Rational Acoustics



Jamie
June 6th, 2008, 01:17 PM
Every now and then, I get a request for the text of this old post from the Smaart forum. Seening as we have much to discuss regarding phase, I figured I post this to begin stirring the pot, and this one certainly did . . .

From: Jamie
E-mail: Jamie@SIASoft.com (no longer active - now Jamie at RationalAcoustics dot com)
Date: 13 Oct 2000

Comments

Dave, (I don't remember who the "Dave" was)

Here is a quick way to look at these two types of crossover time alignment (best impulse response vs best crossover response). I have posted a Smaart .rgp file of my data (crossalign.rgp (http://support.siasoft.com/Downloads/binary/Misc/crossalign.rgp)) for those of you playing along at home.

Start with a simple mid-high crossover. Add the same ammount of delay to both drivers (in the case of the data I am posting, I used 50 ms). Start with just the high driver on. Measure the delay, set Smaart's measurement delay, and make a Transfer Function (TF) measurement. The high driver's phase response should look like a smile with the majority of it's range having a relatively flat phase response. (Ref trace A1 in my data) Notice that the phase trace has a downward (read left to right) angle at crossover. This downward angle is mostly due to the crossover filters and is normal/standard in most analog and digital crossovers. This downward angle also indicates that the energy in the crossover range is lagging behind the rest of the driver's energy. (Don't believe me? Add some delay to Smaart's measurement delay setting - you will see that down-angle part of the phase trace level out as the rest of the trace begins to take on a significant up-angle - which indicates leading energy.)

Save the TF trace for your high driver.

NOTE: For the rest of the measurements, DO NOT CHANGE THE SMAART DELAY. We will be looking at phase/timing relationships so we want to keep the same time reference for all of our subsequent measurements.

Now turn off the High driver and turn on the low driver. Vary the delay on the low driver and watch it's effect on the low driver's TF phase reponse. Start by taking away delay (this is why I had you add extra delay to both drivers - its hard to do negative delay) until the low driver's phase response is basically flat with much the same smile shape as the high driver. Adjust the delay as necessary so that the two phase traces cross at crossover while trying to keep as much of the low driver's phase response "flat" (this is shown as trace B1 in my data).

Now combine the two drivers and note what happens at crossover. (this is shown as trace C1 in my data) You should see addition right at crossover but not so much above and below. This makes sense because the phase traces are diverging above and below crossover. In general, the speaker's phase response will be relatively flat over it's pass band.

Now, mute the high driver again, bring up the stored high driver trace, and now add delay to the low driver until it's phase trace OVERLAPS the high driver's phase trace THROUGH CROSSOVER. (This I stored as trace D1). What you have just done is time aligned the two drivers through crossover.

---------------

Well, that's the entirety of the post. I will follow this with a PowerPoint on the same subject. If the link to the data is bad (old SIA support site), I have also attached the file.

discuss,

-j

Jamie
June 6th, 2008, 01:46 PM
This is a PowerPoint that follows sorta along the same lines as the "To Jamie RE:Phase" post above. This was done to illustrate the same practice of reading timing in the phase trace at crossover - this time with cleaner data since it was created with purely electronic measurements.

It was also done as a cautionary tale to people who try to "Time-A-Line" the drivers in their speakers using the peak of the Impulse Response or the auto-delay locate function.

Again, this is stuff that's been around for a while - I 've been using in my class for years.

This is NOT a recommendation to re-time your crossovers - at least, not on site of a gig or install. It is a phun thing to play with/listen to in the comfort of your own lab/shop/home. (Although it does come in handy when timing subs to main crossovers)

Proper crossover setting requires a detailed knowledge of each driver's power handling and the speaker's HF and LF polar responses (among other factors). Even though you may come up with something that sounds gud on axis at moderate drive levels, playing with this stuff could quite possibly screw up the speaker's off axis response and possibly the power handling (particularly if you shift x-over) - and possibly damage your speaker:eek:. The reason you buy high Kwality speakers from reputable manufacturers is that they have the engineering resources to make all of the necessary measurements to make the proper x-over settings decisions.

