Timeline for Frequency estimation and Cramér-Rao Bound
Current License: CC BY-SA 4.0
11 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Oct 5 at 10:21 | vote | accept | Balasana | ||
| Oct 5 at 10:21 | comment | added | Balasana | Thank you for the source code. When looking at it, I more or less immediately found my dumb mistake: I scaled the noise by $\sigma^2 / 2$ instead of $\sigma / \sqrt{2}$. | |
| Oct 4 at 16:26 | comment | added | Balasana | Let us continue this discussion in chat. | |
| Oct 4 at 11:12 | history | edited | Royi | CC BY-SA 4.0 |
added 299 characters in body
|
| Oct 4 at 11:09 | comment | added | Royi |
@Balasana, Look at github.com/RoyiAvital/StackExchangeCodes in SignalProcessing/Q76644/Q76644C.m. I will update the answer.
|
|
| Oct 4 at 10:48 | comment | added | Balasana | You mean the MATLAB snippet of the Steven Kay algorithm!? I also have to try this. - I still suspect that I am doing something wrong, maybe when converting SNR to noise variance or something. | |
| Oct 4 at 10:38 | comment | added | Royi | @Balasana, Have you looked at the code from the link I posted? For me they collide perfectly with the CRLB I wrote. | |
| Oct 4 at 10:32 | comment | added | Balasana | Yes, of course. This was just a toy example. However, the CRB for known and for unknown phase have the same slope on a log scale. They would just be shifted up or down. So that would not explain, why the MSE curve of the ML estimator (and as I tried also of the MUSIC estimator) has, on a log scale, a steeper negative slope than the CRB. Or at least, I do not see it. | |
| Oct 4 at 10:24 | comment | added | Royi | I don't think in real world you expect to know the phase :-). | |
| Oct 4 at 10:18 | comment | added | Balasana | According to the publication mentioned above, this is the CRB for the case that the phase in unknown. However, in my case the phase is known (i.e., zero). I also tried the derivation on my own and also came more or less to the same CRB (a different constant factor of: 3 instead of 6; see the Q variable in the paper.). | |
| Oct 4 at 5:59 | history | answered | Royi | CC BY-SA 4.0 |