Abstract
Maximum posterior probability (MAP) demodulation with delay of an FM signal in the presence of noise is realized and compared with a) Griffiths' linear prediction method, b) a complex-valued version of Griffiths' method, c) three new modifications of Griffiths' method which can improve the performance of this method, and d) maximum-likelihood (ML) frequency estimation. These first quantitative results on MAP demodulation with delay show an advantage of a few db in output SNR at high input CNR and much larger advantages at low input CNR, relative to the other techniques.