Detection of coherent and noncoherent signals via the stochastic signals model

A novel method for detection of coherent and noncoherent signals, based on the application of Rissanen´s minimum description length principle for model selection to the stochastic signals model, is presented. In this method, the detection and localization are done simultaneously, with the location estimator coinciding with the maximum likelihood estimator derived by Bohme. The proposed method outperforms the recently proposed method of Wax and Ziskind (1989), especially in the threshold region. Another important factor in the improved performance is the maximum likelihood estimator for the stochastic signals model which, unlike the maximum likelihood estimator of the deterministic signals used in the solution of Wax and Ziskind, is efficient. Simulation results demonstrating the improved performance are included