 10 15 20 25 30
time/Tb time/Tb

Figure 70: Waveform for Gaussian Minimum Shift Keying (BT=0.3) Figure 71: Waveform for Gaussian Minimum Shift Keying (BT=0.3)




VI-83 VI-84
 


1 60

0.5 40

0 20

-0.5 0

-1 -20
0 5 10 15 20 25 30

-40
1

-60
0.5

-80
0

-100
-0.5

-1 -120
0 5 10 15 20 25 30 0 1 2 3 4 5 6 7 8 9 10

Figure 72: Waveform for Gaussian Minimum Shift Keying (BT=0.3) Figure 73: Waveform for Gaussian Minimum Shift Keying (BT=0.3)




VI-85 VI-86


Data Waveform
2

Real(x(t))
0

π 4 QPSK −1

−2
As mentioned earlier the effect of filtering and nonlinearly amplifying a QPSK waveform 0 5 10 15 20
time
25 30 35 40

causes distortion when the signal amplitude fluctuates significantly. Another modulation Data Waveform
scheme that has less fluctuation that QPSK is π 4 QPSK. In this modulation scheme every 2

other symbol is sent using a rotated (by 45 degrees) constellation. Thus the transitions from 1

one phase to the next are still instantaneous (without any filtering) but the signal never makes Imag(x(t))
0
a transition through the origin. Only 45 and 135 degree transitions are possible. This is



shown in the constellation below where a little bit of filtering was done. −1

−2
0 5 10 15 20 25 30 35 40
time

Figure 74: Data Waveforms for π 4 QPSK




VI-87 VI-88
 


2
2

1 1.5

0
1
−1

−2
0.5
−3
0 0.5 1 1.5 2 2.5 3 3.5 4
0

−0.5
2

1
−1
0

−1 −1.5

−2

−3 −2
0 0.5 1 1.5 2 2.5 3 3.5 4 −2 −1.5 −1 −0.5 0 0.5 1 1.5 2

Figure 75: Eye Diagram for π 4 QPSK Figure 76: Constellation for π 4 QPSK.




VI-89 VI-90

General Modulator
Lecture 6b: Other Modulation Techniques


Orthogonal Signals
A set of signals ψi t : 0 t T 0 i M 1 are said to be orthogonal (over the interval








0 T ) if b1 t






T
ψi t ψ j t dt 0 i j b2 t


"

Select







In most cases the signals will have the same energy and it is convenient to normalize the one of ut st
b3 t
M 2k




signals to unit energy. A set of signals ψi t : 0 t T 1 i M are said to be





  












orthonormal (over the interval 0 T ) if unit energy


 

signals
T 0 i j
 

ψi t ψ j t dt
 

1 i j bk t



0
2P cos 2π f c f0 t



"







Many signal sets can be described as linear combinations of orthonormal signal sets as we
will show later. Below we describe a number of different orthonormal signal sets. The signal
∞
sets will all be described at some intermediate frequency f 0 but are typically modulated up to bi t ∑ bl pT t lT i 12 k




 
"" 

the carrier frequency f c . l ∞





VI-91 VI-92
 General Coherent Demodulator


t lT
ψ0 T

  
t







X1 l T
∞
ut ∑ ψil t lT t lT






l ∞ ψ1 T

  
t









where for l 1T t T X2 l T
  



1 b1 t b2 t bk t bk t 1 rt
1 Choose
" " "
" " "
  
   
  



  
  



  



2 b1 t b2 t bk 1 t 1 bk t 1 Largest

 
il


  

M b1 t b2 t bk 1 t bk t 1

2 T cos 2π fc f0 t




t lT
ψM

  
1 T t








XM l T




VI-93 VI-94

The symbol error probability can be upper bounded as

ψm T t is the impulse response of the m-th matched filter. The output of these filters





(assuming that the il -th orthogonal signal is transmitted is) given by E

1 N0 ln M

 

 
  

2
E E
ηm m il Pe s exp ln M ln M 4 ln M


N0 N0
  !

 !

