Print

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

[tboooe]

I am a long time MC user.  At the moment I am still using version 20 and was considering upgrading but wanted to ask first if anyone has had a similar experience as me with inconsistent results using convolution?  I use Room Eq Wizard to measure my room and generate filters which I export as wav mono files for each speaker.  I then created a convolver text file pointing to the mono wav files.  Most of the time this works great.  However, I recently noticed a that I am getting strange measurements.  When I create a filter using a large negative gain, I will get different measurements over a period of time.  When I output and measure just 1 speaker, the results are as expected. 

Here is a link to my post on the REW forum for more info:
http://www.hometheatershack.com/forums/rew-forum/158473-monster-null-out-no-where-2.html#post1515601

I have attached pics of the measurements I took in REW.  In the image, Both Speakers Filters Applied two separate measurements are shown taken only mins apart.  Nothing changed in the setup or my system.  As you can see, in one measurement, there is a trough between 60 and 80hz.  In another measurement, there are two deep nulls at 62 and 77hz.  These nulls correspond to the where the negative gain filter was applied. 

I have also attached images to show the measurement of the right and left speaker by themselves, using the same filters.  You can see how nice and flat the measurements are with no major nulls or peaks in the 60-80hz range. 

Finally, I show the measurement with convolution turned off.  In this image, there is a trough between 60-80hz but no deep nulls.

So I am very perplexed why when using convolution, there are different meausrements???

[mwillems]

I've never seen that kind of variation in the convolution module myself.  You say the measurements were taken with no changes between them; was the microphone in the exact same position?  Have you tried using less or different smoothing to see if the raw data looks similar?

Nulls of that size are generally only caused by cancellation effects (two sound sources exactly out of phase). Examples of phenomenon that could cause that kind of cancellation: room nulls (caused by out of phase reflections hitting each other), two sound sources interacting with each other in a destructive way, or a misaligned crossover.  For example, a well known phenomenon called "comb filtering" occurs when two speakers are playing at the same time, which creates a jagged looking measurement where the speakers cancel, don't cancel, cancel, don't cancel so it looks like a comb.  And the interaction between the two will differ at different room positions based on how far the speakers are apart, what they're playing, etc.  Pro audio is mastered with the effect in mind, but test signals are clearly not.  This seems likely as you don't see it with the speakers playing by themselves, but do see it in a composite measurement.  FWIW, I don't recommend that anyone measure all speakers at the same time for anything other than getting a sense of total volume levels for exactly these reasons.

Two theories:

1) If the microphone wasn't in the exact same spot, my guess would be that you may have inadvertantly moved it into an area of the room that had a different interaction with those frequency bands, or in which the speakers themselves were interacting differently.  Room nulls or comb filtering can be very location dependent, the "anti-sweet spot" can be physically very small, and the roll off is very steep, so even moving the mic 6 inches or so can be enough to change things if you happen to land at the center of the null.  If you're seeing these results a few minutes apart without touching the microphone, feel free to disregard.

2) Alternatively, smoothing can sometimes mask very narrow nulls, such that I've seen two measurements where the smoothed graphs looked pretty different, but the raw data looked fairly similar, it's just that one had a null that was slightly narrower than the other, if you see what I mean.

It's posible there's something wonky with the convolution, but I've never seen that myself in a few years of use.  Maybe other convolution users can chime in too?

EDIT: I just saw in the update to your REW thread that you used REW to mathematically sum the individual speaker measurements as a test and got the same null.  You're almost certainly seeing comb filtering or inter-speaker phase cancellation. 

[tboooe]

I've never seen that kind of variation in the convolution module myself.  You say the measurements were taken with no changes between them; was the microphone in the exact same position?  Have you tried using less or different smoothing to see if the raw data looks similar?

Nulls of that size are generally only caused by cancellation effects (two sound sources exactly out of phase). Examples of phenomenon that could cause that kind of cancellation: room effects/standing waves, two sound sources interacting with each other in a destructive way, or a misaligned crossover.  For example, a well known phenomenon called "comb filtering" occurs when two speakers are playing at the same time, which creates a jagged looking measurement where the speakers cancel, don't cancel, cancel, don't cancel so it looks like a comb.  And the interaction between the two will differ at different room positions based on how far the speakers are apart, what they're playing, etc.  This seems likely as you don't see it with the speakers by themselves, but do see it in a composite measurement.

Two theories:

1) If the microphone wasn't in the exact same spot, my guess would be that you may have inadvertantly moved it into an area of the room that had a different interaction with those frequency bands, or in which the speakers themselves were interacting differently.  Room nulls or phase nulls can be physically very small and the roll off is very steep, so even moving the mic 6 inches or so can be enough to change things if you happen to land at the center of the null.  If you're seeing these results a few minutes apart without touching the microphone, feel free to disregard.

