REW + multisub - my journey begins

[paulspencer]

This is more of a "sharing the journey" than a "please help" post.

In the past I've used Ultracurve DEQ2496 to EQ flat at the listening position. Current system DIY omni speakers with two Rythmik servo subs. I used the subs as stands running stereo. Good results, but always looking to improve.

The multisub approach makes a lot of sense to me, and I've found REW is a brilliant tool to help with it.

So now I'm getting into this multisub, here's how I've started. First, I put one of the subs in the listening position with the driver at ear level, set crossover to 160 Hz 2nd order LP to see the effect up as high as possible. I'd never actually run them that high. Now I move the mic around to every thinkable sub position and measure. Then I use the averaging feature to see how different positions sum together. Ultimately I think I'll get even more serious and add a 3rd sub.

So far in this first test, I find that the best two positions are dead centre of the room, and the rear left corner. After measuring, I set up the system like this, then used DEQ to do the final adjustment. Well, Ultracurve was doing less work than ever, and subjectively I found the EQ works over a wider area than before. The corner seat in front of the sub now sounds right in the bass - before it was terrible, too boomy.

Please note: these are unsmoothed measurements! If I had 1/3 octave smoothed then the result at the end would probably be flat from 20 - 120 Hz +/- 3db! A more typical result would be +/- 15 db.

This is how the subs look when placed in the position of the mains:




Around 40 Hz I get a persistent mode, below this there are no modes and above quite a few. The top end roll off is NOT LP filtering.

This is showing variations of the sub height when placed in the left rear corner about 2m from my listening position side ways (I'm on the back wall).



Orange floor level, magenta ceiling level, blue midpoint (the worst). As soon as I move up from the floor, things go downhill.

Here are the 3 corners I can use:



Left and right are similar on the front wall behind the speakers, but the rear wall is clearly better. Note the flatness from 70 - 120 Hz. Sadly I had to low pass it and lose some of that benefit.

Now, coffee table sub - always thought this was a good idea to make the sub "nearfield" and reduce room interaction and put the listener in the "direct field." These measurements seem to indicate that this doesn't work. The nearfield range requires subwoofer headphones much closer than a coffee table!

These positions are all slight variations, allowing for a 1.5 x 0.5 x 0.5 m coffee table where the driver could be downfiring, or in a bandpass with vents on either end. I measured these positions, and also showed the dead centre of the room in black. Here:



The centre looks the best by far.

Here it gets interesting:


The two best positions are shown and averaged together. Rear left corner + dead centre in the room. Black is average, blue is the centre (the only position that seems to miss out on the 40 Hz peaks) and orange corner.

These are the locations that most would say "don't do it!" But I had to try. So I set the subs up this way, then ran ultracurve eq, and found I got the flattest response ever, with less eq required than usual. Also, sitting in front of the corner sub, it still sounds balanced. Before it was just too boomy. I still need eq for this to work. I needed to still cut around 40 Hz, and boost around 100 Hz. It would have been better had I been able to run the subs higher, but I actually have to low pass them lower than in the measurements. Otherwise, I may not have needed eq at all.

The mains are doing more work, and the result works over a wider area than before, and I won't call it a final result, more of a beginning.

[brucek]

I move the mic around to every thinkable sub position and measure. Then I use the averaging feature to see how different positions sum together.

I hadn't thought of that method for multiple subs. Really good idea..

brucek

[glaufman]

Actually, I would've thought dead center as being "rule of thumb" for achieving flatness... many/most times though it cuts the output so far that the tradeoff (especially in conjunction with interior design concerns) is not viewed as being worth it...

[paulspencer]

It turns out in my room it's a good spot for a coffee table, but rule of thumb? Definitely not.

[glaufman]

paulspencer wrote:

It turns out in my room it's a good spot for a coffee table, but rule of thumb? Definitely not.

Maybe it would be if more people were willing to give it a try...

Again, I'm only talking flatness here.

BTW, WRT to using a sub as a coffee table, watch out for rolloff

[paulspencer]

Good point! lol

The great thing is that with a mic and REW you don't have to - you can find out for yourself what works best.

[Ricci]

Good stuff Paul. Did you do an actual measurement with both subs one in each position to see if it resulted in the simulated average response? Should be close.

