2.1 Stereo Calibration - what am I missing?

[moedra]

Hello Forum

I'm coming here to ask some questions after watching the tutorials of @Moadra on Youtube. They are great, thanks for that.
Best description of myself would be "All the gear, but no idea". DYI Speakers as a FAST of a Vifa fullrange Speaker and a Visaton TIW200XS with an crossover at 500Hz.
MM-1 with individual calibration file, UMC-22 and REW as measurement setup. After using a IMG Stage Line DSM-240 for the last 10 years, I switched to a MiniDSP Flex for optimization purposes.
I would like to try to setup the crossover and the room correction with FIR filters for my speakers following the instructions of @Moadra.
It provides 4x 2048 taps on the outputs and in addition PEQs for the two inputs.
1. Does it make sense to use FIR filters for crossover (LR4) and linearization of magnitude optimized with an measurement at 1m distance between the drivers and than use the PEQ of the inputs to correct the room over all at hearing position?
Should be the goal here to use minimum phase or zero phase approach (I do not understand, what each implies ;-))?
Should I adjust the GD of each driver (BB and Bass) between left and right individually or as a sum at the hearing position?
2. Using FIR for crossover and room correction: As above corrected at 1m only in the area of crossover and baffle step plus using the second banks on all outputs to correct room modes and modeling room curve at the top? This would mean to use the room correction filter twice on bank2 for the bass and the fullrange speaker to cover complete frequency spectrum.
3. Using PEQs for crossovers and FIR for room correction.

So much for today. Help would be highly appreciated.

Thanks

Jojo

The way I do this doesn't involve implementation of an equalizer in the signal path. Granted, the rePhase eq is used to generate the inversion of the measured response, but the final product is an impulse response which contains the adjustments for magnitude and phase. The signal is processed with convolution in the end, just as it is done with packages like Acourate, Audiolense, and Focus Fidelity. The filter is a minimum phase IR wav file, not a set of eq parameters. I use 65536 taps, which is a bit more than the hardware will allow you to use. I use Hang Loose Convolver within JRiver Media Center to run my correction.

 

[moedra]

FIR filtering is a very inefficient way to implement a crossover filter, better using the crossover filters the Flex offers. Room effects at low frequencies are minimum phase phenomena that are properly countered by parametric EQ filters. FIR filtering can be used to change the phase response of a speaker independently of its magnitude response though there's little consensus on the audibility of the result.

I can at least offer this observation:
When merely correcting the magnitude response, the sound does indeed improve, but not as much as it does when I also work to align the group delays with minimum phase response data. The result of process always sounds better when the additional steps needed to align the step response are taken, and the rephase filters are compiled into a stereo IR in REW.

 

[Dave the Rave]

How about a minimum phase FIR filter:

View attachment 59429

 

[Dave the Rave]

@serko70 I managed to try out the Phase Inversion technique and played around with the Target slope curve and Target Level.
The Impulse measurements allowed me to slightly adjust my speaker toe-in and physical distance from the LP.
I have a big dip in the bass section so tried with a lower Target Level than REW calculated (T68), the calculated one (T70.1) and finally the initial measurement SPL level(T71).
My observations - T68 was really detailed and probably overly clear which had some long term slight fatigue. T70 was really better but I got the best result with T71 (i.e initial SPL level for measurements). It had the details/clarity of T68 but also way better tonality with realistic low end and so enjoyable for long hours listening.
In the end, I am amazed how good that technique performs (and WITHOUT using filters) - another dimension for musicality, clarity, instrument separation and transcient . I could see the effect on the corrected measurements in REW but the real test was actual listening which it delivered with flying colours.
Thought I share my $0.02 words and commend your findings.

 

[Jojo10]

Hello Guys

I hear you, but I only understand 10%, at best.
I used the PEQs and Crossovers of the four MiniDSP channels for adjusting Magnitude and Crossover for both my two way speakers left (blue) and right (red).
1m Measurement:

Good enough for me,
Then I measured at listening position both left and right at 7 positions as in Davids Tutorial and got the RMS Averages.

With that I followed Davids tutorial but I did not get satisfying results as the 1024 taps of the MiniDSP are not sufficient to correct in the low frequency area.
Question 1: Is that the limitation of only 1024 taps and could not be overcome?
So I used the PEQ of the Input channels and got this, aiming for the Harman room curve:

I also aligned the Impulse of both speakers with a small delay 0,23ms for the left speaker:

Question 2: Do I need to adjust impuls responses in general to the beginning of the impuls (RampUp begin) or to the Peak?
Now my problem begins. If I measure the Sum left+right (green) I got this:

The dips at 106Hz and 250Hz are result of phase cancellation. The phases are nearly 180° offset in this areas.
This is clearly due to room effects, as the phases in the 1m measurements would be fine.
I also tried the REW alignment tool for left to right but did not find any solution with delays which is assumingly not the right way to solve this anyways.