Ok, hopefully that covers us for warnings and liability purposes.

It is interesting to listen to though . . .

-j

Harry Brill Jr.
June 6th, 2008, 02:57 PM
I made a pdf a couple months ago.

Harry Brill Jr.
June 6th, 2008, 03:06 PM
I'm kind of leaning toward crossover alignment then using allpass filters to flatten out the phase of the entire box.

Arthur Skudra
June 6th, 2008, 05:13 PM
What's everyone's thoughts of Charlie Hughes' recent article in LSI about using group delay for crossover alignment?

Harry Brill Jr.
June 6th, 2008, 08:21 PM
If that's the June issue I haven't received my electronic copy yet.

Arthur Skudra
June 7th, 2008, 11:22 AM
Hey Harry,

The article appeared early this year in LSI. I'll see if I can dig up the exact issue amongst the piles in my office. I can't log into the digital archives for some reason.

Arthur

ewanmcdonald
June 8th, 2008, 04:58 AM
Hi There,
If its the using group delay rather than impulse for subwoofer alignment article- it was in the April 2008 edition of LSI and Volume 35 of the Synaudcon newsletter last year. Hope that helps
Cheers
Ewan

Arthur Skudra
June 8th, 2008, 09:32 AM
Yes, that's the one! Thanks Ewan!

I did try following Charlie's group delay technique for aligning the subs on a system that I was optimizing, but being under some time constraints and not having any luck using this technique for the alignment of some subs, I gave up and went back to the alignment technique using phase response. Maybe I was doing something wrong, or perhaps the venue/system was not cooperating/unstable? I dunno, and since haven't had the chance to revisit this. Just wondering if anyone else had experience in using group delay for the alignment of a sound system, particularly the subs to the mains?

One clue that a good friend mentioned to me in passing about another measurement issue is that the pink noise stimulus that is digitally generated through software or several of the popular portable noise generators out there, does not have sufficient "randomness" in the low frequencies considering the period of the LF frequencies (hence the popularity of using swept sine waves for improved stimulus of the system under test). Before thinking I'm heretical in my thoughts here, I have yet to test this hypothesis out, but have access to an older vacuum tube/diode based pink noise generator that supposedly offers an improved "randomness" in the LF. Maybe in a couple of weeks I can report back when I have had a chance to test it out.

ewanmcdonald
June 8th, 2008, 11:24 AM
Hey Arthur,
I mucked around with this when I read the paper back last year. I really thought the theory made perfect sense so thought I would give it a go... Just tried it in our factory though, never on a gig. I think it was with some Adamson Y10s and T21s . I could never get any real usable group delay trace. I would have to set the averages to inf to get anything remotely usable (otherwise it would just jump around). There were some time though when I would get things to line up and it worked a treat but the results weren't at all consistent for different mic positions. Didn't seem like anything I could use on a gig. I would go back to the phase method then switch to group delay and see if they lined up. Sometimes yes, sometimes no. I had always been told that group delay didn't really work out in the field. Im not too sure if winmls or some other swept sine or MLS measurement system that does group delay would give more consistent results... Im running Smaart 6 on a mac by the way. Thats interesting about pink at low frequencies- i never new that. Id be real keen to hear of anyone getting this to work too, It is MORE THAN LIKELY I was just doing something wrong!!!
Cheers
Ewan

Harry Brill Jr.
June 8th, 2008, 11:26 AM
Yes, that's the one! Thanks Ewan!

I did try following Charlie's group delay technique for aligning the subs on a system that I was optimizing, but being under some time constraints and not having any luck using this technique for the alignment of some subs, I gave up and went back to the alignment technique using phase response. Maybe I was doing something wrong, or perhaps the venue/system was not cooperating/unstable? I dunno, and since haven't had the chance to revisit this. Just wondering if anyone else had experience in using group delay for the alignment of a sound system, particularly the subs to the mains?

