Font Size: a A A

Research On User Activity Detection And Channel Estimation For Massive MIMO Grant-free Access

Posted on:2021-10-05Degree:MasterType:Thesis
Country:ChinaCandidate:X QiuFull Text:PDF
GTID:2518306476450094Subject:Communication and Information System
Abstract/Summary:PDF Full Text Request
The main characteristics of the massive machine-type communication(m MTC)scenario for the future Internet-of-Things(Io T)communication are: massive connectivity,that is,huge number of terminals;short data packets transmitted by terminals;the activity pattern of terminals is data transmission by terminals is sparse.The traditional authorized access method applied in the Io T communication scenario will bring huge access signaling overhead,reduce the transmission efficiency of the system;and when a large number of terminals request to establish a connection at the same time,the preamble sequence will collide,which will cause network congestion and greatly increase the transmission delay of the system.Therefore,related research proposes to apply the grant-free access scheme based on the shared collision domain to this scenario.When the base station(BS)is equipped with a large-scale antenna array,the grant-free access in the m MTC scenario will face the following challenges: first,when multiple users transmit data on a shared wireless resource block,the data superposition between users makes the BS unable to distinguish which users are transmitting data,it is necessary to study the user activity detection;second,if orthogonal pilots are allocated for massive users,this will bring huge pilot overhead,it is necessary to study the design of non-orthogonal pilots and efficient channel estimation methods with achievable complexity.In this article,we focus on the massive MIMO grant-free access scenario with large-scale antennas on the BS side,and studies joint user activity detection and channel estimation problems,non-orthogonal pilot design and scheduling problems.The specific content includes:First,review the approximate message passing(AMP)algorithm for sparse signal reconstruction in compressed sensing theory.Using the central limit theorem and Taylor's second-order expansion approximation under the constraints of large systems to approximate the message passing rules of the BP(Belief Propagation)algorithm,the TAP-AMP algorithm can be obtained.Base on the TAP-AMP algorithm and further assuming that the elements of the sensing matrix obey the Gaussian distribution CN(0,1/m),the BAMP(Bayesian optimal AMP)algorithm can be derived,where m is the number of rows in the sensing matrix.Replacing the denoiser function of the BAMP algorithm with the soft threshold function can obtain the BPDN-AMP algorithm,and the BPDN-AMP can be converted to the BP-AMP algorithm in the special case of zero noise.In addition,in view of the limitation that the AMP algorithm can only solve the problem of single measurement vector(SMV)in CS,the vector AMP algorithm generalizes the AMP algorithm to multiple measurement vectors(MMV)under the condition that the prior statistical characteristics of the sparse signal are known and the noise is additive white Gaussian noise.In view of the limitation that the AMP algorithm can only solve the linear SMV problem in CS,the GAMP algorithm can solve sparse signal reconstruction problems in generalized linear compressed sensing models with arbitrary input and output distributions,and it can provide the approximation of the MMSE and MAP estimates of the sparse vector,which are called the Sum-Product GAMP algorithm and the Max-Sum GAMP algorithm,and when the noise is additive white Gaussian noise,the Sum-Product GAMP algorithm will degenerate to the TAP-AMP algorithm.Afterwards,we review the Bethe free energy theory framework,under which different message passing algorithm variants can be derived by designing different constraints on auxiliary beliefs,such as: BP algorithm,AMP algorithm,Expectation Propagation(EP)Algorithms,etc.Second,focus on the joint user activity detection and channel estimation problem in massive MIMO grant-free access systems,unified modeling the massive MIMMO grant-free access problem with scene universality(suitable for multiple channel statistical models,suitable for known or unknown user priori active statistical characteristics,suitable for known or unknown priori parameters of channel statistical model).On the basis of this,from the perspective of variational Bayesian inference,the problem of user activity detection and channel estimation in grant-free access scenarios is transformed into a variational free energy minimization problem,and use the Bethe free energy framework to design the auxiliary beliefs with marginalization consistency constraints,mean and variance consistency constraints,and factorization constraints,transform the problem into the Bethe free energy minimization problem under hybrid constraints,finally the general form of a hybrid messaging passing algorithm for joint active user detection and channel estimation in massive MIMO grant-free access systems is derived.Furthermore,the HMP algorithm framework is applied to the user activity detection and channel estimation in independent and identically distributed(i.i.d.)complex Gaussian fading channels,massive MIMO spatially correlated channels,and scenarios where the user priori active statistical characteristics are known or unknown.Simulation results show that the proposed method not only has the scene universality,but also can obtain better performance than the existing methods in various scenarios.Finally,focus on the problem of non-orthogonal pilot design and scheduling in massive MIMO grant-free access systems,the research on sequence design is carried out from two perspectives: compressed sensing and Welch Bound Equality(WBE)sequences.From the perspective of CS,the pilot matrix should meet the restricted isometry property(RIP)to ensure that the information of the sparse vector to be estimated is not lost during the measurement.This performance can be measured by the correlation of the pilot matrix,and the correlation is smaller,the pilot matrix performance is better.Therefore,a deterministic non-orthogonal pilot matrix design method is proposed,and a ZC pilot matrix is designed based on the ZC(Zadoff Chu)sequence.From the perspective of minimizing the mean square error of the LS and LMMSE channel estimates,the optimal non-orthogonal pilot sequence should be the WBE sequence that meets the Welch Bound Equality.Therefore,a WBE sequence-based non-orthogonal pilot matrix design method is proposed.In addition,based on the special case of the WBE sequence: Maximum Welch Bound Equality sequence,a MWBE pilot matrix is designed.The MWBE matrix can not only satisfy the WBE equation,but also reach the theoretical lower bound of matrix correlation.For the joint user activity detection and channel estimation problem in grant-free access systems,the most ideal non-orthogonal pilot matrix should be the MWBE pilot matrix.For grant-free access systems where the physical channels are the massive MIMO spatially correlated channel,pilot scheduling is considered using orthogonality between channel covariance matrices to improve user activity detection and channel estimation performance of the hybrid messaging algorithm.Simulation results show that when pilot scheduling is not considered,the hybrid messaging algorithm achieves the best user activity detection and channel estimation performance under the MWBE pilot matrix;after pilot scheduling,the performance of the hybrid messaging algorithm can be further improved.
Keywords/Search Tags:massive MIMO, grant-free access, compressed sensing, user activity detection, channel estimation, non-orthogonal pilot design
PDF Full Text Request
Related items