EE538 Digital Signal Processing I

Due Date: Friday, Nov. 22, Fall 2013

Matlab Homework 2

Note: You should make use of the matlab codes PR4chan.m and PRRC4chan.m posted at
the course web site.
Background. See relevant notes at course web site.

x [n]
H0 ()

x [n]
H1 ()

H3 ()

x[n]

H4 ()

G1 ()

x [n]
5

G2 ()

G3 ()

x [n]
4

x [n]
H5 ()

x [n]
3

x [n]

G0 ()

x [n]
2

x [n]

x [n]
1

x [n]
H2 ()

x [n]

G4 ()

G5 ()

x [n]
6

H6 ()

x [n]

x [n]
7

H7 ()

Figure 1(a). Analysis Filter Bank, M = 8.

G6 ()

G7 ()

x [n]
7

y[n]

Figure 1(b). Synthesis Filter Bank, M = 8.

x [n]
(2)

H0 ()
(2)

H0 ()

x [n]

(2)

H1 ()
(2)

H0 ()

(2)

x [n]

(2)

H0 ()

H1 ()

x [n]

(2)

H1 ()

x[n]

x [n]
4

(2)

H0 ()
(2)

H0 ()

x [n]

(2)

H1 ()
(2)

H1 ()

x [n]
6

(2)

H0 ()
(2)

x [n]

H1 ()

(2)

H1 ()

Figure 2(a). Analysis Section of Three-Stage Tree-Structured Filter Bank.

x [n]
0

(2)

G0 ()
(2)

x [n]
1

G0 ()

2
2

(2)

G1 ()
(2)

x [n]
2

G0 ()

(2)

G0 ()
(2)

x [n]
3

(2)

G1 ()

y[n]

x [n]
4

G1 ()

(2)

G0 ()
(2)

x [n]
5

G0 ()

2
2

(2)

G1 ()
(2)

x [n]
6

G1 ()

(2)

G0 ()
(2)

x [n]
7

G1 ()

2
2

(2)

G1 ()

Figure 2(b). Synthesis Section of Three-Stage Tree-Structured Filter Bank.

2

Synthesizing M=8 Channel Perfect Reconstruction Filter Banks from TreeStructured Filter Banks. This Matlab assignment is centered on synthesizing an M=8
channel uniform PR filter bank from a three stage tree-structured PR filter bank. As discussed in class, if the number of channels, M, is a power of two, an M-channel uniform
PR filter bank may be synthesized via an equivalent (in terms of I/O relationship) treestructured PR filter bank having log2 (M) stages with each stage formed from a two-channel
QMF filter bank as depicted in Figure 2. That is, the combination of the analysis filter pair,
(2)
(2)
(2)
(2)
{H0 (), H1 ()}, and synthesis filter pair {G0 (), G1 ()}, form a two-channel PR filter
bank.
Part I. Deriving the Uniform Filter Bank Equivalent to Tree-Structured Filter
Bank. Using Nobles First Identity to express each analysis filter, Hm (), m = 0, 1, ..., 7,
(2)
(2)
in terms of H0 () and H1 (). In each case, express the corresponding impulse response
(2)
(2)
hm [n], m = 0, 1, ..., 7, in terms of h0 [n] and h1 [n].
Next, use Nobles Second Identity to express each synthesis filter, Gm (), m = 0, 1, ..., 7,
(2)
(2)
in terms of G0 () and G1 (). In each case, express the corresponding impulse response
(2)
(2)
(2)
(2)
(2)
hm [n], m = 0, 1, ..., 7, in terms of g0 [n] and g1 [n]. Note g0 [n] = h0 [n] and g1 [n] =
(2)
h1 [n]; this is assumed throughout.
(2)

(2)

Part II. Matlab Calculations. For each pair of h0 [n] and h1 [n] specified below,
use Matlab and the results derived above to compute the numerical values of the analysis
filters hm [n], n = 0, 1, ...., N 1, for m = 0, 1, ...., 7. Plot all of the corresponding DTFTs
Hm (), m = 0, 1, ..., 7, superimposed on a single graph using (at least) a 1024 pt. FFT
of each hm [n], m = 0, 1, ...., 7. Next, compute the numerical values of the synthesis filters
gm [n], n = 0, 1, ...., N 1, for m = 0, 1, ...., 7. Plot all of the corresponding DTFTs Gm (),
m = 0, 1, ..., 7, superimposed on a single graph using (at least) a 1024 pt. FFT of each gm [n],
m = 0, 1, ...., 7. Also, using Matlab, for each case below place each impulse response hm [n],
n = 0, 1, ...., 7, as the row of a matrix called H and compute HHH . Put the elements of the
resulting 8 8 matrix in a Table. Finally, generate a sample function of Gaussian random
process with zero mean and unity power of length 128 as the input signal x[n]. Plot the
magnitude of the DTFT of the x[n] using at least a 1024 pt. FFT. In addition, plot the
magnitude of the DTFT of the corresponding output of the M=8 channel uniform PR filter
bank y[n], using at least a 1024 pt. FFT.
(2)

