Machine Learning–Based Mobility Robustness Optimization Under Dynamic Cellular Networks

Abstract

In this paper, we propose a machine learning−based mobility robustness optimization algorithm to optimize handover parameters for seamless mobility under dynamic small-cell networks. Small cells can be arbitrarily deployed, portable, and turned on and off to fulfill wireless traffic demands or energy efficiency. As a result, the small-cell network topology dynamically varies challenging network optimization, especially handover optimization. Previous studies have only considered dynamics due to user mobility in a specific static network topology. To optimize handovers under dynamic network topologies, together with user mobility, we propose an algorithm consisting of two steps: topology adaptation and mobility adaptation. To adapt to a dynamic topology, the algorithm obtains prior knowledge, which presents a belief distribution of the optimal handover parameters, for the current network topology as coarse optimization. In the second step, the algorithm fine-tunes the handover parameters to adapt to user mobility based on reinforcement learning, which utilizes the knowledge obtained during the first step. Under a dynamic small-cell network, we showed that the proposed algorithm reduced adaptation time to 4.17% of the time needed by a comparative machine-based algorithm. Furthermore, the proposed algorithm improved the user satisfaction rate to 416.7% compared to the previous work.INDEX TERMS Transfer learning, distributed reinforcement learning, small cell on/off, self-organizing network, handover optimization.

ZEUS: Handover algorithm for 5G to achieve zero handover failure

“…One of the objectives is to observe multiple HO events with associated parameters, use this information to train its machine learning model, and try to identify sets of parameters that lead to successful HOs and sets of parameters that lead to unintended events. Some recent literatures have shown that machine learning‐based mobility robustness optimization can improve the user satisfaction rate and increase network performance significantly by optimizing HO parameters [51,52]. We anticipate that the mobility optimization enhanced by the use of machine learning can find the optimal HO parameters to maximize the MASE solving the various complex trade‐offs.…”

Advanced Mobility Robustness Optimization Models in Future Mobile Networks Based on Machine Learning Solutions

“…In [40], a distributed reinforcement learning was proposed to adapt the UE mobility along with proposing ML-based algorithm (i.e., transfer learning based algorithm) for a dynamic network topology adaption. The main objective of the proposed algorithm is to optimize the HOM, TTT and CIO over a dynamic small BSs to minimize the ratio of the HOF and HOPP.…”

“…Article [31] applied random walk mobility model. Manhattan grid mobility model and random way point mobility model are presented in [40], the random way point mobility model used in pedestrian environment at user speed of 5 km/hr. Constant velocity mobility model is introduced in [41].…”

“…The common types of ML have been implemented to optimize the HCPs of MRO functions, such as the supervised ML [21], [22], [23], [24], [25], [26], [27], unsupervised ML [28], and reinforcement learning [14], [29], [30], [31], [32], [33], [34], [35], [36], [37], [38], [39], [40], [41], [42]. Several ML techniques under each type have been addressed as a solution method.…”

“…The fuzzy Q-learning technique was used throughout numerous works [14], [29], [30], [33]. The research of [31], [32], [35], [38], [39], and [40] presented a Q-learning technique to acquire optimal HCP settings. The Q-learning technique was integrated with other methods such as Q-learning and Analytic hierarchy process technique for order of preference by similarity to ideal solution (AHP-TOPSIS) [34], deep Q-learning [42].…”