A Grimani tuning technique... All pass filters for improved stereo imaging. Thoughts?

[jEDGEc0m]

I've enjoyed listening to chats by Anthony Grimani, generally on room acoustics. I feel there's not much to disagree with him about, nor the way he presents things. He makes good sense to me.

In a recent presentation, titled "How to Achieve Incredible Stereo Imaging from Two Speakers", he mentions in passing a few things, including a method to tune and apply all-pass filters.
Discusses it in this bit, at this time stamp...

Technique - Play pink noise with speakers out of polarity. (eg use REW's pink noise and tick "Invert second output").

You would then listen for parts of frequency spectrum that still seem to be in the phantom centre.

I felt I could hear what he means.

Then technique is to sweep around with an all-pass filter (on one channel of course), still with pink noise out of polarity, and try get those centre-sounding frequencies to move away from the centre. (I am applying eq/processing via REAPER DAW, using the stock ReaEQ)

I felt I achieved some success - in the past I've messed with all-pass filters, but never been confident the result was an improvement. Of note is that the frequencies targeted were not bass - the low end is behaving and measuring quite well in here - the main all pass difference was at ~2800hz.

In playing actual music, flicking it on and off, a difference exists, but is generally quite subtle, But is kinda interesting. preference might be hard to pick in a blind AB. But generally, it is not worse, and sometimes I think better. Early impressions.... in-room moving mic freq response doesn't show any real difference from the all-pass, but I guess that's to be expected.

I've often had a perception that things are a bit off with stereo in this room - like mono mid-range elements are perceived a touch left leaning. speakers are AsciLab F6B, and in a fairly crude measurement, they seem well matched as a pair (link to post where I measured that...) and in-room at listening position, the frequency response measures well/pretty even.

I just changed the arrangement in the room yesterday which should have helped improve symmetry (main part of the change was to try 2x subs mid-wall either side of room - previously one was in front corner, and other ~1/3 along front wall, and elevated off ground. Room is square 3.5x3.5x2.4m.).

Then I played just a mono voice-over... the all-pass enabled does better focus the mono voice, and moves it better towards the centre. It is all fairly sensitive to head position, but I suppose such is the reality of stereo speakers and phantom centre/"soundstage".

Thoughts?

If you give it a try/have or done all-pass things, would be keen to hear about it. Didn't seem much discussion of it in a bit of a forum search.

ps: Grimani also mentions comparing step responses... I don't really know what I'm looking at. Does this mean much?? He said the curves should follow each other. They seem to do so reasonably well to me. I set it to %, as this seems easier to visualise. This was with the all-pass filters, but difference without them seemed pretty similar on that graph. Also of note, I didn't put the mic in a stand - I was holding it pretty steady braced on the couch, but still - held by hand.

edit: I'd written Grimani's name incorrectly everywhere, including thread title.

 

[DVDdoug]

That can create a "widening effect" but it's not really "correct stereo". Sometimes it's used as a creative production technique but it can foul-up mono compatibility.

and try get those centre-sounding frequencies to move away from the centre.

Why would you want to foul-up the phantom center? Usually the lead vocals and other sounds are supposed to be in the center.

At low frequencies out-of-phase soundwaves cancel. In the mid frequencies things can get weird and perhaps interesting, especially as you move around.

At high frequencies the phase relationships are pretty random already because of the short wavelengths. The direct and reflected waves have random phase relationships and there is enough delay between your left & right ears to create phase differences between left and right. We don't usually notice it with music but it becomes very obvious with constant test tones when you move your head and the loudness changes drastically.

2800Hz has a wavelength smaller than the difference between your ears and at 1/2 wavelength the waves are 180 degrees out-of-phase so I wouldn't expect a whole lot of "effect" in that range and above.



BTW - Floyd Toole says:

The important localization and soundstage information is the responsibility of the recording engineer, not the loudspeaker.

 

[TimoJ]

Why would you want to foul-up the phantom center? Usually the lead vocals and other sounds are supposed to be in the center.

Because while adjusting, you should use : "Play pink noise with speakers out of polarity."