 "

 

Xm l T
 



 




E ηm m il E E


exp ln M 4 ln M



2N0 N0



where ηm m 0 1 2 M 1 is a sequence of independent, identically distributed Normally a communication engineer is more concerned with the energy transmitted per bit


"" 



Gaussian random variables with mean zero and variance N0 2. rather than the energy transmitted per signal, E. If we let E b be the energy transmitted per bit

To determine the probability of error we need to determine the probability that the filter output then these are related as follows
E
corresponding to the signal present is smaller than one of the other filter outputs. Eb


"
log2 M
The symbol error probability of M orthogonal signals with coherent demodulation is given by Thus the bound on the symbol error probability can be expressed in terms of the energy
∞ transmitted per bit as
M 1 2E M u2 2
Φu Φ 2
 

Pe s ue du

 





2π ∞ N0 1 Eb
ln 2


 

N0


 
 2
where Φ u is the distribution function of a zero mean, variance 1, Gaussian random variable  Pe s exp2 log2 M Eb
ln 2 ln 2 Eb
4 ln 2

 
N0 N0




given by




 





Eb Eb
u
exp2 log2 M 2N0 ln 2 N0 4 ln 2
1



x2 2
Φu



e dx




"

2π ∞ where exp2 x denotes 2x . Note that as M ∞, Pe 0 if Eb

ln 2 = -1.59dB.


N0



!




VI-95 VI-96
 1


Pe , s
1 0- 1

1 0- 2

1 0- 3
M=2
So far we have examined the symbol error probability for orthogonal signals. Usually the 1 0- 4

number of such signals is a power of 2, e.g. 4, 8, 16, 32, .... If so then each transmission of a 1 0- 5 4

signal is carrying k log2 M bits of information. In this case a communication engineer is 8


1 0- 6
usually interested in the bit error probability as opposed to the symbol error probability. These 16
1 0- 7
can be related for any equidistant, equienergy signal set (such as orthogonal or simplex signal Shannon
Limit M=32
sets) by 1 0- 8

1 0- 9
2k 1 M
 
Pe b Pe s Pe s

 M=1024
2k 1

"
2M 1 1 0- 1 0



-4 -2 0 2 4 6 8 10 12 14 16
Eb/N0 (dB)

Figure 77: Symbol Error Probability for Coherent Demodulation of Orthogonal

Signals




VI-97 VI-98

1
General Noncoherent Demodulator


Pe,b
1 0- 1
t lT
1 0- 2
ψ1 T t 2






Xc 1 lT


1 0- 3


Z1 lT

 
1 0- 4


M=2 2 T cos 2π fc f0 t θ t lT

ψ1 T

1 0- 5 t 2






4 Xs 1 lT




1 0- 6
8

   

1 0- 7
Shannon 16 Choose
Limit  


1 0- 8 Largest
rt
32
1 0- 9
t lT

    
M=1024

1 0- 1 0 ψM T t 2






-4 -2 0 2 4 6 8 10 12 14 16
θ Xc M lT



2 T sin 2π fc f0 t


Eb/N0 (dB)





ZM lT

 


t lT
Figure 78: Bit Error Probability for Coherent Demodulation of Orthogonal ψM T t 2






Xs M lT






Signals




VI-99 VI-100
If signal 1 is transmitted during the interval l 1 T l T then





E cos θ ηc 1 m 1



 
Xc m l T

 


 
ηc m m 1


E sin θ ηs 1 m 1 The symbol error probability for noncoherently detection of orthogonal signals is



 
Xs m l T

 


 
ηs m

m 1 M
1 M
Pe s e Eb log2 MN0
∑ 1 m
e Eb m log2 MN0

 

 




M m


The decision statistic then (if signal 1 is transmitted) has the form m 2

Z1 l T E 2 E ηc 1 cos θ ηs 1 sin θ η2c 1 η2s 1 As with coherent demodulation the relation between bit error probability and symbol error
  



  




probability for noncoherent demodulation of orthogonal signals is
Z2 l T η2c 2 η2s 2
2k 1 M

