Re: [Discuss-gnuradio] Burg spectral estimation

[Martin Braun]

On Fri, Mar 26, 2010 at 01:25:41PM -0400, Brian Padalino wrote:
> If you don't mind, I am curious of your experience of the Burg versus
> Welch implementation and results - which do you prefer?  It looks like
> Burg translates to frequency domain first, and Welch stays in the time
> domain?  Any idea how much of the bandwidth can be occupied before the
> algorithms are not relatively accurate anymore?  In the presence
> high-noise, are the algorithms still able to detect the tones in your
> current setup?

Welch is quite similar to what the GR FFT sink does, and will work with
any kind of spectral content (it is "non-parametric"). Burg needs some
a-priori knowledge of what you're about to expect ("parametric") and
will only operate well if your a-priori assumption reflect the
signal(s).
Actually, it estimates the filter coefficients of an AR process
(so, technically, it operates in the z-domain). The spectral estimation
output is gathered by running that result through an FFT. 

In high-noise, and with little knowledge, you have no choice but use
Welch. A cool application of Burg is to track narrow-band interferers
from only a handful of samples. The pictures on 
https://www.cgran.org/wiki/SpecEst were made with a few thousand samples
for Welch, and only 512 samples for Burg (and way less calculations).

Cheers
MB

-- 
Karlsruhe Institute of Technology (KIT)
Communications Engineering Lab (CEL)

Dipl.-Ing. Martin Braun
Research Associate

Kaiserstraße 12
Building 05.01
76131 Karlsruhe

Phone: +49 721 608-3790
Fax: +49 721 608-6071
www.cel.kit.edu

KIT -- University of the State of Baden-Württemberg and
National Laboratory of the Helmholtz Association

pgp0TDixbcIk8.pgp
Description: PGP signature