Estimation of future occupation of spectral channels by licensed users in cognitive radio networks applying neuro-fuzzy models

摘要

Cognitive radio is a paradigm that proposes the dynamic management of the radio spectrum, through the integration of sensing, decision making, sharing and spectral mobility. The decision-making phase is in charge of deciding the best channel available to transmit secondary user data (SUs) opportunistically; Its success depends on how efficient the characterization (modeling and estimation) model of primary users (PUs) in spectral bands is. A combination of prediction methodologies based on the ANFIS-GRID Neuro-Fuzzy Inference System (which divides the input data sequence into rectangular sub-spaces) and ANFIS-FCM (based on the use of the Fuzzy C- Means Clustering method) is proposed to reduce the forecast error in spectrum usage behavior by PUs in wireless cognitive radio networks. The results show that the proposed algorithm has the robustness necessary to improve the prediction percentage far above that obtained against methodologies such as Long Short-Term Memory (LSTM) yielding a success percentage greater than 90% in the GSM frequency band. In conclusion, although the proposed ANFIS presents a better performance in the generation of forecasts for time series that represent the dynamics of the PUs, the computational complexity is higher, thus making its implementation in cognitive radio systems based on centralized network topologies feasible.