 
Z3 l T η2c 3 η2s 3 Pe b Pe s Pe s



2k 1

"
2M 1



ZM l T η2c M η2s M






VI-101 VI-102

S y m bol Error P robability forN oncoherent D etection 1


of O rthogonal S ignals
Pb
10 0
10 1

         
P e,s

10 -1

10 2
M 2


10 -2

10 3 4
10 -3

10 4 8
10 -4

M =2

10 -5
10 5 16
10 -6

10 6
10 -7
M =8
10 7
10 -8

M =3 2
10 8
10 -9 M =4
M =1 6

10 -1 0 10 9 M ∞ M 32


0 2 4 6 8 10 12 14 16

10 10
Eb/N 0 (dB )
5 0 5 10 15
E¯b N0 (dB)

Figure 79: Symbol Error Probability for Nonocherent Detection of Orthogonal
Figure 80: Bit error probability of M-ary orthogonal modulation in an additive
Signals.
white Gaussian noise channel with noncoherent demodulation




VI-103 VI-104
 


A. Time-orthogonal (Pulse position modulation PPM)
B. Time-orthogonal quadrature-phase

2M
sin 2π f 0t iT M t i 1T M
ψi t T


 
  





 
sin 2π f 0t

0 elsewhere 2M 2iT
  "  t 2i 1T M
ψ2i t T M









 
M


0 elsewhere


i 01 M 1 f0 n

  

2T
"
2M
cos 2π f 0t 2iT
t 2i 1 T
ψ2i T M M

 




i 1T M 1


  
2M 2M


sin 2π f 0t sin 2π f 1t dt 0 elsewhere


  " 

T T



iT M
M M




i 01 1 M even f0 n
2 2T

"

M M
f0 n f1 n 1



2T " 2T






VI-105 VI-106


D. Frequency-orthogonal quadrature-phase
C. Frequency-orthogonal (Frequency Shift Keying FSK)

2E i
ψ2i t sin 2π f 0 t 0 t T

  "
T T

2E i



 



ψi t sin 2π f 0 t 0 t T


 

T 2T 2E i


 "


 


ψ2i t cos 2π f 0 t 0 t T


nM 1
T T








i 01 M 1 f0





2T
"" 

nM


f0
 2T




VI-107 VI-108
 E. Hadamard-Walsh Construction Example (M 4):


H2 H2
The last construction of orthogonal signals is done via the Hadamard Matrix. The Hadamard H4



matrix is an N by N matrix with components either +1 or -1 such that every pair of distinct H2 H2




rows are orthogonal. We show how to construct a Hadamard when the number of signals is a 1 1 1 1
power of 2 (which is often the case).


1 1 1 1



 



Begin with a two by two matrix

"



1 1 1 1


 



1 1 1 1 1 1


H2





"
1 1 Example (M 8):




Then use the recursion
H4 H4
H8



H2 l 1 H2 l 1 H4 H4





H2l






"
H2 l 1 H2 l 1

H2 H2 H2 H2








Now it is easy to check that distinct rows in these matrices are orthogonal. The i-th modulated H2 H2 H2 H2




 



signal is then obtained by using a single (arbitrary) waveform N times in nonoverlapping time





H2 H2 H2 H2


 
intervals and multiplying by the j th repetition of the waveform by the jth component of the




H2 H2 H2 H2


i-th row of the matrix.






VI-109 VI-110

ψ1 t


1






T 8 T 4 3T 8 T 2 5T 8 3T 4 7T 8 T t
1

ψ2 t



1


1 1 1 1 1 1 1 1 T


t


1 1 1 1 1 1 1 1 1
ψ3 t






 
   
  





1 1 1 1 1 1 1 1 1


 

   





T


1 1 1 1 1 1 1 1





T 8 3T 8 5T 8 7T 8 t





"

1 1 1 1 1 1 1 1 1




   





1 1 1 1 1 1 1 1




 





1 1 1 1 1 1 1 1



 


