Zero-forcing Solution

It is done by disposing off top and bottom rows, thus, making X to be square
matrix. This zero-forcing solution minimizes the peak ISI distortion by selecting
the C n  weights so that the equalizer output is forced to zero at N (say)
sample points on either side of desired pulse. In other words:
Z k   1 for k  0
0 for k  1,2,....,  N

Example: Consider that the tap weights of an equalizing transversal filter are to
be determined by transmitting a single impulse as a training signal. Let the
equalizer filter be made up of just three taps. Given a received distorted set of
pulse samples xk  , with the voltage values 0.0,0.2,0.9,0.3,0.1 as shown in the
following Figure we use a zero forcing solution to find the weights c1, c0 , c1
that reduce the ISI values of the equalized pulse samples zk  have the value
z1  0, z0  1, z1  0. Using these weights calculate the ISI values of the
equalized pulse at the sample times k  2,3 . What is the largest magnitude
sample contributing to ISI and what is the sum of all ISI magnitudes?

Sol.

Figure 9.40: Distortion in pulse after passing through channel

Sample values are  0.0, 0.2, 0.9,  0.3, 0.1

  0.0 0 0 
 0.2 0.0 0 

 0.9 0.0 0.0 
 
 X   0.3 0.2 0.2 
 0.1 0.9 0.9 
 
 0  0.3  0.3
 0 0.1 
 0.1

Hence, in order to make it square, the upper two rows and bottom two
rows are disposed off. This process will ensure ISI free samples at one sample
position at either side of center (note that if we do not want to dispose off the
top and bottom rows, then, we need a filter with seven tap coefficients, then,
we will ensure ISI free condition at three sample positions at either side of the
center).

Hence,

 0.9 0.2 0.0

X   0.3 0.9 0.2

 0.1  0.3 0.9

Now, Z  XC

0  C 1 
 Z  1  , X is given, C   C 0  is to be found.
0  C1 

0  0.9 0.2 0 C 1 

 1   0.3 0.9 0.2 C 
     0
1  0.1  0.3 0.9  C1 

C 1   0.2140
  C    0.9631   This value of coefficient will make the received
 0  
 C1   0.3448 
pulse free from ISI only at two time instants, that is, at  T & T .

Hence, ISI at all sample times k  3,2,1,0,1,2,3 are computed by using above
coefficient values.
  z  3  0 0 0 
 z  2  0.2 0 0 
  
 z  1   0.9 0.2 0   0.2140
     0.9631 
 z 0    0.3 0.9 0.2   
 z 1   0.1  0.3 0.9   0.3448 
   
 z 2   0 0.1  0.3
 z 3   0 0.1 
   0

This comes out as

z 3  0, z 2  0.0428, z 1  0, z0  1, z1  0, z2  0.0071, z3  0.345 .

Note

(i) Here, sample of greatest magnitude is 0.0428 and total ISI magnitude
is equal to  0.0844.
(ii) Here, it should be clear that this three-tap equalizer has forced the
sample points on either side of equalized pulse to be zero. If
equalizer is made longer than three-taps, more of the equalized
sample points can be forced to a zero value.

Adaptive Equalization
 Figure 9.41: Adaptive Equalization

In this, we develop algorithm for adaptive equalization of linear channel of

unknown characteristic. In this, to adjust the tap as per unknown
characteristics of channel, we send the trainee sequence (PN Sequence) of
known characteristic. This received trainee sequence is passed through
equalizer and its output is subtracted from the same PN sequence (the trainee
sequence that was transmitted) and error is fed back to equalizer to adjust its
taps. When the trainee process is completed, the PN sequence generator is
switched off and the adaptive equalizer is ready for normal data transmission.

(i) Least-Mean Square Algorithm

Let
x nT   xn 
y nT   yn 

where, N  1 is the total number of taps.

Now, let an be the desired response. Let en be error defined as

en  an  yn

 In the least-mean-Square (LMS) algorithm for adaptive equalization, the
error signal actuates the adjustment applied to the individual tap weights of
the equalizer as the algorithm proceeds from one iteration to the next.

