Access Delay Optimization of M2M Communications in LTE Networks

　

M2M communications has experienced a remarkable growth over the past few years. Many emerging M2M use-cases, such as industrial automation, rely on the realtime control, which imposes stringent latency requirements on wireless connectivity. However, recent studies have demonstrated that as a key enabler for M2M communications, the LTE networks would suffer from severe congestion at the random access channel when a massive number of Machine-Type Devices (MTDs) simultaneously send access requests for connection establishment. In that case, excessively long access delay occurs due to low chances of successful access. How to optimize the access delay performance has thus become a significant challenge for supporting M2M communications over LTE networks.

Despite extensive studies, effects of key system parameters on the access delay performance have not been well understood. For instance, the data arrival rate of each MTD, which determines the traffic load of the network, has been ignored in previous studies, where MTDs were usually assumed to be bufferless. How the mean access delay varies with the total number of MTDs or the number of preambles under different traffic conditions thus remains largely unknown. More importantly, due to the lack of accurate information of the number of access requests, the mean access delay has to be numerically calculated based on the estimated or iteratively updated number of access requests. The implicit nature renders it extremely difficult to further study how to optimize the access delay performance of each MTD by properly tuning backoff parameters.

In [Zhan-Dai'18], a new analytical framework was proposed for M2M communications in LTE networks to maximize the network throughput. In [Zhan-Dai'19-WCL], the proposed analytical framework is further extended to optimize the access delay performance of MTDs. Specifically, based on the discrete-time Markov process of each access request, the probability generating function of the access delay is derived, from which the first and second moments of access delay can further be obtained. The analysis shows that when the UB window size W = 1, the minimization of the second moment of access delay is equivalent to the minimization of the first moment of access delay, i.e., the mean access delay. Explicit expressions of the minimum mean access delay and the corresponding optimal ACB factor are obtained, which show that the minimum mean access delay linearly increases with the number of MTDs and is inversely proportional to the number of preambles when the total data arrival rate is large.