Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[Mark Wilkinson]

Another thing that surfaced here.

I've tried a few things, and to my ears it seems like using IIR filters to adjust the response of each frequency band sounds better than using auto mag, even if the result looks similar in supplementary responses. My best sounding preset to date is done using IIR to correct as much as possible before I apply a LF/HF linear phase crossover, a couple of manual linear phase eq to adjust summation and a couple of manual filters to adjust the phase response as needed.
I then apply a tiny bit of auto mag/phase to the HF, leaving the LF manually adjusted.


Does this sound familiar to others using FIR Designer?
I still have a couple of things I need to work out, there is a bit of "jumpyness" in the horn, some part of the spectrum stick out a bit om some recordings. I suspect I might have overcorrected a bit on the HF, going to do a A/B with a version without auto mag correction.
BTW that is an andvantage of sticking to IIR filters, it's really easy to fine-tune a filter directly in the DSP, no need to load a new FIR filter on the outputs.

Hi Helge, very nice work/observations imo.   
And i owe you a small apology....one of the first questions i should have asked before offering any auto-EQ advice, is how many taps at what sample rate are you using?

Apart from delay, there are two shortcomings/dangers with FIR, or at least two I've found.
First is the frequency resolution.  Like Pat Brown illustrates in this article you may have seen......   www.prosoundtraining.com
An example he gives shows that that 1024 taps at 48kHz, used as linear phase with impulse centering, only has a frequency resolution of 95Hz. Fine for high frequency work, but clearly insufficient for low frequency work. Even if the 1024 tap FIR filter is used for IIR replication by moving impulse peak to the front of the filter, the resolution only improves to 47.5Hz....still insufficient, and no phase correction at all, so why bother...
IIR filters on the other hand, have theoretically infinite resolution.

You may/probably already know all this...i'm writing as much for general knowledge's sake as anything.
So lot's of taps (and delay) are needed for any kind of EQ resolution capable of handling full range....which is why FIR is really only applicable for live sound, maybe stating at around 500Hz ime.

The second issue with FIR, becomes more true as tap count increases when using any form of auto EQ.....it demands good data, good measurements, to keep from correcting what should not be corrected.  I personally believe, the only valid EQ corrections occur at the individual driver level, and are always IIR / minimum phase. Once drivers sum together in acoustic space, they quit acting as minimum phase, and auto EQ FIR will do a beautiful job correcting to a single point in space, but at the expense of everywhere else.
Which makes your use of IIR first, such a great strategy.

It took me a long time, learning how to make spatially average measurements that are suited for auto EQ...and still learning. i also must stress that this is at the individual driver section level.
I'd never apply FIR auto-EQ to an entire speaker, unless it really was for perfect sound for one small spot.

Where i think FIR and linear phase really shine, is when used for complementarity linear phase xovers.  All gain, no loss, if the delay can be tolerated.  Bye bye unnecessary xover induced phase rotation.  The jury is out on how audible that is, but if nothing else, lord knows it helps nailing down timing alignments between drivers. Getting flat phase traces to overlap is easy peasy.

I think the best tuning practice i know, is to first tune each driver section with IIR, flattening both in-band response, and out-of-band response through the critical summation region (i like to achieve that to at least -15dB for each driver).
Then apply FIR and complementarity linear phase xovers. (the complementary part is important, at least to measurements.  If not complementary, crap showing pre-ringing shows up in measurements away from the reference tuning axis.)
You can apply lin phase steeper xovers, to undo the out-of-band IIR flattening that could be potentially dangerous to the drivers lower end response. And the steeper filters also reduce the width of the critical summation range, reducing the task of out-of-band flattening. Best benefit of all, is by reducing the range of summation, lobing range goes goes down.  This all works (steep) simply because of the lack of phase rotation. I think it may be the biggest factor in why i like what i hear so much from my DIYs.

I get away with using firD's auto-EQ essentially imbedding the IIR EQ's into a linear phase file, because i do a lot of spatial averaging work to acquire a reference measurement i'm willing to auto-correct, and i use as the saying goes here, a metric shit ton, of taps.  16,384 at 48 kHz to be exact...170ms delay....not exactly good for live, huh?

Hope i've been of some help.

