Documentos de Académico
Documentos de Profesional
Documentos de Cultura
com
UNIT I
SIGNALS AND SYSTEMS
1. What are the basic elements of DSP and its requirements
Ans. The basic elements of digital signal processing system are shown
in fig. below.
is measured as
www.5starnotes.com
Page 1
www.5starnotes.com
Ans.
S.Leena
Page 2
www.5starnotes.com
www.5starnotes.com
S.Leena
www.5starnotes.com
Page 3
www.5starnotes.com
S.Leena
www.5starnotes.com
Page 4
www.5starnotes.com
The input x(n) will be bounded if I x(n) is less than some finite
number. Let us denote this finite number by Mx. Thus for input signal
to be bounded
Here Mx is finite number, so its values should be less than infinity.
Thus eq. (2) can be written as
Now taking absolute value of both sides of eq. (1)
S.Leena
www.5starnotes.com
Page 5
www.5starnotes.com
We will read R.H.S. of eq. (4) as absolute value of summation of
terms. If we take
sign outside then the term become
Thus:
values of terms. Always absolute values of sum of terms is less
than or equal to the sum of absolute value of terms.
Here h(k)= h(n) is the impulse response of LTI system. Thus eq. (9)
gives the conditions
of stability in terms of impulse response of the system.
S.Leena
Page 6
www.5starnotes.com
www.5starnotes.com
S.Leena
www.5starnotes.com
Page 7
www.5starnotes.com
S.Leena
www.5starnotes.com
Page 8
www.5starnotes.com
. Find
5. An LTI system is described by
the response of this system for an input of x(n) = 10 cos (0.05
3m).
Ans. Given
S.Leena
www.5starnotes.com
Page 9
www.5starnotes.com
S.Leena
www.5starnotes.com
Page 10
www.5starnotes.com
S.Leena
www.5starnotes.com
Page 11
www.5starnotes.com
6. Find the signal energy of the signal x(t) = U(t) U(10 t)?
Ans. Energy of input sequence x(t) is given by
S.Leena
Page 12
www.5starnotes.com
www.5starnotes.com
Linearity
S.Leena
www.5starnotes.com
Page 13
www.5starnotes.com
1.
Shifting:
1.
5.Time Reversal:
6. Conjugation:
7. Convolution:
8. Initial value:
9. Final value:
S.Leena
www.5starnotes.com
Page 14
www.5starnotes.com
9. Determine the z-fransform of the signal
Ans. By using Eulers identity the signal x (n) can be expressed as
S.Leena
www.5starnotes.com
Page 15
www.5starnotes.com
S.Leena
www.5starnotes.com
Page 16
www.5starnotes.com
Ans. Given:
S.Leena
www.5starnotes.com
Page 17
www.5starnotes.com
By using long division method.
Ans. Given:
S.Leena
www.5starnotes.com
Page 18
www.5starnotes.com
Ans.
S.Leena
www.5starnotes.com
Page 19
www.5starnotes.com
S.Leena
www.5starnotes.com
Page 20
www.5starnotes.com
Thus the ROC is
because the ROC is the common region where
both sum
are finite. The fig shows the ROC of the z-transform of each of the
individual terms & for combined signal.
S.Leena
Page 21
www.5starnotes.com
www.5starnotes.com
(b)When the ROC is interior of the circle, the signal x (n) is anticausal
signal. Thus we divide so as to obtain a series in power of z as follows.
S.Leena
www.5starnotes.com
Page 22
www.5starnotes.com
Ans. Given
S.Leena
Page 23
www.5starnotes.com
www.5starnotes.com
S.Leena
www.5starnotes.com
Page 24
www.5starnotes.com
Ans. Given
S.Leena
www.5starnotes.com
Page 25
www.5starnotes.com
S.Leena
Page 26
www.5starnotes.com
www.5starnotes.com
UNIT II
FREQUENCY TRANSFORMATIONS
1. Discuss various properties of DFT.
Ans. Properties of the DFT.