 

[jEDGEc0m]

Why would you want to foul-up the phantom center? Usually the lead vocals and other sounds are supposed to be in the center.

Yeah, the point is to do entirely the opposite. As TimoJ said, and glad I hopefully was clear enough initially - the evaluation and adjustment is while pink noise is played with the speakers intentionally set out of polarity.

As Grimani discusses, there could be any number of reasons it might happen - the point is to improve it. And I am hearing some merit in the result, even though I'm sure I haven't identified and tuned the filters perfectly.

Definitely does something in those higher frequencies. Just now, with a song with some moments of more sparse arrangement and moments of mono guitar in the mix, the guitar the guitar becomes quite notably better focused and more prominent with the all-passes enabled. (Porcupine Tree - Anesthetize, remaster).

It's a positive change, and does seem to help address the stereo weirdness I feel I experience sometimes in here.

edit: I'd written Grimani's name incorrectly everywhere, including thread title.

 

[Joe Bloggs]

So basically, the APFs are there to make sure that all frequencies sound out of phase when they ARE out of phase--
playing pink noise out of phase and picking out frequencies that still sound like coming from the middle and try to pull them apart somehow?

I wonder what he thinks of algorithmic calculation of room correction impulses.

 

[Joe Bloggs]

It just looks to me like the OP's system's bass could really use some speeding up judging from the waveform

 

[Keith_W]

One look at that step response and i'm guessing you are using DLBC. Am I right?

Anyway, let us look at his claim - that an All-Pass Filter can improve stereo imaging. First he starts off with a very brief description of the ITD (Interaural Temporal Difference) - sound that arrives in one ear before the other is perceived as coming from that direction. This is the mechanism that is thought to be responsible for localisation of sound between 100-200Hz up to about 1500-2000Hz.

If there is phase asymmetry between left and right speakers within this freq range, it will mess with the ITD. It will produce a phenomenon called "phantom centre image drift" - for example, a singer may be perfectly centred in the upper notes, but may wander to left/right for the lower notes. This phenomenon can also be caused by poor positioning of stereo microphones when the recording was made. I found one such example:

For a long time I wondered whether it was my speaker that seemed to make Bostridge suddenly jump to the left or right in a most distracting fashion, then I listened with headphones. It is indeed in the recording.

Anyway, the phenomenon is real. The question is whether an AP Filter is the correct solution. We should think about what creates the phase asymmetry in the first place. Culprit 1: it may be in the recording. Culprit 2: it may be your speakers, for e.g. one driver may be wired out-of-phase. This would be a very rare manufacturing defect, but it's possible. Culprit 3 (most likely): it's how you set up your speakers in your room.

If one speaker is closer to the side wall than the other, it will skew imaging. The most common audible defect from this is a smaller soundstage on the side which is closer to the wall, with the other more open side sounding more spacious. Since not all speakers have smooth directivity, some speakers may have stronger reflections in certain frequency bands.

In theory, where you can spot this is in the excess phase group delay plot in REW. Sweep your L/R speaker, then right click and "Make Excess Phase Copy". Then go to overlays, select the EP versions of both measurements to compare them. Then click on "GD" to see the Group Delay. You will see something like this:

In a EP GD plot, flat areas are the minimum-phase response, and the spikes are excess phase. If you want to design an AP filter to correct EP, use this plot to find the centre frequency and tailor the Q to match the height and width. If you are using linear phase filters, you can time-reverse it and put it in the same speaker. This will flatten the graph. Or if you are using minimum-phase DSP, you have to put it in the other speaker. This will make the left and right speaker more symmetrical.

This technique is used in DSP to remove dips in the frequency response by using the other speaker to fill dips. Acourate's "ICPA" (Interchannel Phase Alignment) tool does this. But I suppose it could be used to "improve" imaging. I have never tried it, but bear in mind that messing around with phase this way, especially at high frequencies, and high Q, will potentially introduce unwanted artefacts into your correction. Like, pre-ringing (if using linear phase) or worsening post-ringing (if using IIR DSP) if high Q corrections are used. Or correcting for an extremely specific location if you are attempting to correct high freq EP.

