Distributed frequency allocation for cellular networks: Trade-offs and tuning strategies

Frequency allocation in cellular networks has a number of optimisation criteria and a number of interesting trade-offs involved in trying to meet them. Analysing them is particularly important for applying frequency allocation algorithms in practice. However, this can be prohibitively complex, even in a sequential setting, and especially in a distributed setting. The latter implies a larger number of parameters to consider, but is more suitable for dynamic solutions. This, besides analytical evaluation, necessitates the use of experimentation in tuning the algorithms’ performance. In [4], it was shown how to analytically measure and provide worst-case guarantees regarding request satisfiability and how to provide fully dynamic frequency assignment (i.e. ondemand) with low communication and time overhead, which is particularly important for adapting to temporal changes in frequency demands in each cell. In applying these methods, the trade-off between connection set-up time and request satisfiability had to be considered. Our conjecture was that, by designing appropriate dynamic tuning strategies, the trade-offs involved could be balanced in a way so that the gains are substantial, while the losses remain small. The key results of our study, confirming the above conjecture, are the following: We propose strategies for applying and fine tuning the performance of the methods in [4], mainly addressing the above decision and without affecting the worst-case guarantees. The proposed strategies are dynamic, i.e. the amount of frequencies allocated at each base station follows the offered load at that station. Next we study the performance of the original algorithms and their tunable versions. The set of experiments model a variety of situations, including dynamic, non-uniform load and base station failures, having the main variable parameter actually be the temporal-change-of-traffic-distribution; it measures an extensive range of performance metrics (including utilisation, frequency reuse, fault-tolerance). Also, our study provides a way to measure the down-side of the trade-off and get some new insights into the issues involved in attempting to optimise key constraints.