1 1 1 1 1 1 1 1
ψ8 t




1


T


T 4 3T 4 t
1



Figure 81: Hadamard-Walsh Orthogonal Signals




VI-111 VI-112
 


Noncoherent Reception of
Hadamard Generated Orthogonal Signals
2 T cos 2π f ct W1




       
t iT M W2

 


2


W1 X1 X2 X3 X4 X5 X6 X7 X8


LPF

   

   
W3

   






2
Xi Y1 Y2 Y3 Y4 Y5 Y6 Y7 Y8
2
W4 W2 X1 X2 X3 X4 X5 X6 X7 X8

 
 
 

rt Choose 2


Process Y1 Y2 Y3 Y4 Y5 Y6 Y7 Y8


W5


Largest
2
W3 X1 X2 X3 X4 X5 X6 X7 X8

   
 

   
 

t iT M W6

   

   
 

2

Y1 Y2 Y3 Y4 Y5 Y6 Y7 Y8
LPF

     

W7


2




Yi W4 X1 X2 X3 X4 X5 X6 X7 X8

 

2
W8 Y1 Y2 Y3 Y4 Y5 Y6 Y7 Y8


2 T sin 2π f ct

Figure 82: Noncoherent Demodulator




VI-113 VI-114


Noncoherent Reception of
Hadamard Generated Orthogonal Signals
X1 WX 1

       

       

       



2
W5 X1 X2 X3 X4 X5 X6 X7 X8 X2 WX 5
   
 
   
 

   

   

 

2
Y1 Y2 Y3 Y4 Y5 Y6 Y7 Y8 X3 WX 2

2

W6 X1 X2 X3 X4 X5 X6 X7 X8


 


X4 WX 6

2
Y1 Y2 Y3 Y4 Y5 Y6 Y7 Y8



2 X5 WX 3

W7 X1 X2 X3 X4 X5 X6 X7 X8
 
 
 
 


 

   


2
Y1 Y2 Y3 Y4 Y5 Y6 Y7 Y8 X6 WX 7

 
2
W8 X1 X2 X3 X4 X5 X6 X7 X8

 
 
 

 

X7 WX 4


2
Y1 Y2 Y3 Y4 Y5 Y6 Y7 Y8 


X8 WX 8

Figure 83: Fast Processing for Hadamard Signals




VI-115 VI-116
 


Biorthogonal Signal Set

A biorthogonal signal set can be described as

s0 t Eφ0 t

 
 
 

 
If we define bandwidth of M signals as minimum frequency separation between two such s1 t Eφ1 t
signal sets such that any signal from one signal set is orthogonal to every signal from a
frequency adjacent signal set are orthogonal then for all of these examples of M signals the
sM 2 1 t EφM 2 1 t






bandwidth is

 

 



M sM 2 t Eφ0 t



W M 2W T !




2T
Thus there are 2W T orthogonal signals in bandwidth W and time duration T .
sM 1 t EφM 2 1 t










That is a biorthogonal signal set is the same as orthogonal signal set except that the negative
of each orthonormal signal is also allowed.. Thus there are 2N signals in N dimensions. We
have doubled the number of signals without changing the minimum Euclidean distance of the




VI-117 VI-118

Symbol Error Probability


signal set. For example:

1 1 1 1 Let H j be the hypothesis that signal s j was sent for j 0 M 1. The probability of correct



" 
is (given signal s0 sent)


1 1 1 1



   
  





1 1 1 1 Pc 0 P r0 0 r1 r0 rM 2 1 r0 H0
   

 


""  
  


 
 

!

∞





1 1 1 1 fs r0 Fn r0 Fn r0 M 2 1
dr0

 


B8






r0 0


1 1 1 1




   

where f s x is the denisty function of r0 when H0 is true and Fn x is the distribution of r1



1 1 1 1





when H0 is true.


1 1 1 1





1 1 2
1 1 1 1 fs x exp x E
 


σ 2π

   
2σ2


   

  

!