What you are essentially trying to do is move walls with DSP. Whilst it is theoretically possible, I think it's better not to f*** around with high freqs ... so I disagree with Grimani on this one. It's better to fix the problem at the source - set up your speakers symmetrically in your room, and leave the high freqs alone!

 

[Keith_W]

Here is a quick AP filter I designed (black) to match the EP peak at 104Hz (purple). The f0 is 104Hz, and Q is 40. Just a quick illustration to show you what an AP Filter looks like on a GD plot.

(Edit) I realize I forgot to answer your step response question. Go to chapter 4.4 (page 50) in Book 2 of the REW eBook, the section on "step response". There are examples of "good" and "bad" step responses in there. Grimani is correct in saying you want the left/right step response to look as identical as possible. The head of the step response is how well your speakers have been integrated with DSP. The tail of the step response tells you what the speakers are doing in the room. You could also read Chapter 4.5 on the IACC, which is how REW compares the step response. I find REW's IACC calculation to be nearly useless for reasons I mention in the book.

 

[jEDGEc0m]

It just looks to me like the OP's system's bass could really use some speeding up judging from the waveform

Of some relevance might be that the all-pass filters that made the improvement in my case were up rather high - i messed around a bit, but best result i could get was a really wide Q one up at like 10k, and another one, narrower at i think it was ~5k - so still rather high - and i guess the two might be interacting with each other - but it certainly seems to help in my case.

By "speeding up bass" - what do you mean? That maybe the subs are too delayed? (The crossover to mains is linkwitz-filey 4 pole/24db/octave, and summation at crossover was good with applying a delay that correlated with physical distance difference - ~1.5ms).

 

[Joe Bloggs]

Of some relevance might be that the all-pass filters that made the improvement in my case were up rather high - i messed around a bit, but best result i could get was a really wide Q one up at like 10k, and another one, narrower at i think it was ~5k - so still rather high - and i guess the two might be interacting with each other - but it certainly seems to help in my case.

By "speeding up bass" - what do you mean? That maybe the subs are too delayed? (The crossover to mains is linkwitz-filey 4 pole/24db/octave, and summation at crossover was good with applying a delay that correlated with physical distance difference - ~1.5ms).

Don't mind me--I thought I was looking at impulse response. You're showing step response.

 

[jEDGEc0m]

One look at that step response and i'm guessing you are using DLBC. Am I right?

Thanks for your detailed response - I look forward to reading and understanding better, and trying some of the mentioned measurements and evaluation and reporting back when I get the chance. Might have to be a weekend thing...

but regarding DIRAC + Bass Contol (DLBC) - funny you mention it- cause since then, I did try the 14-day trial of DIRAC Live + Bass Control, via PC/VST3 - but no, all this was before that - the L/R step response graph in first post was measured with no DIRAC.

Setup is pair of "subs" + the AsciLab F6B, with LR4 xover + basic PEQ below ~500hz. no linear phase stuff. no latency added.
"Subs" are a re-purposed pair of ported tower speakers each with 2x 8" woofers wired to bypass xover. They do plenty well down a little below 30hz in room.

My DIRAC experience was generally very negative - initially with just the AsciLab run full range, DIRAC seemed alright - fixed the peaky bass regions. from Freq Response REW measure, it also kinda seemingly had some success filling in a pretty substantial and likely room dip. sounded ok - say improved from flat signal to speakers in room. but not better than a quick REW PEQ tune and then the all-pass filter Grimani technique. DIRAC still kinda sounded weirdly phasey with some content/freq range, like before my all-passes.

When I added the subs and "BC" to DIRAC measure and calibrate, it was completely ridiculously messed up results. tried a bunch of things - including restricting it to 500hz correction. couldn't figure out how to not make it sound terrible, nor understand what I could possibly be doing that might have been confusing it. also, I thought it was choosing 70hz xover for subs to mains - not an unreasonable value - but apparently it doesn't choose - 70hz is just a default. it uses no intelligence to set sub-main xover, and leaves that to the user, for some reason. it did however choose 1.4ms delay to subs, basically the same as I had determined manually on distance - so that was something. It grouped the subs together - seemed it wouldn't/couldn't delay them separately - though in my trials of that manually, it doesn't help in this setup anyway - they are symmetric in the room - mid side walls.

