Question on Averaging

[JoachimStrobel]

I have taken 10 measurements for each loudspeaker around my listening position, using a 20sec 10-20Khz sweep with default windowing (no FDW)
I have done an RMS and an Vector average and wonder about the difference. I am sure it has to do with the phase and am looking for a simple explanation and how to move forward.
I show all the measurements for the left channel below, the RMS average is in lime, the vector average is in red, some lower.

I am showing the averages again to be seen more clearly. I added the centre LP in black

Moving forward into EQ, I assume that I use the RMS average, but then I have no phase information, is that right? To move into RePhase, I would use the vector average. This seems kind of Ok if I stay blow 200 Hz (avoiding the large room mode) as all curves are the same there. However, I might want to do an EQ up to 13Khz as I am relatively close to the speakers and they are hence a bit too bright (and I might try to do some other stuff with HF EQ later).
I want to try that in REW first and then later in RePhase. As the averages above 1Khz differ almost by 10 dB, the EQ results will be different. So, which is the correct one to use? Or does the difference show, that I do have problems with the phase, and I should rather not attempt Eqing above 200 Hz?

On another line of though, I noticed that the centre measurements fit the RMS average quite well (after removing the two extreme measurements that are behinm my LP close to a tilted roof) and of course still contains phase information. Should I rather skip averaging and simply use that centre LP measurement?

I am thankful for any comments.
(The filters will be used in ROON)

 

[jtalden]

FYI:
For manual IIR PEQ I prefer using a RMS average rather than the vector average and address the entire curve. My process uses the moving microphone measurement (MMM) technique for EQ purposes. It provides the same response as the RMS average of many individual measurements in the same area, but is much quicker and is more repeatable. There is a bit of a learning curve however to do it correctly. I EQ the bass to about 250 Hz and also address a wide 450 Hz sag in SPL response. Above 600 Hz I only shape the rolloff to my preferred house curve.

I see that it is more popular to use the vector average for PEQ in the guidelines provided in the rePhase thread at diyAudio.

I use the LP sweep measurement to create the phase correction FIR. I just use enough FDW (maybe 4-5 cycles) to smooth the phase response to determine the direct sound phase rotation. I correct phase according to that. I ignore remaining room disturbances still present in the direct sound phase rotation.

The rePhase thread guideline I have seen only correct for the excess phase portion. I don't understand why that extra work is done as there is probably some additional phase rotation added during the recording process. The difference between the correcting for the measured phase response vs only the calculated excess phase response is probably meaningless in practice so I don't think it matters either way.

I would not try to correct for the center measurement at the higher frequencies. It's fine to use that up to 250 Hz. Above that the wiggles are probably not representative of what we hear.

This is just another opinion based on what has worked best in my room given my equipment constraints having evaluating many different PEQ approaches. I do not know if my process is likely to be better or worse in other applications. I attached some charts to give you some idea of the results.

I suppose you either just pick a process to use or experiment yourself to see which process works better for your application.

Good luck!

 

[Breeman]

Beatiful graphs

FYI:
For manual IIR PEQ I prefer using a RMS average rather than the vector average and address the entire curve. My process uses the moving microphone measurement (MMM) technique for EQ purposes. It provides the same response as the RMS average of many individual measurements in the same area, but is much quicker and is more repeatable. There is a bit of a learning curve however to do it correctly. I EQ the bass to about 250 Hz and also address a wide 450 Hz sag in SPL response. Above 600 Hz I only shape the rolloff to my preferred house curve.

I see that it is more popular to use the vector average for PEQ in the guidelines provided in the rePhase thread at diyAudio.

I use the LP sweep measurement to create the phase correction FIR. I just use enough FDW (maybe 4-5 cycles) to smooth the phase response to determine the direct sound phase rotation. I correct phase according to that. I ignore remaining room disturbances still present in the direct sound phase rotation.

The rePhase thread guideline I have seen only correct for the excess phase portion. I don't understand why that extra work is done as there is probably some additional phase rotation added during the recording process. The difference between the correcting for the measured phase response vs only the calculated excess phase response is probably meaningless in practice so I don't think it matters either way.

I would not try to correct for the center measurement at the higher frequencies. It's fine to use that up to 250 Hz. Above that the wiggles are probably not representative of what we hear.

This is just another opinion based on what has worked best in my room given my equipment constraints having evaluating many different PEQ approaches. I do not know if my process is likely to be better or worse in other applications. I attached some charts to give you some idea of the results.

I suppose you either just pick a process to use or experiment yourself to see which process works better for your application.

Good luck!
View attachment 37375

View attachment 37376

View attachment 37378

Beautiful graphs, there is quite a lot of similarities between all your channels based on your first picture. If I may ask, is this a function of your room's dimensions/speaker placement or is it because you used the same speakers for all your channels? Do you know what is causing the phase rotation at around 260Hz for the two front speakers? Just to ensure that I understand your phase correction process; did you correct the resultant speaker phase response, after removing the time of flight to the LP, and never bothered with excess phase?

