Research On Symbol Detection And Channel Estimation In OTFS System With Message Passing Algorithm

With the development of the mobile communication technologies,China has officially enter the 5G era.Thanks to the characteristics of 5G which are knowns as higher speeds,lower latency and larger capacity,more and more devices in various industries have acccess to the mobile internet,gradually building an interconnected intelligent world.However,when the terminals are under high mobility conditions,the carrier frequency of the transmitted signal will shift due to the Doppler effect.In such a scenery,the orthogonal frequency division multiplexing(OFDM)scheme currently deployed in4 G and 5G mobile system suffers from inter carrier interference(ICI),thus causing synchronization problem and performance loss.Therefore,the recently proposed orthogonal time frequency space(OTFS)received tremendous attention.OTFS mainly works in delay-Doppler(DD)domain and performs data transformation in DD domain,time-frequency domain and time domain,this shceme can efficiently resist Doppler effect and multi-path time delay effect,showing better data transmission performance.In the current research of OTFS system,the symbol detection algorithms still generally have the problems of performance loss due to the approximation processing,limited applicability and poor robustness due to the sensitivity to the channel parameter changes? the channel estimation algorithms mostly ignore the fractional Doppler shift to simplify the system model,thus there is still a lack of research on fractional Doppler related channel parameter estimation.Based on factor graph(FG)and message passing(MP)algorithms,this thesis aims to address the above problems.The main contents are as follows:1.To fully exploit the diversity brought by the multi-path propagation of signals in OTFS,unitary approximate message passing(UAMP)based detectors are proposed.1)When the pluse-shaping function at the transmitter side and the matched filter function at the receiver side satisfy the bi-orthogonal property(bi-orthogonal waveform scenery),this thesis studys the input-output relation in DD domain and diagonalizes the effective channel matrix by exploiting it structure,thereby simplifing the system model.Then UAMP with some modifications is performed for symbol detection.2)When the functions at both transmitter side and receiver side do not satisfy the bi-orthogonal property,this thesis studys the typical rectangle waveform scenery.By appending a cycle prefix(CP)to each sub-block of the singal in time domain,the effective channel matrix in time domain can be block diagonalized,then UAMP is utilized for symbol detection on the corresponding model in DD domain.Simulation results demonstrate that the proposed detectors for both bi-orthogonal scenery and rectangle scenery exhibit lower bit error rate(BER)than the existing detectors,and are more robust.2.Considering that the research on fractional Doppler related channel estimation is still insufficient,the structured sparse singnal recovery based channel estimators are proposed.This thesis first considers the bi-orthogonal scenery and studys the pilot and guad symbols arrangement scheme.Then according to the input-output relation in DD domain,the channel estimation problem is reformulated as a structured sparse signal recovery problem followed by a belief propagation(BP)-variational message passing(VMP)based algorithm,the channal matrix is finally reconstruced by using the estimated channel parameters.Next,by comparing the differences of input-output relations in bi-orthogonal scenery and rectangle scenery,the research for the former is extended to the latter.In addition,to assess the accuracy of the estimation of channel parameters,this thesis develops the Cramer-Rao Lower Bound(CRLB).Simulation results demonstrate that the normalized mean squared error(NMSE)performance of the proposed algorithm outperforms the existing ones and is close to the CRLB.When using the estimated channel matrix for symbol detection,the BER performance only has a small gap with the case that having perfect channel state information.3.To solve the overhead problems caused by the CPs in symbol detection and pilot,guard symbols in channel estimation,low-overhead symbol detection algorithm with joint channel estimation followed by symbol detection algorithm are proposed.When handling the overhead problem in symbol detection,this thesis focuses on the time domain and partitions the large channel matrix into a number of small sub-matrics with the all-zero columns removed,deriving a series new system models with lower diemension.Then UAMP is applied on each model for detection and the results are iteratively updated between time doamin and DD domain.This scheme significantly reduces the overhead but in order to achieve better permance,the size of partitioned block is required to be properly large,leading to higher computation complexity.As for the overhead problem in channel estimation,this thesis considers the superimposed pilot scheme,which is able to avoid overhead completely.Meanwhile,the joint channel estimation and symbol detection scheme is performed to eliminate the interferences between pilot and data symbols.The proposed joint algorithm has no extra overheads,however,the computations in channel estimation require all of the data in DD plane,therefore it has a higher complexity and is suitable for the case with small OTFS frame.