Automatic Extraction of Minimum-Phase Response

Member
Joined 2009
Paid Member
No this is the same artifact for me.
But intuitively i would consider them from different origins one being 'inherent' to the driver, the other a question related to cabinet.

I wonder about suspension diffraction since i listen the most to coax for 2 years now. I'm sure it is part of the 'ragged' high freq they have but in practice i can't identify it by ear or i'm not sensitive to it?
Maybe it is because i can't compare to a coax driver which doesn't have this issue (or very low like Genelec's or Cabasse's or Kef's) but even then directivity behavior would be different to the Tannoy and could hide the point.
I remember Mark100 described the high of his coax experience as 'diffuse' iirc and wonder if this is not related.



So i take it as part of the sound of the driver... a compromise for the others quality i like.
 
Member
Joined 2009
Paid Member
No.
It crossed my mind to investigate it but i didn't had time allowed to this between kids and my business to run in this weird time, and had not find a way ( a protocol) to identify the different source which exist on the loudspeaker i have ( Tannoy System800).
How would you proceed Allen?
 
diyAudio Moderator
Joined 2008
Paid Member
I haven't tried it either.. I guess I'd want to get back a few times the surround dimension, mount the driver flush in a baffle larger than the measuring distance and do a fine set of polars. The information would be in amongst breakup and lobing. If it were hard to spot I could experiment by adding a larger ring around the driver and indentifying that in the differences.
 
http://www.excelsior-audio.com/Publications/Phase_Response_&_Receive_Delay.pdf

This is a great article by Charlie Hughes and I believe the answer to your question Mark is found in pages four and five.

I guess I have taken having TEF for granted all these years for it’s ease of use in this respect.

Barry.

Thanks Barry. Yes, Charlie does a great job of explaining where to reference phase. Been a fan of his clarity for some time...

I kinda knew the "best science" answer is the Heyser approach.

Other than maybe TEF (which i've only read about), i don't know how we can realistically make measurements for frequencies well outside the passband of the DUT.

Because it think alot of the time, any measurements that are well past the intended passband are pretty much garbage. And get make sense of the high end of the phase curve. (for those of us relying on Hilbert Transform...Smaart, REW, Arta, etc)

I'm hoping the "measure with a well-higher-low-pass than than the intended-passband", is a pragmatic and at least a more accurate step towards mathematically precise. (using existing tools/software)


Thanks again...much to learn :)
 
Bohdan - it is a very cool paper, and a useful tool.. Thanks!

Your paper answered a small mystery I had with some of my measurements:

When I used a 2-channel audio interface + ARTA + Audix TM1 mic, I was able to measure (directly measure) a TOF difference between the mid driver and the tweeter of 84 u-sec, which is 28.8 mm difference in the acoustic centers.

When I used an OmniMic USB mic, I of course do not capture a valid "time of flight" measurement, so I use the Jeff Bagby method: Box
This method uses inference to iteratively march toward the offset values which most closely reproduce the responses of the tweeter and mid driver.

Sometimes these two methods agreed to within 1 mm. But recently the best agreement I could get was 7.5 mm.

Now I know why thanks to the Bohdan paper. The sampling rate is 48k. This means I get a sample every 20.8 usec, which is equivalent to 7.1 mm… this is the resolution limit of my measurement system, both for the OmniMic and the 2-channel audio interface + ARTA + Audix TM1.

Nice to have that mystery cleared up :)

j.

If your method for determining offset using 2-channel system was to simply measure sample time difference between the peak of the impulse, that will create an error +/- 1 sample time which can be significant as you've found out. However, when measuring using a 2 channel system this process is completely unnecessary, just lock the FFT window start, the TOF differences will be captured in the measured phase. Even though 1 sample may be 7.1mm, you may find that a movement of the mic by 1mm is still captured in the measured phase, it is accurate even at 48kHz sample rate but I recommend 96kHz rate for measuring regardless, most any modern equipment can do it.

