Cascaded learning system for SR noise induced spectrum sensing in cognitive radio network

In this paper artificial neural network (ANN) based cascaded learning for stochastic resonance (SR) noise driven spectrum sensing is proposed for cognitive radio (CR). In the first phase of a two-stage cascaded learning system Backpropagation gradient ascent learning is used to estimate optimal noise standard deviation which has an outstanding performance improvement for the discovery of the existence or absence of PU signal in a specified channel. In the second stage, based on results of local detection, binary classification method is used to finally predict the behaviour of the PU channel as either idle or busy channel. The non-robustness of energy detectors (ED) to common noise uncertainty and unable to detect in fading environments is improved by use of adaptive SR noise through learning so that high detection probability is achieved with minimum sensing time under low SNR environment. Therefore, through the cascaded learning system performance parameters such as probability of detection, detection time and false alarm probability of ED in multiple antenna aided CR systems is improved significantly. Moreover, simulation results of our proposed system reveal that throughput and spectral efficiency of the CR (aka secondary user) is improved as compared to the conventional ED algorithms.