1.
2. Periodicity
S.Leena
Page 27
www.5starnotes.com
www.5starnotes.com
and
i.e.,
S.Leena
www.5starnotes.com
Page 28
www.5starnotes.com
Here we wish to determine the sequence x3 (n) for which the DFT is,
S.Leena
Page 29
www.5starnotes.com
www.5starnotes.com
S.Leena
Page 30
www.5starnotes.com
www.5starnotes.com
S.Leena
www.5starnotes.com
Page 31
www.5starnotes.com
Proof.
S.Leena
Page 32
www.5starnotes.com
www.5starnotes.com
If we change the index from n to m = N + n - 1, then
S.Leena
Page 33
www.5starnotes.com
www.5starnotes.com
where
Proof.
2. Parsevals theorem:
For the complex-valued sequence x (n) and y (n), if
S.Leena
Page 34
www.5starnotes.com
www.5starnotes.com
www.5starnotes.com
Page 35
www.5starnotes.com
S.Leena
Page 36
www.5starnotes.com
www.5starnotes.com
S.Leena
www.5starnotes.com
Page 37
www.5starnotes.com
www.5starnotes.com
Page 38
www.5starnotes.com
means N = 8 So considermg (17) & (18) (that means N/2=4) we can
obtain N (8-point)
DFT. This is first stage of decimation. Note that eq. (17) indicate 4
(N/2) point OFT of IN\
g(n) and eq (18) indicates 4 (N/2) pomt DFT of h(n) For 8 pomt OFT
eq (15) becomes
Here we are computing 4 point DFT, So range of n is n=0 to n=3.
Putting these
values in eq. (19), we get
Using equations (20) & (22) & eq. (17) & (18) we can draw the flow
graph of the
first stage of decimation as shown in fig. below.
S.Leena
Page 39
www.5starnotes.com
www.5starnotes.com
Second stage of decimation: In the first stage of decimation we have
used 4-point DFT. We can further decimate the sequence by using 2
point DFT. The second stage of decimation is shown in fig below.
4.Discuss DIT and DIF algorithms and also compare the two
algorithms.
S.Leena
www.5starnotes.com
Page 40
www.5starnotes.com
Ans. Radix-2 Decimation in Time (DIT) Algorithm (DIT FFT) : To
decimate means to break into parts. Thus DIT indicates dividing
(splitting) the sequence in time domain. The different stages of
decimation are as follows:
First stage of decimation Let x(n) be the given input sequence
containing N samples. Now for decimation in time we will divide. x (n)
into even and odd sequences.
Since we have divided x(n) into two parts, wq can write separate
summation for even and odd sequences as follows:
S.Leena
www.5starnotes.com
Page 41
www.5starnotes.com
Now F1(k) and F2(k) are 4-point (N/2) DFTs. They are periodic with
period N/2.
Using periodicity property of DFT we can write,
S.Leena
Page 42
www.5starnotes.com
www.5starnotes.com
S.Leena
www.5starnotes.com
Page 43
www.5starnotes.com
www.5starnotes.com
Page 44
www.5starnotes.com
We can further decimate f1(m) into even and odd samples. Let g11(n)=
f1(2m), which contains even samples and let g12(n) = f1(2m +1),
which contains odd samples of f1(m).
Note that here range of n and m is from 0 to N/4-1
Now recall equations (15) and (16). We obtained sequences X(k) and
X(k+N/2)
from F1(k) and F2(k). The length of each sequence was N/2. Here in
the second stage of
decimation. We are further dividing the sequences into even and odd
parts. So similar to Equations (15) and (16) we can write; For F1(k),
Here the values of K are 0 and 1. That means it is 2-point DFT. Thus
Equations (23) and (24) shows that we can obtain 4-point DFT by
combining two 2-point DFTs. The graphical representation is shown in
Fig.
