Random Access for Machine-to-Machine Communications: Unified Analysis and Optimal Design

Abstract

Driven by the proliferation of machine-type devices (MTDs), wireless communication networks are undergoing a paradigm shift from traditional Human-to-Human (H2H) communications to Machine-to-Machine (M2M) communications. To support massive access of MTDs in M2M communications, random access, with which nodes decide their transmissions in a distributed manner, has been considered an appealing solution thanks to its scalability, flexibility and low-cost nature.

Nevertheless, it has long been observed that random access networks are prone to suffer from poor performance due to uncoordinated transmissions of nodes. For next-generation wireless communication networks, which are expected to accommodate a myriad of MTDs and meet their stringent quality-of-service requirements, it is of utmost significance to understand how to design random access schemes for optimizing the network performance. A series of crucial issues on the optimal design of random access, however, remain unresolved. First, it is not fully understood how to optimally tune access parameters for some widely used random access strategies, such as connection-based Aloha. Second, across various random access strategies, which one should be adopted is still obscure due to the lack of unified analysis. Third, in addition to access parameters, the access strategy can indeed be optimized by leveraging learning-based approaches, with which nodes can find the optimal access strategy based on their observations and experience. Yet learning-based access schemes are often designed empirically owing to the absence of theoretical guidance.

In this thesis, a comprehensive study on the optimal design of random access is conducted to address the above issues. It begins with the modeling and optimization for a given random access strategy, connection-based Aloha. Specifically, a request-queue model is established for connection-based Aloha, based on which the network throughput and mean queueing delay of data packets are characterized and further optimized by tuning the transmission probability of each node. By comparing the optimal throughput and delay performance of connection-based Aloha with connection-free Aloha, conditions for beneficial connection establishment are discussed, and further applied to cellular networks to investigate whether grant-free access is preferable over grant-based access for M2M communications.

To further compare the optimal performance of different random access strategies, a unified analytical framework is established, which incorporates various design features, including sensing-free or sensing-based, connection-free or connection-based, and backoff. Based on that, the mean queueing delay of data packets with various random access strategies is characterized and optimized in a unified manner. A comparative study on the optimal delay performance is further presented between sensing-free Aloha and sensing-based Carrier Sense Multiple Access (CSMA) to obtain the criteria for beneficial sensing. The analysis is also applied to the random access schemes in cellular networks to demonstrate when and how significantly their delay performance can be improved by sensing.

In the last part, we leverage learning-based approaches for throughput optimization, and demonstrate that the queueing-theoretical analysis can serve as a powerful tool for optimal design of learning-based access. Specifically, based on a Multi-Armed-Bandit (MAB) framework, two random access schemes, MTOA-L with local rewards and MTOA-G with global rewards, are proposed and shown to both achieve the maximum throughput of 1, though with different short-term fairness performance. By identifying the access strategies learned via MTOA-L and MTOA-G and feeding them into the proposed queueing-theoretical framework, the optimal tradeoff between throughput and short-term fairness and the corresponding parameter setting are obtained for both MTOA-L and MTOA-G, which sheds important light on optimal design of learning-based access.

Date of Award	3 Sept 2025
Original language	English
Awarding Institution	City University of Hong Kong
Supervisor	Lin DAI (Supervisor)