[paulspencer]

I now have them in those positions, but the result is different as I had to set the low pass filter lower in both subs, also one of them is set to 4th order. As as result, it's not as flat, but before I adjusted the crossover, I could see in the RTA that it was very flat already just looking at pink noise. So in practice I can't actually achieve that result without eq right now.

[Wayne A. Pflughaupt]

paulspencer wrote:

Here it gets interesting:

The two best positions are shown and averaged together. Rear left corner + dead centre in the room. Black is average, blue is the centre (the only position that seems to miss out on the 40 Hz peaks) and orange corner.

Is the black line REW’s averaging? If so you won’t get that response with an actual measurement. The measurement will more-or-less track whatever sub is dominating at whatever frequency range - “blue sub” below 30 Hz, “orange sub” between 30-70 Hz, etc.

Re the three-corner graph, while the blue trace was flatter across the 90-100 Hz range, overall it’s not as good as the orange trace, IMO. In the subwoofer range, below 90 Hz, the orange trace is flatter overall, and has better extension to boot. An broad EQ filter at about 40 Hz would get you good extension all the way out to 20 Hz. So you might want to reconsider that position, unless there’s just no other way to fill that 90-100 Hz hole (from your mains-position graph it appears you’re having problems in that region from the mains as well).

There seems to be at least a couple of thoughts about using multiple subs. One is to increase overall gain (read headroom). However, you only get this benefit with corner placement. Some people like corner placement, others think it is wretched. I expect the difference depends on the particular room. Works great in my room, across most of the seating area, but at a buddy’s house, corner placement got one of the worst readings I’ve ever seen.

Another thought on multiple subs is what you’re trying to accomplish, evening out response to a wider number of seating positions. Again, how successful it is will be largely room-dependant. But it sounds like you’re having some good success. I wouldn’t worry about having to use some equalization to achieve your objective. It will only make things better - subs are pretty forgiving.

Regards,
Wayne

[paulspencer]

Hi Wayne and thanks for the feedback. My goal is to get it as flat as I can in as many useful locations as I can, and then use EQ to improve further. I've been using eq the lazy way for years - Ultracurve eq for the seating position only. This works fine for the main couch (3 seats) but not for the corner. Now with the current setup it works in the corner as well.

When I had it set up with the centre and rear corner, with pink noise seen on the RTA it was dead flat up to the xo point around 80 - 100. But I had to set the lowpass lower to make it work! On paper (the RTA screen actually) it looked just like the black average (not that I checked with REW which is less forgiving).

This is just the beginning!

[brucek]

The measurement will more-or-less track whatever sub is dominating at whatever frequency range

So if one sub has a peak from 30-70Hz and the other has an identical dip from 30-70Hz, the result will be a peak from 30-70Hz (as you're proposing), or will the peak and dip simply cancel each other?

brucek

[paulspencer]

Bruce, next time I get a chance to measure I plan to find out! It's easily tested.

[glaufman]

Wayne A. Pflughaupt wrote:

Re the three-corner graph, while the blue trace was flatter across the 90-100 Hz range, overall it’s not as good as the orange trace, IMO.

Would love to see these with the xover engaged, to see how much the flatness in sub response in the xover range will really affect the smoothness crossing over to the mains...
Similarly, I'd love to see some scans of the mains with and w/o the xover engaged...

[glaufman]

paulspencer wrote:

Bruce, next time I get a chance to measure I plan to find out! It's easily tested.

Would love to see those as well..
When I first read it from Wayne, I questioned it, but upon reflection, what he says is perfectly logical... consider:

If one sub is producing X dB at Y Hz, there's only one way a second sub can subtract SPL at that freq... that's if the 2 subs are out of phase at that freq at that listening location. But in such a case, the second sub, when scanned on its own, wouldn't show a dip at that freq (without the first sub cancelling it out), it would still show significant SPL. If the second sub's dip was being cause by a room mode, the first sub would show that same room mode the same way at the same location, meaning it would also show a dip when scanned on its own.

Just in case, I'm putting up my blast shield...

This also explains why Geddes, who champions this strategy over on AVS, tends to favor the multiple sub approach utilizing subs with different characteristics... i.e., one sub for lower mid-bass, another for low-bass, perhaps a third for ULF...