S.Leena
www.5starnotes.com
Page 45
www.5starnotes.com
Now similar to Equations (21) and (22) we can write equations for
F2(k) as follows:
S.Leena
Page 46
www.5starnotes.com
www.5starnotes.com
The graphical representation of Equations (28) and (29) is shown in
Fig.
S.Leena
www.5starnotes.com
Page 47
www.5starnotes.com
We will use Equation (31) to compute 2-point DFT. From Fig. consider
the first block of 2-point DFT. It is separately drawn as shown in Fig.
S.Leena
Page 48
www.5starnotes.com
www.5starnotes.com
varies from 0 to 1. Now the output sequences are G11(0) and G11(1).
We can denote it by G11(k); where k varies from 0 to 1. Here G11(k)
is DFT of g11(n).
Thus for G11(k) we can write Equation (31) as,
www.5starnotes.com
Page 49
www.5starnotes.com
Similarly for other 2-point DFTs we can draw the butterfly structure.
Total signal flow-graph for 8-point DIT FF1
The total signal flow graph is obtained by interconnecting all stages of
decimation. In this case, it is obtained by interconnecting first and
second stage of decimation. But the starting block is the block used to
compute 2-point DFT (butterfly structure). The total signal flow graph
is shown in Fig.
S.Leena
Page 50
www.5starnotes.com
www.5starnotes.com
5. Draw a 8 point radius 2 FF1 DIT flow graphs and obtain DF1
of the following
sequence
Ans.
S.Leena
Page 51
www.5starnotes.com
www.5starnotes.com
S.Leena
www.5starnotes.com
Page 52
www.5starnotes.com
S.Leena
Page 53
www.5starnotes.com
www.5starnotes.com
The three sample sequence and its periodic extension are shown in
Fig. below.
S.Leena
www.5starnotes.com
Page 54
www.5starnotes.com
S.Leena
www.5starnotes.com
Page 55
www.5starnotes.com
Ans. The input sequence has length L = 4 and the impulse response
has length M = 3. Linear convolution of these two sequences produces
a sequence of length N = 6. Consequently, the size of the DFT must be
at least six.
For simplicity we compute eight point DFTs. We should also mention
that the efficient computation of the DFT via the Fast Fourier
Transform (FFT) algorithm is usually performed for a length N, that is
power of 2. Hence eight point DFT of x(n) is
S.Leena
www.5starnotes.com
Page 56
www.5starnotes.com
S.Leena
www.5starnotes.com
Page 57
www.5starnotes.com
First stage of computation
The input sequence = {2, 2, 2, 2, 1, 1, 1, 1)
The phase factors involved in first stage
of
computation
are
S.Leena
Page 58
www.5starnotes.com
www.5starnotes.com
S.Leena
www.5starnotes.com
Page 59
www.5starnotes.com
For 8-point DFT by radix 2 FFF we require 3 stages of computation wits
4 butterfly computations in each stage. The sequence rearranged in
the bit reversed order forms the input to the first stage. For other
stages of computation the 0/P of previous stage will be I/P for current
stage.
First stage computation:
The I/P time sequence {2, 1, 2, 1, 2, 1, 2, 1)
The butterfly computations of first stage are shown below.
S.Leena
www.5starnotes.com
Page 60
www.5starnotes.com
S.Leena
Page 61
www.5starnotes.com
www.5starnotes.com
S.Leena
Page 62
www.5starnotes.com
www.5starnotes.com
On the other had, the DFT does lend itself to computation on a digital
computer. In the discussion that follows, we describe how the DFT can
be used to perform linear filtering in the frequency domain. In
particular, we present a computational procedure that serves as an
alternative to time-domain convolution. In fact, the frequency-domain
approach based on the DFT, is computationally more efficient than
time-domain convolution due to the existence of efficient algorithms
for computing the DFT. These algorithms, which are described in
Chapter 6, are collectively called fast Fourier transform (FFT)
algorithms.
Use of the DFT in Linear Filtering: In the preceding section it was
demonstrated that the product of two DFTs is equivalent to the circular
convolution
of
the
corresponding
time-domain
sequences.