One clue that a good friend mentioned to me in passing about another measurement issue is that the pink noise stimulus that is digitally generated through software or several of the popular portable noise generators out there, does not have sufficient "randomness" in the low frequencies considering the period of the LF frequencies (hence the popularity of using swept sine waves for improved stimulus of the system under test). Before thinking I'm heretical in my thoughts here, I have yet to test this hypothesis out, but have access to an older vacuum tube/diode based pink noise generator that supposedly offers an improved "randomness" in the LF. Maybe in a couple of weeks I can report back when I have had a chance to test it out.

How about making a 2 or 3 minute recording of that and posting it? We could loop it ourselves. It would be most appreciated.

Arthur Skudra
June 8th, 2008, 11:32 AM
Hi Ewan,
I was running Smaart 6 on my mac as well. Couldn't get any stability with the group delay trace particularly in the very low frequencies. Similar to what I experience when looking at the phase trace unwrapped when trying to line up subs with the mains. Glad I'm not the only one wondering if I missed something!
Arthur

Arthur Skudra
June 8th, 2008, 11:35 AM
Absolutely Harry! I plan to visit the guy in the coming weeks, and make a couple of WAV recordings of this pink noise to test out and load up to my iPod. It's stereo pink noise as well (which is helpful in setting up stereo systems). I'll get the rational folk to offer it as a download.

Jamie
June 11th, 2008, 12:02 PM
A few things here:
First, the show phase as group delay pretty much sucks in any sort of real environment with reverberence. This is because the function works by calculating the slope of the phase trace at each measurement frequency and then calculating the group delay represented by that slope. In environments with reverb and reflections, you will see some (or a lot) ripple in the phase trace - which translates into some steep slopes and hence lots o group delay. The phase trace is HIGHLY preffered for this application because the object is to read the trend of the slope - not the instantanous slope at individual frequencies. Slope as group delay is really best suited for purely electronic measurements or in anechoic environments.

Second, swept sines are often the preffered source and this is a great resource for explaining why. Sweeps paper (http://www.anselmgoertz.de/Page10383/Monkey_Forest_dt/Manual_dt/aes-swp-english.PDF).
Pink noise is often preffered for system tuning for a few reasons. 1. It excites the system with a complex source - which is normally what is listened to through the system. Sure, the noise floor of the measurement comes up a bit when the harmonic distortion energy is folded into our measurement, but that energy is part of the tonal response of the system and should be considered in the voicing. 2. Once you are used to listening to pink noise, you can get your ears into the measurement process and also actually listen to the complex interraction of system(s) and environment. 3. It is a lot less annoying to listen to, so if you have to play nice with others in the space . . .
No doubt, the swept sine source is by far the best choice for IR measurements and lab-work (provides the most stable, deterministic measurements.), but that does not make it the best for all measurement applications.

Third, the pink noise source from the Smaart generator is based on a 1.024 M(!) FFT - @ 21 seconds long at 48k. It was set at that so that the noise source record was at least twice as long as the longest FFT Smaart uses (512K in IR mode). Sure, "truely random" pink noise source based off of something like filtered Johnson noise (http://en.wikipedia.org/wiki/Johnson%E2%80%93Nyquist_noise) is better. But the not "random enough" stuff is a bit of a red herring.

yow,
-j

DaveG
June 11th, 2008, 01:10 PM
The phrase, "sufficiently random", is a bit of a semantic oversimplification. There are several potential problems with digital samples of noise.
If the noise loop is shorter than the measurement FFT, then there won't be any stimulus energy in some of the bins. For example, if you use a 4096-point noise sample and a 16384-point FFT, there will only be energy in every 4th bin. At low frequencies, the response will have gaps in its smile.:D
Another potential problem is that the average spectrum may not be flat - which is really only a problem for RTA measurements. It takes a random source a long time to average out to flat. A pseudo-random (digitally generated) source can be exactly flat (for a particular FFT length), if it is well constructed.
Then there's randomness. An impulse is as un-random as you can get, and is the worst possible stimulus, because it has the least amount of energy for a given amount of headroom. Random noise is spread out in time, so it is possible to obtain much higher S/N in a measurement. Pseudo-random noise can actually be a little better than random noise, if the criteria is peak-to-average ratio. But, it's not a big enough difference to worry about.
I can't think of any good reason to value "randomness" for its own sake. What we need is a sufficiently dense spectrum, a sufficiently low peak-to-average ratio, and a "soundtrack" that doesn't irritate human listeners. Smaart's pink noise generator is about as good as it gets on all three counts.

