Another "critque my EQing with MSO" thread

[klettermann]

So, I finally managed to work through Dave Boswell's excellent MSO 2.0 tutorial twice - first time to just listen and second time to follow along and apply his process to my system. My system is 2 ported subs in opposite corners of the room and I'm really just concerned with MLP. My previous attempts at this were with REW but I could never quite figure out how to get phase/delay set up. Certainly part of the problem is that I don't entirely understand it in the first place. Anyway, I did manage to get some nice looking results from MSO. But are they nice really? Any comments and interpretation would be very welcome.

The traces are a bit busy, but here's what they are:

• Straight pink: target level
• Heavy green: Natural response
• Heavy black: MLP
• Heavy Grey: slightly above MLP, so kind of an MLP for a taller person. It's merged with MLP in many places.
• Light red and blue: 2' right and left of MLP.

My main questions are about phase and target level. Is the higher target level a bad idea? And does phase look reasonable? I'd love to see those perfect, overlapped phase traces that you guys all seem to achieve, but I'm not getting that. But maybe it's actually pretty good in the real world? Any other comments? Thanks as always and cheers,

Run 2: target leve raised 5dB from 65 to 70dB.

 

[ucmike11]

Were your measurements all focused around the MLP? Also, did you 1-2 all pass filters per sub?

What does REW show after applying the MSO filters?

 

[kemmler3D]

I think you've got some notches / peaks in the position 3/4 traces that look a little iffy, I wonder if moving one or both of the subs might help... A target level of 65 or 70dB is kind of low if you intend to play louder most of the time, but as far as it goes it looks pretty good. I guess you stopped the EQ above 165hz or so?

The phase is obviously a bit of a wreck, but not having used MSO not sure what the procedure there would be.

 

[Keith_W]

1. It looks as if you are optimising over a 4' area. You can absolutely expect a wide variation when trying to optimise for a wider area like that. MSO tries its best, but it can not overcome physics. I would be pretty happy with that outcome. Those dips that @kemmler3D are high Q and should not be audible. If you want it to look even more perfect, try a smaller optimising area, maybe 2'. It depends on whether you listen alone, or with friends and family.

2. Don't worry about the phase. Those perfect looking phase charts are usually taken at one measurement position. MSO nudges the phase here and there to try to optimise over a larger listening area. Phase does not matter unless (a) it creates a cancellation, or (b) the group delay exceeds audible thresholds. You are in the wrong view to look at the latter, you want the group delay view.

 

[andyc56]

Run 2: target level raised 5dB from 65 to 70dB.

As long as there is a shared sub gain block and its allowed range is not exceeded, the solution should be independent of the reference level chosen in the Optimization Options dialog. Exceptions can occur (such as normal MSO variations and the final value of the shared sub gain block itself. The gain block value is discarded at the end anyway though). What matters is the so-called "preliminary reference level", which is only an intermediate calculation. I need to come up with a better name for it. The preliminary reference level is illustrated below.

The preliminary reference level is also described here. It is not entered into any MSO dialogs, but it matters a lot because it determines the maximum allowable PEQ cut, which is the crux of the biscuit. It's specified on the Sub Input PEQ Limits property page of the Optimization Options dialog. Its default value is very large - too large for most applications in fact, but a large value ensures that even extreme cases of room modes can still be equalized (like the example project, which was deliberately chosen to demonstrate that).

In your case, it looks like the preliminary reference level should be about 67-70 dB, meaning that the maximum allowable PEQ cut should be about 20-23 dB (based on the MLP natural response).

If you change the reference level in the Optimization Options, it will ideally just result in a different value for the shared gain block, not a different response. However, choosing a different preliminary reference level means that you calculate a different maximum allowable shared PEQ attenuation, which must be entered into the Optimization Options, and will result in a different response.

Also, the low-frequency response optimization limit should be the intersection of the line at the preliminary reference level with the MLP natural response, which in your case should be about 24-25 Hz or so. It looks like you've chosen 20 Hz, which is giving you about 10 dB boost at 20 Hz (going from 60 dB to 70 dB). If you're using vented subs, this could overload them, depending on the box tuning frequency.

Finally, it might be worth checking to see if there were any undesired low-pass filters in the signal path when the measurements were done. MSO's sub EQ algorithm was designed under the assumption that in the absence of room modes, the subs exhibit a high-pass characteristic per the Thiele/Small theory. Your data shows some signs of the possible presence of a low-pass filter, as evidenced by roll-off above 100 Hz. With room modes, It's difficult to tell though.

 

[klettermann]

I think you've got some notches / peaks in the position 3/4 traces that look a little iffy, I wonder if moving one or both of the subs might help... A target level of 65 or 70dB is kind of low if you intend to play louder most of the time, but as far as it goes it looks pretty good. I guess you stopped the EQ above 165hz or so?

The phase is obviously a bit of a wreck, but not having used MSO not sure what the procedure there would be.