Unfortunately, circular convolution is of no use to us if our objective is
to determine the output of a linear filter to a given input sequence. In
this case we seek a frequency-domain methodology equivalent to
linear convolution.
Suppose that we have a finite-duration sequence x (n) of length L
which excites an FIR filter of lenth M. Without loss of enerality, let
then
..(3)
S.Leena
www.5starnotes.com
Page 63
www.5starnotes.com
where {X (k)) and {H (k)) are the N-point DFIs of the
corresponding sequences x
(n) and h (n), respectively. Since the sequences x (n) and h (n) have a
duration less than N, we simply pad these sequences with zeros to
increase their length to N. This increase in the size of the sequences
does not alter their spectra
which are continuous
spectra, since the sequences are aperiodic. However, by sampling their
spectra at N equally spaced points in frequency (computing the Npoint DFTs), we have increased the number of samples that represent
these sequences in the frequency domain beyond the minimum
number (L or M, respectively).
Since the N = L + M - I point DFT of the output sequence y (n) is
sufficient to represent y (n) in the frequency domain, it follows that
the multiplication of the N-point DFTs X (k) and H (k), according to
(3), followed by the computation of the N-Point IDFT, must yield the
sequence {y (n)). In turn, this implies that the N-point circular
convolution of x (n) with h (n). In other words, by increasing the
length of the sequences x (n) and h (n) to N points (by appending
zeros), and then circularly convolving the resulting sequences, we
obtain the same result as would have been obtained with linear
convolution. Thus with zero padding, the DFT can be used to perform
linear filtering.
Filtering of Long Data Sequences : In practical applications involving
linear filtering of signals, the input sequence x (n) is often a very long
sequence. This is especiffy true in some real-time signal processing
applications concerned with signal monitoring and analysis.
Since linear filtering performed via the DFT involves operations on a
block of data, which by necessity must be limited in size due to limited
memory of a digital computer, a long input signal sequence must be
segmented to fixed-size blocks prior to processing. Since the filtering
is linear, successive blocks can be processed one at a time via the DFT
and the output blocks are fitted together to form the overall output
signal sequence.
We now describe two methods for linear FIR filtering a long sequence
on a block- by-block basis using the DFT. The input sequence is
segmented into blocks and each block is processed via the DFT and
IDFT to produce a block of output data. The output blocks are fitted
together to form an overall output sequence which is identical to the
sequence obtained if the long block had been processed via timedomain convolution.
The two methods are called the overlap-save method and the overlapadd method. For both methods we assume that the FIR filter has
S.Leena
www.5starnotes.com
Page 64
www.5starnotes.com
duration M. The input data sequence is segmented into blocks of L
points, where, by assumption, L>> M without loss of generality.
Overlap-save method : In this method the size of the input data blocks
is N = L + M -1 and the size of the DFTs and IDFT are of length N.
Each data block consists of the last
m - 1 data points of the previous data block followed by L new data
points to form a
data sequence of length N = L + M - 1. An N-point DFT is computed
for each data block.
The impulse response of the FIR filter is increased in length by
appending L 1 zeros
and an N-point DFT of the sequence is computed once and stored. The
multiplication of the two N-point DFTs
for the mth
block of data yields..
Then the N-point IDFT yields the result
are
Since the data record is of length N, the first M 1 points of
corrupted by aliasing and must be discarded. The last L points of
are exactly the same as the result from linear convolution and, as a
consequence,
To avoid loss of data due to aliasing, the last M 1 points of each
data record are saved and these points become the first M - 1 data
points of the subsequent retord, as indicated above. To begin the
processing, the first M - 1 points of the first record are set to zero.
Thus the blocks of data sequences are
and so forth. The resulting data sequences from the IDFT are given by
(8), where the first M - I points are discarded due to aliasing and the
remaining L points constitute the desired result from linear
S.Leena
www.5starnotes.com
Page 65
www.5starnotes.com
convolution. This segmentation of the input data and the fitting of the
output data blocks together to form the output sequence are
graphically illustrated in fig.
