DBN based automatic modulation recognition for ultra-low SNR RFID signals
Liang Ma, Yang Yang, Hongjun Wang · 2016
Deep belief network (DBN) is a powerful tool for extracting high level features of sequential data. This paper proposes a DBN based automatic modulation recognition method of RFID signals in ultra-low signal-to-noise ratio (SNR) conditions. First, we perform FFT with simulated data to get its spectrum. The spectrum and amplitude of the original signal are then used as features to train a DBN. After the training, the DBN is deployed to process modulated signals with low SNR. Experiments were conducted to calculate the performance of the proposed method with simulated ASK, single subcarrier modulation, dual subcarrier modulation, PSK, carrier wave and white Gaussian noise signals. We found that the proposed DBN based scheme is able to achieve better performance with ultra-low SNR signal (around -10dB) than traditional BP artificial neural networks.