Performance of Q-learning Algorithms for Channel Allocation in Small Cell Networks

The small cell networks (SCNs) architecture can greatly increase data capacity. After LTE-A (Long Term Evolution-Advance), the SCN has become the main architecture of the Fifth Generation (5G) communication systems. In SCN, the increasing interference between adjacent base stations can be effectively controlled through an adaptive channel allocation. Therefore, this paper investigates the adaptive channel allocation of small cell networks by using Q-learning via ε-greedy exploration strategy and Softmax exploration strategy. Simulation results show that the Softmax-LC (Logarithmic Cooling) discovery strategy can obtain throughput of 52 Mbps at 70 time steps, while the Exponential ε-greedy strategy obtains 52 Mbps at 200 time steps.