Overlap-add method : In this method the size of the input data block
is L points. and the size of the DFTs and IDFT is N = L + M - 1. To
each data block we append M - 1 zeros and compute the N-points DFT.
Thus the data blocks may be represented as:
and so on. The two N-point DFTs are multiplied together to form
The IDFT yields data blocks of length N that are free of aliasing since
the size of
the DFTs nd IDFT is N = L + M - 1 and the sequences are increased to
N-points by appending zeros to each block.
S.Leena
Page 66
www.5starnotes.com
www.5starnotes.com
first M - I points of the succeeding block. Hence this method is called
the overlap-add method. This overlapping and adding yields the output
sequence.
The segmentation of the input data into blocks and the fitting of the
output data
blocks to form the output sequence are graphically illustrated in fig.
11.
At this point, it may appear to the reader that the use of the DFT in
linear FIR filtering is not only an indirect method of computing the
output of an FIR filter, but it may also be more expensive
compositionally since the input data must first be convected to the
frequency domain vai the DFT, multiplied by the DFT of the FIR filter,
and
finally, converted back to the time domain via the IDFr. On the
contrary, however, by
using the fast Fourier transform algorithm, as will be shown in Chapter
6, the DFTs and
IDFT require fewer computations to compute the output sequence than
the direct
realization of the FIR filter in the time
S.Leena
www.5starnotes.com
Page 67
www.5starnotes.com
domain. This computational efficiency is the basic advantage of using
the DFT to compute the output of an FIR filter.
IIR FILTER DESIGN
UNIT III
Cascaded Realization:
Ans.
Parallel realization:
In terms of positive powers of z, H(z) can be written as
S.Leena
Page 68
www.5starnotes.com
www.5starnotes.com
S.Leena
Page 69
www.5starnotes.com
www.5starnotes.com
S.Leena
www.5starnotes.com
Page 70
www.5starnotes.com
Ans.
S.Leena
www.5starnotes.com
Page 71
www.5starnotes.com
Ans. Given
S.Leena
Page 72
www.5starnotes.com
www.5starnotes.com
S.Leena
www.5starnotes.com
Page 73
www.5starnotes.com
Ans. Given
S.Leena
www.5starnotes.com
Page 74
www.5starnotes.com
S.Leena
www.5starnotes.com
Page 75
www.5starnotes.com
Ans. Given
S.Leena
www.5starnotes.com
Page 76
www.5starnotes.com
1.
Direct form-I
S.Leena
www.5starnotes.com
Page 77
www.5starnotes.com
the order of
1. Direct form-Il : To obtain direct form-Il realization, we
change H1(z) and H2(z), then resulting structure is shown below.
1.
S.Leena
Page 78
www.5starnotes.com
www.5starnotes.com
S.Leena
www.5starnotes.com
Page 79
www.5starnotes.com
7. For given analog filter system function
into
digita IIR filter by means of bilinear z-transformation. Digital
filter is to have resonant frequency
Ans. The given transfer function is:
.(1)
From eq. (1) we can say that f = 4
The value of
is given as
Now we will find out the value of sampling time (T5) using we relation.
Using
bilinear
transformation
H(z)
can
be
obtained
by
puffing
in the eq. of H (s).
S.Leena
www.5starnotes.com
Page 80
www.5starnotes.com
1.
S.Leena
www.5starnotes.com
Page 81
www.5starnotes.com
Given
and R
5. Calculate poles:
S.Leena
www.5starnotes.com
Page 82
www.5starnotes.com
S.Leena
Page 83
www.5starnotes.com
www.5starnotes.com
is given as:
is not obtained by
Ans. Given
S.Leena
Page 84
www.5starnotes.com
www.5starnotes.com
S.Leena
www.5starnotes.com
Page 85
www.5starnotes.com
Assume T =1 sec.
S.Leena
www.5starnotes.com
Page 86
www.5starnotes.com
UNIT IV
FIR FILTER DESIGN
1. Obtain a cascade realization using minimum number of
multiplications for the system.