Dave

Jamie
June 11th, 2008, 02:07 PM
Pretty deep thinking for a "Junior" member. :D

I think we can all agree, a periodic signal source does not have to be as long as the period of all the frequencies in contains to have usable content at those frequencies.

So what I am picking up from what Dave is laying down . . .

The requirement for "randomness" in our FFT-based measurements really pertains to the signal source's length (periodicity) in relation to the measurement FFT's being used. In effect, the signal needs to be random to the FFT (every measurement cycle grabs a slightly different time segment of the source.)

With suede'o-random noise (or nagahide noise), measurement benefits are derived from the signal length being exactly = to the FFT length.

As Dave points out, the measurement problems occur when the signal length (period) is < FFT length.

Arthur Skudra
June 11th, 2008, 03:43 PM
I agree with the above comments on period length, particularly it's length in relation to the size of the FFT window. I always wondered about the period length of the noise generator in Smaart, good to know it's as long as it is.

Perhaps I should have been a bit more clear in the point I was trying to make, the difference between "real" random noise versus "pseudo" random noise. Digital noise generators (including Smaart, NTI Minirator, etc) make pseudo random noise, which depending on how it's done, has equally spaced components every "x" Hz, repeated over a period of time (as excellently described above). For instance, it *may* be every 10 Hz, so it will have energy at 20, 30, 40, 50, 60, 70, 80, 90, 100, 110, and so on up to 20,000 Hz. Notice it does not contain any component say at 65 Hz. "Real" random noise on the other hand, if averaged over an infinite period of time, will contain all possible frequencies.

So my questions to the experts here are:
1. What is the spacing of the pseudo random noise generator in Smaart?
2. Does the spacing of the components of psuedo generated noise really matter when we're dealing with measurements in the extreme low end? (Under 100 Hz)
3. Why does the RTA trace dance around like a monkey on a bed of hot coals under 100 Hz? Likewise, why when the phase trace is set unwrapped, does it dance around like the same monkey under 100 Hz? Or perhaps this is a completely unrelated issue? BTW, I compared a pseudo random noise generator to a real noise generator (yeah, I know how unstable old diodes can become), and couldn't believe how stable things were below 100 Hz! Why is that so?

Stirring the pot a bit more! ;)

Arthur

Arthur Skudra
June 11th, 2008, 03:56 PM
A few things here:
First, the show phase as group delay pretty much sucks in any sort of real environment with reverberence. This is because the function works by calculating the slope of the phase trace at each measurement frequency and then calculating the group delay represented by that slope. In environments with reverb and reflections, you will see some (or a lot) ripple in the phase trace - which translates into some steep slopes and hence lots o group delay. The phase trace is HIGHLY preffered for this application because the object is to read the trend of the slope - not the instantanous slope at individual frequencies. Slope as group delay is really best suited for purely electronic measurements or in anechoic environments.
I was suspecting this was the case. While the group delay graphs in the article were generated under "lab" conditions, when I tried to do the same trick in a live venue, it was really difficult to get decent data, hence reverting back to the regular phase display to finish things up. Are there cases in room acoustical measurement where group delay might be an advantage?

DaveG
June 11th, 2008, 04:43 PM
The spacing of the frequency components is 1/record_length. Since the record length is 1048576 samples, the frequency bins are spaced at 1/(1048576*20.8333us) = 0.0458 Hz. Only - it doesn't really matter. If you use a 16k FFT, the bins of the FFT are spaced at 2.93 Hz. So, having a 0.04 Hz spacing in the stimulus doesn't improve the resolution of the measurement.

