Rational Acoustics



merlijnv
January 23rd, 2010, 12:59 PM
How reliable is trace averaging, or maybe I should say “representative”? I ask this, because Bob McCarthy wrote an entire paragraph about the subject in his book. About the dangers of mathematical averaging in relation to the difference in magnitude between peaks and dips as a result of comb filtering. The latter often seen in combination with low but stable coherence. Is coherence weighted trace averaging a better solution? McCarthy proposes optical averaging, saying that the peaks and survivors create the audible character of the response, the envelope. To complicate things even more. I’d like to add that psychoacoustic experiments have proven that, most people (10%) aren’t able to hear dips or cuts of 10 dB and less than half (40%) dips or cuts up to 20 dB. Whereas everyone (100%) hears peaks or boosts of 10 dB.

Kind regards,

Merlijn

Ferrit37
January 23rd, 2010, 03:50 PM
Hey Merlinjnv,
try this... http://www.rationalacoustics.com/forums/showthread.php?t=176


But I still use "eyeball-averaging" one extra refinement is to take my glasses off for some extra smoothing :)

merlijnv
January 23rd, 2010, 08:01 PM
Hey Merlinjnv,
try this... http://www.rationalacoustics.com/forums/showthread.php?t=176


But I still use "eyeball-averaging" one extra refinement is to take my glasses off for some extra smoothing :)

Hi Ferrit,

Thank you for your reply. I've read the specific thread. But I’m still wondering about the implications of regular and coherence weighted averaging in regard to McCarthy’s “arguably” envelope. IMHO his theory makes sense to me. I’m curious about your experiences.

Kind regards,

Merlijn

Ferrit37
January 23rd, 2010, 09:40 PM
Hey merlinj,
I've only played around with trace averaging a few times and it always seemed to end up looking like the worst of the bunch, although coherence averaging was superior.
I've also used the "weighting" technique by having two measures of say FOH position.
I much prefer to see the differences between positions rather than looking for the similarities. I think that applying EQ decisions to an overall average is a nice idea but in practice not realistic.
I do like the idea of seeing multiple traces and seeing what is happening throught the coverage zone as you effect changes, trying to do the least harm.
I recently made multiple measurements of a 12 element "Mutli-cellular array" where we had seperate control of each individual driver (six per element) allowing us to produce a sound field with a +/- 2dB variation from a central reference point at 35m from 16m out to 75m. cool stuff :)

Ferrit37
January 24th, 2010, 01:02 PM
Hey Pepe,
At Martin-Audio we have developed a computer optomisation program that uses a hybrid model based on complex directional point sources and boundary diffraction loading.
This allows us to set targets for various factors such as, in coverage smoothness, SPL variations and out of coverage leakage, as well as compensate for air losses over distance.
First the optimal array curvature and height is calculated using a recursive algorithm and
then coefficents for FIR filters for each driver in the array are calculated and down loaded to the DSP for each driver. A further set of allpass filters are also applied to produce the final result.

To reverse engineer this from measurements first would take quite a long time. my role was to confirm that the process was working properly. :)

Harry Brill Jr.
January 25th, 2010, 01:38 PM
Isn't it great that we can do it either way?

Hey Ferrit, haven't they changed that company name to Martyn Audio yet?


Hey merlinj,
I've only played around with trace averaging a few times and it always seemed to end up looking like the worst of the bunch, although coherence averaging was superior.
I've also used the "weighting" technique by having two measures of say FOH position.
I much prefer to see the differences between positions rather than looking for the similarities. I think that applying EQ decisions to an overall average is a nice idea but in practice not realistic.
I do like the idea of seeing multiple traces and seeing what is happening throught the coverage zone as you effect changes, trying to do the least harm.
I recently made multiple measurements of a 12 element "Mutli-cellular array" where we had seperate control of each individual driver (six per element) allowing us to produce a sound field with a +/- 2dB variation from a central reference point at 35m from 16m out to 75m. cool stuff :)

Ferrit37
January 25th, 2010, 02:33 PM
Hey Harry,
No, but I answered the phone "Martyn from Martin-Audio" and the other person said "I thought you were dead!......"

Jamie
January 28th, 2010, 07:16 PM
With all due respect to 6o6, the coverage of spatial averaging techniques in his book (1st ed.)is specifically slanted towards his desired conclusion in favor of "optical" or "eyeball" averaging. The discussion makes many assumptions concerning equipment, mathematical technique, and most importantly, user capability and involvement, that are pretty misleading. Or rather, they are chosen to lead to his foregone conclusion.