Ans. We
can
factors
consider
that
H(z)
is
product
of
and
S.Leena
www.5starnotes.com
Page 87
www.5starnotes.com
is coefficient.
S.Leena
www.5starnotes.com
Page 88
www.5starnotes.com
In the simplest way Equation (6) can be realized as shown in Fig. (a).
Now the same output can be obtained by using the structure shown in
Fig. (b). This
structure is called as single stage lattice structure. This structure
provides two outputs
namely A1(n) and B1(n).
S.Leena
Page 89
www.5starnotes.com
www.5starnotes.com
S.Leena
www.5starnotes.com
Page 90
www.5starnotes.com
S.Leena
www.5starnotes.com
Page 91
www.5starnotes.com
2. Ones complement form: In this form the positive number is
represented as in the sign magnitude notation. But the negative
number is obtained by complementing all thebits of the positive
number.
For example:
S.Leena
www.5starnotes.com
Page 92
www.5starnotes.com
where
is magnitude response and
is phase response
For FIR filters with linear phase we can define
where a is a constant phase delay in sample. Symmetric Condition
We can write
S.Leena
www.5starnotes.com
Page 93
www.5starnotes.com
which gives us
Therefore, FIR filters will have constant phase and when the impulse
response is symmetrical about
, then its phase is piecewise
linear. Fig. below shows the property of. eq. (5), for N = 6 and N = 5.
In both the cases when N = 6 and N = 5 in the general case, the unit
impulse
response sequence satisfying eq. (5). is symmetrical about
N odd, there is
one sample
For
S.Leena
www.5starnotes.com
Page 94
www.5starnotes.com
Antisymmetric Condition:
For this
Then
Therefore FIR filters have constant group delay and not constant phase
delay when
impulse response is antisymmetric about
S.Leena
www.5starnotes.com
Page 95
www.5starnotes.com
The filters that satisfy the above conditions and have a delay of Ni
samples but
their impulse response are antisymmetric around the centre of the
sequence, as opposed to the true linear phase sequence that are
symmetric around the centre of the sequence.
Magnitude Specifications : It is shown in fig below:
Hence
S.Leena
www.5starnotes.com
Page 96
www.5starnotes.com
Form the desired frequency response, we can find that the given
response is symmetric N odd
Step 2. To find
S.Leena
www.5starnotes.com
Page 97
www.5starnotes.com
S.Leena
www.5starnotes.com
Page 98
www.5starnotes.com
Sol. Step 1. Draw the ideal desired frequency response of band reject
filtr.
From the desired frequency response, we can find that the given
response is symmertric, N-odd.
Step 2. To find
In general,
www.5starnotes.com
Page 99
www.5starnotes.com
For symetric response,
S.Leena
www.5starnotes.com
Page 100
www.5starnotes.com
From the desired frequency response, we can find that the given
response is
symmetric.
Step 2. To find
We know that,
S.Leena
www.5starnotes.com
Page 101
www.5starnotes.com
S.Leena
www.5starnotes.com
Page 102
www.5starnotes.com
S.Leena
Page 103
www.5starnotes.com
www.5starnotes.com
S.Leena
www.5starnotes.com
Page 104
www.5starnotes.com
S.Leena
Page 105
www.5starnotes.com
www.5starnotes.com
Find the values of h (n) for N 1. Also find the filter transfer and
frequency magnitude frequency function.
Sol. Step 1. Draw the desired frequency response:
Given,
From the frequency response, we can find that the given response is a
symmetrical
N odd response.
Step 2. To find
In general,
S.Leena
www.5starnotes.com
Page 106
www.5starnotes.com
So
But we know that
S.Leena
www.5starnotes.com
Page 107
www.5starnotes.com
S.Leena
www.5starnotes.com
Page 108
www.5starnotes.com
use 10 tap filter and obtain the impulse response of the desired
filter.