    
Let H j be the hypothesis that signal s j was sent for j 0 M 1. The optimal receiver does x E


Φ
"" 

Fs x


a correlation of the received signal with each of the M 2 orthonormal signals. Let r j be the σ

correlation of r t with φ j t . The decision rule is to choose hypothesis H j if r j is largest in 1 1 2

fn x exp x



absolute value and is of the appropriate sign. That is, if r j is larger than ri and is the same σ 2π 2σ2


!

x
sign as the coefficient in the representation of s j t . Fn x Φ
σ





VI-119 VI-120
 Bit Error Probability


The bit error probabiltiy for birothogonal signals can be determined for the usual mapping of
bits to symbols. The mapping is given as

000000 000 s0 t
where σ2 N0 2. The error probability is then

       


∞
000000 001 s1 t
M 2 1
Pe 0 1 fs r0 Fn r0 Fn r0 dr0


011111 111 sM t





2 1






r0 0

  


111111 111 sM 2 t s0 t


 

Using an integration by parts argument we can write this as 111111 110 sM 2 1 t s1 t


∞ 100000 000 sM 1 t sM 2 1 t
z2


2E 1

" 

Φz 1M 2 2


Pe s M 2 2Φ z exp dz

 




N0 2π 2


 





!
0


The mapping is such that signals with furthest distance have largest number of bit errors. An
error of the first kind is defined to be an error to an orthogonal signal, while an error of the
second kind is an error to the antipodal signal. The probability of error of the first kind is the
probability that H j is chosen given that s0 is transmitted ( j M 2) and is given by

Pe 1 P rj r0 r j r1 rj rM 2 1 rj 0 H0

"" 









!





VI-121 VI-122

∞
M 2 2
Fs r j Fs rj Fn r j Fn rj fn r j d r j











 





0

It should be obvious that this is also the error probability to H j for j M 2. The probability


of error of the second kind is the probability that HM 2 is chosen given that s0 is transmitted


and is given by

Pe 2 P r0 0 r1 r0 r2 r0 rM 2 1 r0 H0
 

""   


 
 
 

!


0
M 2 1 M 2 ∞ 2E 2E 1 z2
fs r0 Fn r0 Fn r0 dr0
Φz Φ 1M 2 2




z 2Φ z exp dz


∞










  
 2 N0 N0 2π 2

  





!
∞ 0


M 2 2
M 2 Fs r0 Fn r0 Fn r0 f n r0 d r 0








Notice that the symbol error probability is Pe s M 2 Pe 1 Pe 2 .


 








The bit error probability is determined by realizing that of the M 2 possible errors (all


equally likely) of the first kind, M 2 2 of them result in a particular bit in error while an



error of the second kind causes all the bits to be in error. Thus
M 2
 

Pe b Pe 1 Pe 2
 

2
M 2 ∞
Fs u Fs u Fn u Fn u M 2 2 fn u d u








2



 









VI-123 VI-124
 1 1


Pe , s
P e,b
1 0- 1 1 0- 1

1 0- 2 1 0- 2

1 0- 3 1 0- 3

1 0- 4
1 0- 4
M=4
1 0- 5
1 0- 5
1 0- 6
2 1 0- 6
1 0- 7 M=2,4
8 1 0- 7
1 0- 8

M=128
32 1 0- 8
1 0- 9 M=128 32 8
1 0- 9
1 0- 1 0
-4 -2 0 2 4 6 8 10 12 14 16 1 0- 1 0
Eb/N0(dB) -4 -2 0 2 4 6 8 10 12 14 16
Eb/N0(dB)

Figure 84: Symbol Error Probability for Coherent Demodulation of Biorthog- Figure 85: Bit Error Probability for Coherent Demodulation of Biorthogonal
onal Signals Signals




VI-125 VI-126


For example

1 1 1 1 1 1 1
Simplex Signal Set


1 1 1 1 1 1 1



 
   
  






1 1 1 1 1 1 1

 

   



Same as orthogonal except subtract from each of the signals the average signal of the set, i.e.





