ELEC 9741: Assignment 1, 2022
Hello, dear friend, you can consult us at any time if you have any questions, add WeChat: daixieit
ELEC 9741: Assignment 1, 2022
Instructions
1 |
|
due in Moodle, Friday June 24, 4pm |
||
2 |
|
Signed School Cover Sheet attached |
||
3 |
|
TYPED PDF only - no microsoft word docs. |
||
|
|
|
||
4 |
|
Follow the Homework Rules. |
||
5 |
|
Computer output |
: no commentary ÷ no marks. |
|
6 |
|
Analytical results |
: no working ÷ no marks. |
|
|
|
|
||
7 |
|
◆ means you can use Matlab; else not. |
||
8 |
|
excep(N). No(N)w(o)eb(Di)ssear(cus)c(si)h(o)in(n)g |
Q1 (14) Theory
(a) If θˆis an estimator of a vector parameter θ show that MSE = E(l θˆ _ θ l2 ) =l bias l2 +trace(var(θˆ))
(a) For the regression problem y = Xn×pβ + ∈ where ∈ ~ N (0, σ2 I) with true value βo , consider the biased (’shrinkage’) estimator
βˆa = (1 _ )βˆ
βˆ = (XT X)-1 XT y = LSE
(i) Find the bias and variance of βˆa .
(ii) Show βˆa has a smaller MSE than the LSE when
2τ
a <
where vsnr = and τ = T 1(p) and λk are the singular values of X .
(b) Noise Model.
Consider the stationary process
Yt = a + φYt -3 + ∈t , t = 1, 2, . . .
where ∈t is a Gaussian white noise sequence of zero mean and variance σ 2 .
(i) Explain what are the stability/stationarity con- straints on φ?
(ii) Derive closed form expressions for the mean and acs of Yt .
Q2(14) (Impulse Response Estimation)
Consider a system with output sequence measured in noise
yt = st + nt , st = (h * u)t , t = 1, . . . , T
with impulse response hr = Aγ sin(2πr ), r = 1, 2, . . . , m. The input sequence is a (0, σu(2)) white noise 1 independent of the observation noise sequence which is a (0, σ2 ) white noise sequence. The variance SNR is vsnr =σ(σ) .
(a) Ignoring start-up transients, and setting ωo = show that for large m
σs(2) = 1(m)hr(2)σu(2) s σu(2) 1(~)hr(2)
= A2 σu(2) [ + ]
(b) ◆Simulation.
Write an m-file to simulate the system for T = 100 with the values: [γ, To , σ, vsnr] = [.8, 7, 1, 1] and m = mo = 20.
(c) ◆Show four displays:
plots of st , yt on the same graph;
histograms of yt , st on the same graph;
What do these plots reveal about the signals? an impulse response plot;
a Bode plot i.e. system frequency response.
(c) ◆Parameter Estimation.
Write an m-file to compute the penalized least squares estimator, its standard errors2 , the singular values of the X matrix.
(d) To compute the standard errors you need to derive the following formula for the variance of the smoothness penalized least squares estimator (SP-LSE).
var(βˆλ ) = σ2 (X\ X+λDT D)-1 XT X(X\ X+λDT D)-1
(e) ◆Using the simulated data from (b) compute the SP- LSE of β for m = mo and a grid of λ values as fol- lows.
Plot the ’loss’ = l βˆλ _ βo l2 versus λ to find the minimizing value of λ . 3
Compute the corresponding ’minimizing’ SP-LSE, its standard errors and plot the estimated IR overlaid with the true IR and a 95% confidence interval. Also plot the singular values of the X matrix and comment on them.
Q3 (8). ◆ Statistical Graphics.
The graphics/plots you display in Q1, Q2, Q3 will earn up to 8 marks.
Q4(14) (Noise Modeling)
Do not use any specialised matlab commands such as zp2tf,
arima, aic, bic etc.
(a) ◆ Write an mfile to simulate a stationary AR(3) time series driven by a zero mean Gaussian white noise of
unit variance. Your mfile should accept as input, three real roots or one real root and a complex root; all non-zero.
It should produce the AR parameters & variance di- rectly as well as the simulated values as output.
Show two simulations (T=200) (on a single page) one for each of the above cases. List the two sets of pa- rameters used. In each case ensure that Vo > 3.
(b) ◆ Using your mfile simulate an AR(3) with roots (.9,. 1,.6) for T=200. List the true parameter values.
Using least squares regression4 produce estimates for the 3 parameters, the noise variance as well as stan- dard errors for the parameters.
Are the estimates within 2 standard errors of the true values?
(c) ◆ Using your mfile simulate new data (T=100) from the same model (ii) compute BIC5 and find its mini-mizing order p* . Show a single plot of BIC together with its two components.
Give the parameter estimates corresponding to p* and their standard errors.
Also do a statistical model diagnosis using just the acs of the residuals. What conclusions do you draw about the quality of the estimated parameters and model or-der?
2022-06-24