A Hybrid Neural Network and Maximum Likelihood Based Estimation of Chirp Signal Parameters

This research introduces the hybrid Multilayer feed forward Neural Network (NN) and the Maximum Likelihood (ML) technique into the problem of estimating a single component chirp signal parameters. The unknown parameters needed to be estimated are the chirp*rate, and the frequen cy parameters. NN was trained with several thousands noisy chirp signals as the NN inputs, where the chirp*rate and the freq uency parameters were embedded into those chirp signals, and those parameters were used as the corresponding NN output. The NN resulted in parameter estimates that were near the global maximum point. ML gradient based technique then used the NN output parameter estimates as its initial starting point in its search of the global point parameters. The ML gradient based search improved the accuracy of the NN parameter estimates and the new estimates were very much near the exact parameter values. Hence it can be said that NN working in corporation with the ML gradient based search results in accurate parameter estimates for the case of large signal to noise ratio.