LSTM-Based Spectrum Sensing for Improved Detection in Low-SNR Cognitive Radio Networks

Abstract

Spectrum sensing is the first step in cognitive radio systems; before transmitting, a secondary user must confirm that a licensed channel is truly free. The classical energy detector is used as a baseline in this work, and it is improved by reconsidering the way the decision threshold is selected. This makes the test less sensitive to environmental conditions, especially when the signal-to-noise ratio is low. To maintain full control over the evaluation process, binary phase-shift keying sequences are generated and transmitted through an additive white Gaussian noise channel over a wide range of signal-to-noise ratio values. Two versions of the energy detector are investigated: a conventional version based on a fixed threshold and an adaptive version in which the threshold is adjusted according to an estimate of the noise level. In parallel, a detector based on a long short-term memory network is also examined. Unlike the energy detector, this model does not reduce the observation to a single energy value, but instead processes the received samples directly and exploits their temporal structure. The performance is assessed in terms of probability of detection, probability of false alarm, and probability of miss. The results show that adaptive threshold selection improves the performance of the energy detector at intermediate signal-to-noise ratio levels, whereas the long short-term memory based approach remains more robust when noise is dominant and energy-based detection becomes less reliable.