ECE 735 Signal Synthesis and Recovery Techniques, Fall 2002
Last Modified: Tue 15 Oct 2019, 01:45 PM, CDT
Instructor Office Hours:
See what we did in Fall 2000. Fall 2002 will be quite similar.
Syllabus
Frank Kelly's Papers,
Charging and Rate Control for Elastic Traffic
M. Unser,
"Splines: A Perfect Fit for Signal and Image Processing,"
IEEE Signal Processing Magazine, pp. 22-38, Nov. 1999.
Available through IEEE Xplore.
Click
here
for pdf file.
Class Schedule for Fall 2002
9/4 Wednesday. Course overview. Discussed span and linear
independence. Worked Problem 2-4. Talked about sinc functions
and the fact that the only function that is both time limited and
bandlimited is the zero function.
Assigned Reading: Section 1, pp. 1-3 (you may skip the proof
of Lemma 1.5); Section 2, pp. 3-5.
Assigned HW #1: Problems 2-5, 2-6, 2-7, 2-8 DUE WED 9/11.
9/6 Distributed syllabus. Finished discussion of sampling theorem
and approximately bandlimited functions. Briefly discussed proof
of Lemma 2.8. Covered Definition 1.8 and Lemma 1.9.
Distributed handout on Piecewise Polynomial Functions and covered
p. 1.
9/9 Continued discussion of Piecewise Polynomial Functions handout.
Discussed Section 1.2 Smooth Subspaces; Section 2.1 Knot Sequences;
Section 2.2 Another Basis (the B-splines). Look over pictures of
B-splines on pp. 5-6 of handout.
9/11 Section 3: Introduced the least upper bound axiom, supremum,
and infimum. Stated the Bolzano-Weierstrass Theorem.
Defined inner product space (beginning of Section 6), normed
vector space (beginning of Section 5). Started discussing
metric spaces (Section 4).
Assigned HW #2: Download the file xyfile.txt into the directory where
you will use Matlab. In Matlab, use the commands
load xyfile.txt
splinetool(x,y)
Use the spline tool with least squares approximation selected and
knots displayed. Click on "Edit" and select "add knot" to add internal
knots until you get a good fit. When you are happy with the result,
click on "File" and select "Print to Figure" to create an ordinary
Matlab figure of what is shown in the spline tool. From there you
can print out your results. Turn in your printout by WED 9/18.
Also do problems 4-2, 4-4, and 4-9 DUE WED 9/18.
9/13 Said a few words reviewing splines. Covered Section 4.1: Open and
Closed Sets and Section 4.2: Some General Results. You may skip
Def. 4.8 through Proof of Lemma 4.9. You may also skip the proofs
of Theorems 4.13 and 4.14.
9/16 Convered Section 4.3 up through Definition 4.24 (Complete Space).
Skipped proofs of Theorems 4.19 and 4.20.
9/18 Proved R and R^d are complete (Th 4.26, Cor. 4.27). Showed that
closed and bounded subsets of R^d are sequentially compact (Th 4.28).
Section 4.4: Continuous Functions. Skip all material following
Def. 4.32 up to Th 4.34. Stated Th 4.34 (f is cont. iff it is
conv. preserving). Proved that a continuous fcn on a seq. compact
set is bounded and achieves its max and min values (Th 4.35).
Assigned HW #3: Problems 4-12, 4-13, 4-14 (=> only) DUE WED 9/25.
9/20 Covered selected parts of Section 4.5: Compact Sets. Discussed
Defs. 4.36 and 4.37. Skipped to Def 4.46 (Uniform Continuity)
and proved Th 4.47 (Continuous fcn on a compact set is uniformly
cont.) Stated Th 4.44 (In a metric space a set is compact if and
only if it is sequentially compact). Started Section 5: Normed
Vector Spaces. Covered introductory material and Section 5.1:
Projections --- Introduction.
9/23 Section 5.2: Finite Dim Subspaces. Proved Lemma 5.4. Discussed
Problems 5-5, 5-6, 5-7, 5-8. Stated Lemma 5.7. Stated and discussed
Theorem 5.8. Proved Th 5.9 and Th 5.10. Skipped Example 5.11.
