EE678: WAVELETS
PRACTICE TUTORIAL EXERCISES AND CHALLENGING PROBLEMS - 1
Tutorial Problem 1-1: - In any multiresolution analysis, we have the following containment
(ladder) relationship between the approximation subspaces:
......V− 2 ⊂ V− 1 ⊂ V0 ⊂ V1 ⊂ V2 .......
Consider, in particular, the Haar Multiresolution Analysis. Prove that, since all functions in any
of these subspaces are in L2 ( R) , the sum of the squared moduli of the constants on the pieces for
piecewise constant approximation in these approximation subspaces is always finite. Obtain an
upper bound for this sum, in terms of the norm of the function in L2 ( R) , and the index of the
approximation subspace. Hence explain movement to the ‘left extreme’ in the ladder.
Tutorial Problem 1-2: - There are two different ways in which the projection on the incremental
subspace ProjW0 x may be calculated in a Multiresolution Analysis (MRA):
(i) Take the dot product (inner product) of x with every orthogonal basis element of W0
namely {ψ (t − n)}n∈ Z
(ii) Find the difference between projections on successive approximation subspaces V0
and V1 .
Verify, using the Haar MRA as an example, that these result in the same projection function. You
may illustrate this for some specific functions in L2 ( R) .
Tutorial Problem 1-3: - Consider the function x(t) = { 1 - |t|; 0 ≤ |t| ≤ 1; and 0 otherwise. }
Obtain a general expression and sketch, for the projection of x(t) on the approximation subspaces
Vm , m ∈ Z + and m ∈ Z − of the Haar MRA. Hence obtain ProjWm x for integer m in general.
Tutorial Problem 1-4: - For the function x(t) in Tutorial Problem 1-3 above, how do the
following change as, m : 0 → + ∞ ; : 0 → − ∞
2
i) x − ProjVm x
2
2
ii) ProjWm x
2
Express these quantities as a function of m.
Challenging Problem 1-5: - Consider the signal,
r
x(t ) = 1 − t
r ∈ R+
0≤ t ≤ 1
= 0.....else
Repeat the analysis of the above Tutorial Problems 1-3 and 1-4, for the Haar MRA, on this signal.
1
Tutorial Problem 1-6: - By inductive reasoning or otherwise, show that H (z ) has a zero of
multiplicity M at z = 1 ⇒ 0th ,1st ,...( M − 1) th moments of the corresponding impulse response
h[n] are zero. What implication does this have for polynomial sequence inputs?
Tutorial Problem 1-7: - Show that for a sequence h[n] whose autocorrelation sequence exists,
and which has a Z- transform H (z ) , with a region of convergence (ROC) R, the autocorrelation
sequence has the Z-transform H ( z ) H ( z − 1 ) with an ROC of at least ( z ∈ R ) ∩ ( z − 1 ∈ R) .
Extend this result with suitable modification to the cross-correlation of two sequences h[n] and
g[n]. To recall, the cross-correlation of two sequences is the dot product of one sequence with
shifted versions of the other, as a function of shift.
Tutorial Problem 1-8: - On the course webpage, observe the evolution of the scaling and
wavelet functions for successive lengths in the Daubechies’ family of MRAs from the computer
program written by our former student, Tanmay Pawar. Obtain, using a suitable computer
program/ package if need be, the frequency response of successive analysis lowpass filters in the
Daubechies’ family. Study how an increase in the length of the lowpass filter implies a movement
towards the ideal frequency response desired.
Tutorial Problem 1-9: - Verify that the length 6 filter in the Daubechies’ family has,
H 0 ( z ) = C0 (1 + z − 1 ) 3 (1 + B0 z − 1 )(1 + B1 z − 1 ) and enunciate the conditions that the constants
B0 , B1 must satisfy in order to meet the autocorrelation constraint. You may use a computer
program/ mathematical package to ease your work.
Tutorial Problem 1-10: - What is the logic behind the choice of the C0 coefficient in the
Daubechies’ family of filters? (Hint: What can we say about the frequency response at ω = 0 ,
. 2 norm, . 1 norm of the impulse response, when the filter has length 4?)
Challenging Problem 1-11: - In the lectures, we discussed in detail how to generate the
Daubechies’ analysis lowpass filter impulse response of length 4. Come out with a
progressive/inductive approach to generating further members of the Daubechies’ family.
Generate the impulse responses of this family of filters, and compare with tabulated filter
coefficients for this family. You may use a computer program/ mathematical package to ease
your work. Generate the resultant functions ψ(t) and φ(t) and study their time moments.
Tutorial Problem 1-12: - In case of the length 4 filter in the Daubechies’ family of filters, show
that both choices of B0 , the one inside the unit circle, and the other outside would result in the
same magnitude response. (Hint: Show that an all pass filter, or a filter with a magnitude response
of 1, relates the two of these). Explain how this can this be generalized to the context of arbitrary
conjugate quadrature (CQF) filter bank design ?
Challenging Problem 1-13: - Show that a discontinuity in the magnitude of the frequency
response of a filter (often or always?) means that the filter is unstable, that is, the impulse
response is not absolutely summable.
Challenging Problem 1-14: - Either prove independent of specific example, or disprove with a
counter example, the following statement: An conjugate quadrature (CQF) filter bank, other than
for length 2 (Haar MRA), cannot comprise of lowpass filters which exhibit symmetry in the
impulse response, or equivalently cannot exhibit a linear phase response.