Question 3: Isn't this the point where FIR filters come in handy to "adjust" the phase in this specific areas?

So I generated a plain linear phase Impulse in RePhase without any EQs and applied it to the two channels for my woofers:

So the phases left and right are now more linear, but did not sum up perfectly as well, better, but not good enough.
The magnitude is not changed and is still corrected by the PEQs of the input channels.
And now it gets fuzzy and I don't know what to do.
I exported the last measurements with the "linear phase" FIR and imported it into RePhase and I would try to adjust the phase more or less to Zero now.

Question 4: Is this the way to go forward?

If I do so with the parametric phase EQs in RePhase I do end up with.....(this is only an example of the adjustment, not done nicely)

Question 5: Why is the magnitude changing, when I generate the impulse and only used the Phase EQ?
Question 6: Why is the adjusted blue phase at around Zero and the generated red one is somewhere else? To less taps?

So, Am I on the right way? Is the MiniDSP to limited to help me? Do I use the tools wrong?
I'm proud of my measurements though.

Help would be highly appreciated.

Thanks

Jojo

 

[serko70]

1. 1024 taps will be quite limited either way especially in the lower frequency band but you can improve it by using rectangular windowing instead of hann and maybe energy centering plus some optimization. You do not need to use FIR filters for low frequencies though, min phase PEQ with low Qs and no dip boosting will do a better job. Another clean way to deal with the first bass peak and dip is to apply a virtual bass array to eliminate the long wall room reflections.
2. REW keeps impulse peak time as the reference for its calculations (John to confirm) so I'd equalize impulse peaks but neither would make an audible difference since the distances involved are smaller than your head movements.
3,4,5. Most efficient use of FIR filters for phase correction would be for your speakers' crossover phase corrections which should be done by using rePhase filter linearization tools. Paragraphic phase equalizer especially the way you have used in the second picture (like Q;4, 91 degrees @ 50Hz) will cause "massive" pre-ringing (any phase filter below 500Hz with Q larger than 1 and absoltue change larger than 45 degrees is likely to cause ringing with increasing effect the lower you go down the frequencies). To apply the correct crossover linearization filters, you need to first produce the "correct phase response" of your speakers which is not that straightforward. For starters with RMS averaging you'll lose all phase information of your measurements. You need to "cross correlation align" all measurements to the central measurement at the LP. After that you will see that vector averaging them will work fine. Then you have to align this average's first (not the largest) impulse peak to t-0 and export it to rephase and start applying different filter orders at your speakers XO points starting from the tweeter's crossover until you get a fairly flat phase graph. Don't expect to see a continious phase curve in the lower bass area unless you have sealed speakers. You can use box filters to fix the phase in that region if you have ported speakers. Beyond that, at best, you can use phase equalizer to equalize the speakers' group delays to each other which will not usually require large phase shifts in either speaker.
6. Yes, it's the inadequate number of taps and wrong window/centering used.

 

[sm52]

1m Measurement:

Good enough for me,

I'm proud of my measurements though.

I agree. And here it would be necessary to turn onto another road. Which leads to a comfortable listening spot with iced whiskey.

 

[sm52]

virtual bass array to eliminate the long wall room reflections.

Could you explain how this is done?

 

[JStewart]

Could you explain how this is done?

From @serko70 ‘s YouTube channel.

He uses it as the first of 4 filter creation steps, put together in another single video, but it can be used alone.

Truly incredible work, @serko70 . I’ll never understand it all, but hopefully can copy. (The excess phase filter has my head spinning )

 

[serko70]

From @serko70 ‘s YouTube channel.

He uses it as the first of 4 filter creation steps, put together in another single video, but it can be used alone.

Truly incredible work, @serko70 . I’ll never understand it all, but hopefully can copy. (The excess phase filter has my head spinning )

I have further optimized virtual bass array filter recently but couldn't find the time to make a video on that, hopefully soon. But basically I create a min phase, polarity inverted, 2.5 x first peak frequency low pass filter with "60" dB/oct cutoff directly in rephase and use that (with a +6dB SPL offset) in REW to add with the Dirac pulse. The impulse peak needs to be slightly readjusted to counteract first peak correctly. That filter will dim the first peak, will raise up the first dip and then disappear without any other effect to the rest of the response. No other filter can boost the first dip without artifacts.

