ELEN90094: LARGE DATA METHODS & APPLICATIONS Assignment 1 Semester 2, 2022
Hello, dear friend, you can consult us at any time if you have any questions, add WeChat: daixieit
ELEN90094: LARGE DATA METHODS & APPLICATIONS
Assignment 1
Semester 2, 2022
Introduction
In the class we worked with random matrices with real Gaussian elements. Here, we consider H be a p × n (n ≥ p) random matrix such that its elements are i.i.d. complex Gaussian random variables with zero mean and unit variance. Thus, for j = 1, 2, . . . ,p, k = 1, 2, . . . ,n, we can
describe the entries of H as hjk ~ NC (0, 1) such that Re(hjk ) ~ NR (0, 1/2) independent of Im(hjk ) ~ NR (0, 1/2). We define the ratio, β = .
Part A
In this part, we will empirically explore the histogram of eigenvalues of a p × p matrix, S,
described as
S = R HHH R
(1)
where R is a p × p diagonal matrix. R comprises of α > 0 followed by all ones on the diagonal,
i.e.,
α
0
R =
0
0 ... 0」
Note that this is an example of the problem of signal detection under the H1 hypothesis, as discussed in class. Here, R is analogous to the covariance matrix of the observations and α relates to the strength of the signal.
We examine how the values of α and β influence signal detection.
1) Using Matlab (or any other computing software) observe the histogram of the eigenvalues of S, for n = 1000 and β = 0.1.
a) For each value of α varying from 0.1 to 3.0 in steps of 0.1, comment on the observed behaviour of the histogram as α varies. Include in your report plots of the histogram for the values of α = {0.1, 0.5, 1.0, 2.0, 3.0}.
[3 pts]
b) For each value of α varying from 0.1 to 3.0 in steps of 0.05, generate 10 random realizations of S, and
(i) plot the scatterplot between the smallest eigenvalue of S and α ,
(ii) plot the scatterplot between the largest eigenvalue of S and α . Comment on the observed relationships in these plots.
[4 pts]
c) Using the results in Part A (1b), identify the range (R) of values of α that can be detected/estimated using the smallest/largest eigenvalues of S. For α ∈ R, do the smallest/largest eigenvalues of S underestimate or overestimate α?
[3 pts] 2) In the following, we study the histogram of the eigenvalues of S, for α = 2.0 and n = 1000.
a) Comment on the observed behaviour of the histogram as β varies from 0.1 to 1.0 in steps of 0.05. Include in your report plots of the histogram for the values of β = {0.1, 0.2, 0.5, 0.9, 1.0}.
[2 pts]
b) For each value of β varying from 0.1 to 1.0 in steps of 0.05, generate 10 random realizations of S. For each β, compute the average (i.e., averaged over the 10 real- izations) of the largest and the second largest eigenvalues of S. As a function of β , plot the (i) averaged largest eigenvalue, and (ii) averaged second largest eigenvalue. On the same figure, plot (1+^β)2 as a function of β . Comment on the relationships between the three curves.
[5 pts]
Part B
In this part, we will study the singular value decomposition (SVD) of the rectangular p × n matrix H. The SVD of H yields the factorization,
H = UΣVH , (2)
where U is a p × p unitary matrix and V is a n × p matrix such that VHV = Ip (identity matrix). The matrix Σ is a p × p diagonal matrix,
lσ 1 0 ... 0 」
0 σ2 ... 0
Σ =
. . .
「
with σ 1 ≥ · · · ≥ σp ≥ 0 the singular values of the matrix H.
1) Comment on any similarity and differences between the SVD defined above and the eigenvalue decomposition (EVD) discussed in the class.
[1 pt] 2) If we construct a p ×p matrix W = HHH , then will the p eigenvalues of W (λ1 ,λ2 , . . . ,λp )
be real or complex?
[1 pt]
3) How are the eigenvalues of W related to the singular values of H?
[2 pts]
4) Plot the histograms of the singular values of H for a fixed n = 5000 and varying p such that β = {0.01, 0.1, 0.5, 0.9}. Comment on how changing β affects the distribution.
[2 pts]
5) Consider the case when p = n. Plot the histograms of the singular values of H for
different values of n = {50, 100, 200, 500, 1000, 2000, 5000}. How big n should be for the histogram to converge to a shape? What is the shape of the histogram?
[3 pts]
6) Let us define a function,
f(x) = π xβ 1[^a,^b]
where β = , a = (1 − ^β)2 , b = (1 +^β)2 and 1[a,b] =〈 0(1)
Plot f(x) defined in Eq. (3) on each of the histograms plotted in Part B (5) above and comment on the observation.
[3 pts] 7) The joint probability density function (PDF) of the entries of H can be expressed as
fH (H) = cp,n exp (−Tr (HHH )) , H ∈ Cp ×n (4)
where cp,n is the normalizing constant and Tr(.) is the matrix trace operation.
a) Derive the joint PDF of the ordered singular values of H, fΣ (Σ). Note that the Jacobian of the transformation H → UΣVH is as follows:
u σn −p)+1 .
1≤i≤p
Show all the steps of the derivation. There is no need to explicitly compute the normalizing constant.
[4 pts]
b) Based on your derivation in Part B 7(a), comment on the statistical dependence between the singular values σi’s and the matrices U and V?
[1 pts]
c) Using the result from Part B 7(a), derive the joint PDF of the ordered eigenvalues (λ1 ≥ λ2 ≥ · · · ≥ λp ) of W = HHH . Show all the steps of the derivation.
[3 pts]
2022-08-20