I usually prefer rectangular for precision in higher frequencies, but I've found Hann to be more forgiving of sample rate mismatches and other issues
I usually prefer rectangular for precision in higher frequencies, but I've found Hann to be more forgiving of sample rate mismatches and other issues
Yes, thank you Amir. I appreciate your input and any recommendation on how to improve the method. I would like to get something like a listening window response.In the first video, the response you captured for the center speaker showed a hole in the middle which means it is capturing average of its response across certain angle. As such, it was not the on-axis response near field response.
Maybe the width of the largest surface would be a good starting point. I'm still working on a method that works on as many systems as possible. I thought I had a good method, but the more testing we do on various systems in our private beta, the more we have to make sure it works in more scenarios. Let me know if you have any recommendations.How would you determine baffle width for a large horn system?
LOL, I was hoping you could make sense of it for me!I'm having a rough time deciphering how this relates to the MB method specifically. I'm all ears though if you're willing to walk me through it a bit.
Ok. Let me answer the easy stuff first. This is information from smarter people than me. REW's trace arithmetic is based on voltages. The other method is calculating gain by decibels. Dividing in REW is the same as subtraction when talking about decibels. So to use my method in REW, you'll want to use division.LOL, I was hoping you could make sense of it for me!
Brainstorming aloud:
1) I believe you when you say that MBM seems to work better than limiting to transition frequency or throwing the standard full Harman curve with Audyssey and Dirac. This wasn't just a one-week experiment and you've shared a lot more data.
2) It is possible that the end result/method is correct even if the rationale or why is incorrect.
One thing that I am still confused about is whether the MBM mathematically divides or subtracts the nearfield from MP. If your near and far field are identical, then dividing equals 1 (so your target is flat) while subtracting is 0 (so you are making no EQ changes). Both conventions work to describe the same overall change in different ways, but subtraction and division will have different magnitudes of difference of correction.
Here's my original post, which you should open in another window to follow the images as I write my explanations.
If we take the Meyer Sound Amie as “close enough to an ideal speaker” (and it’s the only one with free software to play with) then we have the following beliefs:
1) Since it’s good anechoic, you should need zero correction in an anechoic home theater. There's no room to correct and we like the speaker.
For the anechoic simulation, the center channel is perfect. No difference from center nearfield and center MLP so you need zero EQ. Great. No surprise.
The left channel would be toed in for a single seat cinema, but it might be straight in a large family room with a wide sofa. In this simulation, we have a wobble in the high frequencies in the anechoic response.
Option A is that this is a true defect in the audio and if I had a JBL M2, it'd also look flat but the Amie isn't perfect.
That's what I showed in the first post which doesn't work.
Option B is that, like the JBL M2, what is measured as this location in a sweep is not reflective of what we hear in room, so you should leave this wobble instead of correcting it.
If your nearfield is attempting to replicate the on-axis anechoic measurement, when you subtract it from the MLP, the wiggle is something to be corrected.
If your nearfield is reproducing the same "window" that Amir pointed out in your center channel measurement then your nearfield measurement matches the farfield (in this anechoic simulation).
View attachment 338146View attachment 338148
If the "True Target" = "Zero Correction" in an anechoic home theater, then you need to have a nearfield that more closely approximates the actual axis that you're listening on. So whether the Amie should or should not have the wiggle is not the question. We need to have no "room" correction in an anechoic room.
Therefore, your center channel measurement may be wrong for getting the on-axis anechoic response but right for the way you're listening to it.
2) Now we turn on room simulation.
When looking at the center channel, in this arbitrarily chosen room, the room effects happen around 550 Hz and lower. This sort of fits the idea of just correcting below the transition frequency. But when we look at the left channel, the room effects go all the way up to 1.5 kHz or even 4 kHz depending how you look at it.
Since those ripples don’t exist in the anechoic simulation, in theory, you would want your room correction software to go all the way up to 4 kHz. But the ripple between 8-16 kHz should be ignored. That's not from the room.
For MBM to work, we need the nearfield in room to reduce the room effect without reducing the speaker. So again, instead of being on-axis with the Left speaker, I just measure closer along the same vector to the MLP
Nearfield
View attachment 338150View attachment 338152
Farfield
View attachment 338153View attachment 338154
This is also telling me that I should correct the Left channel all the way to that 4 kHz range and if I could really measure the nearfield closer than 1m, it would only get better.
