ARMA Model Order Estimation

Hi everyone,

I have a problem and I hope to receive your help: 

I am applying Adaptive Prony's Method to identify spectrum of a non-stationary signal (the signal I am processing is piece-wise stationary). Adaptive Prony's Method models each of the stationary intervals as an AutoRegressive-Moving-Average process and from that, we can find the power spectrum for each stationary interval. However, the method requires the model order for each stationary interval to be known in advance, and this is the problem perplexing me. I hope that you could introduce me a method that can help to estimate the model order for a set of samples, and the related matlab command, if it is available. 

Thank you so much and sorry for my English.

On 7/11/2013 10:14 AM, Anh Tuan Phan wrote:

> I am applying Adaptive Prony's Method to identify spectrum of a
> non-stationary signal (the signal I am processing is piece-wise
> stationary). Adaptive Prony's Method models each of the stationary
> intervals as an AutoRegressive-Moving-Average process and from that,
> we can find the power spectrum for each stationary interval. However,
> the method requires the model order for each stationary interval to
> be known in advance, and this is the problem perplexing me. I hope
> that you could introduce me a method that can help to estimate the
> model order for a set of samples, and the related matlab command, if
> it is available.

Increase model order until you reach desired accuracy; or until 
increasing order substantially improves accuracy.

Vladimir Vassilevsky
DSP and Mixed Signal Designs
www.abvolt.com