Countermeasuring Aggressors via Intelligent Adaptation of Contention Window in CSMA/CA Systems

Abstract

To coordinate channel access and reduce collisions over unlicensed bands, wireless technologies implement a listen-before-talk (LBT) strategy, a variant of Carrier Sense Multiple Access (CSMA) with Collision Avoidance (CA). In LBT, a node backs off for a randomly selected amount of time, upper-bounded by the minimum contention window (CWmin) which is specified by standard settings. However, an aggressive node can choose a lower CWmin value, deviating from standards settings, to gain an unfair throughput advantage at the cost of compliant nodes performance. To address this problem, we propose a framework called Intelligent Contention Window (ICW) that allows compliant nodes to adapt their CWmin values to counter aggressive nodes and achieve their fair share of the channel’s airtime. The adaptation process is based on a random forest, a machine learning model that includes a large number of decision trees. We train the random forest in a supervised manner to recommend the possible best CWmin over a large number of spectrum sharing scenarios. Our results show high generalization performance of the random forest for diverse aggressive spectrum sharing settings. We validate our design using over-the-air hardware experiments as well as simulations. Our results suggest that under ICW, nodes receive their fair shares of the channel airtime and achieve multi-fold boosting in throughput and reduction in latency in both static and dynamic aggression settings. Our SDR experiments show $5.62\times $ throughput improvement when ICW is used relative to the Wi-Fi protocol.

Keywords

• Aggressive behavior
• CSMA/CA
• CWₘᵢₙ
• distributed MAC
• fairness
• machine learning