Thanks, yeah, i cut off EQing at 160Hz cause XOing will cut it off anyway. I'm kinda locked into positioning at this point. Fortunately @Keith_W states below that phase isn't such a big deal. We shall see.... Cheers,

 

[klettermann]

1. It looks as if you are optimising over a 4' area. You can absolutely expect a wide variation when trying to optimise for a wider area like that. MSO tries its best, but it can not overcome physics. I would be pretty happy with that outcome. Those dips that @kemmler3D are high Q and should not be audible. If you want it to look even more perfect, try a smaller optimising area, maybe 2'. It depends on whether you listen alone, or with friends and family.

2. Don't worry about the phase. Those perfect looking phase charts are usually taken at one measurement position. MSO nudges the phase here and there to try to optimise over a larger listening area. Phase does not matter unless (a) it creates a cancellation, or (b) the group delay exceeds audible thresholds. You are in the wrong view to look at the latter, you want the group delay view.

1. Yes, it was about +\-2' either side of MLP, so 4'. If MLP is decent I'm not so concerned about more positions, except that..... POS2 was about 5" directly above MLP for taller sitting. I tend to slouch in the couch.

2. That's good news. As for GD, I don't think MSO it. If it does I'd love to know how. All files got scrambled so I had to start over and I haven't remeasured\evaluated with REW or listened yet.
Thanks and cheers,

 

[klettermann]

As long as there is a shared sub gain block and its allowed range is not exceeded, the solution should be independent of the reference level chosen in the Optimization Options dialog. Exceptions can occur (such as normal MSO variations and the final value of the shared sub gain block itself. The gain block value is discarded at the end anyway though). What matters is the so-called "preliminary reference level", which is only an intermediate calculation. I need to come up with a better name for it. The preliminary reference level is illustrated below.

View attachment 437789

The preliminary reference level is also described here. It is not entered into any MSO dialogs, but it matters a lot because it determines the maximum allowable PEQ cut, which is the crux of the biscuit. It's specified on the Sub Input PEQ Limits property page of the Optimization Options dialog. Its default value is very large - too large for most applications in fact, but a large value ensures that even extreme cases of room modes can still be equalized (like the example project, which was deliberately chosen to demonstrate that).

In your case, it looks like the preliminary reference level should be about 67-70 dB, meaning that the maximum allowable PEQ cut should be about 20-23 dB (based on the MLP natural response).

If you change the reference level in the Optimization Options, it will ideally just result in a different value for the shared gain block, not a different response. However, choosing a different preliminary reference level means that you calculate a different maximum allowable shared PEQ attenuation, which must be entered into the Optimization Options, and will result in a different response.

Also, the low-frequency response optimization limit should be the intersection of the line at the preliminary reference level with the MLP natural response, which in your case should be about 24-25 Hz or so. It looks like you've chosen 20 Hz, which is giving you about 10 dB boost at 20 Hz (going from 60 dB to 70 dB). If you're using vented subs, this could overload them, depending on the box tuning frequency.

Finally, it might be worth checking to see if there were any undesired low-pass filters in the signal path when the measurements were done. MSO's sub EQ algorithm was designed under the assumption that in the absence of room modes, the subs exhibit a high-pass characteristic per the Thiele/Small theory. Your data shows some signs of the possible presence of a low-pass filter, as evidenced by roll-off above 100 Hz. With room modes, It's difficult to tell though.

Very very helpful, more grist for the mill . There wasn't any LP or HP prefiltering, so no concerns there.

I did indeed allow a little boost at 20hz and I do have ported subs. In other experiments I've found that the subs can handle it but I'll take a closer look. All the garbage above about 100hz will presumably go away after XOing and integrating the mains. So i guess I'm making progress. Thanks and cheers,

 

[andyc56]

Don't worry about the phase. Those perfect looking phase charts are usually taken at one measurement position. MSO nudges the phase here and there to try to optimise over a larger listening area. Phase does not matter unless (a) it creates a cancellation, or (b) the group delay exceeds audible thresholds. You are in the wrong view to look at the latter, you want the group delay view.

It's worth clarifying the different phase responses and what they affect. I've made a block diagram of two subs at a single listening position to illustrate.

The input on the left is an analog voltage going into a combination of shared and per-sub filters. The measurements you take correspond to the combination of the Thiele/Small response of the subs and the room transfer function, with the shared and per-sub EQ set flat before measuring. At the output, there are two acoustic pressures going into a summer. These two quantities are plotted in MSO as "Filtered Measurements" in the Graph Properties dialog. SPL is maximized in part by keeping these two acoustic pressures as close as possible to being in phase with one another. It gets messy when there are multiple positions. MSO deals with that by trying to maximize the average SPL over position at each frequency, so it ends up in effect splitting the phase matching errors over position.

OTOH, when speaking of group delay as above, we're interested in the output of the summer, the combined pressure at the listening position taking into account the phase relationships at the input of the summer. This is plotted in MSO as a "Measurement Group". That's bad naming, but MSO was originally created to do more general things than what people mostly use it for nowadays. It is the group delay of this combined response that may be of interest.

In real-world scenarios, there are of course multiple listening positions and often more than two subs, but a block diagram of that becomes very messy. I've kept it to two subs and one listening position above for clear illustration.