 

[serko70]

From @serko70 ‘s YouTube channel.

He uses it as the first of 4 filter creation steps, put together in another single video, but it can be used alone.

Truly incredible work, @serko70 . I’ll never understand it all, but hopefully can copy. (The excess phase filter has my head spinning )

Excess phase filter has also evolved after the recent "Invert A phase" & "1/magn A" arithmetic operations addition to REW but this is a bit too complicated to explain here in text. I should really make a new stereo calibration video at some point.

 

[Jojo10]

Hello
Thanks for helping.
I would see the "produce the "correct phase response"...." section as my next step.
What I understood is, that I cannot simple align my 7 measurements with "Align IR Start" in REW, right?
Is it correct, that I have to bring the 6 measurements with their IR peak to the peak of the one LP measurement?
Then I can "vector average" them and this average I than need to adjust with IR Peak to t=0, right?
Please confirm, thanks.

 

[serko70]

Hello
Thanks for helping.
I would see the "produce the "correct phase response"...." section as my next step.
What I understood is, that I cannot simple align my 7 measurements with "Align IR Start" in REW, right?
Is it correct, that I have to bring the 6 measurements with their IR peak to the peak of the one LP measurement?
Then I can "vector average" them and this average I than need to adjust with IR Peak to t=0, right?
Please confirm, align

Multiple mic position measurements should be aligned with each other with "cross corr align" which aligns IRs taking care of every peak and dip. Select all different measurements for the same speaker including the one at LP, make sure LP measurement is on top of other selected measurements in REW measurements column on the left and feel free to press "cross corr align" multiple times until you can no longer see time shifts in responses (ticking "show points when zoomed in" helps). Acoustic reference speaker measurements shouldn't need any of that but you can cross corr align vector average of the non-acoustic reference speaker to that one for optimal left & right speaker sync. And finally use vector average of these two for your phase graph.

You need to align the first IR peak (not the real IR peak) to t=0. This is purely to get the right end of the phase graph correctly in rePhase and better use a copy of the vector average for that purpose. The original vector average doesn't need to be realigned as it will already be correctly aligned for central LP.

 

[sm52]

serko70,​

I watched the video. Thank you. At the end, a wav file is created. How to use it further? If I play music in foobar, can the resulting wav file be used in foobar?

 

[ddude003]

If I may suggest looking into Convolvers to host your FIR .wav file in what ever Application on what ever Operating System... It is helpful to fill out your More info so we have a better idea as to your end to end signal chain/kit so we can make better suggestions... You might look into Camilla DSP or Hang Loose Convolver for instance...

 

[sm52]

You might look into Camilla DSP or Hang Loose Convolver for instance

That is, if there are no additional hardware devices, but there is a computer, a sound card, a power amplifier, speakers, then this wav file cannot be applied?

 

[serko70]

That is, if there are no additional hardware devices, but there is a computer, a sound card, a power amplifier, speakers, then this wav file cannot be applied?

You can install Equalizer APO to your Windows PC which will convolve any FIR filter (wav file). You will have almost no limits in terms of number of taps depending on your CPU. I have never used Foobar but I am sure it has a convolution engine (maybe as an add on) to process wave files. EQ APO is free but of course the engines ddude003 suggested have better speed, accuracy and latency.

 

[ddude003]

That is, if there are no additional hardware devices, but there is a computer, a sound card, a power amplifier, speakers, then this wav file cannot be applied?

I suggested CamillaDSP and the Hang Loose Convolver because they are both software convolution engines and both have versions that run on Windoz and MacOS systems... Both may have, or soon to have, Unix versions... CamillaDSP is free and the Hang Loose Convolver is priced very reasonably... No additional hardware is required to run these software convolvers beyond a reasonably modern PC... I have used both convolvers on a 5 year old, Intel i5 MacBook Pro with no issues... And as @serko70 suggested, Equalizer APO should work as well... There are also some Convolution Reverb plugins that can read FIR .wav files and be used for Room Correction...

This thread at ASR makes for an interesting comparison between some hardware DSP and CamillaDSP software... I would expect similar performance from the Hang Loose Convolver...

