Deep Blind Demodulation of Binary Modulated Signals

Demodulation is a fundamental and critical function of communication systems. Traditional demodulation methods are designed for specific modulation schemes, which require knowledge of the modulation used in the received signal to achieve normal demodulation. In this paper, we propose a deep learning-based blind demodulation method (DBD) for binary modulated signals without the knowledge of modulation schemes. Four modulations including binary amplitude shift keying (2ASK), binary frequency shift keying (2FSK), differential binary phase shift keying (DBPSK) and minimum-shift keying (MSK) are considered in experiments. Results show that in AWGN scenario, the proposed DBD can approximate the theoretical BER performance. In Rayleigh fading channel, DBD model has better BER performance than traditional demodulation methods. Comparison with modulation recognition-based method also reveals that DBD has better performance.