1 1 1 1 1 1 1





S8


1 M 1


M i∑0
 

1 1 1 1 1 1 1


si t si t si t i 01 M 1

   




 
"" 





1 1 1 1 1 1 1



 






When the orthogonal set is constructed via a Hadamard matrix this amounts to deleting the 1 1 1 1 1 1 1


 



first component in the matrix since the other components sum to zero. 1 1 1 1 1 1 1




These are slightly more efficient than orthogonal signals.




VI-127 VI-128
 1 1


Pe , s Pe,b
1 0- 1 1 0- 1

1 0- 2 1 0- 2

1 0- 3 1 0- 3

1 0- 4 1 0- 4

1 0- 5 1 0- 5

1 0- 6
1 0- 6

1 0- 7
1 0- 7

1 0- 8
1 0- 8 M=2
8 4 M=2
16 8 4
1 0- 9 32
32 16 1 0- 9

1 0- 1 0
1 0- 1 0
0 2 4 6 8 10 12 14 16 0 2 4 6 8 10 12 14 16
Simplex.data
Eb/ N0 (dB) Eb/ N0 (dB)

Figure 86: Symbol Error Probability for Simplex Signalling Figure 87: Bit Error Probability for Simplex Signalling




VI-129 VI-130


Multiphase Shift Keying (MPSK)

2π
si t A cos 2πf0t i λ 0 t T
 

M For this modulation scheme we should use Gray coding to map bits into signals.




Ac i cos 2π f ct As i sin 2π f ct


M 2 BPSK M 4 QPSK


This type of modulation has the properties that all signals have the same power thus the use of
where for i 01 M 1, nonlinear amplifiers (class C amplifiers) affects each signal in the same manner. Furthermore


" 
"

2πi if we are restricted to two dimensions and every signal must have the same power than this
Ac i A cos λ
 

M signal set minimizes the error probability of all such signal sets.
 
 

2πi
As i A sin λ (QPSK and BPSK are special cases of this modulation).
M

π M e E N0 4πE 2E


cos θeγ cos θ 1

2
cos θ



Pe s 1 1 Q dθ





π M 2π N0 N0











VI-131 VI-132
 P erform ance of M P S K M odulation B it Error R ate for M P S K


S y m b o l Erro r P ro b ab ility 10 0 10 0

B it Erro r P ro b ab ility
1 0 -1 1 0 -1

1 0 -2 1 0 -2

M=2 M=4 M=8 M = 16 M = 32

1 0 -3 1 0 -3

M = 2 ,4 M=8 M = 16 M = 32
1 0 -4 1 0 -4

1 0 -5 1 0 -5
0 4 8 12 16 20 24 0 4 8 12 16 20 24
E b /N 0 (d B ) Eb /N 0 (d B )

Figure 88: Symbol Error Probability for MPSK Signalling Figure 89: Bit Error Probability for MPSK Signalling




VI-133 VI-134


Pe,s
10-1
M-ary Pulse Amplitude Modulation (PAM)
10-2

si t Ai s t 0 t T 10-3






where 10-4

Ai 2i 1 M A i 01 M 1
10-5






"" 



Ei A2i
 

10-6

1 M 1 A2 M 1
M i∑0 M i∑0
  
 

2 10-7
E Ei 2i 1 M






10-8
M2 1 2
A


3 10-9

2M 1 6E 10-10


Pe s Q





M M2 1 N0 0 5 10 15 20 25 30



Eb/N0 (dB)

Figure 90: Symbol Error Probability for MPAM Signalling




VI-135 VI-136
 1


Pe,b 10-1

10-2

10-3
Quadrature Amplitude Modulation
10-4

10-5 For i 0 M 1



"" 
M=4 M=8 M=16
10-6 si t Ai cos 2π f ct Bi sin 2π f ct 0 t T




10-7
Since this is two PAM systems in quadrature. Pe 2 1 1 Pe 1 2 for PAM with M signals
10-8







10-9

10-10
0 5 10 15 20 25

Eb/N0 (dB)

Figure 91: Bit Error Probability for MPAM Signalling




VI-137 VI-138

where u t is called the lowpass signal. For general CPM the modulation is nonlinear so that
Bandwidth of Digital Signals:



the below does not apply. Also
∞
∑
In practice a set of signals is not used once but in a periodic fashion. If a source produces
ut In g t nT


 
symbols every T seconds from the alphabet A 0 1 M 1 with be representing the l th ∞


n




" 
"

letter ∞ l ∞ then the digital data signal has the form



where In is possibly complex and g t is an arbitrary pulse shape.