My speculation on why MBM may really work and how it fits with Amir and Dr. Toole's observations thus far
All of this is predicated on Meyer Sound's MAPP XT being accurate. It's supposed to be validated using both math and real-measurements. Their original paper from 2000 has some of the math, and their marketing boasts about anyone being able to independently verify the predictions with on-site measurements (and that they have also done that type of verification).
If the MAPP XT predictions are true, then a 5 x 6 meter room generates room effects as high as 4 kHz with the speaker layout I chose and wall profiles I chose to simulate.
a) If I limit my correction to the bass (<400-500Hz), I get a pretty good room correction. It's a safe choice.
b) If I go full range to an off-the-shelf target curve, it would correct the problems all the way into the 4 kHz range, but incorrectly try to correct a wiggle in the ~10 kHz range that is present anechoically and may cause colorations if it was corrected. With the Meyer Amie, the on-axis is relatively flat, so correcting with an off-the-shelf target won't introduce too many errors, but if your own speaker is irregular, you run the risk of introducing colorations, which is what Dr. Toole has written about.
c) A perfect room correction tool removes the room, and if compare the difference between anechoic and in-room, there are wiggles in the frequency response going up into the 4 kHz range which should be corrected.
So how do we figure out we need to go up to 4 kHz if you don't have a Meyer Sound speaker? Maybe that's what MBM is doing. It lets you figure out what wiggles in the frequency response above the transition frequency to leave in and what to correct.
If this brainstorming actually makes sense, then the trick is NOT to try to get a nearfield on-axis measurement but simply try to get a nearfield measurement representing the general angle of audio that you're listening to, that is just far enough to integrate the drivers but reflects the degree of toe-in or toe-out you have. The less accurate you are (the less nearfield you are), the less of the room you can detect but the less of the correction you get above the transition frequency. That is, as you lose accuracy of your nearfield measurement, the consequence is that you end up correcting less of the region above the transition frequency as long as you're getting a good representation of the axis of listening.
The wildcard is now in Audyssey and Dirac. There will always be measurement errors -- so they won't correct to +/- 0.01 dB. They inherently have some tolerance in their computations. For this to work, that tolerance must be tighter than the room effects above the transition frequency which maybe only be 1-2 dB or so?
So what you need your beta testers to try is to review
a) who is toe'ing in and out -- is it perfectly to the MLP or just a little bit less so that more people on the sofa can enjoy the sound?
b) how does the nearfield change when you're trying to do that 1x the baffle measurement versus following the vector from MLP to speaker.
If your subjective experiences are that you still get better sound with your "true target", you can rename Magic Beans to being a tool to helping you correct above the transition frequency without causing too much damage
I guess the other question is that when you say MBM sounds "better" what do you mean? Do sound effects travel more seamlessly across the room since the timbre isn't messed with? Does it sound more natural because the coloration that you normally add with an off-the-shelf full range target is reduced?
This is awesome of you to take the time to try this out. I do think you should use pink noise. I prefer 16K periodic pink noise myself.I had some time to experiment this evening, so I spent a few hours trying this method. I used Acourate, so the method posted below is for other Acourate users.
Method (for Acourate).
1. Copy all previously created and linearized crossovers into a new directory. Create a multiway filter.
2. Do 2-3 MMM's of the same speaker with the multiway filter loaded to verify that your technique is correct and the result is repeatable. If all is OK we move on to actual measurement. I used white noise instead of pink noise (discussion below)
3. Perform a nearfield MMM of the left and right speakers. I saved these as NF-MMLeft/Right.
4. Perform a MLP MMM of the left and right speakers. I saved these as MLP-MMLeft/Right.
5. Go to FD-Functions - Magnitude Difference and calculate the difference between MLP-MMLeft/Right and NF-MMLeft/Right. The purpose of doing this is to obtain the "room transfer function". We will set this as our target curve. I saved these as Target-Left/Right. This was what my target curve (prior to inversion) looked like:
View attachment 338185
6. Perform a single point sweep at the MLP. Go through Macro 1 ("Amplitude Preparation") as normal.
7. Skip Macro 2 ("Target Curve Design"). You have already designed the target curve.
8. Run Macro 3 with these settings - remember to UNCHECK "use default target" and enter your own target into the dialog box. Make it do a MONO correction. Repeat this step for the right channel (i.e. Macro 3 is run twice).
9. Run Macro 4 ("Filter Generation") and the other Macros as per normal. From here we can proceed to verification measurements.
