Abstract

The Internet of things (IoT) is going to shape the future of wireless communications by allowing seamless connections among wide range of everyday objects. Machine-to-machine (M2M) communication is known to be the enabling technology for the development of IoT. With M2M, the devices are allowed to interact and exchange data without or with little human intervention. Recently, M2M communication, also referred to as machine-type communication (MTC), has received increased attention due to its potential to support diverse applications including eHealth, industrial automation, intelligent transportation systems, and smart grids.

M2M communication is known to have specific features and requirements that differ from that of the traditional human-to-human (H2H) communication. As specified by the Third Generation Partnership Project (3GPP), MTC devices are inexpensive, low power, and mostly low mobility devices. Furthermore, MTC devices are usually characterized by infrequent, small amount of data, and mainly uplink traffic. Most importantly, the number of MTC devices is expected to highly surpass that of H2H devices. Smart cities are an example of such a mass-scale deployment. These features impose various challenges related to efficient energy management, enhanced coverage and diverse quality of service (QoS) provisioning, among others.

The diverse applications of M2M are going to lead to exponential growth in M2M traffic. Associating with M2M deployment, a massive number of devices are expected to access the wireless network concurrently. Hence, a network congestion is likely to occur. Cellular networks have been recognized as excellent candidates for M2M support. Indeed, cellular networks are mature, well-established networks with ubiquitous coverage and reliability which allows cost-effective deployment of M2M communications. However, cellular networks were originally designed for human-centric services with high-cost devices and ever-increasing rate requirements. Additionally, the conventional random access (RA) mechanism used in Long Term Evolution-Advanced (LTE-A) networks lacks the capability of handling such an enormous number of access attempts expected from massive MTC. Particularly, this RA technique acts as a performance bottleneck due to the frequent collisions that lead to excessive delay and resource wastage. Also, the lengthy handshaking process of the conventional RA technique results in highly expensive signaling, specifically for M2M devices with small payloads. Therefore, designing an efficient medium access schemes is critical for the survival of M2M networks.

In this thesis, we study the uplink access of M2M devices with a focus on overload control and congestion handling. In this regard, we mainly provide two different access techniques keeping in mind the distinct features and requirements of MTC including massive connectivity, latency reduction, and energy management. In fact, full information gathering is known to be impractical for such massive networks of tremendous number of devices. Hence, we assure to preserve the low complexity, and limited information exchange among different network entities by introducing distributed techniques. Furthermore, machine learning is also employed to enhance the performance with no or limited information exchange at the decision maker. The proposed techniques are assessed via extensive simulations as well as rigorous analytical frameworks.

First, we propose an efficient distributed overload control algorithm for M2M with massive access, referred to as M2M-OSA. The proposed algorithm can efficiently allocate the available network resources to massive number of devices within relatively small, and bounded contention time and with reduced overhead. By resolving collisions, the proposed algorithm is capable of achieving full resources utilization along with reduced average access delay and energy saving. For Beta-distributed traffic, we provide analytical evaluation for the performance of the proposed algorithm in terms of the access delay, total service time, energy consumption, and blocking probability. This performance assessment accounted for various scenarios including slightly, and seriously congested cases, in addition to finite and infinite retransmission limits for the devices. Moreover, we provide a discussion of the non-ideal situations that could be encountered in real-life deployment of the proposed algorithm supported by possible solutions. For further energy saving, we introduced a modified version of M2M-OSA with traffic regulation mechanism.

In the second part of the thesis, we adopt a promising alternative for the conventional random access mechanism, namely fast uplink grant. Fast uplink grant was first proposed by the 3GPP for latency reduction where it allows the base station (BS) to directly schedule the MTC devices (MTDs) without receiving any scheduling requests. In our work, to handle the major challenges associated to fast uplink grant namely, active set prediction and optimal scheduling, both non-orthogonal multiple access (NOMA) and learning techniques are utilized. Particularly, we propose a two-stage NOMA-based fast uplink grant scheme that first employs multi-armed bandit (MAB) learning to schedule the fast grant devices with no prior information about their QoS requirements or channel conditions at the BS. Afterwards, NOMA facilitates the grant sharing where pairing is done in a distributed manner to reduce signaling overhead. In the proposed scheme, NOMA plays a major role in decoupling the two major challenges of fast grant schemes by permitting pairing with only active MTDs. Consequently, the wastage of the resources due to traffic prediction errors can be significantly reduced. We devise an abstraction model for the source traffic predictor needed for fast grant such that the prediction error can be evaluated. Accordingly, the performance of the proposed scheme is analyzed in terms of average resource wastage, and outage probability. The simulation results show the effectiveness of the proposed method in saving the scarce resources while verifying the analysis accuracy. In addition, the ability of the proposed scheme to pick quality MTDs with strict latency is depicted.