Research On Equalization Technology In HF And GPU Implement

HF communication has characteristic of communication distance,anti-destruction.it is widely used in military communication,emergency rescue and so on.However,HF communication conditions are relatively poor,HF channel not only has noise interference,there are serious multipath effects and fast time-varying.Aiming at the problem of high communication bit error rate under HF bad channel condition,this paper mainly studies the equalization technology and its implement in GPU.In the equalization algorithm,Per-survivor Processing(PSP)equalization has excellent performance,and can adapt to various complex HF channels.This paper introduces the basic concept of PSP equalization,and then according to the needs of soft decision decoding,study soft decision output PSP equalization algorithm based on BCJR.And taking into account the timing synchronization error of the larger case,the study the PSP algorithm of oversampled channel estimation.As the PSP equalization algorithm is too complex,in the implement of the choice of GPU.Aiming at the problem that the PSP equilibrium is too large,the PSP balance of reducing the complexity is studied,and put forward the adaptive M-PSP algorithm.Depth analysis of the parallel characteristics of the PSP equalization algorithm,the segment of each data is divided into multiple equal length sub-data blocks,and the GPU with different stream processors at the same time,and then each sub-data block in the task of fine-grained,Single thread block within a reasonable thread division and memory configuration,and finally gives the PSP equalization algorithm in the GPU to achieve the program.The performance of the PSP equalization algorithm on the GPU and the performance comparison between MMSE and DFE are analyzed by simulation.Aiming at the signals with low SNR thresholds,we study the technology of joint equalization and decoding algorithm.According to the most commonly used Error correction code method and the characteristics of GPU,the joint equalization decoding based on the source space search is mainly studied.This method is more parallel and therefore easier to implement in the GPU than the turbo equalization that typically requires multiple iterations.This paper proposes a PSP equalization based on Chase algorithm for linear block codes,and then proposes a PSP equalization based on LVA for convolutional codes.According to the characteristics of GPU analysis of the parallelism of the two algorithms designed on the GPU parallel algorithm.Finally,the performance of the joint equalization decoding algorithm is analyzed by simulation to verify its performance advantage over independent equalization decoding.