International Journal of Applied Engineering Research ISSN 0973-4562 Volume 12, Number 19 (2017) pp. 8202-8208 © Research India Publications. http://www.ripublication.com 8202 Neuro-Fuzzy Model for the Prediction of Spectral Occupancy Angie Cordoba 1 , Sergio Castiblanco 1 , Cesar Hernandez 1* and Luis Pedraza 1 1 Universidad Distrital Francisco José de Caldas, Technology Faculty, Calle 68 D Bis A Sur No. 49F – 70, Bogotá, Colombia. * Corresponding author, 1* Orcid: 0000-0001-9409-8341 Abstract Modeling spectral occupancy in cognitive radio networks facilitate the prediction of the primary user’s activity and contribute to an efficient use of the radio-electric spectrum. The purpose of this article is to develop a neuro-fuzzy model to predict the spectral occupancy in a Wi-Fi network (2.4 to 2.5 GHz). To achieve this, the ANFIS algorithm is implemented and its performance is evaluated for two types of membership functions, through the comparison of their results. The obtained results validate the performance of the neuro-diffuse model and its usefulness within cognitive wireless networks. Keywords: ANFIS, cognitive radio, inferential fuzzy system, membership function, radio-electric spectrum, spectral handoff. INTRODUCTION As wireless technologies rapidly evolve, the spectral resource is becoming scarcer due to the prominent user demand. Therefore, it is crucial that the new communication systems use the frequency bands efficiently in terms of time, frequency and space [1]. However, the increase in wireless services requires more spectral resources which leads to scarcity within the available frequencies [2]. This makes it necessary to have intelligent, innovative and efficient models that satisfy all users that require one or several spectral opportunities [3]. Over the last years, the radio-electric spectrum has been much subdivided to give opportunity to different types of wireless services. As a collateral effect, the spectrum has become scarcer [4]. In contrast, several studies have identified that primary users (PU) of the licensed bands are not using it efficiently which would allow a secondary user (SU) to opportunely use the spectrum unused by the PU. This is known as spectral handoff [5]. There are currently several handoff strategies so that a SU profits from spectral opportunities. One of the most interest strategies is the proactive strategy since it minimizes the level of interference between the PU and the SU [6],[7]. In proactive handoff, it is necessary to predict the PU’s spectral occupancy to determine the moment when the channel has to be changed. Hence, cognitive radio (CR) must have the capacity to learn and adapt to the radio environment [8],[9]. The recent development of computational models and their application in diverse areas of science has increased the implementation of problem-solving tools that also estimate unknown parameters. Cognitive radio has not been detached to this process and has required the development of methodologies within the criteria of sharing and flexibility of the spectrum. One of them is artificial intelligence in which the models based on fuzzy logic (FL) and artificial neural networks (ANN) that have the advantage of handling expert knowledge, the faculty of reasoning in the case of FL and the ability of ANN to learn and adapt. The combination of these advantages gives place to neuro-fuzzy models [10]. According to this, the present article intends to develop a neuro-fuzzy model that allows the prediction of the spectral occupancy in a cognitive radio network for the 2.4 – 2.5 GHz frequency band. Such model combines the advantages of FL and the learning ability of ANN. This work chose the ANFIS model proposed by Jang in 1993 due to the existence of tools focused on the implementation of these type of models such as MATLAB [11]. RELATED WORKS According to Hong [2], with the purpose of creating a prediction model for the radio-electric spectrum’s behavior with an efficient detection system and access for CR users, the ANN are adopted as an alternative. They develop a practical learning method applied to the spectrum activity and establish an effective prognostic concerning the states of the channels (inactive or occupied) in future time intervals. This prognostic is obtained initially through a supervised learning process so that the parameters created in the CR nodes define an a priori list with the resulting mobility data without interrupting the current transmission. The empirical results show that the method proposed in this document can conveniently match with the spectrum’s future behavior through a RMSE (root mean square error). Additionally, due to the ANN’s generalization capacity, the generated model can be harnessed in a service radio to predict the another service’s spectrum