10. Create a Multiway filter for verification, then run a nearfield MMM:
I was pleased when I saw this result. The correction is working as it should, it is supposed to flatten the speaker's nearfield response, and it does that pretty well. Except for >15kHz. I am not sure what happened there. I did see the massive rise in bass, but this was below the Schroder frequency so I ignored it. It should correct itself in the MLP. Except that there was one concerning feature here - this is a nearfield measurement of the bass. Why does it look so smooth? I would expect massive peaks and dips because the bass correction is supposed to work at the MLP and not nearfield. So let's check the MLP MMM:
Here is the MLP measurement (unsmoothed). Some comments:
- the bass does not correct itself as expected. Still the same rising response. The fact that the output is so high at 10Hz makes me suspect this is a measurement issue rather than actual bass output. My subs are incapable of putting out that much bass energy at 10Hz, equalized or not! And we can see the target curve which I posted (prior to inversion) does not instruct Acourate to massively boost the bass. I will have to think hard about this one.
- the mid freqs (100Hz - 1kHz) are very slightly scalloped out compared to treble.
- Treble freqs > 1kHz does not exhibit the falling Harman-like curve that I would expect.
This is a before (green) and after (red) MMM of the left channel at MLP with smoothing. The "before" curve only has driver corrections, time alignment, and level matching between drivers. Ignore the gain difference, I deliberately staggered the gain to separate the curves:
View attachment 338197
We can see that this technique has mostly smoothed out the wiggles in the FR but it has made the bass even more pronounced.
Listening. I was surprised to find that the listening impression does not match what I would expect from what I see from measurements. From a curve like that, I would expect massively overboosted bass with recessed mids and a thin, shrill, shrieky top end. That is not what I hear at all. The bass is tubby (to be expected) but it is not overwhelming like the curve would suggest. The mids do sound a little scalloped out, but it's not too bad. And the top end seems "polite", in fact it is too polite and the sound seems to lack detail ... as if there wasn't enough top end! It is almost as if my measurement is lying to me - my ears say one thing, the measurement says another.
Issues.
- I am convinced that the measurement/listening discrepancy was because I was using white noise instead of pink noise. I knew I was supposed to use pink noise, but Acourate gives three options for pink noise - 10Hz, 100Hz, and 1000Hz. I wasn't sure which one to choose, so I simply went with white noise.
- I think the bass freqs look like that because of my measurement technique. For some reason it produces a rising bass response. I have other measurements from a single point sweep (not shown) that shows a more accurate picture of what is going on with the bass. Now I have to figure out what I did wrong.
This is a preliminary attempt. I still have a lot of work to do, but I can report that the technique seems to work.
The "true target" is to make the speaker's NF response as flat as reasonably possible while also fixing the bass at the MLP. We also want to transition smoothly between the two types of corrections.
Your first idea should work. It will get you close. The other part of MB is that it determines the bass rise you should have based on the room response curve and it will help you set your sub gain so it matches the room curve.Hmmm. For my HTP-1, I have Dirac and PEQ and I can run the PEQ before or after Dirac.
So, if I measured NF and then used PEQ to flatten the response and then with this PEQ enabled then measured at my MLP and let Dirac handle the bass alone, I’d get close to the result of MBM?
Alternatively, if you had $2000 worth of MiniDSPs in between a processor and amp, and you corrected NF to flat (within reason) and then used regular Audyssey or Dirac up to the transition frequency, that would be also be similar?
The beauty of Magic Beans is that the “PEQ plus room EQ to bass” approach cannot correct the phase response above the bass and Magic Beans is really the only way you could do it. Also, Magic Beans works with AVRs if you’re not running separates, and it will clearly be a lot cheaper than $2000 of MiniDSP PEQ even though you haven’t finalized the pricing yet
You can flatten NF with Dirac by loading a perfectly flat target curve in Dirac, measuring NF, inverting the result of the measurement, and using the inverted result as the target curve. I’ve done this above transition, and the result is almost perfectly flat. I have not done it full range using MMM.So, if I measured NF and then used PEQ to flatten the response and then with this PEQ enabled then measured at my MLP and let Dirac handle the bass alone, I’d get close to the result of MBM?
This is something that continues to confuse me whenever MB is discussed - in any non MB discussion, the inverted curve is the correction curve, not a target curve - the flat target is the target curve. I get so confused whenever there is MB target discussion.You can flatten NF with Dirac by loading a perfectly flat target curve in Dirac, measuring NF, inverting the result of the measurement, and using the inverted result as the target curve. I’ve done this above transition, and the result is almost perfectly flat. I have not done it full range using MMM.
