Research On Achievable Rate And Improvement Of Guessing Scheme Of Decoding Method Based On Guessing Noise

Ultra-reliable low-latency communication(URLLC)is an important application scenario of 5G.The physical layer of URLLC relies on efficient and reliable decoding schemes of short block-length codes,which has drawn attention to high-precision universal short-code decoders.The complexity of traditional decoding algorithms based on the most reliable independent positions grows exponentially with the increase of code rate.In recent years,the emergence of guessing noise decoding algorithms,represented by guessing random additive noise decoding(GRAND),soft GRAND(SGRAND)and ordered reliability bits GRAND(ORBGRAND),has effectively solved the above problem.These algorithms decode by checking the error patterns of the received sequence,and are applicable to any short block-length codes with high code rate.Among them,GRAND is a hard-decision decoding algorithm.SGRAND fully combines soft information to obtain the optimal error performance,but it cannot be implemented in hardware parallel.ORBGRAND combines soft information with an idea suitable for hardware parallel implementation,and achieves a balance between error performance and complexity to a certain extent.But there are still some limitations.Firstly,ORBGRAND is a suboptimal decoder at the information theory level,and there is currently no theoretical analysis of its asymptotic performance.Secondly,the error performance of ORBGRAND still has a large gap with that of SGRAND.Our focus in this thesis is to develop an information theoretic understanding of guessing decoders.Perhaps equally importantly,our information theoretic analysis also yields a way of understanding the asymptotic performance behavior of ORBGRAND.Furthermore,motivated by the information-theoretic study,we also propose improved guessing schemes that are capable of attaining better error performance than ORBGRAND at high signal-to-noise ratio.The main contributions of our work can be summarized as follows.1.Study the asymptotic performance difference of the above three guessing noise decoders.Firstly,by analyzing the relationship among GRAND,SGRAND and ORBGRAND,a general form of guessing noise decoding criterion is established.Based on this general form,achievable rates of three decoding methods are studied.To address the problem of mismatch between the ORBGRAND decoding criterion and channel transition probabilities,we use generalized mutual information(GMI)to evaluate an achievable rate of ORBGRAND.Then,we compare the asymptotic performance differences of the three algorithms through simulations.Finally,a new guessing noise decoding algorithm called cdf-GRAND is designed,which has been shown to be identical to ORBGRAND’s GMI.And we explain the behavior of ORBGRAND’s asymptotic performance approaching that of SGRAND based on cdf-GRAND.2.Improve the error performance of the guessing noise decoder under the finitecode-length.Firstly,the decoding metric function of ORB GRAND is improved by proposing two algorithms,UP-ORBGRAND and B-ORBGRAND,which optimize the error pattern scheduling order and achieve better error performance.In order to further improve the error performance of UP-ORBGRAND,an improvement method of error pattern continuous checking is proposed.Numerical simulations show that the combination of UP-ORBGRAND with this method provides an algorithm that approaches the error performance of soft-decision maximum likelihood decoding with acceptable complexity.