The reason the RTA dances below 100 Hz is because the averaging time constant is the same for all frequency bands, but the rate of level variation scales inversely to the frequency. In other words, the high frequency variations are too fast for the "meter needle" but low frequency variations are slower, so the "meter needle" can track them.

Dave

Arthur Skudra
June 11th, 2008, 06:34 PM
The spacing of the frequency components is 1/record_length. Since the record length is 1048576 samples, the frequency bins are spaced at 1/(1048576*20.8333us) = 0.0458 Hz. Only - it doesn't really matter. If you use a 16k FFT, the bins of the FFT are spaced at 2.93 Hz. So, having a 0.04 Hz spacing in the stimulus doesn't improve the resolution of the measurement.
Fascinating! Certainly 0.04 Hz spacing is more than sufficient enough for acoustical tests and measurement. What is the 20.8333us constant and how is it derived?


The reason the RTA dances below 100 Hz is because the averaging time constant is the same for all frequency bands, but the rate of level variation scales inversely to the frequency. In other words, the high frequency variations are too fast for the "meter needle" but low frequency variations are slower, so the "meter needle" can track them.

DaveWow, this makes a lot of sense! This begs the question, are there instruments out there that would allow us to see those rapid high frequency fluctuations? Would the ability to measure these high frequency fluctuations and compensating accordingly translate into better sounding loudspeaker systems (ignoring HF loss over distance depending on % humidity)? Uh oh, am I treading on proprietary stuff here? :eek:

DaveG
June 11th, 2008, 06:43 PM
20.8333us is just 1/48000Hz - or the sample spacing at 48 kHz.

D

Ferrit37
June 12th, 2008, 05:50 PM
Hry Guys,
for those interested in a really random noise generator try this...
http://www.acopacific.com/noisegen.html

6.5 days...yikes

ferrit

Jamie
June 13th, 2008, 11:46 AM
The first hardware piece that will be sold under the Rational name is nearing release. It will be a pink noise generator - much along the lines of the old White Pink Noise tube, which has been a staple of my measurement kit for years. Stay tuned . . .

GoranO
July 3rd, 2008, 04:40 PM
Yes, that's the one! Thanks Ewan!

I did try following Charlie's group delay technique for aligning the subs on a system that I was optimizing, but being under some time constraints and not having any luck using this technique for the alignment of some subs, I gave up and went back to the alignment technique using phase response. Maybe I was doing something wrong, or perhaps the venue/system was not cooperating/unstable? I dunno, and since haven't had the chance to revisit this. Just wondering if anyone else had experience in using group delay for the alignment of a sound system, particularly the subs to the mains?

One clue that a good friend mentioned to me in passing about another measurement issue is that the pink noise stimulus that is digitally generated through software or several of the popular portable noise generators out there, does not have sufficient "randomness" in the low frequencies considering the period of the LF frequencies (hence the popularity of using swept sine waves for improved stimulus of the system under test). Before thinking I'm heretical in my thoughts here, I have yet to test this hypothesis out, but have access to an older vacuum tube/diode based pink noise generator that supposedly offers an improved "randomness" in the LF. Maybe in a couple of weeks I can report back when I have had a chance to test it out.

Hi,
in the same magazine, but few years earlier, there's interesting article on the same topic, by John Murray.. name of the article is Proper Signal (Time) Alignment, and he explains one other technique for driver alignment, can you please comment that...

p.s. i'm new here, and new in live sound bussines, and so far i think i'll never understand "phase" and "group delay" thing...:confused:

Harry Brill Jr.
July 3rd, 2008, 09:29 PM
p.s. i'm new here, and new in live sound bussines, and so far i think i'll never understand "phase" and "group delay" thing...:confused:

Trust me when I tell you, if you keep plugging away at it, you will get it. It's a steep learning curve. WELCOME. :D

This article http://www.livesoundint.com/archives/2003/july/align/align.php is saying pretty much the same thing in a different way.