In MB, that correction curve is created by the MLP measurement conforming to the “true target” curve that it determines by looking at the difference between the MLP and NF measured response. You can say it another way that a correction curve to flat NF and a correction curve to smooth out the bass response at MLP creates the target curve. A+B=C, B=C-A, A=C-B. All say the same thing. Software like Dirac ask for a target curve, so it makes more sense to talk in those terms. A MiniDSP and MultEQ-X (with all measurements excluded) require a correction curve (or correction PEQ filters) to achieve the target response. The correction is what you do to the signal. The target response curve is the result of the correction. The reason it’s confusing is that many people incorrectly think that there’s an arbitrary target curve that everyone should target as a measured response at MLP regardless of the speaker, room, and MLP.This is something that continues to confuse me whenever MB is discussed - in any non MB discussion, the inverted curve is the correction curve, not a target curve - the flat target is the target curve. I get so confused whenever there is MB target discussion.
I’m in the same page with the terminology here. @GXAlan mentioned measuring to get flat NF. I was sharing a relatively easy way to do that with Dirac.This is something that continues to confuse me whenever MB is discussed - in any non MB discussion, the inverted curve is the correction curve, not a target curve - the flat target is the target curve. I get so confused whenever there is MB target discussion.
Yes, what he applied is a correction curve to achieve a flat target response near fieldI get that you're calling what you do something else. I'm just never sure I understand if what you're saying and what other people are saying are on the same page. For example, johnp said he used an inverted result as his target curve - I think we're in agreement that this is a correction curve, right?
Don't feel bad about getting confused. We had a whole long conversation about correction curves vs. target curves during the MB testing. That conversation actually resulted in a way to use the more accurate correction curves in MultEQ-X rather than a series of PEQs, with both ways excluding Audyssey's measurements so that MultEQ-X basically acts like a dumb filter bank (whereas if you're using Audyssey's measurements, it needs a target curve so it can create its own corrections to meet it).This is something that continues to confuse me whenever MB is discussed - in any non MB discussion, the inverted curve is the correction curve, not a target curve - the flat target is the target curve. I get so confused whenever there is MB target discussion.
I'm excited to see and hear about your results. I guess that shock mount I use really helps. I don't really pick up much handling noise unless I bump something with my arm or the cable hits something.Well, I figured out what's causing the rising bass response in my MMM's. It is the speed which I am whooshing the microphone around.
View attachment 338445
This is an MMM taken in the middle of the room with the speakers muted. No noise at all except for the noise floor of the room. Red = very slow MMM movement. Green = very fast.
I am using an Earthworks M30 mic on a microphone tripod. I pick the whole tripod up and wave it around, trying not to let the cable clang on the tripod. I am not using a microphone isolator that @joentell shows in his videos.
View attachment 338450
Compare the appearance of a slow moving MMM (red) with fast (brown). The other difference is that I increased the speaker loudness + reduced mic sensitivity to get further away from the room noise floor. I was sailing so close to the wind that I had to keep an eye to make sure I wasn't clipping the measurement with the red response. I am pleased to see that it is so repeatable. But what it really shows is if you move your mic quickly, it produces a rising bass response. And when we are talking "slow" mic movement, I am moving it at a glacial speed ... VERRRRRRRRY slow.
Lesson: MMM technique matters. Move your mic slowly!
Interesting. I would be interested in seeing the NF MMM response with the Dirac curve on, and another one with MB. I am also curious what your subjective listening experience is like using both the MB target response curve and the default one Dirac provides.I find this thread fascinating.
Here are some measurements taken in my lounge comparing the Magic Beans (MB) to Dirac Live (DL) measurements, targets and filters.
View attachment 338536
It is remarkable how close today's MLP MMM is to an old SPECTRUM measurement by DL (imported into REW via VituixCad SPL trace tool).
The peak at 50hz seems to be new - possibly a ground loop issue from using REW via HDMI?
Here are the MMM measurements and the Room transfer "target" that they produce (REW A/B).
View attachment 338537
View attachment 338539
Now we can look at the filter that MB should produce (Target / MMM MLP). I have compared it to the actual measured DL filter (MMM with DL on / MMM DL off).
View attachment 338540
I am struck by how similar the MB implied filter and the DL measured filter is other than the big <60hz bass boost that MB gives. I find that the closer NF MMM is measured, the bigger the bass boost. Moving from c. baffle width distance to c. 50cm out creates a 5db step down in the NF MMM <100Hz, transitioning from <300hz. That brings the MB filter back in line with the DL one in the bass region.
View attachment 338541
View attachment 338542
Again, I would ignore the 50 hz peak which wasn't in the original DL measurements. Note that where DL appears not to be corrected as claimed at MLP (140hz, 375hz), the NF MMM looks overcorrected.