ECE565: Estimation Detection and Filtering (Fall 2013) - Dr.

Raviv Raich Project Assignment 1 (TBD) This project deals with frequency estimation. We refer to the following model as multi-tone:
K

y (n) =
k=1

ck ej 2fk (n

N +1 2 )

+ (n),

n = 1, 2, . . . , N

(1)

where ck = ak + jbk is complex amplitude coecient, ak and bk are the real-valued, real and imaginary components of the complex coecient ck , fk is the frequency of the k th component and (n) for n = 1, 2, . . . , N are i.i.d. complex Gaussian random variables: (n) CN (0, 2 ) representing the additive noise. 1. Dene the parameter vector and the observation vector y for this problem. Write an explicit form for the PDF of y given , fY (y| ). 2. Compute the FIM and the CRLB. To simplify the derivation, use the well-known result of the CRLB for a Complex Gaussian process (see Fundamentals of Statistical Processing, Volume I: Estimation Theory by Steven M. Kay). 3. For K = 1 (the single-tone) case nd the ML estimator of f1 . 4. For K = 1 set N = 50, f1 = 0.25 and c1 = 1. Using 500 Monte-Carlo runs, compute the ML estimate of f1 for the realization at each run. From the 500 estimates, nd the empirical MSE. Repeat this process while varying 2 from 0.1 to 10. Plot the empirical MSE of the ML estimator of f1 against 2 on a logarithmic scale (for both 2 and the M SE ). Include on the same graph the CRLB as a function of 2 . Separately, plot the empirical bias of the ML estimator against 2 . For what range of 2 would you consider, the ML estimator approximately unbiased?
N 1 1 1 5. Consider the following MOM estimator f MOM = 2 arg R(1), where R(1) = n1 n=1 y [n]y [n + 1]. Show 2 j 2f that E [R(1)] = A e . As in the previous question include the MSE and bias of the MOM estimator on the previous plots.