Cognitive Spectrum Decision via Machine Learning in CRN

In this research, we propose cognitive spectrum decision model comprised of spectrum adaptation (via Raptor codes) and spectrum handoff (via transfer learning) in Cognitive Radio Networks(CRN), in order to enhance the spectrum efficiency in multimedia communications. Raptor code enables the Secondary User (SU) to adapt to the dynamic channel conditions and maintain the Quality of Service (QoS) by prioritizing the data packets and learning the distribution of symbols transmission strategy called decoding-CDF through the history of symbol transmissions. Our scheme optimizes the acknowledgement (ACK) reception strategy in multimedia communications, and eventually increases the spectrum decision accuracy and allows the SUs to adapt to the channel variations. Moreover, to enhance spectrum decision in a long term process, we use Transfer Actor Critic Learning (TACT) model to allow the newly joined SU in a network to learn the spectrum decision strategies from historical spectrum decisions of the existing ‘expert’ SUs. Experimental results show that our proposed model works better than the myopic spectrum decision which chooses the spectrum decision actions based on just short-term maximum immediate reward.