Stated Th 6.7 (Orthogonality Principle).
9/25 Derived the OP (but you should read the uniqueness proof on p. 20.)
Derived the Cauchy-Schwarz inequality. Derived (6.7), (6.8), (6.9),
(6.11), and Bessel's inequality (6.14). Section 6.2: Projections onto
Infinite-Dimensional Spaces.
Assigned HW #4: Problems 5-5, 5-6, 5-7, 5-8, 6-3, 6-5, 6-6, 6-10.
DUE WED 10/2.
9/27 Section 6.3: Def of convex set. Stated and proved the Projection Thm.
Sums and direct sums of subspaces. Section 6.4: Applications of the
Projection Thm: Prop. 6.13 generalizes OP to convex sets.
9/30 Worked Problems 6-15, 6-16, 6-26. Stated and proved the Strict
Separation Th., 6.14.
10/2 Proved Farkas' Lemma 6.15. Discussed Problem 6-23, and the following
Remark on theorem's of the alternative. Discussed the Karhunen-Loeve
expansion and related eigenvalue problem for integral operators.
Discussed ill-posed equations of the form Ax=y.
Assigned HW #5: Problems 6-18, 6-20, 6-21, 6-24, 6-25 DUE WED 10/9.
10/4 Section 7.1: Linear Functionals. Section 7.2: Linear Operators through
proof of Proposition 7.12.
10/7 Went over solutions to Problems 5-5, 5-6, 5-7, 5-8. Covered Def 7.13,
Example 7.14, Def 7.15. Worked Problem 7-17. Started Section 7.3:
Adjoint Operators.
10/9 Worked Problem 7-21 computing an adjoint operator. Derived Prop. 7.17.
Discussed Problem 7-25 and example following it. Skipped Lemmas 7.18
(Farkas) and 7.19 (Motzkin). Proved Th 7.20. Discussed Problem 7-29.
Proved Lemmas 7.21 and 7.22. Announced Exam 1 will be on 10/23.
Assigned HW #6: Problems 7-3, 7-9, 7-13, 7-22, 7-29, 7-30, 7-31 DUE WED 10/16.
10/11 Discussed Problems 7-30 and 7-31. Covered Section 7.3.1 Ellipses,
Section 7.3.2 Projections onto range A, Started Section 7.4 Solving
Linear Equations.
10/14 Finished Section 7.4. Discussed the problem of estimating the
unknown deterministic x from the observation Y = Fx + W. Discussed
Section 7.5: The Pseudoinverse.
10/16 Mentioned important results we have covered.
Surveyed results we will cover in Sections 8 & 9.
Review Problems: 4-10, 4-16, 6-4, 6-11(b), 6-17, 6-22, 6-28, 6-29,
6-31, 6-32, 7-33, 7-34, 2 problems on board.
10/18 Worked Problems 7-33, 6-28, 6-29. Talked a bit about Section 8 and
developed the notion of a compact operator and working Problem 8-4,
which equates a compact operator to the fact that the closure of the
image of the unit sphere is sequentially compact.
10/21 Worked some of the review problems.
10/23 Exam 1 in class in 2317 EH.
10/25 Went over Exam 1. Proved Lemmas 8.2 and 8.3.
10/28 Proved the invariance Lemma 8.4 and the Spectral Theorem. Also
discussed the Example on p. 37.
10/30 Mentioned Matlab commands for finding eigenvalues and eigenvectors
of matrices, p. 37. Discussed square roots of positive-semidefinite
operators and matrices. Discussed Fredholm equations of the 2nd
kind when A is self adjoint, p. 38. Discussed a special class of
positive definite operators, p. 39.
11/1 Discussed Problems 8-1, 8-2, and 8-3. Proved Theorem 8.6.
11/4 Section 8.2 The Singular-Value Decomposition (SVD), application to
Fredholm equations of the 1st kind, ill-posedness of 1st-kind eqs.
