Research On Intelligent Channel Prediction For Underwater Acoustic OFDM Communications

As the commanding elevation of international technical competition,high-data-rate underwater acoustic(UWA)communication is related to the development of national marine strategic equipment.Attributed to powerful abilities in improving the spectral efficiency and combating the delay spread,orthogonal frequency division multiplexing(OFDM)leads the development of physical layer technology for high-data-rate UWA communication.To implement an UWA-OFDM communication system,acquiring precise channel state information(CSI)is vitally important,as it is the basis of accurate signal recovery at the receiver and adaptive modulation at the transmitter.However,traditional channel estimation algorithms which sacrifice pilot resources in exchange for estimation accuracy have great pilot overheads.In addition,the traditional paradigm which detects the received data based on channel estimation would undoubtedly increase the estimation times,and thus aggravating the consumption of time-frequency resources.The above reasons make it difficult for tradition channel estimation algorithms to be applied in UWA-OFDM communications with extremely limited bandwidth(only on the order of tens of k Hz).Aiming that,we innovatively combine deep learning(DL)and UWA-OFDM communications to explore the theory and method to reduce estimation overheads and times from the perspective of channel prediction.Specially,the detailed contents and contributions are summarized as follows.1.To solve the problem that the accuracy and the complexity of channel estimation algorithms are restricted by the physical sparsity,we design an innovative channel estimation algorithm based on the recovery sparsity.Specifically,we first analyze the mechanism of the noise on the estimation accuracy,and propose the concept of recovery sparsity.Based on above,we develop a channel reconstruction model according to the relation between the recovery sparsity and the noise.Then we derive an exact closed-form expression for the recovery sparsity from the transmission function of UWA communication systems and principles of compressed sensing(CS)algorithms.Based on these,we design an adaptive orthogonal matching pursuit algorithm(A-OMP)whose iteration termination conditions change adaptively with the noise level.Simulation results revel that,compared with traditional orthogonal matching pursuit algorithm(OMP),the A-OMP can increase the estimation accuracy by 80.62% with only 13.48% CPU running time when SNR is 0 d B.2.To deal with the problem that the accuracy of traditional adaptive predictive algorithms is usually limited due to ignoring the block sparsity of UWA channel,we propose a prediction method which takes the spatial correlation of UWA channel into consideration.Specially,we analyze stems of traditional channel prediction algorithms,and propose the idea of constructing the prediction network based on the Conv LSTM.However,due to the special characteristics of UWA channel matrix data(complex data &small value),the Conv LSTM cannot directly be applied to predict the UWA channel.Thus,we carefully devise the network,and design suitable loss function and input format.Based on these,we build the channel prediction network(CPNet)which can predict the next frame channel time response.Simulation results show that,compared with the recursive least square algorithm(RLS),the CPNet can reduce the prediction error by up to 81.77%when SNR is 0 d B.In view of above research,we propose an intelligent channel prediction method based on CS and DL,which realizes the transformation of UWA-OFDM communications from passive estimation to active prediction.Based on this transformation,we can significantly reduce the number of channel estimation and the pilot cost during each estimation,and thus dramatically releasing the time-frequency resource.The above research can provide technical and theoretical basis for the construction of high-data-rate UWA-OFDM communications.