Abstract: Current spectrum sensing methods have some fatal problems, such as poor performance in low SNR (signal to noise ratio) situation, sensitivity to noise uncertainty, and so on. To solve these problems, we propose a fast blind spectrum sensing method using Sevcik fractal dimension in frequency domain. It calculates the Sevcik fractal dimension in frequency domain of the received signal. Then according to the different Sevcik fractal dimension characters between signal and noise, it compares the calculation result with the predefined threshold to determine whether the primary user exists. The proposed method has some excellent advantages, such as insensitive to noise uncertainty, not requiring priori knowledge, and high detection probability in low SNR situation. Furthermore, it consumes less calculation time. So the proposed method can be applied to fast blind spectrum sensing in situation of low SNR and noise uncertainty. Monte Carlo simulation results show that its average detection probability of signals with 7 modulation types can reach 100% when SNR is larger than -12 dB in situation of ±5 dB noise uncertainty.

Key words: communication, spectrum sensing, fractal dimension, noise uncertainty