Prediction of Channel Availability in Cognitive Radio Networks Using a Logistic Regression Algorithm

—The capacity of predicting spectral occupancy in cognitive radio networks offers the possibility of developing better policies in channel assignment to secondary users, according to the predicted spectral opportunities. This work develops a prediction model to determine and exploit spectral opportunities while avoiding the continuous search for channel availability in cognitive radio networks. The proposed scheme creates an availability prediction matrix for every available channel in the GSM band that includes their times of availability. By using this information, there is a potential to improve channel allocation policies. The model contains two processes: the first one performs a training process in order to prepare the prediction algorithm so that it can make more reliable predictions and the second one uses the logistic regression algorithm to estimate the availability in every available frequency which can be profited by secondary users, who intend to start transmissions. Measurements were made for average bandwidth, average delay and prediction error. The results obtained were evaluated with real spectral occupancy data in the GSM frequency band. The developed model shows a low prediction error which enables optimal channel assignment mechanisms, hence minimizing failed handoffs through the channel occupation of primary users.

entered the model, and how was the input data summarized. The results show that the SVM models were capable of detecting the distraction with an average accuracy of 81.1% surpassing the logistic regression model.
In [8], a two-stage cognitive process is proposed with the purpose of learning from the capacities of the physical layer under different channel conditions which leads to an optimization of the package delivery in a multi-antenna radio. In the first step, it learns the characteristics of the available techniques and in the second step the configuration is defined according to the radio's goals and channel conditions. A design is proposed based on the Bayes rule which will be used as a baseline for future comparisons. In this work, the Naïve Bayes, Semi-Naïve Bayes and binary search models are studied since they offer several learning techniques and optimize the design used as a reference. It requires the estimation of fewer parameters but sacrifices optimization in terms of performance to save speed and memory.
In [9] the authors mention the main goals of cognitive radio (CR) pointing out that one of them consists on improving the inefficient use of the spectrum. It also states how cognitive radio can dynamically perceive the spectrum to gather information. Such gathering can be used to describe spectral opportunities as well as determining the future occupancy of the spectrum. In the work developed by the authors, binary time series are used to characterize and predict the occupancy of the spectrum. The deterministic and non-deterministic occupancy data is then examined showing the results over both scenarios.
In [10] the self-regressive linear approach for binary time series is adopted to study the performance in the prediction of the channel occupation based on the spectrum measurements carried out in a synchronized fashion in four different places. Through the modeling, the dependence of the adjacent frequencies in the frequency domain is factored. The order of the model is selected in terms of the measured residual magnitudes and the Akaike information criteria, by tabulating the results and considering the time of observation for each location. The performance of the proposed linear system is compared with the Markov chain model in continuous time for one of the locations.
In the research carried out by [11] the objective is to improve the detection probability using the regression of binary decision-making obtained from individual cognitive radio nodes between the primary user and the cooperative center of spectrum detection. It is an innovative technique that adds upgrades to the probability system using the deterministic nature of propagation loss from long distance radio in a distributed detection center. The simple model of Log-distance path loss is considered in this work and the Log-normal shading effect is assumed. The established framework can be extended to other variations and combinations of path loss models. It can also implement an advanced pattern of adaptation techniques.
In [12] the importance of the efficient use of the spectrum is pointed out due to the omnipresence of wireless technologies that are currently at disposal. The detection of the spectrum is a key stage towards an efficient use of the spectrum. The detection of energy is a fast and simple method to detect the spectrum but its accuracy is limited by the dependence on a threshold value. This article describes a new detection method for spectrum energy in real time using the logistic regression classifier. The implementation is performed using USRP and GNU-Radio, and reaches 98.6% accuracy in classification with a dataset collected over the commercial FM band. Fig. 1 shows the block diagram of the proposed prediction model. The first block called "Spectrum occupancy database" includes real occupancy data corresponding to the GSM band (824 MHz -874 MHz). This block is the input for the spectrum data processing that has the task of defining the occupancy or availability of each channel in the GSM band according to the false alarm probability equation.

III. METHODOLOGY
The rectangular area corresponds to the proposed model that consists of two algorithms (1) Logistic regression algorithm and (2) Channel assignment prediction. The first method's function is to train the algorithm for a ten-minute period through the use of training variables such as PSINR, availability and average availability time. This leads to the calculation of parameters known as cost and gradient which are necessary to adjust the predictor. The second method assigns the channel occupancy by setting "1" and "0" which outputs an availability prediction matrix regarding the bandwidth.

A. Defin
The m intervals lasts 10 m Additi which si differenti and low character

B. Logis
The