Research On Multi-user Detection Algorithm Based On Gradient Pursuit For Uplink Grant-free NOMA System

In massive machine type communication(m MTC),the signal received in the base station are all coupled together.How to efficiently and accurately recover the user's transmission data from the coupled data has become a difficult problem.Due to its sporadic communication characteristics,compressed sensing(CS)has been applied in joint user activity and data detection in the uplink grant-free non-orthogonal multiple access(NOMA)system.Nowadays,greedy algorithms which based on CS are widely used in m MTC to solve multi-user detection(MUD)problems because of their low complexity and high reconstruction accuracy.However,the MUD algorithms based on the greedy algorithm use the least square method to estimate the signal value.This process needs to solve the inverse matrix of the observation matrix,which bring high computational cost.In order to solve this problem,we designed and implemented two types of low-complexity MUD algorithms by using gradient pursuit(GP)algorithms and combining the characteristics of users transmitting data in actual communication scenarios.In order to reduce the complexity of the multi-user detection algorithm,the structured GP MUD algorithms are studied.The gradient descent-based gradient pursuit MUD(GDGP-MUD)algorithm,which based on Structured Matching Pursuit(SMP)algorithm,uses the structure of active users in one frame and GP algorithm to reduce the complexity.Next,a multi-step quasinewton MUD(MSQN-MUD)algorithm use multi-step quasi-newton method to update direction of the estimated signal value which based on the GDGP-MUD algorithm to enhance the detection accuracy while maintaining low complexity.In order to speed up the convergence,a new decision condition is used to terminate the iteration by judging whether the revised residual energy is reduced.Experimental results show that the GDGP-MUD algorithm reduces the complexity of the MUD algorithm,and the MSQN-MUD algorithm improves the detection accuracy while maintaining a low complexity.The correlation gradient pursuit multi-user detection algorithms are studied to further reduce the complexity.From the perspective of reducing the number of iterations of some time slots,a correlation-assisted GP(CAGP)algorithm is proposed to improve the structured GP MUD algorithms,which use the correlation of active users in one frame and the GP algorithm to further reduce the complexity of the MUD algorithm.Then,inspired by stagewise weak GP(SWGP)algorithm,a correlation-assisted group gradient pursuit multi-user detection(CAGGPMUD)algorithm which based on the CAGP-MUD algorithm is proposed to reduce the iterations of each slot.Experimental results show that the CAGP-MUD algorithm and CAGGPMUD algorithm further reduce the complexity of the multi-user detection algorithm.