[brucek]

.....The measurement will more-or-less track whatever sub is dominating at whatever frequency range

If one sub is producing X dB at Y Hz, there's only one way a second sub can subtract SPL at that freq

You're both forgetting that the resultant signal when two subs are playing would require normalization after the addition of the two signals. When two signals combine, the overall level is certainly increased, and this allows increased headroom. A second sub can give you ~6dB extra headroom. When the second one is added, the overall level increases and you can turn them down to achieve the same SPL level, so headroom is realized.

Consider two 50Hz sub signals at 75dB. The resultant mix produces an SPL level that is 6.02dB higher.
It allows you to turn the resultant signal level down by 6.02dB and so your headroom has been increased with the use of the two sub signals as opposed to the one..

I think everyone would agree with that, and the math expression to derive it is: = 20LOG(10^(SPL1/20)+10^(SPL2/20)).

So for two 75dB signals = 20LOG(10^(75/20)+10^(75/20)) = 81.02dB (which is 6.02dB higher than the initial 75dB). We can turn them both down now and obtain the original 75dB, and we've gained headroom. This would be known as normalizing the signal level.

Now consider that one of those same subs has a dip down to 67dB at 60Hz and the other one has a peak up to 79dB at 60Hz.

Wayne and glaufman contend that the result signal will be the dominant peak level of 79dB.

So for a 67dB and a 79dB signal mixed = 20LOG(10^(67/20)+10^(79/20)) = 80.95dB

Remember that we turned the signal down after the two were combined by 6.02dB to normalize them.

If we do this with the 80.95dB signal, it becomes 74.83dB result.

So, the dip pulled down the peak to be around the 75dB target. The result is not the dominant peak level. The dip helped cancel the peak. I contend that the REW average line will be quite accurate.

brucek

[Wayne A. Pflughaupt]

brucek wrote:

Consider two 50Hz sub signals at 75dB. The resultant mix produces an SPL level that is 6.02dB higher.

It allows you to turn the resultant signal level down by 6.02dB and so your headroom has been increased with the use of the two sub signals as opposed to the one..

It’s generally and widely accepted that this is only true in the case of corner loading, when you add a second sub in the same corner. Separate placement in symmetrical corners (e.g. the two front ones) only gets a 3 dB increase. With asymmetrical placement, like Paul is experimenting with – all bets are off. You probably won’t realize any increase in overall gain.

I’ve experienced this first hand, way back in the mid ’90s when I was dead-set on stereo subs.

That’s them, the two tall back things at the far sides of the picture with the little speakers sitting on top of them. As you can see, the right one is in a corner. The left one is right next to an opening (a little hall leading to a couple of bedrooms). To the left of the left sub is a long hallway that is the entryway. IOW, asymetrical locating, at least as far relations to the room’s boundaries go. When I finally gave up on the stereo subs thing and co-located both in the right corner, I saw an instant gain of 6 dB. In other words, when the left sub was in its previous location, it was contributing absolutely nothing to overall gain!

(That’s not to say it was a totally useless arrangement. It did deliver a “wall” of bass to the listening postion, while I could localize when they were both in the right corner. Yes, I know that lows are omnidirectional, but it was more of an air pressure thing than anything else – I could “feel” that the bass was coming from the right. Once I got used to that, I fully enjoyed the improved headroom and extension.)

So if one sub has a peak from 30-70Hz and the other has an identical dip from 30-70Hz, the result will be a peak from 30-70Hz (as you're proposing), or will the peak and dip simply cancel each other?

I thought I had some graphs on file that would show how with two subs, lower levels at some frequencies on one sub won’t show up in a combined graph if the second sub has good output in that range. I don’t seem to have them, so I’ll have to generate some new ones, but it’ll be later this afternoon before I’ll be able to do that.

Regards,
Wayne

[brucek]

When I finally gave up on the stereo subs thing and co-located both in the right corner, I saw an instant gain of 6 dB

Of course you did, because the signal received at the listening position was the same from each sub. You realized a wholesale 6.02dB increase across the entire bandwidth.The math that I outlined in my post agrees and reinforces this observation. But, it's no less accurate if the signals are not identical.

