LSTM-based ACB scheme for machine type communications in LTE-A networks

Abstract

Access Class Barring (ACB) scheme has been proposed to control the number of machine-to-machine (M2M) devices accessing Long Term Evolution-Advanced (LTE-A) networks. In the ACB scheme, dynamically adjusting barring factor is crucial for massive M2M communications because inappropriate barring factor may degrade the performance of LTE-A networks. Many studies have proposed to adjust the barring factor dynamically by estimating the number of retransmissions or using reinforcement learning (RL) approach. However, because base stations know nothing about the traffic load in the next period of broadcasting barring factor, the barring factor broadcasted by the base stations may not fit the upcoming traffic load. In this study, we proposed a long short term memory (LSTM)-based ACB scheme to predict the traffic load before adjusting barring factor. A LSTM network continuously detected the traffic conditions, and it predicted the upcoming traffic load to adjust the barring factor. With the proposed scheme, we achieved a high access success rate and did not significantly increase access delay and the number of access attempts. We compared our approach to a dynamic ACB scheme and an RL-based ACB scheme in terms of access success rate, average number of access attempts, and average access delay. The simulation results revealed that the access success rate of the proposed approach was above 99%. The average number of access attempts decreased by about 6% as compared to the dynamic ACB scheme, and the average access delay of M2M devices dropped by about 68% as compared to the RL-based ACB scheme.