Research On Deep Learning Method Of Polarcode Decoding Method In ACO-OFDM Free Space Optical Communication System

Free space optical communication technology has become a research hotspot in the communication field because of its large transmission capacity,high rate,good security,no spectrum limitation and so on.However,due to the existence of atmospheric turbulence effect,the optical signal of FSO will be affected by external factors,and the received signal will have inter-symbol interference,resulting in high bit error rate（Bit Error Rate,BER）,which will affect the communication quality.In order to restrain the influence of atmospheric turbulence on optical communication system,asymmetric limiting optical orthogonal frequency division multiplexing is adopted in this paper,which can also effectively overcome multipath effect and inter-channel interference.Polarization code is the only known new channel coding scheme which can achieve channel capacity,and its encoding and decoding complexity is low,so the application of polarization code to free space optical communication is helpful to improve the performance of communication system.However,the traditional polarization decoder has some defects such as high decoding delay and large amount of computation,the deep learning technology can effectively reduce the delay and complexity of the decoding algorithm,and the decoding algorithm of polarization coding is essentially a classification problem.Therefore,the deep learning technology is applied to the decoding of polarization codes in this paper,and compared with the traditional decoder,the specific research work is as follows:Firstly,the atmospheric channel is modeled,the double gamma distribution channel model of atmospheric turbulence characteristics is simulated,and the channel parameters of three different turbulence intensities are obtained.At the same time,the characteristics of ACO-OFDM technology are studied,and its signal time is non-negative real number,which meets the transmission requirements of FSO system,so it is a highly efficient single polarization OFDM modulation method.On this basis,an ACO-OFDM FSO system model based on polarization code is built.For the network structure,this paper proposes to design two kinds of deep learning decoders of polarization codes,which are based on multilayer perceptrons（Multilayer Perceptron,MLP）and convolutional neural networks（Convolutional Neural Network,CNN）respectively.The bit error rates under different turbulence intensities and based on 4QAM,16QAM and 64QAM mappings are compared.This decoder can directly achieve decoding and demodulation,and the decoding results of the two decoders are compared with those of traditional SC and BP decoders.The simulation results show that the bit error rate of the MLP decoder trained in weak,medium and strong turbulent atmosphere channel can reach 5×10^-7,2×10^-6 and 0.3under the condition of 4QAM,16QAM and 64QAM mapping.Among them,the performance of MLP decoder under strong turbulence in 4QAM mapping is better than that of traditional decoder,and the gap of 10⁴ can be achieved when the signal-to-noise ratio of characters is 20d B.Based on 16QAM and 64QAM mapping,the bit error rate of MLP decoder is lower than that of traditional decoder under three kinds of turbulence conditions,16QAM can reach 10⁵,and 64QAM gap is small.The bit error rate of the CNN decoder trained in the weak,medium and strong turbulent atmosphere channel can be reduced to about 6×10^-5,0.2 and 0.4 under the above three mapping conditions.Similar to the MLP decoder,the BER difference between the CNN decoder and the traditional decoder can reach 10² at 4QAM,and the BER of 16QAM and64QAM is between SC and BP decoders.The effects of different turbulence intensity,different mapping order and polarization code length on the performance of MLP and CNN decoders under the same mapping conditions are compared.The simulation results show that the bit error rate（BER）is the lowest in weakly turbulent atmospheric channel,and the smaller the mapping order is,the smaller the BER is,and the better decoding performance of MLP and CNN decoders is in short codes.The polarization code decoder based on deep learning proposed in this paper realizes polarization code decoding and signal demodulation in pilot-free ACO-OFDM space optical communication system,and has excellent performance and robustness.However,under the condition of weak turbulence,the performance of polarization code decoder based on deep learning technology needs to be further improved,and it is greatly affected by the code length of polarization code,so it is lack of training in weak,medium and strong mixed turbulence.