“Advancements in Cognitive Radio Systems: A Spectrum Occupancy Prediction Model Development”
Abstract:
Cognitive Radio (CR) systems have two main users: the Primary User (PU) and the Secondary User (SU). The SU must refrain from transmitting on a channel until it senses and determines its occupancy state to avoid interference. However, this process introduces significant delays and inefficient spectrum utilization. To address this issue, a channel predictive system is proposed.
In this study, a machine learning model is developed for spectrum occupancy prediction using Power Spectrum Density (PSD) data collected for 24 hours in Minna, Niger State, and FCT Abuja in Nigeria, within the VHF band (30-300 MHz). Exploratory Data Analysis (EDA) using power density plots is employed to reduce the dataset dimensionality, making it suitable for machine learning. The power density plots reveal 12 distinct frequency sub-bands in the dataset.
A Back-Propagation Neural Network (BPNN) model is created to predict spectrum occupancy using time-series data converted into feature vectors representing the occupancy of all frequency sub-bands. Twenty-four input parameters, capturing hourly spectrum occupancy, are used, resulting in a single output predicting spectral occupancy. The neural network’s initial weights are improved using an Auto-Regressive (AR) model, with coefficients obtained from the synaptic weights and adaptive coefficients of the nonlinear sigmoid activation function in a hidden layer of a ten-neuron Real-Valued Neural Network (RVNN). A linear activation function is used in the output layer.
To obtain the AR coefficients, training data and corresponding expected occupancy, estimated from raw data, are passed to the neural network alongside the number of neurons in the hidden layer. The neural network then trains itself to generate optimal weights for the AR model to predict unseen data.
The prediction model’s performance is evaluated on the entire training data, validation dataset, and test dataset. Results for the Minna dataset in band 1 (30-47 MHz) show the highest actual spectral occupancy of 40.59% with a prediction accuracy of 99.06%, while band 7 (137.05-144 MHz) has the lowest occupancy of 25.24% with a prediction accuracy of 99.31%. Similarly, for the Abuja dataset, band 1 (30-47 MHz) has the highest actual spectral occupancy of 39.11% with a prediction accuracy of 98.59%, while band 11 (230.05-267 MHz) has the lowest occupancy of 22.13% with a prediction accuracy of 99.40%.
The results suggest that band 1 has the highest spectral occupancy in both locations and should be avoided for Cognitive Radio (CR) deployment. The Neural Network prediction model demonstrates an accuracy of 91.51% on an unseen test dataset, 99.02% on the training dataset, and 91.63% on the validation dataset.
“Advancements in Cognitive Radio Systems: A Spectrum Occupancy Prediction Model Development”