The same math holds true for any measured set of signals. Paulspensers graphs of his two subs are the actual listening position measurements. The math can be applied to any specific frequency and will produce the mixed results as accurate as if you actually measured them. This is what Johns software does in his averaging routine. You can take the averaging results that REW produces to the bank....

This is the same theory we use when someone uses a great sub and a challenged sub together, with the mis-guided belief that it will help in the resultant mix. The resultant mix uses the same math that I outlined in my post and "dumbs down" the great sub...... It doesn't help out such that whatever sub is dominating at whatever frequency range.

brucek

[Ricci]

You will get the full fold 6db increase as long as the subs are in close proximity to each other, which is considered to be 1/4 wavelength of each other at the frequency being produced. At 30hz that gives you quite a bit of room to play around, but not so much at 70 or 80 hz.

[glaufman]

brucek wrote:

You're both forgetting that the resultant signal when two subs are playing would require normalization after the addition of the two signals. When two signals combine, the overall level is certainly increased, and this allows increased headroom. A second sub can give you ~6dB extra headroom. When the second one is added, the overall level increases and you can turn them down to achieve the same SPL level, so headroom is realized.

Consider two 50Hz sub signals at 75dB. The resultant mix produces an SPL level that is 6.02dB higher.
It allows you to turn the resultant signal level down by 6.02dB and so your headroom has been increased with the use of the two sub signals as opposed to the one..

I think everyone would agree with that, and the math expression to derive it is: = 20LOG(10^(SPL1/20)+10^(SPL2/20)).

So for two 75dB signals = 20LOG(10^(75/20)+10^(75/20)) = 81.02dB (which is 6.02dB higher than the initial 75dB). We can turn them both down now and obtain the original 75dB, and we've gained headroom. This would be known as normalizing the signal level.

Now consider that one of those same subs has a dip down to 67dB at 60Hz and the other one has a peak up to 79dB at 60Hz.

Wayne and glaufman contend that the result signal will be the dominant peak level of 79dB.

So for a 67dB and a 79dB signal mixed = 20LOG(10^(67/20)+10^(79/20)) = 80.95dB

Remember that we turned the signal down after the two were combined by 6.02dB to normalize them.

If we do this with the 80.95dB signal, it becomes 74.83dB result.

So, the dip pulled down the peak to be around the 75dB target. The result is not the dominant peak level. The dip helped cancel the peak. I contend that the REW average line will be quite accurate.

brucek

Well Bruce, yes, I wasn't thinking about normalizing. That being said, and not knowing the exact algorithm for REWs averaging function, your math shows the resultant being closer to the peak (-4dB)than to the dip (+8dB). That could get taken has having tracked the peak "closer" than tracking the dip, and not being a strict arithmetic mean of the two (is geometric more appropriate? If it's already in dB, I don't believe so, but it's been a while...). Take this to the extreme of a more severe dip on an otherwise same signal (so it offers the same -6dB to normalize) and it will have tracked the peak even closer...

Then, the exact way they combine at the frequency in question (assuming not co-located) as heard at the listening position, will depend on the exact phase difference at that frequency... which assuming is non-trivial, would, I suppose, bring the result closer to your position...

And of course, I have neither the knowledge nor the experience to back up or refudiate Wayne's claim of 6dB being correct for only colocated corner loading, but certainly in the case of two subs designed ot handle different frequency bands, they would certainly not need 6dB to normalize, regardless of their placement... (I'm not saying that's what Paul's doing, but...)

[paulspencer]

Well, gentlemen, some very interesting posts in here! My understanding goes back to high school physics class where we learnt that two different SPL levels were summed based on a formula. Of course the relationship to wavelength wasn't mentioned. I can't see how corner placement would make anything different any more than any location which will have an unpredictable and different impact on modes that are excited.

One thing I noticed about my current setup. The mains are working harder than ever. My quick and dirty omnis have just one midbass in a sealed box and are working harder than previously in open baffle where I had a high pass and worked the subs harder. At the moment I'm mostly interested in experimenting with how to best set up the subs, and how it will influence my sub plans - I've been meaning to get a third sub (already have a spare Rythmik servo plate amp). I've been toying with the idea of some active woofers placed closer to the mains and crossed fairly high. I've also got two out of action AE AV12s collecting dust, a pair of H frame baffles, hmmmm ... many options.

