Research On Communication Technologies Of Underwater Acoustic OFDM Based On Deep Neural Network

In recent years,orthogonal frequency division multiplexing(OFDM)as a high-speed communication technology is widely used in the field of underwater acoustic communication.Due to the complex multipath structure of underwater acoustic channel and serious Doppler effect,two key technologies in underwater acoustic OFDM communication: channel estimation and Doppler estimation and compensation are the key factors affecting the performance of underwater acoustic OFDM system.The performance of the traditional channel estimation algorithm is very dependent on the proportion of the pilot in the subcarrier,and its cyclic prefix needs to be greater than the maximum delay spread of the channel,which will undoubtedly greatly reduce the communication rate of the system.The traditional Doppler compensation algorithm usually can not compensate the Doppler perfectly in the process of practical use,and the residual Doppler will affect the bit error rate performance of the system and reduce the reliability of the system.In order to solve the above defects of underwater acoustic OFDM communication system,this paper applies deep neural network to OFDM system.The main research contents and work are as follows:Because the deep neural network needs a large number of data sets for training,and the underwater acoustic channel is complex and changeable,it is difficult to determine the current channel condition(such as multipath delay and time-varying speed of channel)during the sea trial.The current sea trial data is difficult to make a comprehensive performance evaluation of the underwater acoustic OFDM communication algorithm based on the deep neural network.For the traditional channel estimation algorithm,the channel multipath delay is the key factor to determine its estimation performance.Based on the above situation,this paper uses the simulation time-varying channel model based on large-scale fading and small-scale fading,and uses the sound velocity profiles of deep sea and shallow sea respectively to model the large delay extended slow time-varying and hour delay extended slow time-varying channels(The large delay spread channel and hour delay spread channel in this paper are determined by the channel length that can be estimated by traditional methods compared with OFDM signal.If it is greater than the estimated channel length,it is the large delay spread channel),in order to obtain a large number of random channel impulse responses as the data set of deep neural network,It is used to evaluate the performance of traditional algorithms and receivers based on deep learning for different delay channels.In order to evaluate the performance of traditional channel estimation algorithms,OFDM symbols with cyclic prefix with different pilot intervals are used to estimate the channel in the South China Sea measured channel,large delay spread simulated time-varying channel and hour delay spread simulated time-varying channel,and its estimation performance is evaluated by normalized mean square error(NMSE),The simulation results show that the estimation performance of OMP(orthogonal matching pursuit)channel algorithm is better than LS(least square estimation)and MMSE(minimum mean square error estimation).Within its estimated channel length,its estimation accuracy is high.To increase the estimated channel length can only increase its pilot proportion,but increasing the pilot proportion will reduce the transmission rate of the system,The receiver based on deep neural network is expected to reduce the impact of the number of pilots on the communication performance.In order to overcome the dependence of traditional algorithms on pilots,this paper studies a joint channel estimation and data detection algorithm based on deep learning.In order to evaluate the performance of the algorithm,this paper uses the measured channel and simulated time-varying channel in the South China Sea to test the BER performance of the two algorithms under different pilot numbers and cyclic prefix lengths.The simulation results show that when the channel changes slowly,using the received signal of the previous period to train the deep neural network can be used to demodulate the received signal of the later period,and its bit error rate performance has a gain of 3-6db compared with the OMP channel estimation algorithm based on interpolation.In addition,the algorithm has good robustness to pilot and cyclic prefix,and can demodulate the received signal when the channel delay spread is large when there are a small number of pilots and no cyclic prefix.The traditional Doppler compensation algorithm can not perfectly compensate the Doppler in the process of practical application,and there will be residual Doppler,which will seriously affect the performance of underwater acoustic OFDM receiver.Therefore,this paper studies an underwater acoustic OFDM receiver based on deep learning to mitigate the impact of residual Doppler on the demodulation performance of the receiver,and uses two schemes to evaluate the performance of the algorithm.The first scheme takes the simulated received signal through the measured channel in the South China Sea plus a small amount of Doppler estimation deviation as the residual Doppler,and compares the bit error rate performance of the traditional underwater acoustic OFDM receiver with the underwater acoustic OFDM receiver based on depth neural network.The simulation results show that the underwater acoustic OFDM receiver based on depth neural network has good robustness to residual Doppler,and has a bit error rate performance gain of more than 6d B before and after compensation at a signal-to-noise ratio of more than 5d B.The second is to use the underwater acoustic OFDM signal including Doppler collected from the South China Sea experiment.After the traditional Doppler estimation and compensation algorithm with low complexity,the depth neural network is used to demodulate it,and its original bit error rate performance is compared with the traditional underwater acoustic OFDM receiver.The results show that the underwater acoustic OFDM receiver based on depth neural network has good robustness to residual Doppler,its bit error rate performance is significantly better than the traditional underwater acoustic OFDM receiver,and can decode the experimental data.It is proved that the application of depth neural network in the actual underwater acoustic OFDM receiver has a certain application prospect.