1 Uncertainty Relation

ECE 551 PSET 4 - Uncertainty, Windows, and Filter Banks September 17, 2020
1 Uncertainty relation
From [1]
Consider the uncertainty relation ∆2ω ∆2t ≥ π/2
2 2
1. Show that scaling
√ does not change ∆ω ∆t . Either use scaling that conserves the L2
0
norm (f (t) = af (at)) or be sure to renormalize ∆2ω , ∆2t .
2. Can you give the time-bandwidth product of a rectangular pulse, p(t) = 1, −1/2 ≤ t ≤
1/2, and 0 otherwise?
3. Same as above, but for a triangular pulse.
4. What can you say about the time-bandwidth product as the time-domain function is
obtained from convolving more and more rectangular pulses with themselves?
Your Solution Here!
1
2 Sinc Expansion Completeness

Adapted from [1].
One method for constructing an orthogonal basis is to take a complementary pair of ideal
low and high pass filters, sinc functions of the form
1 sin((π/2)n))
g0 [n] = √ ·
2 (π/2)n
n
g1 [n] = (−1) g0 [−n + 1],
and, combined with an offset, form a basis using all the even shifted versions. Show that
without the offset “+1”, the basis is not complete in `2 (Z). That is to say, find a sequence
that cannot be represented by
1 sin((π/2)n))
g0 [n] = √ ·
2 (π/2)n
n
g1 [n] = (−1) g0 [−n]
and all even shifts.

Hint: Consider trying to represent an odd function such as y[n] = sin(π/2n)
2
3 Truncation as an Orthogonal Projection

Fall 2018
Let I ∈ Z be an index subset. Define the truncation operator TI : CZ → CZ by (TI h)[t],
where w is the window indicator function of I.
1. For h ∈ `2 (Z), use the orthogonality principle to show that ĥ := TI h is the least-squares
approximation (of h) whos support is limited to I.
2. Show that TI is an orthogonal projection in `2 (Z).
3. Filter truncation is a lossy operation: the output ĥ ∗ x is perturbed from h ∗ x. But

by how much ? Show that the deviation sequence D = ĥ ∗ x − h ∗ x is bounded by the
bound below:
|D[n]| ≤ kxkkĥ − hk (1)
and conclude that the ratio M := kDk
kxk
∞
is bounded. You may assume that x and h are
absolutely summable (and bounded). Optional: determine whether M attains its
maximum.
3
4 Parseval’s relation for Nonorthogonal Bases

From [1]
Consider the space V = Rn and a biorthogonal basis, that is, two sets {αi }, {βi } such that
hαi , βj i = δi,j i, j = 0, 1, ..., n − 1 (2)
1. Show that any vector v ∈ V can be written in the following two ways:
n−1
X n−1
X
v= hαi , viβi = hβi , viαi (3)
i=0 i=0
2. Call vα the vector with entries hαi , vi and similarly vβ with entries hβi , vi. Given kvk,
what can you say about kvα k and kvβ k?
3. Show that the generalization of Parseval’s identity to biorthogonal systems is
kvk2 = hv, vi = hvα , vβ i (4)
and
hv, gi = hvα , gβ i (5)
4
5 Range-Finding Pulses (Computational)

In the previous homework, you observed that finding the delay between two signals is fairly
straightforward using the cross-correlation. In this problem, you will explore different wave-
forms for a range-finding application. Frequency shifts from the Doppler effect introduces
significant additional complexity (frequency shifts and time shifts don’t commute), for the
purpose of most of this problem, we will ignore the issue and assume no Doppler shifts.
In a radar environment, there may be multiple targets that reflect a given pulse, resulting
in a system generating a multipath response represented by
N
X −1
H= αi σti ti ∈ R+ , αi ∈ (−1, 1), (6)
i=0
where σd is a delay by d, and αi represents some attenuation of the signal. The set of {αi }
simply lump together the different losses from both propagation and the imperfect reflection
from the targets.
Goal: Identify each element of {σi } in the system.
Q1: Suppose N = 1 (representing a single reflection), how would you apply your
algorithm from homework 3 to identify σ0 ? What happens when ti ∈ / Z?
Your Solution Here!
Q2: Now assume N > 1, how would you extend your algorithm to identify all ti ? This
doesn’t need to be optimal in any sense, just pick something that seems reasonable.
Your Solution Here!
Q3: Implement your algorithm from Q2 and apply it to data generated by the provided
function. How well does it work? What seem to be the issues? What happens as you adjust
the distribution of ti such that the delays are clustered more tightly?
Your Solution Here!
There are actually a couple of issues at play. While we haven’t formally covered noise
yet, we can see that the noise seems to blur together the peaks of the cross-correlation.
Q4: Given no physical constraints, what is the optimal waveform for time localization
(assuming operation in continuous time)? What is the associated time-frequency trade-off?
Your Solution Here!
Q5: Ultra-wide bandwidth electronics are very complicated, can be very expensive to
produce, and can interfere with other items sharing the spectrum. For this reason, consider
limiting your bandwidth to a fixed width B. How would you adjust your answer to Q4 to
account for this constraint?
Your Solution Here!
Q6: Now that we have a finite bandwidth, we may as well operate in discrete time. Fix
the amount of energy in the signal, and simulate the application of your algorithm from Q2
applied to pulses of the form of your response to Q5.
5
Your Solution Here!

Unfortunately, you may additionally have some form of instantaneous power constraint,
or, related, a maximum voltage constraint on the electronics.
Q7: How would you preserve the desired time-bandwidth trade-off while satisfying a
maximum amplitude constraint (i.e. x(t) ≤ P for all t)? Hint: What happens to the phase
when we compute an autocorrelation of a signal? Alternatively, think about the relationship
between an STFT and the Fourier Transform of the entire sequence.
Your Solution Here!
Q8: Write a script that generates pulses of the form described in Q7, include a plot of
the waveform.
Your Solution Here!
Q9: Finally, apply the algorithm to the pulses in your response. Compare the behavior
to sinusoids with rectangular, triangular, and Hanning windows. Include plots of the cross-
correlation in simulation for all four.
Your Solution Here!
6
References
[1] Martin Vetterli and Jelena Kovac̆ević, Wavelets and Subband Coding, Prentice Hall,
Englewood Cliffs, NJ, 1995.

1 Uncertainty Relation

Uploaded by

Copyright:

Available Formats

1 Uncertainty Relation

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

1 Uncertainty Relation

Uploaded by

Copyright:

Available Formats

ECE 551 PSET 4 - Uncertainty, Windows, and Filter Banks September 17, 2020

3. Same as above, but for a triangular pulse.

Your Solution Here!

2 Sinc Expansion Completeness

and all even shifts.

3 Truncation as an Orthogonal Projection

2. Show that TI is an orthogonal projection in `2 (Z).

3. Filter truncation is a lossy operation: the output ĥ ∗ x is perturbed from h ∗ x. But

4 Parseval’s relation for Nonorthogonal Bases

hαi , βj i = δi,j i, j = 0, 1, ..., n − 1 (2)

3. Show that the generalization of Parseval’s identity to biorthogonal systems is

kvk2 = hv, vi = hvα , vβ i (4)

5 Range-Finding Pulses (Computational)

Your Solution Here!

You might also like