Demodulator Of MPPSK Signals Based On Convolutional Neural Networks

With the rapid development of information technology, many wireless broadband access technologies have been widely used. Hence the radio spectrum has become the most essential source for all the countries. Ultra Narrow Band (UNB) technology can make the most of spectrum resources and the key to realize UNB technology is to design the optimal demodulator. There are three kinds of demodulation methods at present:adaptive amplitude threshold decision, classification via BP neural networks and the nonlinear detector combining SVM and impacting filter. Each of these methods is limited and especially in the condition of severe ISI, the demodulation performance is very poor. This paper proposes a detection method, which applies the deep learning-convolutional neural network (DL-CNN) in machine learning to demodulate UNB signals. Much research work has been done about the detection of the M-ary position phase shift keying (MPPSK) modulated signals based on DL-CNN.Firstly, this paper puts great emphasis on introducing the Extended Binary Phase Shift Keying and its special case-bipolar small pulse EBPSK signal, then this paper gives the analysis in time and frequency domain of its general form-bipolar small pulse MPPSK modulation signals.Secondly, this paper explicitly introduces the system model of EBPSK signal decision system based on CNN, including the design and function of the narrow band filter in the sender, faster-than-Nyquist signaling theory, AWGN channel and the design of filters in the receiver.Thirdly, deep research about the network and its parameters of the DL-CNN based EBPSK signal demodulator is discussed and through the results, it is obvious that DL-CNN demodulator improves the spectrum efficiency?transmitting code speed and the suitability. CNN detection method learns and uses the whole characteristics and the inner information of the output wave of EBPSK signals after impacting filter or band-pass filters. Under the circumstance of more severe ISI, this method outweighs the traditional amplitude integral method and the multi-symbol united-decision method is superior than the single symbol decision method.Lastly, the DL-CNN network is used in the demodulation of MPPSK signals. The system model of MPPSK signal demodulation based on DL-CNN is introduced and through theoretical analysis and simulation research, results show that the performance of ?R band-pass filter in the receiver is superior to that of impacting filter in this system. The double symbol united-decision method performs better than single symbol decision system and when Kernel=121, the performance of double symbol united-decision method is 9dB higher than that of single symbol decision system.