functionality of REW filter creation.

[loommidom]

Hello guys,

I have already searched a while for a similar thread but i did not find any so i opened this one. If there is any similar thread i would be grateful, if one could show me this thread.

I am using REW for an university project and need to know what happens exactly, when i press "match response to target". I read in the "help file" that "Match Response to Target starts REW's automated filter assignment and adjustment process. REW assigns filters to match the Predicted response to the Target Response, beginning with the area within the Match Range where the measurement is furthest from the target". But how are these adjustments calculated? How does REW knows how much one frequency has to be intensified or softened.
Also interesting to know would be, how many different filters can be created and after how much created filters the adaption process is finished. So, is there something like a quality criterion implemented in REW that limits the filter creation and interrupts the filter creation process when the quality criterion is reached.

Thanks in advance.

 

[John Mulcahy]

Filters are assigned to correct closed areas above or below the target. If there isn't a sufficiently large area or there are no more filters available filter assignment ends. Settings are adjusted using a nonlinear optimiser, either UNCMIN if the filter coefficients are continuously variable or a stochastic optimiser (Simultaneous perturbation stochastic approximation) if they are more granular. The optimisation aims to minimise the mean squared error of the response compared to the target over the selected range. Sharp dips in the response are ignored.

 

[loommidom]

Thank you very much for your answer John, it helped me a lot but i have got some further questions.


1. As I understand REW stops creating further filters when the mean squared error of the response compared to the target is in a minima (and these minima should be global minima, because the optimization tools are capable to find global minima), so creating more filters would be counterproductive, right?

2. What is meant with, whether the filter coefficients are continously variable or more granular? How does REW knows which of the two cases is true or not?

 

[John Mulcahy]

Filter assignment and overall optimisation are separate steps. Filter assignment stops when there are no more meaningfully large enclosed areas in the response relative to the target.

Equalisers vary in the control they offer over their filter parameters. Some only offer coarse control (gain in 1 dB steps, for example, or a limited range of Q values or filter centre frequencies). Some do not constrain the step size, or offer small enough steps that they can be treated as unconstrained. The equaliser selection contains that information.

 

[loommidom]

okay thank you.

So basically the EQ filtering with REW is made in two steps. First the filter assignement is made and then dependent on how my filter parameters are preset the optimization of these filters is made, right?
So this means that I could smooth my predicted response with a filter more and more dependent on how much filters i use and this would imply that i could get for example a linear line as my predicted response if i use infinite filters?