The formula for LMS algorithm in words is as follows;

Old value  Input signal 

 Updated Value   th  Step - Size   
 th   of k tap      applied to   error Signal
of k tap weight   weight  parameter   th 
  k tap weight 

If   Step size parameter

xn  k   The input for k th tap at the time clock n

Then,

ˆ k n  1  ˆ k n  xn  k en, k  0,1,2,.... N

where,

̂ k n  Tap value at n th clock

ˆ n 1  Updated tap value at n  1th clock.

Here from the Figure 9.41, it is clear that

N
en  an   ˆ k n.xn  k 
k 0

The above may be simplified in the following way:-

Let

x n  xn,..., xn  N  1, x[n  N ]T

 n  ˆ 0 n, ˆ1n,...., ˆ N nT

Then, yn  x T n  n

Where, xT n. n is called inner product of vectors xn and  n.

Hence, we can summarize the LMS algorithm for adaptive equalization as

follows:
 (1) Initialize the algorithm by setting w1  0 (That is, set all the tap-weights
of the equalizer to zero at n  1.
(2) For n  1,2,.... compute
yn  x T n.wˆ n
en  an  yn

Wˆ n  1  Wˆ n  enxn

(3) Continue the iteration until equalizer reaches a steady state.

Unit III

Spread Spectrum

In spread spectrum, a signal (Let us say, binary wave) is multiplied with PN sequence (to be

discussed later) which has a very large bandwidth. Hence, the spectrum of resulting signal

will occupy a very large BW. Since total power remains fixed, thus, PSD of signal falls down

and its level goes near noise level which has spread in the complete range of frequency.

Hence, it appears that signal has hidden itself or submerged into an ocean of large BW

spectrum of noise. Since the PN noise has got very wide BW and hence, this transmitted

signal blends itself into this AWGN noise background so as to remain undetected by

Jammer.

One more justification can be given as to why spread spectrum is wide. We know

that from convolution theory if we multiply the two signal in time domain, then they will be

convolved in frequency domain. Now bt  is narrowband and ct  is wideband PN

Sequence. The multiplication of both the signals in time-domain results in convolution of their

frequency counterpart given as

bt .ct   B f   G f  (1)

So, convolution of narrow band with the wideband signal results in wideband signal.

Therefore, we can divide the complete process of spreading and dispreading of signal into

three steps given as

Figure 1: Spreading and de-spreading

mt   bt ct  (2)

r t   mt   it   bt ct   it 

This is the received signal r t  . Now, for detection purpose, just reverse process takes place

i.e. r t  should, first, be multiplied by ct  .Hence, we have

zt   r t   ct 

 bt ct   it ct   bt c 2 t   it ct 

zt   bt   it ct  (3)

We see that bt  is a narrow band signal and it  is a wideband signal (as it ct  is also a

kind of spread spectrum on it  ). Hence, low pass filter should be used to filter out bt  .

Here we note that ct  is a PN sequence having amplitude+1 and -1. Hence, c 2 t   1 .

Consider the general case of communication system in which message signal is occupying

only certain narrow BW and it is added with AWGN noise.

Figure 3: Spreading and De-spreading

Hence, signal can easily be detected using suitable receiver. Now If the signal is spreaded

by multiplying with PN sequence, the situation looks like as shown in Figure 3(a). It may be

noted that if the signal b(t) is multiplied in time-domain with C(t), then, in frequency domain,

this signal will be convolved with PN sequence as shown in Figure 3(a). Hence, signal

spectrum has spread in a wide range spectrum and after being added with AWGN noise, as

shown in Figure 3(b), it appears that it has hidden itself into the noise background just like

grasshopper camflouges itself in a grass. Had the grasshopper been different from grass
color, it could easily be detected. Further, in Figure 3(c), this signal plus noise is being

multiplied with the de-spreading code and resultant signal will consist of base-band signal

(low BW) plus noise (Wide BW) as given in (3). Using Low pass filter, this baseband signal

can easily be filtered out.