In ARTA, instead of simply observing the impulse peak, run your measurements, then convert to FR and observe the excess phase. Adjust the delay for phase estimation until the excess phase looks the same in both measurements, and then the delay value here becomes the difference in acoustic distance and should be more agreeable than simple observing the impulse peaks.

The real question when it comes to minimum phase and 2 channel measurement systems is why do I need it? HBT, IHBT, guiding filters, it all seems rather elaborate for a process that isn't needed at all for loudspeaker design using a 2 channel system. For any measurement of decent SNR, just use measured phase as-is and keep the FFT window start locked and a constant distance fro baffle surface to microphone. Minimum phase is not important or necessary, only relative phase between drivers.
 
I have produced two papers...

Bohdan, nice work.
A while back I tried to identify excess phase in LTspice.
The obvious way is to calculate the minimum phase from the amplitude response and then compare with the actual LTspice determined phase.
I believe you have Spice expertise, not sure if this is LTspice.
Any advice you can provide?
In any case, thanks for the papers, very educational.

Best wishes
David
 
> > The number one misconception about woofers is that moving mass has
> > something to do with "speed" of response or its high frequency limit.
> > It does not, at least directly.
> > What mass effects is efficiency and also the shape of the low frequency
> > roll off.

......

> > If one had a driver which had a very strong motor or a normal motor but
> > very low moving mass, one gets an "over damped" response.
> > This term is from filter design meaning that it is not optimally flat,
> > excessively damped, rolling off too soon and gradually
> > Should the slope of the response reach 6 dB per octave, the driver is
> > operating in the Velocity controlled mode, while the response is not
> > flat, the acoustic phase DOES track the input signal (zero degrees) and
> > the different frequency components are not spread out in time.
> > The waveshape of the input signal is more closely replicated as the
> > frequency components are in the original "time" although the amplitudes
> > are off 6 dB/oct. Each 3 dB /oct change in the slope produces a 45
> > degree change in phase.
> >
> > An over damped response more closely retains the time information where
> > a flat amplitude response cannot.
> >
> > A proper LF horn can have flat acoustic response AND roughly zero degree
> > acoustic phase.
> >
> > For a person more sensitive to "time errors", they will likely find an
> > over damped system more realistic.
> > For a person more sensitive to "amplitude errors" the traditional "flat
> > response" system will be more satisfying.
> > For the person lucky enough to have heard a proper lf horn system, you
> > have heard that one can have "lightning fast" sounding bass and still
> > make your pant legs flap.

......

> > Additional reactance's can alter this phase relationship.
> > For example above midband at the point in the impedance called Rmin, the
> > electrical series "L" is equal but opposite the reflected moving mass
> > (capacitive reactance) of the driver thus canceling each out and being
> > resistive (no phase shift).
> > Above that frequency, the series Inductance dominates and produces a
> > roll off with an inductive reactance.
> > At the bottom end of the response, for a simple sealed box, there is a
> > point where the parallel spring or "compliance" of the box and driver
> > are equal but opposite the moving mass and this point is also resistive
> > (at box resonance). Below that frequency, the spring constant dominates
> > (an inductive reactance) and the acoustic phase leads
> >
> > More complications from non perfect drivers and alignments..
> >
> > In pro sound it is a common practice to evaluate subwoofers with a kick
> > drum signal.
> > Such a signal has a wide spectrum however it makes a bad test signal for
> > a subwoofer alone.
> > Nearly always, the subwoofer with the best "snap" or attack is the one
> > with the greatest distortion and or the highest low cutoff.
> > Clearly, the ear hears the added hf content and judges it to be more
> > lifelike.
> > In actual operation, the subwoofer is mated with midbass speakers who's
> > job it is to produce the spectrum above the subwoofer.
> > Now, with the actual drum signal being produced in the upper ranges, the
> > subwoofers formerly desirable distortion spectrum interferes with the
> > real drum signal. Now, the subwoofer with the least distortion will
> > often have the best subjective sound.
> >
> > A subwoofer can be made with a under damped low frequency corner, this
> > puts a bump in the response right before roll off.
> > Subjectively, this can make a woofer sound even slower and the "decay"
> > of the too high Q takes more time.
> >
> > Remember that it is acceleration which produces sound, it is the
> > amplifier current which produces the force which produces the
> > acceleration.