 

[jtalden]

Hi Breeman,
Per my equipment list I am using 5 identical DIY mains and 2 DIY passive woofers/SWs all managed using two DCX2496 boxes for XO, delays and PEQ. I consider my room and choice of LP to be a bit problematic and have very little acoustic treatment to help. The picture below shows the AV section of the room. It's about 11 x 24 x 9 ft and it has a 12 ft wide opening to another similar sized space. My surrounds are located as shown. This ~60° location is as effective in creating spaciousness as as the normal 120° setup as per Muroka and Nakazato (2007) (charts referenced in Floyd Toole's book). It is was more convenient for placement and wiring in my case.

I am not sure about the causes of the strong phase flips at 60 Hz and 250 Hz. I have had trouble assigning them to any single particular room effect, but they both shift or disappear with different mic locations. They are not in the direct sound.

Yes, I corrected the phase using rePhase using probably a 5 cycle FDW on one of the surrounds. I used that as it was the cleanest of the 5 phase responses to work with. I could have averaged all 5 mains to get or the FL and FR to get a smoother response, but all are basically identical so that was not necessary in this case. If we just look at the FL at 3 different FDW settings shown below (2, 5, 8 cycles), we can see I could have easily used the FL instead. A 2 cycle FDW results in a pretty smooth phase rotation in this case. I routinely just use 5 cycles to see a little more detail and just ignore the 2 large room disturbances. The FIR filter setting are chosen based on the acoustic XOs I have created at approximately 250 and 2k Hz. I also have a subsonic filter to protect my woofers and a low pass filter at 20 kHz that adds to the total phase rotation and thus the needed FIR correction. The rePhase filter settings are shown below.

The spectrogram of the vector average of the 5 channels shows the room/EQ problems at 60 and 250 Hz.

 

[Breeman]

Thanks for the detailed explanation, it makes sense now that I can see your setup. Just out of interest pertaining to your subwoofers; there is significant delayed energy from 50 to 80 Hz. Can you pick this up during critical listening or does everything just sound coherent? At which range are the SUBs crossed over to the mains?

In any case, the spectrogram is great at showing these anomalies?

 

[jtalden]

XOs are at 250 and 2k Hz.

Yes, the strong room effect around 60 Hz leaves a ~6-7 dB dip in the response. The choice is to leave it there or EQ it. In this case I EQ'ed it almost flat. The down side is that that the EQ cannot boost the direct sound. It just puts more energy into the room that eventually returns to the LP at a later time. Another unwanted impact is the long decay time at that occurs at those frequencies. I have tried no EQ, a partial/compromise EQ and full EQ as this is. I have no advice as to which sounds better as it probably a personal choice and this is a choice that I have not even decided for myself. I'm leaning in toward a more compromise EQ setting. Only room acoustic adjustments can really address this problem.
My SWs are actually DIY JBL 2235H woofers in 6 cu ft vented boxes and are capable and flat 20-600 Hz with reasonable output at 20 Hz for a small room. This is the first time in 30 years I have crossed them above 100 Hz and thus far I like the change. It eliminated a 120 Hz null due to the mains locations and I am not aware of any sound quality or differing spatial effects in my program material even though they are still running as mono.

Yes, the spectrogram provides a good view of the arrival time and decay as well as helpful indication of reflections.

 

[JoachimStrobel]

FYI:
For manual IIR PEQ I prefer using a RMS average rather than the vector average and address the entire curve. My process uses the moving microphone measurement (MMM) technique for EQ purposes. It provides the same response as the RMS average of many individual measurements in the same area, but is much quicker and is more repeatable. There is a bit of a learning curve however to do it correctly. I EQ the bass to about 250 Hz and also address a wide 450 Hz sag in SPL response. Above 600 Hz I only shape the rolloff to my preferred house curve.

I see that it is more popular to use the vector average for PEQ in the guidelines provided in the rePhase thread at diyAudio.

I use the LP sweep measurement to create the phase correction FIR. I just use enough FDW (maybe 4-5 cycles) to smooth the phase response to determine the direct sound phase rotation. I correct phase according to that. I ignore remaining room disturbances still present in the direct sound phase rotation.

The rePhase thread guideline I have seen only correct for the excess phase portion. I don't understand why that extra work is done as there is probably some additional phase rotation added during the recording process. The difference between the correcting for the measured phase response vs only the calculated excess phase response is probably meaningless in practice so I don't think it matters either way.

I would not try to correct for the center measurement at the higher frequencies. It's fine to use that up to 250 Hz. Above that the wiggles are probably not representative of what we hear.

This is just another opinion based on what has worked best in my room given my equipment constraints having evaluating many different PEQ approaches. I do not know if my process is likely to be better or worse in other applications. I attached some charts to give you some idea of the results.

I suppose you either just pick a process to use or experiment yourself to see which process works better for your application.