Direct-sequence spread spectrum with Coherent BPSK

Figure 4: DS-Spread Spectrum with Modulation

Here, we note that in the receiver stage, there are two-stage modulations. First stage is to

bring from pass band signal to baseband signal. Hence at (6), we get the distorted version of
spread sequence mt  . To get the distorted version of binary wave, this is to be multiplied

by PN sequence. The reason for using LPF in first stage demodulation is that even mt 

which is also equal to ct bt  is also a baseband signal (or low pass signal), though, it has

got wide spectrum, yet it is centered at zero. Hence, LPF is required.

Here, we note a very important point. The spreading is followed by modulation (Like BPSK,

QPSK). Now these two operations can also be interchanged to each other. The reason is

that these two operations (i.e. spreading and modulation) are linear processes. This is

shown as follows:

Let m1 t   b1 t c1 t  and m2 t   b2 t ct 

Hence,

mt   b1 t   b2 t ct 

 b1 t ct   b2 t ct 
 m1 t   m2 t 
(1)

x1 t   m1 t Cos c t x2 t   m2 t Cos c t
In a similar way, and

 xt   m1 t   m2 t Cosc t  m1 t Cosc t  m2 t Cosc t  x1 t   x2 t 

Hence, it proves that spreading and modulation both are linear process. Now, we know that

linear process can be interchanged without affecting the result. Hence it is concluded that

these two processes can be interchanged. In a similar way, it can be shown that at receiver

side, dispreading and demodulation both the process can also be interchanged.

Hence, the resultant block diagram is as follows

 Fig. 5

Here corresponding equations are

yt   xt   j t  (2)

Here, let us ignore the presence of noise as we are concerned about Jammer’s interference

and hence we are trying to see the effect of Jammer’s interference only.

ut   yt .ct 

  x(t )  j (t ) c  t   x  t  c  t   j  t  c  t 
  s  t  c 2  t   j  t  c  t  

 st   j t ct  (3)

Here, c t   1
2

Here note that Jammer interference j t  is a pass band signal. Now, if j t  be multiplied

with ct  , the resulting term j t ct  will also be passband. Now, S t  is baseband. Hence,

st  could easily be filtered out using low pass filter.

Pseudo-noise sequence
What is Pseudo-noise sequence? It is the periodic sequence of binary wave which appears

like random binary wave. Its appearance as binary random wave but not exactly equal to

random wave results in the word like pseudo noise “.

How is it generated? It is generated using the shift register with some kind of feedback

combinational logic circuit which decides the kind and period of PN sequence.

Consider the following feed back shift register:

Fig. 7: PN Sequence generator

This is shift-register with length m. We first choose the some initial state of shift register then,

after clock is applied, logic circuit computes the output which will shift the content of first FF

when clk is applied . In this way , this process goes on till the same state of FF is repeated .

At the best, the number of possible state of shift register is 2 m  1 where m is the number of

FF in shift register. Here, 1 is subtracted because we have excluded one possible state of all

zero. If the initial state of shift register is all zero then next state will also be zero. In this way

only this zero loop will go on till infinite so, this situation is avoided.

If the state of the FF is given as s1 k , s 2 k ....s m k   where k  0

Then, as per definition of shift register, the state of j th FF at k  1 clock, is given by

th
 k 0
S j (k  1)  S J 1 (k ) (6)
1 j  m

Hence, PN sequence is the output of last F-F (that is, mth FF) of shift Register.

Note:

 A feedback shift Register is said to be Linear when the feedback logic consists

entirely of Modulo-2 adders.

 A period of shift Register will always be less that or equal to 2 m  1. When the period

is exactly 2 m  1, the PN sequence (obtained at output of mth FF) is called a

Maximal-Length Sequence or simply m-sequence. It may be noted that the period

of PN sequence may be less that 2 m  1 depending on the type of combination logic

used. Hence, there are only specific number of combinational logic circuits that

corresponds to maximal-length PN sequence. The list of such Filter will be given in

next section.