Ans. The filter co-efficients are given by :
Therefore
S.Leena
www.5starnotes.com
Page 109
www.5starnotes.com
UNIT V
APPLICATIONS
1.Describe about multirate signal processing
The signals of interest in digital signal processing are discrete
sequences of real or complex numbers denoted by x(n), y(n), etc. The
sequence x(n) is often obtained by sampling a continuous-time signal
xc(t). The majority of natural signals (like the audio signal reaching
our ears or the optical signal reaching our eyes) are continuous-time.
However, in order to facilitate their processing using DSP techniques,
they need to be sampled and converted to digital signals. This
conversion also includes signal quantization, i.e.,discretization in
amplitude, however in practice it is safe to assume that the amplitude
of x(n) can be any real or complex number. Signal processing analysis
is often simplified by considering
S.Leena
www.5starnotes.com
Page 110
www.5starnotes.com
is defined as
andX(ej) is nothing but X(z)
evaluated on the unit circle z = ej.
Multirate DSP systems are usually composed of three basic
building blocks, operating on a discrete-time signal x(n). Those are the
linear time invariant (LTI) filter, the decimator and the expander. An
LTI filter, like the one shown in Fig.1.1, is characterized by its impulse
response h(n), or equivalently by its z-transform (also called the
transfer function) H(z). Examples of the M-fold decimator and
expander for M = 2 are shown in Fig.1.2. The rate of the signal at the
output of an expander is M times higher than the rate at its input,
while the converse is true for decimators. That is why the systems
containing expanders and decimators are called multirate systems.
Fig.1.2 demonstrates the behavior of the decimator and the expander
in both the time and the frequency domains. In the z-domain this is
described by
The systems shown in Figs.1.1 and 1.2 operate on scalar signals and
thus are called single inputsingle output (SISO) systems. The
extensions to the case of vector signals are rather straightforward:
thedecimation and the expansion are performed on each element
separately. The corresponding vector sequence decimators/expanders
are denoted within square boxes in block diagrams. In Fig.1.3 this is
demonstrated for vector expanders. The LTI systems operating on
vector signals are called multiple inputmultiple output (MIMO)
systems and they are characterized by a (possibly rectangular) matrix
transfer function H(z).
2. Explain Application in channel equalization
In the following, we consider the case where an FIR LBP is used as a
MIMO channel equalizer. We will showthat the flexibility in the choice
of H(z) can be exploited in order to reduce the undesirable
amplification of the channel noise. But, before proceeding to these
results, we give a brief overview of some equalization techniques for
the vector channels.
S.Leena
www.5starnotes.com
Page 111
www.5starnotes.com
www.5starnotes.com
Page 112
www.5starnotes.com
The idea behind an FSE is to let the equalizer work at a higher rate.
Because of this additional redundancy, FSEs are both more flexible and
more robust than SSEs. In a continuous-time communication system,
FSE is realized by sampling the received waveform at M times the
symbol rate, and feeding such oversampled signal to the equalizer,
which now operates at the rate M/T. In this chapter, the oversampling
24 ratio M is assumed to be an integer. In discrete time, this is
modeled as shown in Fig.2.3(b). The discrete transfer functions G(z)
and C(z) are obtained after sampling the corresponding continuoustime impulse responses at the rate M/T. Thus, the equivalent channel
F(z) in this case is such that F2(z) = [F(z)]M
and the simplified scheme is shown in Fig.2.3(c).
Note that the noise also needs to be modified, but this is not
the main point of discussion here. We recall from Section 2.3 that a
zero-forcing FSE H(z) in Fig.2.3(c) is nothing but an LBP of the
channel matrix F(z). In this section we will exploit the non-uniqueness
of this biorthogonal partner with the aim of minimizing the noise power
at the receiver. The concept of fractionally spaced equalization is by no
means new [57]; however, the original contribution of this section is
the attempt to parameterize the FIR FSE solutions, making the search
for the best solution analytically tractable. On the other hand, the
optimization of MIMO systems of the type shown in
has been considered by several authors in many different contexts
(derive the optimal transmitter and receiver for a given channel in the
sense of minimizing the overall mean squared error. This MMSE
solution clearly outperforms any zero-forcing equalizer, however the
price is paid
in terms of complexity: the solution in involves ideal filtering. Here we
have taken a simplistic approach of decoupling the problems of ISI and
noise suppression.
