Spectrum Prediction Based on Echo State Network and Its Improved Form

Cognitive Radio (CR) is an efficient solution to spectrum scarcity as it can sense the spectrum Based on previous information about the spectrum evolution in time, thus predicting the future occupancy status. Framed within this statement, the method of spectrum prediction Based on a new type of recurrent neural network which called echo state network (ESN) and its improved form are proposed in this paper. In view of ESN problems in practical applications, a new ESN structure is constructed. The proposed ESN is constructed by a cycle reservoir with fixed feedback connections. In order to compare its comprehensive properties to the traditional ESN, benchmark series with different origin and characteristics are simulated, experimental results show that the performance of the improved ESN can be comparable to the traditional ESN. In addition, an improved particle swarm optimization (θ-PSO) algorithm is used to select parameters for the optimal design of ESN and the improved ESN. Finally, the state duration of the authorized spectrum occupied and idle is predicted by ESN and its improved form, and it shows satisfactory results.