DSP Measurements and Rising Noise Floor

In several threads related to DSP platforms I've seen requests for measurements with filters engaged. I thought it unlikely that implementing filters in a DSP would have measurable effect on noise / distortion other than the obvious change in the fundamental, however @Sokel recently mentioned...

www.audiosciencereview.com

 

[sm52]

Thanks to ddude003 and serko70. I once tried the Equalizer APO but didn't get any benefit in sound. It was better without him. Any actions with the Equalizer APO led to a change in the sound, but not to an improvement. The software player reads the music track, passes it to the convolver with our wav file, the convolver changes it so that the amplitude at the standing wave frequency is much smaller, so the amplitude of the standing wave is the same as the rest of the frequencies. Right?

 

[Jojo10]

Hello @serko70

First, thanks for your patience and willingness to help starters like me to understand.
To the method:
After searching for the "cross corr align" button for 30min and thinking being totally blind, I figured out, that there is a newer version than the "up to date" version which includes this new feature. Using it and importing this to RePhase it looks like this:

Not sure if the Box correction is correct for my closed box.
The alignment left to right of the Vector averages I already did in the past (0,23ms) and it change the sound stage massively.
Let's assume I do the same now with the other speaker and I can create this impulse response with RePhase, what's next?

To the overall procedure:
I understood that because of my limitations of taps and the negative effects (preringing), I should not do aggressive corrections (e.g. like Q;4, 91 degrees @ 50Hz).
I can do the method above and use the result as a FIR filter in my MiniDSP.
But how does the phase correction of the loudspeaker output (vector average adjustment to t=0) will change the phase cancellation problem at 100Hz, if the phase shift is produced by the room between the speaker and the listening position?

"Beyond that, at best, you can use phase equalizer to equalize the speakers' group delays to each other which will not usually require large phase shifts in either speaker." This is done with the procedure of David's tutorial? No, because there the EQ is used to adjust the GD. How I equalize the group delay with the phase equalizer? In RePhase?

Thanks

 

[moedra]

Hello Guys

I hear you, but I only understand 10%, at best.
I used the PEQs and Crossovers of the four MiniDSP channels for adjusting Magnitude and Crossover for both my two way speakers left (blue) and right (red).
1m Measurement:
View attachment 59875
Good enough for me,
Then I measured at listening position both left and right at 7 positions as in Davids Tutorial and got the RMS Averages.
View attachment 59876
With that I followed Davids tutorial but I did not get satisfying results as the 1024 taps of the MiniDSP are not sufficient to correct in the low frequency area.
Question 1: Is that the limitation of only 1024 taps and could not be overcome?
So I used the PEQ of the Input channels and got this, aiming for the Harman room curve:
View attachment 59877
I also aligned the Impulse of both speakers with a small delay 0,23ms for the left speaker:
View attachment 59879
Question 2: Do I need to adjust impuls responses in general to the beginning of the impuls (RampUp begin) or to the Peak?
Now my problem begins. If I measure the Sum left+right (green) I got this:
View attachment 59878
The dips at 106Hz and 250Hz are result of phase cancellation. The phases are nearly 180° offset in this areas.
This is clearly due to room effects, as the phases in the 1m measurements would be fine.
I also tried the REW alignment tool for left to right but did not find any solution with delays which is assumingly not the right way to solve this anyways.

Question 3: Isn't this the point where FIR filters come in handy to "adjust" the phase in this specific areas?

So I generated a plain linear phase Impulse in RePhase without any EQs and applied it to the two channels for my woofers:
View attachment 59881 View attachment 59880

So the phases left and right are now more linear, but did not sum up perfectly as well, better, but not good enough.
The magnitude is not changed and is still corrected by the PEQs of the input channels.
And now it gets fuzzy and I don't know what to do.
I exported the last measurements with the "linear phase" FIR and imported it into RePhase and I would try to adjust the phase more or less to Zero now.
View attachment 59882 View attachment 59883
Question 4: Is this the way to go forward?

If I do so with the parametric phase EQs in RePhase I do end up with.....(this is only an example of the adjustment, not done nicely)
View attachment 59884
Question 5: Why is the magnitude changing, when I generate the impulse and only used the Phase EQ?
Question 6: Why is the adjusted blue phase at around Zero and the generated red one is somewhere else? To less taps?

So, Am I on the right way? Is the MiniDSP to limited to help me? Do I use the tools wrong?
I'm proud of my measurements though.

Help would be highly appreciated.

Thanks

Jojo

Remember that my method assumes that you will be using a convolver within the operating system, such as in a DAW or system-wide application, to process your correction filters.
I have further optimized the procedure I showed in those videos but I just haven't had time to make new videos yet. Soon...

