An order optimal policy for exploiting idle spectrum in cognitive radio networks

In this paper a spectrum sensing policy employing recency-based exploration is proposed for cognitive radio networks. We formulate the problem of finding a spectrum sensing policy for multi-band dynamic spectrum access as a stochastic restless multi-armed bandit problem with stationary unknown reward distributions. In cognitive radio networks the multi-armed bandit problem arises when deciding where in the radio spectrum to look for idle frequencies that could be efficiently exploited for data transmission. We consider two models for the dynamics of the frequency bands: 1) the independent model where the state of the band evolves randomly independently from the past and 2) the Gilbert-Elliot model, where the states evolve according to a 2-state Markov chain. It is shown that in these conditions the proposed sensing policy attains asymptotically logarithmic weak regret. The policy proposed in this paper is an index policy, in which the index of a frequency band is comprised of a sample mean term and a recency-based exploration bonus term. The sample mean promotes spectrum exploitation whereas the exploration bonus encourages for further exploration for idle bands providing high data rates. The proposed recency based approach readily allows constructing the exploration bonus such that it will grow the time interval between consecutive sensing time instants of a suboptimal band exponentially, which then leads to logarithmically increasing weak regret. Simulation results confirming logarithmic weak regret are presented and it is found that the proposed policy provides often improved performance at low complexity over other state-of-the-art policies in the literature.

A Sensing Policy Based on Confidence Bounds and a Restless Multi-Armed Bandit Model

A sensing policy for the restless multi-armed bandit problem with stationary but unknown reward distributions is proposed. The work is presented in the context of cognitive radios in which the bandit problem arises when deciding which parts of the spectrum to sense and exploit. It is shown that the proposed policy attains asymptotically logarithmic weak regret rate when the rewards are bounded independent and identically distributed or finite state Markovian. Simulation results verifying uniformly logarithmic weak regret are also presented. The proposed policy is a centrally coordinated index policy, in which the index of a frequency band is comprised of a sample mean term and a confidence term. The sample mean term promotes spectrum exploitation whereas the confidence term encourages exploration. The confidence term is designed such that the time interval between consecutive sensing instances of any suboptimal band grows exponentially. This exponential growth between suboptimal sensing time instances leads to logarithmically growing weak regret. Simulation results demonstrate that the proposed policy performs better than other similar methods in the literature.

Design of Spectrum Sensing Policy for Multi-user Multi-band Cognitive Radio Network

Finding an optimal sensing policy for a particular access policy and sensing scheme is a laborious combinatorial problem that requires the system model parameters to be known. In practise the parameters or the model itself may not be completely known making reinforcement learning methods appealing. In this paper a non-parametric reinforcement learning-based method is developed for sensing and accessing multi-band radio spectrum in multi-user cognitive radio networks. A suboptimal sensing policy search algorithm is proposed for a particular multi-user multi-band access policy and the randomized Chair-Varshney rule. The randomized Chair-Varshney rule is used to reduce the probability of false alarms under a constraint on the probability of detection that protects the primary user. The simulation results show that the proposed method achieves a sum profit (e.g. data rate) close to the optimal sensing policy while achieving the desired probability of detection.

Reinforcement learning based sensing policy optimization for energy efficient cognitive radio networks

This paper introduces a machine learning based collaborative multi-band spectrum sensing policy for cognitive radios. The proposed sensing policy guides secondary users to focus the search of unused radio spectrum to those frequencies that persistently provide them high data rate. The proposed policy is based on machine learning, which makes it adaptive with the temporally and spatially varying radio spectrum. Furthermore, there is no need for dynamic modeling of the primary activity since it is implicitly learned over time. Energy efficiency is achieved by minimizing the number of assigned sensors per each subband under a constraint on miss detection probability. It is important to control the missed detections because they cause collisions with primary transmissions and lead to retransmissions at both the primary and secondary user. Simulations show that the proposed machine learning based sensing policy improves the overall throughput of the secondary network and improves the energy efficiency while controlling the miss detection probability.