LSTM model for Channel Occupation Prediction in GSM Band

Abstract

Radio frequencies are a limited resource and hence sold by the governments at premium prices. At any point of time, the licensed users may or may not be using the channel. The objective of Cognitive Radio (CR) is to predict unused time slots in licensed channels at a given time so that others can use such unused slots. However, channel usage patterns may be predicted only in a statistical sense and are essentially random in nature. Therefore, we need a standard data set for comparison of CR techniques. We have created a data set that can be used for simulation, training and testing of CR over GSM band (890-960MHz). A typical file with two hour of observations will have about 1.2 million samples. More than 1000 sets of data samples have been captured from urban and rural areas in India. We have used this data set for prediction using a neural network called Long Short Term Memory (LSTM). The model achieves Lowest Mean Square Error (MSE) of 0.0319 for1024 LSTM units when trained with 100 epochs.