> >
> > Tom Danley


Thanks Mark, for posting this gem ;)

A higher Q (box tuning) doesn't necessarily result in a under damped response, it also depends on the driver's characteristics in relation to the cabinet.

The first paragraph regarding Mms and the remarks on horn loaded bass are spot-on.
 
are members familar with the excel sheet " Frequency Response Combiner "
from FRD?

Yes - I still use this when evaluating propsective driver combinations... to apply with baffle diffraction and step response.

I trust it is correct :)

Now tools like VituixCAD and XSim can do a HBT or some other minimum phase derivation on the fly.

Acoustic centres of course then need to come into play.

PS: I have had bad experiences on Windows with USB external sound cards. the impulse response can vary significantly.

If using measured phase - then a hardwired sound card with loopback reference is the only consistent option I've found (msec accuracy).

My new laptop doesn't have this... but I have yet to see if Windows 10 / hardware speed has resolved my USB timing volatility.

Otherwise - I'll use my USB mic, derived phase and work out the offset using the method I learned from Jeff Bagby.
 
Otherwise - I'll use my USB mic, derived phase and work out the offset using the method I learned from Jeff Bagby.

I have confirmed that both methods yield the same result, to the extent that it matters. Using a 96kHz sampling rage, the difference between the two methods is ~ 3 mm... close enough for any practical purpose.

We have a tendency to view current best-practices methods as "correct", and any other method as "incorrect", including older methods which were best-practices at the time. This is a logical fallacy and we all need to keep in mind that there is more than one way to achieve our goals.

j.
 
I have confirmed that both methods yield the same result, to the extent that it matters. Using a 96kHz sampling rage, the difference between the two methods is ~ 3 mm... close enough for any practical purpose.

We have a tendency to view current best-practices methods as "correct", and any other method as "incorrect", including older methods which were best-practices at the time. This is a logical fallacy and we all need to keep in mind that there is more than one way to achieve our goals.

j.

My point was using measured phase and derived phase with acoustic offset should get me the same outcome with crossover design. My measured phase wasn't when using the USB sound card which on investigation I found the impulse response was not consistent. Therefore my setup couldn't be relied upon. Both methods are fine.
 
Dave - my point (which was not 100% clear) was that the Jeff Bagby method and the measured absolute phase method both result in the same acoustic offset... so the same crossover will result. I was agreeing with you.

Sometimes people will criticize those who use the minimum phase + offset method for being not up to date or for making an error. Not being up to date is a fair statement, but being in error is not correct. The old method works well.
 
diyAudio Moderator
Joined 2008
Paid Member
It's not that simple, Jim. Jeff used that technique where he mentioned he did not have reliable access to a measurement system that preserved time between measurements.
hifijim said:
to the extent that it matters.
Qualifiers like this could become necessary, in a case where it weren't also taken into consideration that time offset is not the only non-minimum phase element in a typical measurement, although it easily could. If the discrepancies attributable to diffraction and other secondary sources are assumed to be minimum phase during the conversion, do they not become errors?
 
The real question when it comes to minimum phase and 2 channel measurement systems is why do I need it? HBT, IHBT, guiding filters, it all seems rather elaborate for a process that isn't needed at all for loudspeaker design using a 2 channel system. For any measurement of decent SNR, just use measured phase as-is and keep the FFT window start locked and a constant distance fro baffle surface to microphone. Minimum phase is not important or necessary, only relative phase between drivers.

Strongly agree.

Phase pragmatically means relative phase, and relative to the highest audio frequency, imho.

Having precise minimum phase curves for each driver (relative to infinite bandwidth) is not needed ime.
What is needed is locking the FFT window start like you mentioned, to capture relative phase.

Besides, what pragmatic point is there in knowing precise absolute phase. Non-PC processing is usually limited to 48kHz timing precision, or perhaps 96k on high dollar gear. That's the best we can adjust phase to....