Back to the averaging question. You guys have got me thinking. When I get back to measuring, I'm going to do some tests to determine how it works. This time I'll put the min in the listening position and the subs in place - measure them individually and combined, then compare the combined to the averaging done by REW. We'll see the fit - if they follow the software, or if the peaks get tracked.

One question - how do I get the absolute levels correct? I don't have an SPL meter, but the mic is ECM8000.

[Wayne A. Pflughaupt]

paulspencer wrote:

One question - how do I get the absolute levels correct? I don't have an SPL meter, but the mic is ECM8000.

What?? How are you running REW without an SPL meter??? How did you do the SPL calibration routine?

Regards,
Wayne

[Wayne A. Pflughaupt]

brucek wrote:

Paulspensers graphs of his two subs are the actual listening position measurements.

I assume you’re referring to the graph with the black trace?

Paul did not say it was the measured response of both subs combined. Remember, he stated that he was using REW’s Averaging feature (emphasis below is mine):

paulspencer wrote:

So now I'm getting into this multisub, here's how I've started. First, I put one of the subs in the listening position with the driver at ear level, set crossover to 160 Hz 2nd order LP to see the effect up as high as possible... Now I move the mic around to every thinkable sub position and measure. Then I use the [REW] averaging feature to see how different positions sum together.

brucek wrote:

So for two 75dB signals = 20LOG(10^(75/20)+10^(75/20)) = 81.02dB (which is 6.02dB higher than the initial 75dB). We can turn them both down now and obtain the original 75dB, and we've gained headroom. This would be known as normalizing the signal level.

Now consider that one of those same subs has a dip down to 67dB at 60Hz and the other one has a peak up to 79dB at 60Hz.

Wayne and glaufman contend that the result signal will be the dominant peak level of 79dB.

So for a 67dB and a 79dB signal mixed = 20LOG(10^(67/20)+10^(79/20)) = 80.95dB

Remember that we turned the signal down after the two were combined by 6.02dB to normalize them.

If we do this with the 80.95dB signal, it becomes 74.83dB result.

Good and fine. We saw an increase of 6 dB-SPL so we turned down the level 6 dB to compensate. Makes sense. But...

brucek wrote:

The same math holds true for any measured set of signals… The math can be applied to any specific frequency and will produce the mixed results as accurate as if you actually measured them.

The only problem here is that the math didn’t do the normalizing. That was done when a warm body adjusting a gain control.

I think an important factor has been overlooked here: averaging and summing are not the same thing. Summed response will typically display the highest peaks of all signal sources, will it not? After all, isn’t that what the math shows?

So for a 67dB and a 79dB signal mixed = 20LOG(10^(67/20)+10^(79/20)) = 80.95dB

Yes, 80.95 dB combined. Averaged would be 74 dB...


Hey, don't feel bad - Paul made the same mistake (emphasis mine).

Then I use the [REW] averaging feature to see how different positions sum together.

Keep in mind that I’m mathematically challenged, so be gentle!

Regards,
Wayne

[Wayne A. Pflughaupt]

Okay, I did some in-room experiments yesterday afternoon; here are the results.


Question: Given the presence of two acoustical audio signals with different dB levels, will: (a) the louder one dominate, (b) they both combine for an increased level, or (c) they average out to a level about halfway in between?

Wayne sez: A measuring device will “see” and display only the highest signal and ignore the one at a reduced level.

Let’s see what happens.


First Test
To see if the lesser of two signals would be essentially ignored by a measuring device, I bypassed the crossover on my main L/R speakers and used my receiver’s balance control to reduce the SPL level of the right speaker. Using a broadband pink noise signal, I calibrated the right speaker to a level of 75 dB, then measured SPL of each speaker playing the pink noise in turn, with the other disconnected.

Results:
Left alone: 82 dB.
Right alone: 75 dB.
Both speakers connected: 82 dB.


Second Test:
Same as the first, only with a 75 Hz sine wave. I had some help from “Mother Nature” for this one - didn’t have to offset the balance control, as the right speaker is in a corner and the left is not. I.e., the right speaker will naturally have louder bass response.

