A soft receiver using recurrent networks

Two different neural network architectures have been used to realize a non-linear adaptive receiver for GSM signals. Using the well-established backpropagation technique we firstly built a recurrent network which has been trained considering different channels corrupted by ISI, fading and Doppler. The network has shown better performances than the a classic coherent receiver. A second recurrent architecture, based on a partially supervised Self Organizing Map, has been proposed in order to perform an effective real time learning.