|
I have recently discovered Larry Bretthorst's article on Bayesian Spectral Analysis and I'm trying it out. It seems like a really cool approach to spectrum analysis with a few neat results, such as the Fourier transform is ideal in finding a single signal within noise if one takes a probabilistic approach. I'm struggling with making the method able for use to detect multiple harmonic signals, though. Can anyone point me somewhere I would be able to get some example code for BSA or be able to help me? Much appreciated. More detail: I'm particularly struggling with harmonic signals that have close frequencies. My input was two frequencies at 0.2 and 0.8 hertz/sample. Here is my output which suggests two very close frequencies at either 0.2 and 0.8 hertz/sample is more likely than two signals at 0.2 and 0.8 hertz/sample respectively. Here is a closeup of 0.2 hertz/sample. It would seem that the probability of two signals on top of one another at 0.2 hertz per sample is low but then the probability shoots up around that area. Very strange and likely not correct. I would expect this output, but with a function that decreases the probability by 0.5 on the diagonal and gradually less as one moves away from the diagonal. |
