On the Analysis and Optimization of Massive Machine to Machine Communication Systems

Abstract

Machine to Machine (M2M) Communication Systems play a critical role in providing ubiquitous Internet of Things (IoT) within the emerging 5G technologies, and the realization of M2M is now possible due to the advances in technology and the market drivers. However, there are several challenges which hinder the realization of M2M, such as the massive number of connections that cannot be accommodated by the traditional technologies and access protocols, and the limited capabilities of the IoT devices. In this dissertation, we consider a cellular-based M2M network with limited uplink access resources and a massive number of devices. To accommodate the massive traffic, we propose a hybrid scheduled and group-based random access framework that utilizes device grouping and per-group random access. While the concepts of hybrid access and device grouping already exist in the literature, our main contributions are in the mathematical modeling of the proposed framework, the design and analysis of suitable clustering algorithms, and the application of our results in multiple realistic deployment and optimization scenarios. We study two deployment scenarios for grouping the devices: one that is based on network densifi cation and another cluster-based scenario.

For the network densifi cation deployment scenario, we consider a massive MIMO scheme where the uplink access resources are the orthogonal pilots for MIMO access. We use Stochastic Geometry (SG) analysis to derive the access and coverage probabilities in the proposed access solution. We also provide an optimization study as an example on how our model can be used for optimal resource planning.

We extend our study to include a cluster-based deployment scenario which has a wider range of applications than network densifi cation, and for which the analysis is generalized to consider different kinds of orthogonal resources other than MIMO. To this end, we propose a clustering approach where devices are grouped based on their physical locations as well as their traffic pattern distributions, and we propose two alternative methods for the the traffic-based clustering. The fi rst method uses a kmeans-based heuristic, while the second is based on submodular-cost function minimization for which we prove a factor-two performance guarantee. We also provide an SG analysis for the proposed framework, where we derive expressions of the access and coverage probabilities in the network. We finally extend the study further to include variable device activity and power control within the cluster-based scheme. With this aim, we conduct a spatiotemporal analysis, where we combine SG, probability theory, and iterative algorithms to derive the frequency of successful transmissions and the expected queue status at the devices. The analytical model is then used to study the stability-scalability trade-off in the proposed scheme. Future research directions following this work include the design of the signaling protocol within and between the device groups and management entities, power analysis, and the study of massive MIMO scenarios with imperfect channel state information.