[Helge A Bentsen]

Thanks, lot of good points

I ended up with this method by studying what manufacturers do in their tunings and looking a lot at supplementary responses.
Pretty sure my LF needs more refinement in the measured data I put in to FIR D, but I can't get outdoors to get better measurements right now (winter...).
Did a few measurement sessions before I was able to capture data that I could EQ in such a way that what I heard correlated to what I measured after EQing.

As an example, I tried a linear phase brick wall for LF/MF crossover, but it introduced some variables in the response off-axis.
Noticed that EAW uses Butterworth 24dB/oct for a speaker with similar xo points, so I tried that. It sums better off axis, good on axis, and I used a out of band EQ point to nudge it a bit into place.

I'm using a fairly short FIR filter (maybe too short?) filter delay of 200 samples and filter length of 400 samples @48K.
My LF/MF xo is right now at 600hz.
I think my next step is to load the HF FIR without auto mag and have a listen to it while I wait for warmer weather.

[Chris Grimshaw]

First is the frequency resolution.  Like Pat Brown illustrates in this article you may have seen......   www.prosoundtraining.com
An example he gives shows that that 1024 taps at 48kHz, used as linear phase with impulse centering, only has a frequency resolution of 95Hz. Fine for high frequency work, but clearly insufficient for low frequency work. Even if the 1024 tap FIR filter is used for IIR replication by moving impulse peak to the front of the filter, the resolution only improves to 47.5Hz....still insufficient, and no phase correction at all, so why bother...
IIR filters on the other hand, have theoretically infinite resolution.

Mark,

What about using a larger number of taps, but moving the impulse centre to be nearer the start of the filter? Would that gain back the frequency resolution?

RePhase lets you choose the delay in terms of taps (number), delay (ms), or as a fraction of the filter length. ie, it can be chosen independently of the number of taps. Of course, filter performance will vary.

Chris

[Mark Wilkinson]

Glad to have helped Helge, and hi Chris,

I'll try to reply to both of your posts, with regard to frequency resolution.
Helge, your filter with 400 samples @ 48k has a time span of 400/48000, or 0.0083 sec, or 8.33ms.
Using the F=1/T rule, 1/.0083 = 120Hz.  Since you are using the file as linear phase with impulse centering, effectively only 1/2 of the samples, 200 of them, make frequency response corrections.
So the resolution drops by half, to 240 Hz.  A rule of thumb Pat Brown gave in an article i tried to link, is that 3X the resolution of the FIR file, is about the lowest frequency we should try to EQ.   So your file becomes effective somewhere around/above 720Hz.
 
(i don't know why the SynAudCon article link got changed to a home page link, but the great article that explains all this is "FIR-ward Thinking – Part 5 Are More Taps Better?")

And Chris, yes, exactly....... more taps and moving the impulse peak to the front of the file for IIR replication, increases frequency resolution (as also very well explained in the article)
I say 'IIR replication' because it is still not true IIR, which has theoretically infinite resolution.  It is 'IIR replication' with the resolution inherent in the size and structure of the FIR file.

So using my 16,384 taps @ 48k for example.  With linear phase and impulse centering i get a resolution of 8,192/48000 or about 6Hz which should be good for work down to 18Hz, or rather all i need ime/imo. 
If i found i needed more resolution for IIR replication, because of maybe a particular high-Q EQ that is needed, and doesn't quite replicate with impulse centered, I could move impulse peak closer to start....all the way to start for 3Hz resolution if desired..... or only as far toward start as needed, and continue to maintain a linear phase xover.

Hope that made sense! I'm not near as good an explainer as Mr Brown !

[Helge A Bentsen]

Glad to have helped Helge, and hi Chris,

I'll try to reply to both of your posts, with regard to frequency resolution.
Helge, your filter with 400 samples @ 48k has a time span of 400/48000, or 0.0083 sec, or 8.33ms.
Using the F=1/T rule, 1/.0083 = 120Hz.  Since you are using the file as linear phase with impulse centering, effectively only 1/2 of the samples, 200 of them, make frequency response corrections.
So the resolution drops by half, to 240 Hz.  A rule of thumb Pat Brown gave in an article i tried to link, is that 3X the resolution of the FIR file, is about the lowest frequency we should try to EQ.   So your file becomes effective somewhere around/above 720Hz.
 
(i don't know why the SynAudCon article link got changed to a home page link, but the great article that explains all this is "FIR-ward Thinking – Part 5 Are More Taps Better?")