(2)

(A) h0 [n] = {1, 1} and h1 [n] = {1, 1}. How does the resulting H compare with a
Hadamard matrix of dimension 8?
(i) Label the plot of all of the corresponding DTFTs Hm (), m = 0, 1, ...., 7 superimposed as Figure 1(a).
(ii) Label the plot of all of the corresponding DTFTs Gm (), m = 0, 1, ...., 7 superimposed as Figure 1(b).
(iii) Label the table containing the values of the 8 8 matrix HHH as Table 1.
(iv) Label the plot of the magnitude of the DTFT of the Gaussian random process
input signal as Figure 1(c).
(v) Label the plot of the magnitude of the DTFT of the corresponding output of the
M=8 channel uniform PR filter bank as Figure 1(d).
3

(2)

(2)

(2)

(B) h0 [n] = hsr [n 16], n = 0, 1, ..., 31, h1 [n] = (1)n h0 [n], and = 0.35 where
hsr [n] =

2 cos[(1 + )(n + .5)/2]

sin[(1 )(n + .5)/2]
+
, n = 16, ..., 1, ..., 15.
2
2
[1 4 (n + .5) ]
[(n + .5) 4 2 (n + .5)3 ]

(i) Label the plot of all of the corresponding DTFTs Hm (), m = 0, 1, ...., 7 superimposed as Figure 2(a).
(ii) Label the plot of all of the corresponding DTFTs Gm (), m = 0, 1, ...., 7 superimposed as Figure 2(b).
(iii) Label the table containing the values of the 8 8 matrix HHH as Table 1.
(iv) Label the plot of the magnitude of the DTFT of the Gaussian random process
input signal as Figure 2(c).
(v) Label the plot of the magnitude of the DTFT of the corresponding output of the
M=8 channel uniform PR filter bank as Figure 2(d).
(2)

(2)

(2)

(C) h0 [n] = hsr [n 24], n = 0, 1, ..., 47, h1 [n] = (1)n h0 [n], and = 0.1 where
hsr [n] =

2 cos[(1 + )(n + .5)/2]

sin[(1 )(n + .5)/2]
+
, n = 24, ..., 1, ..., 23.
2
2
[1 4 (n + .5) ]
[(n + .5) 4 2 (n + .5)3 ]

(i) Label the plot of all of the corresponding DTFTs Hm (), m = 0, 1, ...., 7 superimposed as Figure 3(a).
(ii) Label the plot of all of the corresponding DTFTs Gm (), m = 0, 1, ...., 7 superimposed as Figure 3(b).
(iii) Label the table containing the values of the 8 8 matrix HHH as Table 1.
(iv) Label the plot of the magnitude of the DTFT of the Gaussian random process
input signal as Figure 3(c).
(v) Label the plot of the magnitude of the DTFT of the corresponding output of the
M=8 channel uniform PR filter bank as Figure 3(d).
(2)

(2)

(D) h0 [n] = {1, j} and h1 [n] = {1, j}.

(i) Label the plot of all of the corresponding DTFTs Hm (), m = 0, 1, ...., 7 superimposed as Figure 4(a).
(ii) Label the plot of all of the corresponding DTFTs Gm (), m = 0, 1, ...., 7 superimposed as Figure 4(b).
(iii) Label the table containing the values of the 8 8 matrix HHH as Table 1.
(iv) Label the plot of the magnitude of the DTFT of the Gaussian random process
input signal as Figure 4(c).
(v) Label the plot of the magnitude of the DTFT of the corresponding output of the
M=8 channel uniform PR filter bank as Figure 4(d).

General Information.
Deliverables for this project include:
the derivation required in Part I
16 plots and 4 tables
a paragraph summarizing your observations and any conclusions you can draw from
this set of computer experiments.
your source code appended to the report
The collection of plots and accompanying explanation should be put together in a cohesive
manner in the form of a brief report. You may use any Matlab command you like in solving
these problems. Each student is expected to do his/her own work and each must turn in
his/her own report. Again, your write-up for this homework should be in the form of a brief
report. Handwriting is acceptable but please be sure it is legible.