 

[serko70]

Hello @serko70

First, thanks for your patience and willingness to help starters like me to understand.
To the method:
After searching for the "cross corr align" button for 30min and thinking being totally blind, I figured out, that there is a newer version than the "up to date" version which includes this new feature. Using it and importing this to RePhase it looks like this:
View attachment 59909
Not sure if the Box correction is correct for my closed box.
The alignment left to right of the Vector averages I already did in the past (0,23ms) and it change the sound stage massively.
Let's assume I do the same now with the other speaker and I can create this impulse response with RePhase, what's next?

To the overall procedure:
I understood that because of my limitations of taps and the negative effects (preringing), I should not do aggressive corrections (e.g. like Q;4, 91 degrees @ 50Hz).
I can do the method above and use the result as a FIR filter in my MiniDSP.
But how does the phase correction of the loudspeaker output (vector average adjustment to t=0) will change the phase cancellation problem at 100Hz, if the phase shift is produced by the room between the speaker and the listening position?

"Beyond that, at best, you can use phase equalizer to equalize the speakers' group delays to each other which will not usually require large phase shifts in either speaker." This is done with the procedure of David's tutorial? No, because there the EQ is used to adjust the GD. How I equalize the group delay with the phase equalizer? In RePhase?

Thanks

What speakers are you using? You need to find the their actual crossover frequencies from their spec sheets and apply linearization filters only at these frequencies. Even then, it's quite hard to determine the exact filter orders to apply although you will see improvements with any order at the correct XO freq. The optimal way is to take near field measurements.

The box frequency is the port frequency for ported speakers. I found mine in a magazine's lab test. I am not too sure which frequency to use for a sealed speaker.

The frequency used seems wrong, I think MiniDSP works at either 48kHz or 96 kHz but I am not sure.

The 100Hz dip is most probably from a large phase difference between the left and right speaker around that freq range (unless your speakers are 85 cm away from the front wall) and you cannot fix that with XO phase correction which will effect both speakers equally anyway. XO phase correction will only fix delays between different drivers of the speakers and this increases the strength of the impulse and lifts the sound stage higher as when woofer and tweeter are time aligned, sound tends to move to the level of the tweeter.

The phase difference between your left and right speaker must be close to 180 degrees for them to cause cancellation. This means around 90 degrees, very sharp (effective only around 100Hz area - high Q) phase equalization filter will be required for each speaker which will obviously cause ringing. You cannot move the walls of a room with digital correction. Try optimizing speaker placement instead. On the positive side, you will not hear the presence of that dip while listenning to your system as much as you will hear the peaks.

My methods manipulating phase are based on accepting a certain average group delay (or a phase response) and correcting each speaker to mimic that. So minimal phase correction (to the limits allowed before ringing) is required for each speaker and cohesion between speakers is improved which makes a positive difference in sound stage.

 

[serko70]

Here's a GD graph from my actual measurements and correction:

LR is the original Left and Right speaker combined GD response. LR3 is after all the magnitude correction in the frequency domain. LR5 is after various highly complex phase corrections

The delay around 45Hz and around 200-300Hz is completely gone but nothing could be done to the delay at 67Hz and I have to live with it as the phase anamoly there is way too large to correct without ringing.

To show how close this is to pre-ringing, check this step response graph:

You will not hear anything after -5ms, maybe -10ms but before that if the ringing is any louder than -30dB, it will be audible. For what it's worth, Dirac Live doesn't go over -40dB in phase correction.

 

[Jojo10]

Hello

"DIY Speakers as a FAST of a Vifa fullrange Speaker and a Visaton TIW200XS with an crossover at 500Hz."
Active speakers with digital X0 with LR4 at 500Hz.

You are right, should be 96kHz

The LP measurements of R and L shows 190° phase shift

The both speakers left and right are time aligned

So you say, I need to live with the dip at 100Hz, because there is no chance to correct 180° phase shift in this area.

Thanks

 

[serko70]

Hello

"DIY Speakers as a FAST of a Vifa fullrange Speaker and a Visaton TIW200XS with an crossover at 500Hz."
Active speakers with digital X0 with LR4 at 500Hz.
View attachment 59944

You are right, should be 96kHz

The LP measurements of R and L shows 190° phase shift
View attachment 59945 View attachment 59946
The both speakers left and right are time aligned
View attachment 59947

So you say, I need to live with the dip at 100Hz, because there is no chance to correct 180° phase shift in this area.

Thanks

I cannot see the graphs properly at the moment (I'm on phone) but you don't need crossover phase correction with active speakers.