Example: 1 Consider the following shift

Register for m=3. Find the PN sequence assuming initial state of the shift register to be (i)

100. (ii) 111.

Modulo 2
Adder

Output, PN
1 2 3
S0 S1 Sequence
S2
 Fig. 8

Hence the following is the list of all state of the shift register:

Clock Sate of Shift

register

1. 100

2. 110

3. 111

4. 011

5. 101

6. 010

7 001

8. 100

…..

Hence the output PN sequence is 0011101=7 bits. We see that this is equal to 23  1  7 .

For initial state 111, the PN sequence is obtained as 1110100. This

signal waveform is as shown in Fig. 9. We note that the waveform repeats itself after seven

chips. This duration is called period of PN sequence. Here chip is the duration of the

individual pulse of this waveform. Thus, we note this example that there are seven chips in

one PN sequence period. Fig. 10 shows the spread sequence which is obtained by

multiplying the message (101) signal with the PN sequence. Here, we have assumed
Tb=NTc. After understanding how a PN sequence is generated, we are going to discuss that

the some PN sequence satisfies some properties of run-length. These sequences are known

as maximal-length sequence.

Fig. 9: PN Sequence with initial state 111

Fig. 10: Plot of message bit [1 0 1], PN sequence for shift Register [3,1] with initial

state 10000 and spread sequence.

Properties of Maximal Length Sequence

1. The number of 1 is always greater than number of 0 by 1. For example in above

case, total number of 1 is equal to 4 and total number of 0 is 3.

2. Count total run-length. Run-Length means a subsequence of identical sumbols (1

and 0) within one period.

Total number of run length of 1 bit=1/2 * Total Run Length

Total number of run length of 2 bit=1/4 * Total Run Length

Total number of run length of 3 bit=1/8* Total Run Length

For example, in above example, total Run Length=4

Hence run Length of 1 bit= 1/2*4=2

Run Length of 2 bits=1/4*4=1

Run Length of 3 bit= 1/8*4 can not be a fraction and hence 1

3. The autocorrelation function of PN sequence is Binary valued (that is having two

levels. One value is 1 and another is –1/N). This is periodic in nature.

Considering the Fig. 6 once again, it may be noted that the capacity (number of users) of the

system depends upon the cross-correlation between the two sequences. Moreover, it is also
important to understand the auto-correlation between two PN sequences of the same type

as the PN code at the Rx of Tx-Rx system, if slightly desynchronized, may results in loss of

transmitted message.

Auto-correlation function of Maximal-length sequence

Example: This is an example which gives the cross correlation between and maximal-length

sequence [3,1] and non-maximal length sequence [3,2]. Generate two PN sequences using

two different shift Registers shown in Fig. 15.

Shift Register I:

Tap=[3,1], Polynomial equation : X 3  X  1 , Initial state=[1 0 0]

Output of PN sequence thus found is : 0 0 1 1 1 0 1

Fig. 15

Shift Register II:

Tap: [3,2], Polynomial Equation: X 3  X 2  1 , Initial state: [1 1 1]

Output of PN sequence: [1 1 1 0 0 1 0];

Find autocorrelation and cross correlation for these two PN sequences.

Solution:

Autocorrelation: For autocorrelation we can chose any of the two PN sequences and result

will be same. Let us chose the first sequence and find its autocorrelation with itself.