And Chris, yes, exactly....... more taps and moving the impulse peak to the front of the file for IIR replication, increases frequency resolution (as also very well explained in the article)
I say 'IIR replication' because it is still not true IIR, which has theoretically infinite resolution.  It is 'IIR replication' with the resolution inherent in the size and structure of the FIR file.

So using my 16,384 taps @ 48k for example.  With linear phase and impulse centering i get a resolution of 8,192/48000 or about 6Hz which should be good for work down to 18Hz, or rather all i need ime/imo. 
If i found i needed more resolution for IIR replication, because of maybe a particular high-Q EQ that is needed, and doesn't quite replicate with impulse centered, I could move impulse peak closer to start....all the way to start for 3Hz resolution if desired..... or only as far toward start as needed, and continue to maintain a linear phase xover.

Hope that made sense! I'm not near as good an explainer as Mr Brown !

Thank you

This probably explains why I like the IIR filtered result better than some of my first versions with a lot of FIR EQ.

[Michael John]

.... If you are using an average of a few angles generate the FIR correction, I'm not exactly sure how FirDesigner's various averaging options handle phase..over my pay grade still.
I've been using Smaart's Averager with default settings, because it makes for a little easier workflow.

If the measurements are simply for displaying polars, where you want to show phase too, I usually reset delay finder for each meas.

Great thread. Mark, thanks for the shout outs.

Sorry I'm lake to the party. :-) I thought I'd throw in some comments re FIR Designer's averaging.

When loading measurements on the "Load and Setup" tab, one measurement can be noted as the "reference" - e.g. the on-axis measurement. The notion of "reference" is used by the following features.

"Move Ref IR Peak to 0" automatically removes any time delay from the "reference" measurement, bringing its IR peak sample to 0 time.  (This is similar to SMAART's "Delay Finder" for TF measurements.) The same delay adjustment is also applied to all other measurements and so the relative time delay, between measurements, is retained. This is useful for complex averaging of turntable measurements of individual drivers in a multi-way, with one caveat. Assuming the time delay - in the measurement system - is locked for all measurements across all multi-way drivers, to maintain relative delay between drivers use "Move Ref IR Peak to 0" for, say, the HF average, then manually apply this delay in the "Apply common delay (ms)" field when averaging measurements for each of the other drivers.

"Time Align all to Reference" automatically adjusts the delay of all other measurements to match the reference and so lines up the IR peaks. Relative time delay or phase between measurements is discarded but cabinet phase characteristics are retained. This is useful when complex, spatially averaging full BW measurements of a loudspeaker to do mag & phase adjustment of the whole cabinet.

BTW on the "Average" tab, there're tools to quickly check each measurement, remove bad measurements by unchecking "Use" and manually tweak time delay.

To ensure no individual measurement dominates the averaging, on the "Load and Setup" tab use the "Normalise magnitudes to max" or "Normalise to frequency range" options.

Re the actual averaging modes

• Complex: This is full transfer function averaging. The phase of the cabinet is retained.
• Power (flat, zero phase): This power averages the magnitude only and discards the phase.
• Power (minimum phase): This power averages the magnitude only and discards the phase, then calculates the minimum phase response from the magnitude response. For a single driver in an anechoic chamber, the resulting phase should be very similar to the full Complex average. For a multi-way system with crossovers, this mode gives the cabinet phase response minus the additional or "excess" phase from crossovers.
• Power (complex phase): This power averages the magnitude only then combines this mag with the phase from the Complex averaging mode.
• Power (min. phase + Ref. excess phase): This is the same as the "Power (minimum phase)" mode but the excess phase from the Reference measurement is added in. (The Reference excess phase is the original phase minus the phase from a minimum-phase calculation from the Reference magnitude.) In a nutshell it's calculating the excess or crossover related phase from the Reference measurement, then adding this crossover phase to a minimum-phase average of all the measurements.
• Power (with complex avg. excess phase): The magnitude is a power average and the phase is the excess phase only. This could be useful for isolating excess or crossover related phase and only correcting this phase. We're not aware of anyone using this but it was easy to calculate so we thought it worth including.

Cheers,
Michael

[Michael John]

Another thing that surfaced here.

