Predictive opportunistic spectrum access using learning based hidden Markov models

To realize opportunistic spectrum access, spectrum sensing is applied to detect the presence of spectrum holes. If secondary radios (SRs) randomly or sequentially sense the channels until a spectrum hole is detected, significant amount of the scarce spectrum resource will be wasted, since SRs transmit only after a decision has been made. On the other hand, with the use of an intelligent predictive method, SRs can learn from the past activities of each channel to predict the next channel state. By prioritizing the order in which channels are sensed according to the channels availability likelihoods, the probability that an SR gets a channel upon its first attempt significantly increases, and thus reduces the possible waste. This paper introduces a learning-based hidden Markov model (HMM) to predict the channel activities. Simulation results show that the proposed HMM can predict the channel activities with high accuracy after sufficient training. Our algorithm predicts the availability of the channels by only making use of the current state of the spectrum. Furthermore, by incorporating the outcome of the actual channel sense, our algorithm is able to make self-regulation before next decision, so that errors will not propagate.