I agree that the question of "sufficient" resolution is the real issue. In that regard I did some more digging.
Now it appears to me that no smoothing of FFT data makes little sense since we know that the ear has far lower resolution than an FFT at HFs, but it could have higher resolution at lower freqs. If we assume a good model of this resolution follows Moore (and I challenge anyone to find a better one) we can find the bandwidth of our hearing as a function of the center frequency (you can find this on Wiki under Equivalent Rectangular Bandwidth).
If for the sake of argument we assume that the resolution of a measurement is that of its window (which I still don't accept, but lets just move on as if I did.) then a 10 ms window would have a 100 Hz. cutoff and 100 Hz bandwidth. According to Moore this would be just slightly wider than our hearing at all frequencies for which the data is valid. This means that a 10 ms. windows effect on the data as perceived by our ears would be minimal to negligible.
A 5 ms. window would have a 200 Hz. resolution and according to Moore our ears are greater than this below about 1.5 kHz. Thus anything above 1.5 kHz would not be affected by the window, but from 200 Hz to 1.5 kHz the window would limit the resolution of the measurement to less than that of our hearing. Clearly 5 ms. windows may cause some limitations to representing a measurement consistent with our hearing. I would not want to look at data with a window less than 5 ms.
In my room I can usually get from 6-7 ms. of clean data so everything above 800-1 kHz is unaffected by the window. From 200 - 800 Hz is questionable, but probably reliable if the impulse response has decayed sufficiently in that time.
I find it no coincidence that 10 ms. can be correlated to the best resolution that our hearing is capable of as this is what I consider to be the window for early reflections. Greater than 10 ms. reflections are not an issue but < 10 ms. are. This is entirely consistent with the concept of hearing bandwidth (ERB).
Now it appears to me that no smoothing of FFT data makes little sense since we know that the ear has far lower resolution than an FFT at HFs, but it could have higher resolution at lower freqs. If we assume a good model of this resolution follows Moore (and I challenge anyone to find a better one) we can find the bandwidth of our hearing as a function of the center frequency (you can find this on Wiki under Equivalent Rectangular Bandwidth).
If for the sake of argument we assume that the resolution of a measurement is that of its window (which I still don't accept, but lets just move on as if I did.) then a 10 ms window would have a 100 Hz. cutoff and 100 Hz bandwidth. According to Moore this would be just slightly wider than our hearing at all frequencies for which the data is valid. This means that a 10 ms. windows effect on the data as perceived by our ears would be minimal to negligible.
A 5 ms. window would have a 200 Hz. resolution and according to Moore our ears are greater than this below about 1.5 kHz. Thus anything above 1.5 kHz would not be affected by the window, but from 200 Hz to 1.5 kHz the window would limit the resolution of the measurement to less than that of our hearing. Clearly 5 ms. windows may cause some limitations to representing a measurement consistent with our hearing. I would not want to look at data with a window less than 5 ms.
In my room I can usually get from 6-7 ms. of clean data so everything above 800-1 kHz is unaffected by the window. From 200 - 800 Hz is questionable, but probably reliable if the impulse response has decayed sufficiently in that time.
I find it no coincidence that 10 ms. can be correlated to the best resolution that our hearing is capable of as this is what I consider to be the window for early reflections. Greater than 10 ms. reflections are not an issue but < 10 ms. are. This is entirely consistent with the concept of hearing bandwidth (ERB).
One needs some method of "scaling" the importance of peaks and dips because the resolution of an FFT at HFs is way beyond what the ear can hear. I know of nothing better to do this than ERB.
Did you ever read the Farina reference? No!? Maybe if you followed up on what I told you it might make more sense. And no, I a not going to look up that reference for you.
Did you ever read the Farina reference? No!? Maybe if you followed up on what I told you it might make more sense. And no, I a not going to look up that reference for you.
Those are some fair points in Earl's post. But doesn't the ERB only work in relation to other frequencies? I mean the ear can detect differences in neighboring bands, but if the resonance lies completely within the band, will it not make it go up (or down) in relation to other frequencies, thereby allowing the ear to detect it?
The ear has an almost perfect ability to discriminate steady state tones, but that in no way tells us what the ear would be able to detect as far as complex transient signals are concerned. One has to look at the data from Toole and Olive regarding the audibility of resonances which clearly shows that some smoothing of narrow peaks and dips is required. But it is impossible to smooth in a way that is consistent with those findings, so one needs to use something else. Floyd used 1/20 th octave or something like that which was independent of frequency. But I find that untenable since we know that the ears abilities are frequency dependent. The ERB are about 1/20th octave in the midrange somewhere - wider at LFs and narrower at highs. It makes sense to me (and Farina) but maybe nobody else.
And NO, Markus this hasn't been "validated" with absolute proof by anyone - has anything?