I've tried a few things, and to my ears it seems like using IIR filters to adjust the response of each frequency band sounds better than using auto mag, even if the result looks similar in supplementary responses. My best sounding preset to date is done using IIR to correct as much as possible before I apply a LF/HF linear phase crossover, a couple of manual linear phase eq to adjust summation and a couple of manual filters to adjust the phase response as needed.
I then apply a tiny bit of auto mag/phase to the HF, leaving the LF manually adjusted.


Does this sound familiar to others using FIR Designer?
I still have a couple of things I need to work out, there is a bit of "jumpyness" in the horn, some part of the spectrum stick out a bit on some recordings. I suspect I might have overcorrected a bit on the HF, going to do a A/B with a version without auto mag correction.
BTW that is an advantage of sticking to IIR filters, it's really easy to fine-tune a filter directly in the DSP, no need to load a new FIR filter on the outputs.

That all makes sense. Just to add to Mark's reply, the "Oct Smooth" setting in the Auto Mag makes a difference. If it's too fine - say 1/24th oct or 1/48th oct - it could over-push some narrow notches, which might be what you are experiencing. I've personally never used finer than 1/6th oct unless I've first averaged a lot of measurements.

We are looking at updates to the Auto Mag process to provide some independent control of pushes versus cuts.

BTW re using IIR's only, I recently tuned FOH in a church install that had a QSC PLD - no FIR. I took about 20 measurements in the room with SMAART, averaged in FIR Designer, then on the FIR Designer IIR tab, selected the QSC IIR mode, placed a few filters to push the measurement to target and manually copied the IIR settings into the PLD. Overall really quick and the result was excellent, despite the limited number of IIR's in the PLD.

[Mark Wilkinson]

Re the actual averaging modes

• Complex: This is full transfer function averaging. The phase of the cabinet is retained.
• Power (flat, zero phase): This power averages the magnitude only and discards the phase.
• Power (minimum phase): This power averages the magnitude only and discards the phase, then calculates the minimum phase response from the magnitude response. For a single driver in an anechoic chamber, the resulting phase should be very similar to the full Complex average. For a multi-way system with crossovers, this mode gives the cabinet phase response minus the additional or "excess" phase from crossovers.
• Power (complex phase): This power averages the magnitude only then combines this mag with the phase from the Complex averaging mode.
• Power (min. phase + Ref. excess phase): This is the same as the "Power (minimum phase)" mode but the excess phase from the Reference measurement is added in. (The Reference excess phase is the original phase minus the phase from a minimum-phase calculation from the Reference magnitude.) In a nutshell it's calculating the excess or crossover related phase from the Reference measurement, then adding this crossover phase to a minimum-phase average of all the measurements.
• Power (with complex avg. excess phase): The magnitude is a power average and the phase is the excess phase only. This could be useful for isolating excess or crossover related phase and only correcting this phase. We're not aware of anyone using this but it was easy to calculate so we thought it worth including.

Cheers,
Michael

Thank you Michael,
like i said earlier, understanding how FirD handles phase and averaging is above my pay grade....and still is until i digest all you wrote here!   

Oh, and +1 to using less fine corrections in Auto Mag, as you mention in next post.  I should have brought that up previously.
1/6th has become my go-to, sometimes less..... with 1/12th tops for super well behaved drivers.
I've slowly weened off trying to get near perfect measurements!  lol

[Helge A Bentsen]

Great thread. Mark, thanks for the shout outs.

Sorry I'm lake to the party. :-) I thought I'd throw in some comments re FIR Designer's averaging.

When loading measurements on the "Load and Setup" tab, one measurement can be noted as the "reference" - e.g. the on-axis measurement. The notion of "reference" is used by the following features.

"Move Ref IR Peak to 0" automatically removes any time delay from the "reference" measurement, bringing its IR peak sample to 0 time.  (This is similar to SMAART's "Delay Finder" for TF measurements.) The same delay adjustment is also applied to all other measurements and so the relative time delay, between measurements, is retained. This is useful for complex averaging of turntable measurements of individual drivers in a multi-way, with one caveat. Assuming the time delay - in the measurement system - is locked for all measurements across all multi-way drivers, to maintain relative delay between drivers use "Move Ref IR Peak to 0" for, say, the HF average, then manually apply this delay in the "Apply common delay (ms)" field when averaging measurements for each of the other drivers.

