Efficient decentralized dynamic spectrum learning and access policy for multi-standard multi-user cognitive radio networks

In this paper, the problem of an efficient decentralized dynamic spectrum leaning and access (DSLA) for multi-standard multi-user cognitive radio networks (CRNs) is considered. Existing Multi-armed Bandit based decentralized DSLA policies allocate M secondary users to the M orthogonal optimum subbands without any explicit information exchanges or prior agreements among them and have shown to outperform conventional random selection based policies. However, total subband switching cost (SWC) of existing DSLA policies is very high. Also, it is assumed that the desired vacant bandwidth of all secondary users are fixed and equal to the smallest primary channel bandwidth. This assumption is not valid in multistandard CRNs such as 5G where the desired vacant bandwidth of each secondary user needs to be tunable depending on the availability of vacant subbands as well as underlining services such as voice, video, data, Internet etc and hence, a generalized decentralized tunable DSLA policy is desired. The proposed tunable DSLA policy improves the performance as well as SWC by allocating M secondary users to the larger subset of orthogonal optimum subbands, the size of which is tunable compared to a fixed subset size of M in existing DSLA policies. Also, when the desired vacant bandwidth is wider than the smallest primary channel bandwidth, a new approach to quickly and accurately identify the subset of orthogonal optimum subbands is presented. The detailed simulation results validate the superiority of the proposed policy over existing policies.