And NO, Markus this hasn't been "validated" with absolute proof by anyone - has anything?
Earl;
Your propensity to take a thread off topic is astonishing.
Resolution of human hearing system has nothing to do with measurement resolution of an external information channel.
?!?
Isn't the end goal to be able to quantify what we hear? And then use that information to build better reproduction equipment? If one is attempting to take something from one frame of reference to another and make the most sense of it, one must understand both systems.
One can zoom in on a picture to the point that all that can be seen is dots or pixels but if you can't see enough of it to resolve what you are seeing, what good is that?
It has long been the struggle to get the information from the subjective domain quantified and properly translated in the objective domain and knowing how and what we hear is half the battle.
Barry.
Actually in this case it was ra7, noting that the topic objectively (the actual measurement in this context) was already finished:
http://www.diyaudio.com/forums/mult...quency-response-resolution-6.html#post3647233
(..the prior "market" stuff, was off-topic.)
-frankly the point was reasonably addressed empirically and theoretically with Toole's ex. (fig. 18.9) and the Stereophile Howard article on testing.
Arguing that Toole's figure is a "worst case scenario" is beside the point. That there was any example of a significant deviation from anechoic to gated, expresses a limitation in the measurement process of gated.
Of course anything like that will show up as a non-linearity in the impedance trace (of the raw driver as loaded).
No impedance prob.? -stick to gated and don't worry about it.
Impedance prob.? -use an anechoic/ground plane/free-field measurement to see how bad the problem is.
(note that Impedance problems are usually *very* small "narrow" bandwidth changes from the average.)
I suspect however that even a non-gated "noisy" in-room measurement would expose a problem. You might have to move/measure the speaker a few times to distinguish modal influence (at low mid.s), but you should see the problem if there is one.
One category of loudspeaker design that would almost certainly require non-gated would be transmission lines with drivers extending above 200 Hz for any useful output, particularly under-damped lines.
Last edited:
A non-critically damped T-line example:
PMC Twenty.24:
gated + nearfield-splice:
http://www.pmc-speakers.com/sites/default/files/attachments/HiFi Critic twenty 24 review.pdf
anechoic:
SoundStageNetwork.com | SoundStage.com | SoundStageNetwork.com | SoundStage.com
..little impedance bump at 180 Hz,
-but a BIG difference between the two measurements.
"worst case scenario" again?
-You decide.
PMC Twenty.24:
gated + nearfield-splice:
http://www.pmc-speakers.com/sites/default/files/attachments/HiFi Critic twenty 24 review.pdf
anechoic:
SoundStageNetwork.com | SoundStage.com | SoundStageNetwork.com | SoundStage.com
..little impedance bump at 180 Hz,
-but a BIG difference between the two measurements.
"worst case scenario" again?
-You decide.
Last edited:
A DIY design that also makes use of a type of T-line:
TQWT-
Planet 10 used to say "hey what about this".. and I kept saying look at the measurement - it's pretty flat in the midrange despite the impedance ripple.
Chances are extremely high that he was/is right, there are probably some significant non-linearity's going on in the mid-band.
-my bad Dave.
TQWT-
Planet 10 used to say "hey what about this".. and I kept saying look at the measurement - it's pretty flat in the midrange despite the impedance ripple.
Chances are extremely high that he was/is right, there are probably some significant non-linearity's going on in the mid-band.
-my bad Dave.
The question WAS asked - so I answered it. And it isn't off topic since ALL audio measurements need to correlate with our hearing or they are pointless. Doing that means understanding how the two things compare.
What is your problem!?
I didn't take the time to calculate it, but I would love to know if someone wants to take on that task.
What is the minimum ERB in octaves and at what frequency?
The ERB in that equation is given in linear frequency not octaves. Its probably just a log2 conversion, but I don't know it off the top of my head.
There are lots of things that can be "helpful", but I hate to see people chasing a non-problem audibly just because its there in some set of data. That's why I think that all measurements need to be made with a serious regard for what they mean audibly.
Its just like THD. I don't do it because it is meaningless from a subjective perspective, so what is the point? Just to show more "data"?
Last edited:
That is a great example. And it well could be that the manufacturer is trying to hide that flaw. As I said, its trivial to do if you want to.
It would be great to see anechoic data on all speakers, but its just not reasonable to expect that. So you do the best that you can with what you have. I have always been able to find the problems, you always can if you want to. But if you want to hide them, that's easy too.
The first one isn't PMC's data.
It's an *independent* measurement (and subjective review) from HiFi Critic, not particularly dissimilar from measurements from Stereophile or DIY'ers in general.
PMC actually lists all the reviews for their products, including the Soundstage review:
twenty.24 | PMC Loudspeakers
Basically I think they don't care about it.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.