This thesis investigates spectrum sensing for Cognitive Radio. First, we deal with the hidden node problem and the selection of sensing frequency in a P2P CR system. Different from existing models where TSU executes sensing and transmitting periodically, a novel detecting model is proposed to consider simultaneous sensing and transmitting. At RSU, BER estimation is applied to detect whether a PU is active or not. Simulation results show that i) simultaneous sensing improves spectrum utilization compared with periodical sensing; ii) The BER estimation improves the probability of detection and spectrum utilization. The second challenge related to decision fusion in cooperative sensing. We propose novel schemes based on machine learning and introduced probability vector to improve system performance. Simulation results demonstrate that the new schemes are better than the traditional ones. The probability vector can shorten the training duration and the classification delay compared with energy vector.