"Time Align all to Reference" automatically adjusts the delay of all other measurements to match the reference and so lines up the IR peaks. Relative time delay or phase between measurements is discarded but cabinet phase characteristics are retained. This is useful when complex, spatially averaging full BW measurements of a loudspeaker to do mag & phase adjustment of the whole cabinet.

BTW on the "Average" tab, there're tools to quickly check each measurement, remove bad measurements by unchecking "Use" and manually tweak time delay.

To ensure no individual measurement dominates the averaging, on the "Load and Setup" tab use the "Normalise magnitudes to max" or "Normalise to frequency range" options.

Re the actual averaging modes

• Complex: This is full transfer function averaging. The phase of the cabinet is retained.
• Power (flat, zero phase): This power averages the magnitude only and discards the phase.
• Power (minimum phase): This power averages the magnitude only and discards the phase, then calculates the minimum phase response from the magnitude response. For a single driver in an anechoic chamber, the resulting phase should be very similar to the full Complex average. For a multi-way system with crossovers, this mode gives the cabinet phase response minus the additional or "excess" phase from crossovers.
• Power (complex phase): This power averages the magnitude only then combines this mag with the phase from the Complex averaging mode.
• Power (min. phase + Ref. excess phase): This is the same as the "Power (minimum phase)" mode but the excess phase from the Reference measurement is added in. (The Reference excess phase is the original phase minus the phase from a minimum-phase calculation from the Reference magnitude.) In a nutshell it's calculating the excess or crossover related phase from the Reference measurement, then adding this crossover phase to a minimum-phase average of all the measurements.
• Power (with complex avg. excess phase): The magnitude is a power average and the phase is the excess phase only. This could be useful for isolating excess or crossover related phase and only correcting this phase. We're not aware of anyone using this but it was easy to calculate so we thought it worth including.

Cheers,
Michael

Just to follow up on this, when you do turntable measurements of a speaker, what do you consider best practice?

Since I'm measuring indoors ATM (weather issues) I got the best resuls with this method:
- Measure delay differences between various pass bands and apply delay/level adjustments first in the DSP. All bands delay adjusted in the processor so they all end up with the same delay time in Smaart using the delay finder. Level set so all bands are roughy equal in level.
- Measure each pass band individually, from 0 to X degrees off-axis.

Secondly, how should one proceed to average each passbands measurements together? Time align to reference + normalise to max?
I noticed that if I used "Time align to refererence" the impulse response of the averaging got sharper compared to Smaarts averager.
And if I normalised to max the level of the average got higher, so the level in FIR D more resembled the on-axis response of the speaker.
This was helpful because in my first tuning I ended up with too much level on the horn on-axis, this was easier to get right when I normalised before averaging.

[Michael John]

Hi Helge,

I think your measurement setup makes sense. You could leave out the initial delay and level adjustments, and just set SMAART delay finder using an on-axis measurement of the deepest driver - usually the horn. Then delay and level adjustments can be done in FIR Designer M as part of the overall preset design, after averaging.

Regarding averaging, I tend to not time align all the measurements to the reference, and then I do a complex averaging for each driver, but I generally work within modest coverage angles (+-45 degrees). It's worked for me in the past but might not work as well for wider angles. Re normalisation, I usually apply a common gain to all measurements across all drivers, so that the bulk level differences between drivers are maintained for FIR Designer's "Gain/Polarity/Delay" tab.

(For horn reflection correction, I'd suggest time aligning all to reference and then doing a complex average. This averaging should tend towards isolating complex, angle-invariant aspects of the horn.)

I think Mark's approach with SMAART - if delay finder is adjusted for each measurement - is analogous to time aligning all to the reference in averaging.

I'd recommend taking a look at FIR Designer M's "supplementary responses." Here sets of measurements for each driver can be loaded, matched and plotted with all gain/polarity/delay/IIR/FIR processing applied. The combined response, across all drivers, is also shown. With this feature you can work on the tuning with on-axis driver measurements, and simultaneously view the processed, combined response for many off-axis directions.

Related this, I'm working on another tool that shows polar and balloon plots of the acoustic sum of the driver measurements with processing applied. With this it's possible to see how the processing/crossover works across many angles and check the average of the combined acoustic response (with processing). I can't say when it will be ready.

Best,
Michael

[ProSoundWeb Community]