 0 C1 (t ) = 0 0 1 1 1 0 1

C1 (t ) = 0 0 1 1 1 0 1

0000000

Hence R11(0)=(7-0)/7=1

 1 C1 (t ) = 0 0 1 1 1 0 1

C1 (t ) = 1 0 0 1 1 1 0

1010011

Hence R11(1)=3-4/7=-1/7

Similarly, we can find that

R11(2)=-1/7, R11(3)=-1/7, R11(4)=-1/7, R11(5)=-1/7, R11(6)=-1/7, and R11(7)=1,

Cross Correlation:

 0 C1 (t ) = 0 0 1 1 1 0 1
 C 2 (t ) = 1 1 1 0 0 1 0

1101111

Hence R12(0)=1-6=-5

 1 C1 (t ) = 0 0 1 1 1 0 1

C 2 (t ) = 0 1 1 1 0 0 1

0100100

Hence R12(1)=5-2=3

 2 C1 (t ) = 0 0 1 1 1 0 1

C 2 (t ) = 1 0 1 1 1 0 0

1000001

Hence R12(2)=5-2=3

 3 C1 (t ) = 0 0 1 1 1 0 1

C 2 (t ) = 0 1 0 1 1 1 0

0110011

Hence R12(3)=3-4=-1
 4 C1 (t ) = 0 0 1 1 1 0 1

C 2 (t ) = 0 0 1 0 1 1 1

0001010

Hence R12(4)=5-2=3

 5 C1 (t ) = 0 0 1 1 1 0 1

C 2 (t ) = 1 0 0 1 0 1 1

1010110

Hence R12(5)=3-4=-1

 6 C1 (t ) = 0 0 1 1 1 0 1

C 2 (t ) = 1 1 0 0 1 0 1

1111000

Hence R12(6)=3-4=-1

 7 C1 (t ) = 0 0 1 1 1 0 1

C 2 (t ) = 1 1 1 0 0 1 0

1101111

Hence R12(7)=1-6=-5
 Fig. C

Hence once again considering the Fig. B, and assuming that C1(t) at Tx and Rx are perfectly

synchronized, the total signal at input of Rx is 1+(-0.7)=0.3

We note that -0.7 is because of cross-correlation between C2(t) and C1(t) at Rx1.

Frequency Hopped Spread Spectrum (FHSS)

Frequency Hopping Spread Spectrum (FHSS) is a transmission technology used in wireless

transmissions where the data signal is modulated with a narrowband carrier signal that
"hops" in a random but predictable

sequence from frequency to frequency as a function of time over a wide band of

frequencies. This technique reduces interference because a signal from a narrowband
system will only affect the spread spectrum signal if both are transmitting at the same
frequency at the same time. If synchronized properly, a single logical channel is maintained.

Advantage of FHSS:

 Fundamentally much simpler to implement

  Better range, due to lower receiver sensitivity
 Good rejection of in band interference
 Good performance in multipath environments
 No "near/far" problems

FHSS can be thought of as a two steps modulation process: the data modulation and

frequency-hopping modulation even though it can be implemented as a single step whereby

frequency synthesizer produces a transmission tone based on simultaneous dictates of PN

code and data for a given hop, the occupied transmission BW is identical to the BW of

conventional modulation scheme, which is typically much smaller than WSS (total frequency

hopped spectrum).

In the following example, we determine the minimum number of chips necessary

required to generate frequency tone if the frequency hopping BW is WSS and minimum

frequency spacing between consecutive hop positions ∆ f .

Example: A hopping BW WSS of 400MHz and frequency step size f of 100Hz. What is

the minimum number of PN chips that are required for each frequency word?

Sol.: W SS =400MHz  4  108 Hz.

f  100 Hz

If n is the number of possible frequency tones, then,

Since number of frequency tones in BW of WSS is given as

4  108 Hz
n  4  106 Hz
100

=No. of possible hops.

Hence, synthesizer output can have 4106 Hz frequency output signals.