The freq response measured pretty normal looking with a typical "room" curve, similar to what i'd end up with - but the sound of it was messed up in a bad weird way.

"Culprit 2" - AsciLab speaker driver xover polarity wrong in one speaker - it'd reflect poorly on my ears if this is the case! will have to check. could it be!!? can i get my forum donor refunded and delete my profile if so??

Culprit 3 - room setup/symmetry - "stuff" in the square room is not like an engineering control room, but mostly not too bad I would think - but speaker distance to each sidewall is measured and set to be pretty darn similar (from memory, about 800mm from sidewalls.). but right speaker has a door next to it in that corner, which i almost always have ajar ~150mm - so an angled pretty flat surface.

Before teh all-pass discovery, I did at mess around with my makeshift basstraps/aborbers, thinking that front upper tri-corner had a bass trap, other doesn't, cause of the door - both rear upper tri corners have simlar bass traps - it didn't notably improve the issue.

For the excess phase group delay check grpahs mentioned - Should i sweep each speaker separately from listening position? (i'd tape-measure such that umik1 is within ~1cm distance from each speaker - umik1 would be pointed vertically, with 90 cal file).

Would i apply any smoothing before looking at that group delay?

I will certainly also close that corner door, and listen with my all-pass filters on/off.
I may also at some point switch back to a previous pair of bookshelf speakers - although i guess i won't do that if it turns out one of the AsciLabs are xover'd out of polarity!!

Anyway, I'll be trying some of the things you've mentioned - thanks again for your detailed response.

I kinda wanted this thread to potentially get some momentum - cause if it isn't a defective pair of speakers, the all-pass improvement has been very beneficial, and could help others.

 

[Keith_W]

My DIRAC experience was generally very negative - initially with just the AsciLab run full range, DIRAC seemed alright - fixed the peaky bass regions. from Freq Response REW measure, it also kinda seemingly had some success filling in a pretty substantial and likely room dip. sounded ok - say improved from flat signal to speakers in room. but not better than a quick REW PEQ tune and then the all-pass filter Grimani technique. DIRAC still kinda sounded weirdly phasey with some content/freq range, like before my all-passes.

Yeah, that's a Dirac issue. They try to shield the user from the complexity of DSP, which is what you have to do if you want to sell a lot of product to people who don't have the ability to read measurements, let alone understand how to DSP. But in the process it removes a lot of flexibility and control. So if it screws up the correction (and it screws up the correction very often!!) the user has no recourse. At least with manual DSP, you can identify what you want to correct and what you want to leave alone. It has its place in the market, and it is usually better than no DSP at all.

"Culprit 2" - AsciLab speaker driver xover polarity wrong in one speaker - it'd reflect poorly on my ears if this is the case! will have to check. could it be!!? can i get my forum donor refunded and delete my profile if so??

Some speaker manufacturers churn out dozens of speakers a day, and each one is hand soldered. I have encountered speakers with faulty wiring VERY OFTEN but those are DIY jobs. One guy held his crossover together with alligator clips and didn't notice that one of his tweeters was disconnected until I told him I would hope that speaker manufacturers don't make this very basic mistake, but you never know!!!

Culprit 3 - room setup/symmetry - "stuff" in the square room is not like an engineering control room, but mostly not too bad I would think - but speaker distance to each sidewall is measured and set to be pretty darn similar (from memory, about 800mm from sidewalls.). but right speaker has a door next to it in that corner, which i almost always have ajar ~150mm - so an angled pretty flat surface.

In that eBook I linked, go to the section on Phase. It's difficult reading but if you are going to mess around with AP Filters, you need to understand what phase is and what it sounds like.

