A learning approach to link adaptation based on multi-entities Bayesian network

Adaptive modulation and coding (AMC) is widely used in modern communications which attempts to predict the best available rate and select the most suitable modulation and coding scheme (MCS) by estimating the real-time channel quality to obtain higher throughput of communication system. However, due to the characteristics of wireless channel fading, there are a lot of uncertainties in the communication process, which makes deviation between the channel estimate and the true value and can affect performance of AMC system. Bayesian network is an important tool to research uncertainty. This paper considers learning with the multi-entities bayesian network (MEBN) as a new framework for adaptive modulation and coding which avoids the flaw of flexibility in traditional Bayesian network (BN). Simulation results show that our algorithm has more validity in the selection MCS and lower bit error rate (BER) by considering estimate deviation in MEBN-AMC system. We also provide the further simulation results by using Bayesian structure learning and parameter learning.