A learning based cognitive radio receiver

In this paper, we introduce a learning based cognitive radio receiver to automatically demodulate several types of modulated signals without sophisticated digital signal pre-processing. Our embedded learning engine can automatically learn the signal features and then achieve signal demodulation through feature-based classification. The proposed demodulator consists of a neural network (NN) structure with one sub-NN for each possible demodulation bit pattern. To capture the temporal behaviors of different modulation types, the sampling features in two consecutive time slots are collected for each decision. The final classification result is jointly decided by ensemble learning. To validate our proposed method, three classical modulation signal (i.e. BPSK, QPSK and GMSK) are investigated with SNR varying from 0 dB to 9 dB. The simulation results indicate that the performance of our proposed method is highly competitive and the system provide much more flexibilities than the traditional demodulation method.