Good luck!
View attachment 37375

View attachment 37376

View attachment 37378

THanks for your answer

FYI:
For manual IIR PEQ I prefer using a RMS average rather than the vector average and address the entire curve. My process uses the moving microphone measurement (MMM) technique for EQ purposes. It provides the same response as the RMS average of many individual measurements in the same area, but is much quicker and is more repeatable. There is a bit of a learning curve however to do it correctly. I EQ the bass to about 250 Hz and also address a wide 450 Hz sag in SPL response. Above 600 Hz I only shape the rolloff to my preferred house curve.

I see that it is more popular to use the vector average for PEQ in the guidelines provided in the rePhase thread at diyAudio.

I use the LP sweep measurement to create the phase correction FIR. I just use enough FDW (maybe 4-5 cycles) to smooth the phase response to determine the direct sound phase rotation. I correct phase according to that. I ignore remaining room disturbances still present in the direct sound phase rotation.

The rePhase thread guideline I have seen only correct for the excess phase portion. I don't understand why that extra work is done as there is probably some additional phase rotation added during the recording process. The difference between the correcting for the measured phase response vs only the calculated excess phase response is probably meaningless in practice so I don't think it matters either way.

I would not try to correct for the center measurement at the higher frequencies. It's fine to use that up to 250 Hz. Above that the wiggles are probably not representative of what we hear.

This is just another opinion based on what has worked best in my room given my equipment constraints having evaluating many different PEQ approaches. I do not know if my process is likely to be better or worse in other applications. I attached some charts to give you some idea of the results.

I suppose you either just pick a process to use or experiment yourself to see which process works better for your application.

Good luck!
View attachment 37375

View attachment 37376

View attachment 37378

Thanks for you answer. I had to sort out a few things before I could progress but am on a good path now. I thought that if I dive into phase and Rephase, I rather do this with the full system response. Hence I now play seeps via Roon so that my music and the REW paths are the same (This only works for 2 channels, I need to confess that for 6 channels I use Dirac anyway; the 2 channel work here I just for auxiliary systems).

So, the mystery continues, but is it fun.

I made 4 measurements around my LP, no FDW and vector average them. The results reflect the phase difference of the measurements and hence the vector average is different to the original measurements.

Then I apply a 1/6 octave FDW processing and then vector average. The phases of the measurements are now all identical which I need to fully understand. The vector average is now identical to the rms average (but the phase has shifted a bit….).

This seems to imply that phase measurements are a bit tricky, are they important? I like to EQ the full bandwidth as I want to manipulate the region above 2 kHz to check how hearing problems could be combated. For that I must manipulate the phase of the filters otherwise the result sounds awful (I tried that already). Dirac does that beautifully, but with Rew & RePhase I should arrive at the same results. And I cannot implement software Dirac in Roon (or only if I use Vcable, Jriver, a Dirac-VST plugin and HDMI out). And once I have mastered that with Roon, I can apply this to any system that I attach to Roon´s RAAT protocol.

 

[jtalden]

Phase is definitely a tricky bit. I do not recommend worrying about unless there is a clear understanding of what it is, how it is manipulated and its overall relative importance. I do not claim (or hear) that there is a sound improvement by just removing the overall phase rotation. I just do it because I can for stereo music by using a rePhase FIR filter in my Foobar 2000 music server. That's what hobbyist do. Dirac, Acourate, DCR and the other more sophisticated programs may be more effective in treating room effects and are more likely to be effective in improving sound quality. I unfortunately have no personal listening experience with those programs and have no judgment about their impact.

 

[JoachimStrobel]

Phase is definitely a tricky bit. I do not recommend worrying about unless there is a clear understanding of what it is, how it is manipulated and its overall relative importance. I do not claim (or hear) that there is a sound improvement by just removing the overall phase rotation. I just do it because I can for stereo music by using a rePhase FIR filter in my Foobar 2000 music server. That's what hobbyist do. Dirac, Acourate, DCR and the other more sophisticated programs may be more effective in treating room effects and are more likely to be effective in improving sound quality. I unfortunately have no personal listening experience with those programs and have no judgment about their impact.

I am interested in EQing frequencies above 1 kHz. I read that phase is important there without having fully understood why. Nevertheless, comparing Dirac‘s full frequency EQs and my homegrown REW attempts I can hear why.
Using FDW as you did, one measures mostly the direct sound from the speaker at higher frequencies. This gives a clean phase that can be vector averaged nicely and worked on with Rephase or else as per your nice examples. Is that the correct way doing this? Without FDW, phase gets very messy, vector averages look strange and Rephase is a bit of a hopeless attempt. On the other hand, FDW does look a bit too smooth for my eyes.
BTW, with growing projector screens, the centre may have to move behind that screen eventually. That makes full frequency Equing a must and these issue highly relevant. One could argue what is worse to EQ: Using a small loudspeaker with a huge subwoofer to stay in front of the screen, or going behind the screen with a full loudspeaker system.