Improved noise characteristics of a SAW artificial neural network RF signal processor for modulation recognition
D. Kavalov, Vyacheslav Kalinin · 2002
A novel training algorithm for the SAW neural network (NN) digital modulation classifier is developed that allows one to improve significantly the noise performance in the case of classification of three modulation schemes. It yields high probability (90-95%) of correct recognition, reducing the required SNR at the input of the processor from 25-27 dB down to 12-15 dB, which is comparable to that of the two-signal classifier. Further improvement is achieved by adding one more layer of hidden neurons. As a result, the number of neurons in the first layer, i.e., the number of SAW filters implementing them, is reduced from seven to five, which considerably simplifies the SAW device topology.