Optimum filter placements for BFD ? (1 Viewer)

[Sankar]

It appears that it should be possible to compute the "optimum" filter placements for a parametric equalizer such as the BFD ... let me brainstorm a bit here ...
1) When we measure the base response using SpectraPlus, we can save this overlay as a text file in the overlay directory .. if one looks at that, its just a text file with frequencies and db levels (which can be imported into any graphing program including Excel).
2) Now if I can describe the filter as an equation based on the "bandwidth", "center freq" and "gain", then its just a matter of picking these optimally to flatten the response as far as possible, right ?
Does anyone offhand know the equation that describes the filter ? Looking at the BFD manual it appears that the filter can be described by some combination of exponential functions ..
Once this equation is prescribed, it should be simple to write a program that optimally picks "n" filters to minimize some objective function (e.g. minimize the max deviation in the 20-100Hz range) .. that way we could choose both the number of filters we wish to use and their optimum description (due to filter interactions, I doubt if this is a problem that can be solved sequentially)
... more stuff to work on !

 

[Deane Johnson]

Sankar:
You're not going to have time to watch any movies.
Deane

 

[MichelF]

If you look at Smaart-Live software from SIA (www.siasoft.com) you will see that this software can work with external devices and it can compute for some of them the "optimal" filters (devices such as XTA, BSS or DBX digital cross-overs have also param EQ and time alignment capabilities). This tool seems to be the reference in Pro market.
Michel

 

[Sankar]

Does SIA have an evaluation copy that we can download ? I could not find anything.

Also does anyone know the relationship betw the octave setting and decimal Q factor for the bandwidth in the BFD ? This is the table on page 9 of the manual ....

I've made an assumption that the bandwith "low" and "high" points that they plot are locations where the filter is a fraction (50%) of the peak. (The "low" and "high" points are easily derived once we recognize that the x-axis is a log scale)
With that assumption the filter at any frequency f0 can be described as
A.exp(B*(f0-C)^2)
.. in this, A is the gain at the center, C is the center freq and B .. well that is the unknown which I'm obtaining from the bandwidth number.
I'm putting this all together on an excel sheet .. once the filter eqn is in there, I can provide it to anyone that is interested ..

 

[Sankar]

It worketh !
Managed to figure out the necessary equations and put them into a spreadsheet which can now be optimized!
So here's what I finally did:
1) Fed a pink noise signal directly to the sub input of the receiver ..
2) Using Spectra Plus took 5 sets of measurements .. one at the sweet spot, and 4 others along the perimeter of a circle of radius about 2 feet around the sweet spot. Saved these to overlay files (5 of them). All readings were with "infinite averaging" to avoid problems of random fluctuations
3) Took the average curve (yes there was quite some variation) ...
4) Used the spreadsheet to figure out the filters ... The resulting gains and bandwidths were not necessarily integer .. so I just rounded them to the nearest integer.
5) Programmed them in ... remeasured response .. significantly flatter !
.. the code still requires some manual intervention (its written in VBA, so its not that fast to allow more than 2 controls at a time).
I'd like to do a "blind" calibration .. if someone could send me the freq overlays (from Spectra Plus) that you'd like to flatten out, I'd be glad to make a suggestion of which filters you should use etc (using the program that I've written etc) .. which you could then try to see if it indeed does do the job ..
What I need is a list of freqs and responses (like the overlays produce in the *.ovl files .. you could send me an average over a bunch of measurement points if you wish) .. and the range that you'd like flattened out ... (e.g. 16Hz - 120Hz).
If it works, then we may have a useful tool ! (I still need to do some cleanup to accomodate the discreteness of the gains, bandwidths etc in the BFD .. but that should be simple)

 

[MichelF]

Sankar,
Yes, they have an evaluation copy (30 days) of the Smaart-live software. It can be ordered via their web site. It is not downloadable, they'll send you the CD. On their home page, there is "Want a demo disc?" with a link to send them an email.
The demo disk comes without the manual...
Note that they just announced a new release (4.5, free for onwer of 4.0 or higher versions) with some more features and also a price increase as of 1st of April.
Michel