∞
st ∑ sbn t nT Note that while u t is a (non stationary) random process u t τ̂ where τ̂ is uniform r.v. on









l ∞ 0 T is stationary.
 

Note: 1) si t need not be time limited to 0 T . In fact we may design si t iM 0 1 so that si t





 !






is not time limited to 0 T . If si t is not time limited to 0 T then we may have intersymbol
Φu f F Eu t τ̂ u t τ̂ τ






 
interference in the demodulaton. The reason for introducing intersymbol interference is to


 
  

 
! 
1
”shape” the spectral characteristic of the signal (e.g. if ther are nonlinear amplifiers or other ΦI f G f 2
T

nonlinearities in the communication system).

where
2) The random variables bn need not be a sequence of i.i.d. random variables. In fact if we are ∞
using error-correcting codes there will be some redundancy in b 2 so that it is not a sequence of ΦI f ∑ E In In e j 2π f mT

 


m

 
  ∞


i.i.d. r.v. m


∞
G f F gt gt e j2π f t
dt


In many of the modulation schemes (the linear ones) considered we can equivalently write the

  
∞

signal as
Example: BPSK In 1 (i.i.d).
st Re u t e jωct











VI-139 VI-140
 4. Noise bandwidth


gt A cos ωct pT t 0 t T WN P S fc


   


  



    
1 m 0 where P is total power and S f c is value of spectrum f fc



δm 0


E In In m ∞
0 m 0



P S f df

 


∞


ΦI f

1
5. Gabor bandwidth
Φu f
4
 
A2 T
sinc2 ω ωc
T
2
sinc2 ω ωc
T
2 ∞
f fc 2 S f d f




∞
σ

 
WG


∞

 



∞S f df


6. Absolute bandwidth

WA min W : S f 0 f W


Definition of Bandwidth





1
1. Null-to-Null bandwidth (in Hz) of main lobe ( 2
for BPSK). 7. half null-to-null 2 null-to-null.


T


 
2. 99% containment bandwidth bandwith such that 1
lies above upper bandlimit 12 % 1 2 3 35dB 4 5 and 6 3 3dB
2%


lies below lower level. BPSK 2.0 20.56 35.12 1.00 ∞ 0.88
3. x dB bandwidth Wx bandwidth such that spectrum is x dB below spectrum at center of QPSK 1.0 10.28 17.56 0.50 ∞ 0.44


band (e.g. 3dB bandwidth). MSK 1.5




VI-141 VI-142

Comparison of Modulation Techniques


2Eb
BPSK has Pe s Q N0





1 1 R
W R 10





"
QPSK for same date rate T bits/sec T T W



Eb N0 9 6dB

"

R
QPSK has same Pe but has W 20


"
2 1
R Ts R T




or
1 1
W
 Ts W 2T







VI-143 VI-144
M-ary PSK has same bandwidth as BPSK but transmits log2 M bits/channel use (T sec).


E b /N 0 (d B )
36

M-ary PSK
30

log2 M A chie v ab le
R 24


T R e g io n
R


log2 M 18


W




1 12
W





T




6
Capacity (Shannon Limits)




0




















U nachie v ab le
R Eb R e g io n







-1
R W log2 1
W N0





















 

-2
or






1 0 -3 1 0 -2 1 0 -1 1 101 102
2R W 1


Eb N0

R W 


R /W (b its/se c/H z)

We can come close to capacity (at fixed R W ) by use of coding (At R W 1 there is a Figure 92: Capacity of Additive White Gaussian Noise Channel.



possible 9 6 dB ”coding gain”)
"




VI-145 VI-146