Results:
Left alone: 78 dB.
Right alone: 82 dB.
Both speakers connected: 86 dB.


Third Test:
Subwoofer measurements. I don’t have two subs that I can position in different places in the room, but my DIY sub installed in some built-in cabinetry in my living room has two drivers. Measured separately, the drivers exhibit different response readings in the 80 Hz and 100 Hz range. This is probably because one of them - we’ll call it the left one, as that’s where it is when you’re standing in front of the cabinet - has a vertical boundary virtually up against the speaker’s magnet, while the other does not. Apparently there is some internal interaction between that driver and the boundary in close proximity.

This graph shows the separate measurements of the two drivers:

Left (green) and Right (blue) Drivers

As you can see, the graphs are essentially the same, except that the left driver is several dB down in the 70-85 Hz region, and a few dB hotter just above 100 Hz.

The next graph shows the two measurements, along with REW’s averaging calculation.

Left (green) and Right (blue) Drivers, w/ REW Averaging (red)

REW logically splits the difference between the readings of the two frequency ranges in question. Which is exactly what averaging should do.

The next graph shows the left and right drivers once again, along with a reading of the combined output of both. Naturally, this reading registered a higher overall level than the two separate readings, so I shifted the trace down 3 dB so that it would better show how the combined reading tracked across the 80 and 100 Hz regions in question.

Left (green), Right (blue) Drivers, w/ Combined Response

The combined reading essentially ignores the lower level between 70-85 Hz from the left driver, and the lower >100 Hz level from the right driver. Combined response reflects the highest peaking frequencies, no matter which source is responsible for them.

Conclusion:
Well, unfortunately for me, two out of three wins doesn’t exactly permit a completely definitive conclusion that a combined reading will always reflect nothing but the peaks of two or more signal sources. But it’s enough to seriously question the idea that combined readings will always “split the middle” between them, or sum to be greater than the two. I’d be interested in seeing what Paul’s graphs of combined response look like.

Regards,
Wayne

[paulspencer]

Thanks for sharing that Wayne! Very interesting. This would suggest that sub positions should be chosen for the desired peaks. If this is true then there will be no way around eq with the 40 Hz peak I get, as I can't get a lower peak to balance it out in any position. Fortunately because it is fairly universal, the eq used here should work fairly well for all positions except the centre. The centre position I thought was desirable as it had a slight dip where the other had a peak.

I'm not sure I'll get a chance to do any of my own tests until next week.

One interesting issue here is how the signal (pink noise vs sine wave) affects the result. Pink noise is probably more like music and I'd expect it to be more relevant.

[glaufman]

Wayne A. Pflughaupt wrote:

SPL of each speaker playing the pink noise in turn, with the other disconnected.

Results:
Left alone: 82 dB.
Right alone: 75 dB.
Both speakers connected: 82 dB.

Same as the first, only with a 75 Hz sine wave. Results:
Left alone: 78 dB.
Right alone: 82 dB.
Both speakers connected: 86 dB.


Naturally, this reading registered a higher overall level than the two separate readings, so I shifted the trace down 3 dB so that it would better show how the combined reading tracked across the 80 and 100 Hz regions in question.

See, I'm afraid this might be a bit misleading here... You shifted -3dB... where did that number come from? Was it now 3dB hotter with the combination? If you want to show the summing function, leave it alone. If you want to show what happens after normalization (where most of us care) then normalize. But an arbitrary 3dB makes comparison invalid. You can't legitimately say whether it tracked the high peak, or summed to a higher value, or sucked out to a lower value.
At any rate, even if you say this graph is now level matched, it looks like the results varied based on freq, which I presume would indicate the phase difference at the freq: in some regions it added to produce gain (20-50Hz, 59Hz, 64Hz). In some regions it tracked the higher one(70-90Hz, 100-103Hz) . In some regions though, it seemed to be some combination, I won't call it an average, but clearly a dip had a subtractive affect (90-95Hz)... do you have a mode there?

Clearly, to properly talk about the results, you have to choose, either 0 normalization, or complete normalization to the original level (within reasonable tolerance)... anything in between just clouds the facts...

But then again, thanks so much for taking the time to take these...