Hence, if the number of chips required is m, then,

2 m  4  10 6  Min No. of Chips, m  22

Definition of chip in FHSS

Referring to DS-spread spectrum, the chip is defined as the shortest duration pulse between

PN sequence and data bit. Since the duration of PN sequence pulse is lesser than that of
data pulse and hence chip is defined as the PN sequence pulse width. Proceeding with the

same line, for FHSS, chip is defined as the shortest duration pulse between the symbol

duration and hopping tone duration. Hence, if the hopping ton duration is less than the

symbol duration (i.e. the case of fast FHSS), then chip is defined as the period of hopping

done. On the other hand, if the symbol duration is less than hopping tone duration (the case

slow FHSS), the chip will be defined as the symbol duration. Hence, in summary,

Tc  min(Ts , Th )

where Ts is the symbol duration and Th hopping tone duration. The chip rate is given as

Rc  max( Rs , Rh )

Consider an example of symbol rate as 40 symbols/s. Hence Ts=1/40 s and let us consider

a hop rate as 10 hopps/s with Th=1/10 seconds. It implies that in one hop duration, there will

be four symbols (slow FHSS case).

Note that the frequency separation between the two consecutive symbols should be Ts or its

multiple to ensure that the symbols are orthogonal to each other. For fast FHSS let us
consider hopping rate as 80 hops/s with Th=1/80 s. Therefore, corresponding to one symbol,

there will be two hopping tone as shown below.

Slow Hopping versus Fast Hopping

Depending on the symbol rate of modulator output at ‘1’ and the rate frequency tone

generation of frequency synthesizer at ‘2’ in Figure 2, frequency hopping can be

categorized in two parts: (i) Slow hopping (ii) Fast hopping.

Slow frequency hopping

If Th is the hopping duration and Ts is the symbol duration, then for slow frequency hopping,

the requisite condition is that

Ts≤Th

Or

Rs≥Rh
where Rs and Rh are the symbol and hopping rate.

That is, in one hopping duration, there will be multiple symbol duration. This is shown in the

Figure 3.

Fast frequency hopping

For fast frequency hopping, the requisite condition is that

Th≤Ts

or

Rh≥Rs

where Rs and Rh are the symbol and hopping rate.That is, in symbol duration, there will be

multiple hopping.

Fig. Example of (a) slow FHSS (b) fast FHSS.

Example:1 In an FHHS system, a total hopping BW of 100MHz and

frequency spacing of 10KHz is used. Calculate the minimum number of chips
required for generation of each frequency tone signal.
Sol. Number of frequency sub-band=100MHz/10kHz=104

If n is the number of chips  2  10  n  13 Chips

n 4

Example 2:

Let us consider a communication system transmitting the message bits at the

rate of 120 kbps and the modulation scheme used 32-FSK. A hopping rate of
2000 hops/s is used over an total available spectrum of 10MHz. Calculate

(a) Data symbol transmitted per hop.

(b) The number non-over-lapping hopping sub-band.

Sol.

Rb  120kbps

If n be the number of bits to be mapped to one symbol of 32 ary FSK,

then, 2n  32  n  5 . Hence, symbol rate=120/5=24ksps=24,000sps

Hopping rate =24,000/2000=12Symbols/hop (a case of Slow-FHSS)

For M-ary FSK scheme, the minimum separation between two frequency tones
such that they remain orthogonal is 1/ TS .Further assuming that the BW occupied
by each symbol is equal to RS is symbol rate, the minimum BW one sub-band of
12 symbol (corresponding to one hop) is given by

Bmin  12  24 kHz  288 kHz

Hence, total number of non-overlapping hopping subband is

10MHz
N  35
288KHz

Example 3:

Let us consider a FHSS where the input data rate is 200 bits/s and the
modulation scheme used is 32-ary FSK. The frequency hopping rate is 200
hops/s. Calculate

(a) Minimum separation between frequency tones.

(b) Number of frequency tones produced by frequency synthesizer.
(c) Processing gain if hopping BW is 10 MHz.
Sol:

200
Rb  200bits/S, 2 k  32  k  5  RS   40 Symbol/S
5
Rh  200bits/S,  Th  1  5 ms
200

Since RS  Rh  a case of fast frequency hopping.

200
Number of hops/symbol   5 hops/symbol
40

25ms
Ts  1 Seconds= 25ms  chip duration   5ms
40 5

Note that chip duration is equal to min TS , Th  . Minimum separation between
tones is 1/5ms=200Hz. Processing gain

 BW h 10  106
   5  104
R 200