Active User Detection And Signal Recovery Algorithm Of Large-scale Asynchronous Grant-free Random Access

Large-scale grant-free random access is one of the bottleneck problems in field of massive Internet of Things(m Io T),where active user detection and the signal demodulation is key to the Io T receiver.Massive machine type communication is characterized by sparse access and massive connections,so many methods based on compressed sensing have been applied to the joint problem of active user detection,channel estimation and signal recovery.It is one of the important research directions in massive machine type communication systems for the use of compressed sensing theory in getting stable,reliable and practical active user detection solutions.In addition,the popularity of machine learning has attracted the interest of many scholars in recent years,and various deep-learning-based approaches have been proposed to deal with the receiver design problems in the traditional communication field.At present,most of the random access approaches assume that the access user is slot-synchronous,which,however,is not practical in general.The performance of active user detection and signal recovery will be seriously degraded and even not work if the received signal is not synchronized.This paper mainly studies the active user detection and signal recovery algorithms under the scenarios of quasi-synchronous and asynchronous grantfree random access,the main contributions can be summarized as follows:Firstly,a low-complexity greedy algorithm is proposed for active user detection and signal recovery in the quasi-synchronous grant-free random access scenario.It first estimates the most active user and its delay,and then uses the BCD algorithm for channel estimation and signal demodulation.Then,the signal components of the currently estimated most active user are eliminated in the received signal,and the control is transferred to the detection and signal demodulation of the next active user.Experimental results show that the proposed quasisynchronous greedy algorithm performs very close to the ideal counterpart.Secondly,the impact of delay on the system was introduced and an asynchronous system model was established for the asynchronous grant-free random access.In order to reduce the impact of delay on active user detection and signal recovery,the problem of joint delay estimation,active user detection,channel estimation and signal recovery is formulated.An asynchronous active user detection and signal recovery algorithm based on DA-AUD algorithm is proposed.The algorithm firstly estimates the most active user and its access delay,and then performs channel estimation and signal demodulation.Then,the signal contribution of the currently estimated most active user is eliminated from the original signal,and the algorithm turns to the detection and signal demodulation of the next active user.The performance of the proposed algorithm is compared with that of the synchronous OMP algorithm and DA-AUD algorithm.Simulation results show that the proposed algorithm performs very close to the ideal counterpart at high signal-to-noise ratios.Lastly,several state-of-the-art(STOA)deep-learning approaches for active user detection are compared and analyzed,and simulation and comparison experiments are conducted to analyze their performance.We show that compared with the traditional algorithm based on compressed sensing,the deep-learning-based approaches have a relatively higher success rate of active user detection and lower computation complexity,while the traditional compressed-sensing algorithm does not require the training process and can be applied immediately.