For the excess phase group delay check grpahs mentioned - Should i sweep each speaker separately from listening position? (i'd tape-measure such that umik1 is within ~1cm distance from each speaker - umik1 would be pointed vertically, with 90 cal file).

That's right. Sweep from the listening position.

Would i apply any smoothing before looking at that group delay?

Standard 1/6 or 1/12 smoothing would be fine. I prefer 1/6. Look at the little legends on the graphs I posted, I am pretty descriptive when I name my curves.

I kinda wanted this thread to potentially get some momentum - cause if it isn't a defective pair of speakers, the all-pass improvement has been very beneficial, and could help others.

IMO adding an AP Filter to high freqs is a recipe for weird sound. It won't be disastrous, but it would sound weird. Don't confuse "I hear a difference" with "it's an improvement". Problems with EP are best fixed at the source. Make sure your speakers perform correctly, and set them up symmetrically in your room.

 

[jEDGEc0m]

I downloaded your REW eBook pdfs when I came across them a while ago, and did do some (attempted) reading - I recall primarily the takeaway from 4.2 Phase Response - "This is an advanced topic" and then "In short, all you need to do is look at the frequency response."

In my case, the sound is weirder without the all-passes on one of the pair. and seems pretty great with it.

Don't think these Excess Phase 1/6 L/R plots are telling us much, if I did it right. it certainly seems to show absolutely nothing interesting up in the higher areas where I am reporting a perceived "problem" being improved with the all-pass.

These were sweeps of just the AsciLab speakers - no subs, no crossover, no eq, no all-pass.

If one of the tweeters/crossovers were out of polarity, surely something would have been pretty telling in all the measurements i've taken along the way so far.

Here is unwrapped phase of each AsciLab speaker measured each in same place in room with mic in same place, ~200mm away between tweeter and mid-woofer. Does that prove both speakers are performing the same/as they should?

I tried closing that door to the hallway that is next to one of the speakers - makes no obvious difference - things still sound quite a lot improved with imaging and stereo balance with the all-passes enabled.
Maybe something still with the room setup - the speakers are up against the wall, with a flat screen tv between them, but the panel is about 150mm in front of the speaker baffle, with the speaker centrelines about 300mm away from the edge of the panel. there are 150mm thick low density insulation batts filling the space between each speaker and the flat screen's edges. Screen could be better centred in room and between speakers. Should be able to adjust that, but it doesn't seem like much in it.

Still does seem curious then how something that maybe shouldn't help/be required really is improving my perceived experience.

edit: i also sometimes put the music playback in mono, then switch the all-pass on off - it's substantially better centred phantom with the all-passes enabled. Just did that then - I am at a loss to understand what could be causing the symptom, but the Grimani technique implemented really does seem to help in my situation.

 

[Keith_W]

I downloaded your REW eBook pdfs when I came across them a while ago, and did do some (attempted) reading - I recall primarily the takeaway from 4.2 Phase Response - "This is an advanced topic" and then "In short, all you need to do is look at the frequency response."

Yup. No need to look at the phase response, unless you want to manipulate phase. There are reasons to manipulate the phase response, primarily to improve the frequency response. But this has the potential to introduce DSP side effects in the time domain, which you have to anticipate and specifically look for. And then you have to ask yourself whether the improvement in the frequency response is worth the price you pay. Very aggressive corrections will introduce very nasty side effects, mostly in the form of pre or post-ringing, or strange "phasiness".

But I guess it's DSP, the cost of screwing around is $0 and you might learn something. I can tell you I have done my share of stupid things which seemed to be brilliant ideas at the time. I ran an idea for phase linearisation past Dr. Bruggemann (author of Acourate) and his reply was "Dear Keith. Please forget this idea." without explaining why. Well, I went ahead and did it anyway. The effect was pretty dramatic, I was shocked to hear my soundstage vertically compress, lift off the floor and squashed down from the ceiling. It sounded like I was listening to music from inside a letter box. I have never heard anything like it. Up till now I still don't understand why manipulating phase should compress the sound vertically. It's not in Toole, not mentioned by Griesinger, or JJ, or anybody else.