2) Alternatively, smoothing can sometiems mask very narrow nulls, such that I've seen two measurements where the smoothed graphs looked pretty different, but the raw data looked fairly similar, it's just that one had a null that was slightly narrower than the other, if you see what I mean.

It's posible there's something wonky with the convolution, but I've never seen that myself in a few years of use.  Maybe other convolution users can chime in too?

Thank you for the reply.  Yes, absolutely no changes in my system, mic position, or filters.  In fact, I took the measurements only minutes apart.   I verified using REW's signal generator and SPL meter.  Sometimes there is a null at 62 and 77 as shown in the response.  Other times, there is no null.  When I remove all filters and again use the REW SPL meter, there are no nulls. 

As for smoothing, I do not use any at the lower frequencies.  That is raw data.

As for cancellation, I agree that makes sense but why would the measured combined response only sometimes show the 2 dips at 62 and 77hz but with nice response in between the dips while other times I see a trough from 60-80hz?

[mwillems]

Thank you for the reply.  Yes, absolutely no changes in my system, mic position, or filters.  In fact, I took the measurements only minutes apart.   I verified using REW's signal generator and SPL meter.  Sometimes there is a null at 62 and 77 as shown in the response.  Other times, there is no null.  When I remove all filters and again use the REW SPL meter, there are no nulls. 

As for smoothing, I do not use any at the lower frequencies.  That is raw data.

As for cancellation, I agree that makes sense but why would the measured combined response only sometimes show the 2 dips at 62 and 77hz but with nice response in between the dips while other times I see a trough from 60-80hz?

I can think of a few reasons it would only affect those frequency bands; my recollection is that REW still generates minimum phase filters when generating convolution files.  In that case any change to frequency will also affect phase, etc.  So if you've got significant adjustments at those frequencies you're also adjusting the phase at those frequencies which may cause just those regions of the two speakers to be "out of step." 

I can't explain the inconsistency though, unless you're taking the measurements so close together that room-loading could be a thing, but I highly doubt that. 

A useful diagnostic test: take the filters you're using to generate the convolution file in REW and manually type them into JRiver's parametric equalizer.  Then measure with the convolver off, but the PEQ on.  If you can reproduce the same issue, it's something about the way your speakers and room interact with those filters specifically.  If you can't reproduce it, there's (likely) something wrong the convolver module.

[tboooe]

A useful diagnostic test: take the filters you're using to generate the convolution file in REW and manually type them into JRiver's parametric equalizer.  Then measure with the convolver off, but the PEQ on.  If you can reproduce the same issue, it's something about the way your speakers and room interact with those filters specifically.  If you can't reproduce it, there's (likely) something wrong the convolver module.

GREAT idea.  Let me try that and report back.

[mattkhan]

difficult to comment without seeing the filter, perhaps you can post the wav somewhere

on a different note, why not just use PEQ in jriver instead of a wav?

[tboooe]

difficult to comment without seeing the filter, perhaps you can post the wav somewhere

on a different note, why not just use PEQ in jriver instead of a wav?

Thank you for the reply.  Attached are the filters in text for each speaker.

The reason why I dont just use the PEQ is that I am trying out another software player that only accepts wav files for convolution. 

[tboooe]

More testing...I manually entered the REW filter settings into MC parametric eq and reran the measurements. The combined graph show the double dip at 62 and 77hz (precisely where I have my negative gain filter on the R and L speaker respectively). The measurements for each speaker is also very similar to when I use MC convolution engine. Therefore, I have to believe the inconsistent results of the combined measurement must be some how related to the MC convolution engine???  Now the question is what's going on??

[mwillems]

More testing...I manually entered the REW filter settings into MC parametric eq and reran the measurements. The combined graph show the double dip at 62 and 77hz (precisely where I have my negative gain filter on the R and L speaker respectively). The measurements for each speaker is also very similar to when I use MC convolution engine. Therefore, I have to believe the inconsistent results of the combined measurement must be some how related to the MC convolution engine???  Now the question is what's going on??

I'm sorry I don't follow; you said that entering the filters into the PEQ shows you the same double dip you see when using convolution?  If that's the case, it's not the convolution engine, it's the interaction between your speakers.

Or do I misunderstand?

[tboooe]

I'm sorry I don't follow; you said that entering the filters into the PEQ shows you the same double dip you see when using convolution?  If that's the case, it's not the convolution engine, it's the interaction between your speakers.

Or do I misunderstand?

Sorry about that...recall from my first post I had 2 questions:
1.  Why was I getting two different measurements using the convolution?
2.  Why was I getting the double dip?

So it seems the double dip is indeed present since it shows up using the PEQ.  However, I am still perplexed why I got different results using convolution.