Machine-Learning-Based Adaptive Approach for Frame-Size Optimization in Wireless LAN Environments

The performance of IEEE 802.11 wireless local area networks (WLANs) is affected by the channel conditions of wireless medium and the contention processes of wireless stations. Under these effects, it is feasible to adjust the frame size to maximize the throughput. The majority of existing frame-size optimization solutions are based on the assumption that the optimal frame size only depends on the channel conditions. These approaches do not consider the contention effects, thus leading to suboptimal solutions. In this paper, we tackle the frame-size optimization problem using a machine-learning-based adaptive approach. This approach takes both the channel conditions and the contention effect into consideration. In this approach, each source station performs the throughput measurement and collects the ldquoframe size-throughputrdquo patterns. The collected patterns contain the knowledge about the effects of channel conditions and contention processes. Based on these patterns, our approach uses neural networks to accurately model the throughput as a function with respect to the frame size. After the knowledge building, we obtain the gradient information to adjust the frame size. The main contribution of this paper is that we take both the channel conditions and the contention effect into consideration, without any assumptions of the channel models or the stations' behavior that are required by the previous analytical solutions. We perform comprehensive simulations to validate that the proposed approach can effectively optimize the frame size under various channel conditions and contention effects. We also implement the proposed approach and perform comprehensive experiments on a real testbed to evaluate the effectiveness and reaction speed of the proposed approach in practice.