A neural net algorithm for multidimensional minimum relative-entropy spectral analysis

摘要

A neural net algorithm is presented to solve the general 1-D or multidimensional minimum relative-entropy spectral analysis. The problem is formulated as a primal constrained optimization and is reduced to solving an initial value problem of differential equation of Lyapunov type. The initial value problem of Lyapunov system comprises the basis of the neural net algorithm. Experiments with simulated data convincingly showed that the algorithm did provide the multidimensional minimum relative-entropy spectral estimator with the autocorrelation matching property with computational efficiency.