Forums

Blind symbol rate estimation references

Started by roblkc3 5 months ago2 replieslatest reply 5 months ago76 views

Hello!
In the context of a signal classification system I need to (blindly) estimate the symbol rate of an incoming signal. I know nothing about the signal (even its presence is going to be detected by a combined energy+cyclostationary detector).
I have been looking around for some "standard" techniques for the estimation of the symbol rate (besides using cyclostationary signal processing, or saying that it is roughly the BW of the signal) but I was unable to find a "solid" reference --- just a ton of quite similar approaches published on dubious journals/proceedings.

Could anyone point me to a good technique for doing this for both single-carrier and multi-carrier systems, please? ("good" = the system will include some machine learning classification, therefore the signals have to be normalized with a precision sufficient enough that the system still works, there are so many dof in the system that it is impossible for me to say something like "within X%" :S --- sorry for that).

Thanks in advance and have a nice day!

Rob

[ - ]
Reply by SlartibartfastApril 19, 2021

It's a difficult problem.   I've built signal classifiers like this, and what we found was that you nearly always need to start with the BW estimate.  That's fundamentally the first-order estimate of symbol rate for a single carrier signal.  At that point trying to close a timing loop based on the original BW estimate will provide a more accurate estimate once it converges, and you can tweak acquisition time based on SNR requirements, etc.

It is naturally much more difficult with OFDM since the BW doesn't tell the symbol rate story.  In that case if you run an autocorrelator over a guessed range of symbol rates, you should see the CP spiking at the symbol rate.

This stuff isn't easy to get working in the real world, especially if there are SNR and interference performance requirements.

[ - ]
Reply by roblkc3April 20, 2021

Thanks a lot for your insightful reply! It confirmed me that the problem is as difficult as it seemed at first sight :(
... but your suggestions are very useful to avoid wasting time in useless directions.

Thanks again and have a nice day!