DSP is a powerful tool and it's a fantastic toy to have. So yeah, if you want to try some experiment, go ahead and do it. Hear for yourself what the effects are. It's educational, and it won't cost you anything but your time. And one day you'll have amusing stories to tell at the pub to your bored friends

In my case, the sound is weirder without the all-passes on one of the pair. and seems pretty great with it.

Don't think these Excess Phase 1/6 L/R plots are telling us much, if I did it right. it certainly seems to show absolutely nothing interesting up in the higher areas where I am reporting a perceived "problem" being improved with the all-pass.

These were sweeps of just the AsciLab speakers - no subs, no crossover, no eq, no all-pass.


View attachment 496332

That is a suspiciously clean looking EP group delay plot. You took it from the main listening position?

Here is unwrapped phase of each AsciLab speaker measured each in same place in room with mic in same place, ~200mm away between tweeter and mid-woofer. Does that prove both speakers are performing the same/as they should?

Your question was about speaker polarity. For that, examine the impulse response and the step response. They should be identical.

edit: i also sometimes put the music playback in mono, then switch the all-pass on off - it's substantially better centred phantom with the all-passes enabled. Just did that then - I am at a loss to understand what could be causing the symptom, but the Grimani technique implemented really does seem to help in my situation.

That's great!

 

[3ll3d00d]

That is a suspiciously clean looking EP group delay plot. You took it from the main listening position?

1/6 octave is pretty heavy smoothing so seems likely to be due to that

 

[jEDGEc0m]

Your question was about speaker polarity. For that, examine the impulse response and the step response. They should be identical.

Thanks again for your detailed response and information.

I guess I deduce the speaker wiring/crossover of each one is ok and consistent. Here's the nearfield measured step response and impulse response graphs of each speaker.

"Dear Keith. Please forget this idea." without explaining why. Well, I went ahead and did it anyway

Great way to learn and experience things, as long as no one gets hurt, and good when there's not financial cost. Interesting outcome. Maybe DIRAC should have been advised not to do certain things, without explanation. Nah, I don't know -but certainly the results in my DIRAC trial were weird and unpleasant.

1/6 octave is pretty heavy smoothing so seems likely to be due to that

I don't know if I've done something wrong, but indeed without 1/6 smoothing, there's more to see in that group delay Excess Phase Version of L & R measured from listening position. But means little to me.

I for now continue to conclude that this Grimani All-Pass filter technique might be worth a try and of some benefit in some instances.

Perhaps in a few weeks I'll take opportunity to switch in different speakers, and see if similar symptom exists, and seems improved with the all-pass filters.
I guess that will conclude it's something with the room/setup. but yeah, things are about as symmetric as is reasonable.

 

[3ll3d00d]

I don't know if I've done something wrong, but indeed without 1/6 smoothing, there's more to see in that group delay Excess Phase Version of L & R measured from listening position. But means little to me.

Completely normal, just assorted cancellations/reflections at higher frequencies due to short wavelengths

 

[Keith_W]

You need to understand what Excess Phase is. I just had a look at my book and realized that my explanation was rather paltry. I'll expand on it a bit more when I update the book next.

Basically: a single reflection is minimum phase. But many reflections may not be. Think of it this way: if you have a ping-pong ball bouncing back and forth in a closed cylinder of approximately the same diameter, its motion and the time of arrival at either end is predictable. That's because each bounce is minimum-phase. Let's say we shrink the size of the ping-pong ball so that it's 1/1000 the size of the original and put it in the same cylinder. Now its motion is much more difficult to predict. Although each bounce is minimum-phase, the cumulative effect of all the bounces is may not be. If the ball were to bounce a regular pattern around the cylinder, e.g. straight up and down, then it's still minimum-phase. But if the angle was off, it's no longer predictable and thus it's not minimum-phase.

In the same way, long wavelengths bounce around your room in a predictable pattern and form room modes. But short wavelengths bounce around all over the place. This is why minimum-phase regions (flat areas of the EP GD graph) are found at low frequencies, and as wavelengths get shorter, you get more excess phase.