Assigned HW #7: Copy the 5 files main.m, Ggen.m, integrand.m,
kernel.m, data.txt. Start Matlab, and run the script file main.m,
and print out resulting figure. Now choose your own k(t,tau) and
x(tau) and compute y(t) theoretically. Do not use k(t,tau) of the form
k(t,tau)=f(t)g(tau). Do not use k(t,tau) and x(tau) so that y(t) is
a polynomial or a piecewise polynomial. Using your formula for y(t),
write a script to generate vectors of t_i and y(t_i). To check your
work, use the command plot(t,y(t)) and print out your figure. For
plotting purposes, you may want to use vectors of length 100. However,
when using the data to recover x, you should start with only a few
samples, say 5, until everything runs without errors. Then you
can increase the amount of data to try to get a better reconstruction.
Write your own version of kernel.m corresponding to your function
k(t,tau). Use the Matlab routines main.m, Ggen.m, integrand.m, and
your version of kernel.m to reconstruct x(t). You may also want to
change the initial number of knots (set to 5) in main.m. But start
small, say 5, until everything runs without errors. This number
controls the number of basis functions. Also Problems 8-3, 8-8(a)(b),
8-9, 8-11 DUE WED 11/13.
11/6 Well-posedness of 2nd-kind Fredholm eqs. Section 8.3 Regularization.
Skipped Section 8.4. Started Section 9: Optimization Using Lagrange
Multipliers. Section 9.1 Sufficient Conditions for Arbitrary Functions.
Theorem 9.1.
11/8 Covered Section 9.2: Convex Functions. Skipped Section 9.3. Covered
Section 9.4: One-Sided Gateaux Derivatives, Section 9.5: Kuhn-Tucker
Sufficiency Theorems.
11/11 Section 9.6: More on Gateaux Derivatives. Skipped Section 9.7.
Section 9.8: Still More on the Gateaux Derivative. Skipped Lemmas
9.25 and 9.26, at least for now. Prove Mean-Value Theorem 9.27.
Worked Problem 9-14.
11/13 Section 9.9: The Frechet Derivative. Skipped Section 9.10.
Section 9.11: The Chain Rule and the Product Rule. Discussed, but
did not derive them.
Assigned HW #8: Problems 9-1, 9-4, 9-6, 9-8, 9-9 DUE WED 11/20.
11/15 Section 9.12: Applications.
11/18 Discussed part of Frank Kelly's paper
Charging and Rate Control for Elastic Traffic.
11/20 Discussed regularization for last Matlab HW. Discussed Matlab
matrix-vector operations for computing the SVD series representation
and the regularized x_alpha in eq. (8.23). Worked some examples
for directly computing Frechet derivatives.
Assigned HW #9: 9-12, 9-13, 9-15, 9-19. Also modify the last Matlab HW to
use regularization, as discussed today in class, to try to improve
your results. DUE WED 11/27.
Announced Exam 2 will be an evening exam on Wed. 12/4.
11/22 Worked another example of computing a Frechet derivative. Started
Section 9.13 Functions Concentrated in Time and Frequency thru (9.40).
11/25 Continued Section 9.13 through Proof of Lemma 9.42.
11/27 Concluded discussion of Section 9.13. Derived the Contraction
Mapping Theorem and discussed some of its applications.
Review Problems: 8-6, 8-7, 8-8(c)(d), 8-10, 8-13, 8-15, 9-2, 9-10,
9-11, 9-20.
11/29 No Class --- Thanksgiving Recess.
12/2 Worked some of the Review problems.
12/4 Worked more of the Review problems.
Also Exam 2 in 3534 EH, 5-7 pm
You may bring 2 sheets of 8.5 x 11 in. paper with any formulas
you believe may be helpful.
12/6 Discussed entropy, log inequality, the forward part of Shannon's
source coding theorem.
12/9 Went over Exam 2 solutions and covered handout on Lloyd-Max
quantizer design.
12/11 Started discussion of Arimoto-Blahut Algorithm for computing
channel capacity.
12/13 Last Class Day.
Web Page Contact: John (dot) Gubner (at) wisc (dot) edu