Recently, the idea of clustering MTDs into smaller groups has emerged as a promis- ing technique to reduce the traffic load on the cellular BS and improve spatial reuse and energy efficiency, while reducing interference in the network, as studied in [18, 26, 46], and [47]. Fig. 2.1 and 2.2 outline an illustrative example of an M2M network topology, without clustering and with clustering. In particular, a clustered M2M network, as outlined in Fig. 2.2, this will effectively reduce the number of MTDs transmitting to the BS. Existing clustering techniques for M2M communi- cations [26, 46, 47, 80, 140–148] have focused on clustering MTDs based on resource allocation, location, load on the random access channel (RACH), and data corre- lation. Clustering has been considered in literature, as an effective approach to alleviate the potential massive congestion caused by MTDs, as done in [140–142] and [143]. These aforementioned works aim to maximise the number of MTDs that attempt to simultaneously access the BS, while minimising network congestion, the load on the RACH and signalling overhead [141, 142]. In [140], an energy-efficient cluster formation (load adaptive multiple access scheme) and cluster head selection scheme was proposed, to maximise network lifetime in a massive M2M network. The work in [142], investigated the problem of random access contention between cooperative groups of MTDs that coordinate their random access channel, while taking into account energy consumption and time varying queue length. However, the algorithm presented in [142] cannot cope with a massive number of MTDs, as its complexity will grow significantly. On the other hand, clustering techniques based
2.2 Data Correlation in M2M Communications 39
Cellular link MTD
Figure 2.1: M2M network topology, without clustering
M2M link Cellular link
MTD
Figure 2.2: M2M network topology, with clustering
on the QoS requirements and locations of MTDs, are proposed in [26], [144–147] and [148], in order to maximise the number of supported MTDs. A cluster priori- tisation scheme for massive access management is studied in [26], where MTDs are clustered based on QoS requirements. The work in [144] proposes a self-organised cluster formation mechanism in which MTDs form clusters with neighboring MTDs. In addition, a number of works such as in [145–147, 149], and [150] have also con- sidered joint clustering and resource allocation. The goal of these works is to maximise MTD data rate, allocate resource blocks efficiently, reduce interference to the cellular network, and optimise the battery lifetime of MTDs. The work in [149] proposes a distributed resource (time) allocation scheme, to address the diverse QoS requirements in an IoT network, while taking into account data rate of CTDs and energy consumption of MTDs. The aforementioned works [26, 140–150] consider machine-centric clustering approaches, that cluster MTDs in order to max- imise MTD data rate and number of supported MTDs, while minimising energy consumption. Such a machine-centric clustering approach does not take into ac- count the individual data/information of each MTD. Additionally, in [151], the authors proposed a clustering approach that uses a coalitional game to optimise the tradeoff between sum-rate gains and power costs.
Meanwhile, the dense deployment of MTDs in M2M networks, will enable MTDs within close proximity to gather correlated data, thus often sending the same infor- mation (redundant bits) to the BS (e.g., see [44,47] and [152]). Hence, adata-centric clustering approach can be used to improve the data quality sent to the BS. Ex- isting work on clustering MTDs with respect to data correlation remains limited.
40 Literature Review
Primarily, the works in [46] and [47] have studied the possibility of MTD clus- tering based on location and correlation, however, these works rely on centralized approaches that are not practical for large-scale M2M networks. Such centralized clustering approaches can lead to significant signalling overhead as they require gathering of global information, such as location and data correlation factors, for a large number of MTDs. Indeed, in practice, centralized clustering approaches are not robust to the dynamic changes in the MTD networking environment, such as the joining of new MTDs, MTD loss of battery, or rapid fluctuations in the sensing environment. Thus, there is a need to introduce new distributed correlation-aware clustering approaches.
To develop such distributed solutions for cooperation in wireless network, it is customary to resort to tools from game theory [61,67]. Particularly, in [80,142,149], and [153], game theory has been used for distributed cluster formation in M2M networks. The work in [153] proposed a distributed correlation-aware cell associ- ation algorithm, that maximises the information sent to the BS while maximising the number of assigned IoT devices to every BS. The game considered in [153] is a two-sided matching game, however the proposed solution cannot cope with dynamic changes in the M2M network environment. Indeed, the existing works in [26, 46, 47, 80, 140–148], and [153] consider only clustering a small, finite number of MTDs. However, this is not the case in practical IoT scenarios as the num- ber of MTDs within the network is massive, which causes substantial interference and impacts the way in which correlation-aware clustering must be performed. Meanwhile, the aforementioned works are also not robust to the dynamics of a large-scale M2M network that results from factors such as the arrival or departure of new MTDs, or the deactivation of MTDs (e.g., due to battery loss, or rapid fluctuations in the sensing environment). In particular, in [80], we have developed an evolutionary coalitional game for correlation-aware clustering, for a finite num- ber of MTDs. However, our proposed distributed algorithm in [80] cannot cope with a massive number of MTDs, as its complexity will grow significantly. Fur- thermore, [80] also relies on a simplistic utility function, that does not capture the real-world deployment of MTDs.
Therefore, this thesis aims to investigate a distributed correlation-aware cluster- ing approach for a finite and massive number of MTDs, while ensuring low signalling overhead and robustness for small stochastic changes in the M2M environment.