Adaptive Filters -get and set coefficients-

[M.Siddeq]

Dear all

Maybe it is about the theory and implementation of adaptive filters more than software problem;

the Idea is to apply the strategy of the example (set and get coefficients) with the system identification setup to my data so I can evaluate the performance of the filter based on the obtained unknown system coefficients;

the difference between the example and my application is:

1- my input signal is non-stationary

2- the targeted noise is non-Gaussian, it is from the material's structure (grains)

my application is ultrasonic testing of materials;

The question:

What will be the input and reference signals for system identification setup that will enable me to get the ideal coefficients of my ultrasonic signal contaminated with grain noise?

Thank you

[M.Siddeq]

Hello-

Any one here can instruct me where I can get help on this?

[M.Siddeq]

I hope that the help is coming 

[jimmy.qi]

Hi,

As far as I know, there are lots of models of system identification toolkit could be used to obtain your unknown system coefficients.  They almost share same inputs of stimulus signal and response signal.

Briefly, system identification setup in your case is that [Input: Stimulus signal] should be your input signal and [Input:Response signal] should be your ultrasonic signal contaminated with grain noise.

It seems that this toolkit is designed for various conditions even noise is non-Gaussian. 

If you do not sure about the orders of your model to be estimated, you could use the [SI Estimate Orders of System Model VI] to optimize the orders.

By the way, you could valid your model by using [SI Model Simulation VI] to make sure that your estimated model do fit for your unknown system.

As your input signal is not stationary, you could adjust your model coefficients recursively until it could fit for your feature of unknown system.

On the other hand, if it is hard to simulate the input signals, you could refer to this article as reference to select proper type of stimulus signal as the input of model estimation VI.

Hope I can answer your question.

If you have any question and I misunderstanding your meanings, please feel free to let me know.

Thank you.

Regards,

Jimmy

[M.Siddeq]

Hi Jimmy

Thank you very much for the detailed informative reply.

In fact it opened new windows for me although that my knowledge in system identification is too weak, I just wanted to use it as a tool to help me evaluate my adaptive filters, I will try to cover all the aspect that you have mentioned then give you feedback.

I found in Labview Signal express in the system identification step an option to estimate the impulse response based on prewhitening -based correlation analysis method. Do you think will be suitable for my application? Can I use it then take the obtained impulse response initialize the adaptive filter?

Thank you

[jimmy.qi]

Hi M.Siddeq,

You can have a try to use this non-parametric Correlation Analysis Method.  Note that in order to obtain a prewhitened input signal that is suitable for this model, the AR order of the filter should be selected carefully.  If prewhitened input signal is not white enough, the result from the correlation method is not reliable.  You can increase the AR order to improve the accuracy of the impulse response if the whiteness test is rejected.  (If most of the autocorrelation is within the confidence region, the input signal is well prewhitened, and the estimation of the impulse response is reliable. If the autocorrelation is outside of the confidence region, the estimation is unreliable.).

Best regards,

Jimmy