3. Derivation of poly phase decomposition
S.Leena
www.5starnotes.com
Page 113
www.5starnotes.com
S.Leena
www.5starnotes.com
Page 114
www.5starnotes.com
S.Leena
Page 115
www.5starnotes.com
www.5starnotes.com
S.Leena
www.5starnotes.com
Page 116
www.5starnotes.com
Traditionally, this structure has been called the system for digital
interpolation since the rate of y(n) is M times higher than that of x(n).
Filter H(z) is usually referred to as the interpolation filter [61].
Suppose the goal is to recover the signal x(n) from y(n). Conceptually
the simplest way to achieve this is shown in Fig.1.5(a). Namely, y(n) is
first passed through the inverse of the interpolation filter 1/H(z). This
recovers the signal at the input of the M-fold expander. The M-fold
S.Leena
www.5starnotes.com
Page 117
www.5starnotes.com
decimator that follows simply discards the zeros inserted by the
expander and the recovery of x(n) is complete. Notice, however, that
this is not the only way to reconstruct x(n), simply because the inverse
filter forces the discarded samples to be zero, while they can take
arbitrary values. Indeed, any filter F(z) with the property that its
output preserves the desired samples of x(n) in the appropriate
locations, with arbitrary values in between x(n)
S.Leena
www.5starnotes.com
Page 118
www.5starnotes.com
The up and down sampler are time variant systems and the
analysis and synthesis filters are chosen to cancel the effect
of aliasing.
Even though the figure title reads general, numerous variations and
extensions are possible. The system focuses on a single user with the
corresponding discrete message s(n). This message is to be
S.Leena
www.5starnotes.com
Page 119
www.5starnotes.com
transmitted to the single receiver, only after sustaining the
perturbations introduced by the transmission medium. These
perturbations are modeled by the continuous-time LTI channel cc(t)
and the appropriate additive noise at the input of the receiver. The
design challenge amounts to ensuring that the received sequence s(n)
resembles the original message under some criteria To this end, the
receiver often introduces some redundancy combined with the
appropriate pre-processing.
The goal of this block is to facilitate the signal reconstruction at the
transmitter (more about this in the following subsection). The obtained
signal x(n) with rate 1/T is converted to analog and sent through the
medium to the receiver, only after the pulse-shaping. The combined
effect of the pulse-shaping and the physical channel is often referred
to as the equivalent channel and denoted by fc(t).
At the receiver, the corrupted signal is first sampled and digitized,
according to the sampling rate q/T. In this thesis we mostly focus on
the case when q > 1 which corresponds to acquiring more information
than absolutely necessary about the signal and the channel. This
further facilitates the signal recovery. The corresponding digital
equalizer works at the higher rate and that combined with its
increased complexity is the price to pay for the improvement in
performance achieved by oversampling. Finally, the signal rate needs
to be reduced back to the rate of s(n). This is usually achieved after
decimating by q and removing the redundancy.
All the signals in Fig.1.9 can be scalars or vectors. Correspondingly,
the LTI systems can be SISO or MIMO. The unconventional notation in
Fig is chosen to reflect these possibilities. Vector signals arise naturally
in situations when there is a number of sensors, but can also be
obtained by blocking scalar signals . The system is further complicated
if there is more than one In this thesis we mainly focus on the blocks
for redundancy insertion and pre-processing together with the
corresponding equalization and redundancy removal as well as
equalization
For different values of the parameter q in the vector and the scalar
case Also, of special interest will be the system modifications for
application in multiuser communications and the corresponding
equalization algorithms.
S.Leena
www.5starnotes.com
Page 120