Why do I believe this?

To be honest, when I first started with Smaart, I was deeply entrenched in refusing to use this feature as well. My arguments fell along exactly the same lines as 6o6's (wonder why), although not as well and systematically stated. It wasn't until Don Pearson cajoled me into showing Smaart's trace averaging feature to students in classes that a funny thing happened. Once I agreed, I set out to specifically show why he was fooling himself in using this feature - basically to illustrate the cautionary tale about mathematical trace averaging. What happened however was interesting. In virtually all cases, the mathematical average matched my eyeball average very closely - and the resulting trace was actually USEFUL. In fact, I really had to bend over backwards to come up with a clear example of it's misleading nature. In short, I had to do things I knew were dumb in the first place to produce the obvious failure I needed for my example.

Here is what I learned:
1) If you are going to average traces (eyeball or mathematically), you need to use valid measurements - you know, ones made properly from useful locations.

2) With any type of data processing, there are many types of mathematical approaches that can be employed to your advantage - coherence weighted avg, dB vs power averaging, vector vs. rms averaging, etc. As one of our past programmers (Hai - a PHD in math) once lectured me, statistical analysis and pattern recognition are not new fields, and there are a lot of tools out there to help us.

3) And most importantly - this is not a zero-sum game - an either / or choice. DUH! When you mathematically average traces, you usually start by viewing all the raw traces together first. In other words, you begin with the eyeball avg anyways. A good engineer uses all the information at hand when making a decision. (Do I use an analyzer OR my ears? Huh? How about both)

In effect, there is an underlying assumption in 6o6's discussion of trace averaging that the engineer is blindly (deafly?) following/accepting the resulting trace as gospel - instead of putting it in context.

So back to your original question: "How reliable is trace averaging."
(I would suggest, that the question should be "useful" instead of "reliable")

My answer would be - Very
. . . for determining an average spatial response for setting EQs. (Assuming your individual measurements are good/useful)

I say this from 17 yrs of experience, with 8 of which in which I have often used the technique - and I would be very surprised if 6o6 has ever seriously, neutrally employed/evaluated it once.

But don't believe me - go and make some measurements yourself, and compare the two. When v7 hits soon, you will be able to do this very easily.


And of course, one final note. I feel that 6o6's book is an essential read for system engineers. I happen to take exception with this specific section

Jamie
January 29th, 2010, 01:12 PM
To add something useful to my post . . .

The reason we added coherence weighted averaging was specifically to create an average that is biased towards the peaks (which typically have higher coherence), and therefore produce a curve that more closely follows the upper envelop of the multiple traces.

The root of this discussion however still must start with derivation of the individual traces you are averaging (either eyeball or mathematically). Assuming that you have a valid measurement, there is still the issue of the basic TF Mag averaging type that you were using in the first place. Smaart by default uses an RMS (Power) average for it's Mag trace (Phase is always calculated from a complex/vector average). The effect of using an RMS average is to allow more of the reverberent energy into the trace, which many users feel produces a TF Mag trace that more closely reflects the listeners' experience of system tonality. By contrast, using a Vector average for Mag rejects more of the reverberent energy, and many users feel that this produces a trace that more closely follows a listener's perception for intelligibility.

So?

Well, in the case of averaging vector averaged magnitude traces, you will definitely want to be more aware of the upper envelope curve, whereas in the case of averaging RMS averaged Mag traces, you will probably be more comfortable using the basic mathematical average.

IMO . . .

In the end however, the effects/differences between your various averaging decisions is most often very subtle - and if you factor a non-zombie engineer into the process - won't be the dominant factor in creating major alignment variations.

Wahoo
-j

adam
January 31st, 2010, 05:42 PM
With v6 that is the process. With v7 you can run multiple live measurements, at the same time, and have as many live averages as you desired, using the live measurements.

adam
February 14th, 2010, 05:09 PM
Hi Adam,

I've been doing during smaart in Barcelona, and James Woods and Jamie Anderson, have presented us Smaart v7.

Is starting a new era.:)

Pepe.

Hi Pepe,

Glad to hear you liked what you saw. We're quite proud of v7 and can't wait to release it to the industry.