Detection and counting of time and spectrum overlapping signals in a NDA context

We propose in this paper a method based on the cyclostationary properties of telecom signals in order to detect and count the real number of signals recorded by a given captor in a NDA (non data aided) context. Moreover, the context is spectrum survey and the recorded signal is supposed as completely unknown, whereas its bandwidth is considered to be roughly estimated in order to verify the sampling theorem. Our method is dedicated to automatic applications: once the parameters have been settled, the detector can work without any operator. We first prove that the performance can be greatly improved by using logarithmic scale and lowpass filter smoothing. Then we show that the modulation can induce false detection, which can be prevented by introducing a well chosen frequency offset. We have also tested the influence of different algorithms, of the channel perturbations (noise, frequency offset, etc.) and finally of the signal parameters (modulation, number of symbols, etc.). Simulations have been made and we show that we can obtain a probability of true detection equal to about 88%.