% maximum entropy spectrum and Burg spectrum
me_spectrum=me(R,A,B,th);
plot(th,me_spectrum);
me_burg=pburg(y,5,th);
hold on;
plot(th,me_burg);
Maximum entropy spectrumThe entropy of a probability distribution function with , quantifies the uncertainty of the corresponding random variable. When is a multidimensional Gaussian distribution with zero mean and covariance matrix , the entropy becomes For a zeromean stationary Gaussian process , corresponding to infinitesized Toeplitz structured covariance, the entropy “diverges”. Thus, one uses instead the ‘‘entropy rate". Let be a dimensional principle submatrix of the covariance matrix Then the entropy rate is The limit of converges to the optimal onestep prediction error variance, and from the SzegKolmogorov formula Given state statistics, the maximum entropy power spectrum is The solution to this problem (see [2]) is where and is the inputtostate filter defined above. Then the maximum entropy spectrum is precisely the solution given above. This formula subsumes the classical Burg method/AR modeling where the is a Toeplitz matrix and is lagdelay filter bank. The maximum entropy spectrum is obtained using the routine me.m. For the example discussed above, the maximum entropy spectrum is shown in blue. There are two peaks detected inside the window. Burg's spectrum is shown in green. The resolution of Burg's solution is not sufficient to distinguish the two peaks. % maximum entropy spectrum and Burg spectrum
me_spectrum=me(R,A,B,th);
plot(th,me_spectrum);
me_burg=pburg(y,5,th);
hold on;
plot(th,me_burg);
