I might as well give my 2 cents as well.
Loss of Dynamics can come from many sources. I think in modern music, the recording is primary culprit, and we can’t do anything about that.
On the stuff we can do, it can come from many sources.
What interests me, Dr. Geddes, is your statement that non-linearity does not matter. I read your paper on the Geddes Metric, and you actually demonstrated good correlation between that metric and sound quality. Why have you changed your mind? On small signals, and if the harmonics are -60db down you may be right that distortion is not a player, but, when we talk about dynamics, a properly recorded album might have 18-22db peak to average signal. I have conducted my own abx tests, and for me, when I set very low crossover points for the drivers I am using, I lose dynamics. When I look a kippel results, I believe they validate my personal abx tests.
As for thermal compression, the JBL tests seem to contradict the Stereophile test. By the way, I trust JBL way more than Stereophile. I definitely would like to see more Thermal Compression studies in JAES.
I think one key aspect is to analyze a broad swath of music to actually determine the necessary peak to average ratios and the required duty cycle. We might have a situation where you have an easily produced result on test tones, but you will never see it in a recording or in a sound reinforcement setting.
I would also like to see if dynamics change in a current drive vs. a voltage drive system.
Finally, I would like to see tests with at least a 2 way loudspeaker, one where the transducers have equal power handling specs, and another where the power handling specs are vastly different.
I don’t ask for much!
Loss of Dynamics can come from many sources. I think in modern music, the recording is primary culprit, and we can’t do anything about that.
On the stuff we can do, it can come from many sources.
What interests me, Dr. Geddes, is your statement that non-linearity does not matter. I read your paper on the Geddes Metric, and you actually demonstrated good correlation between that metric and sound quality. Why have you changed your mind? On small signals, and if the harmonics are -60db down you may be right that distortion is not a player, but, when we talk about dynamics, a properly recorded album might have 18-22db peak to average signal. I have conducted my own abx tests, and for me, when I set very low crossover points for the drivers I am using, I lose dynamics. When I look a kippel results, I believe they validate my personal abx tests.
As for thermal compression, the JBL tests seem to contradict the Stereophile test. By the way, I trust JBL way more than Stereophile. I definitely would like to see more Thermal Compression studies in JAES.
I think one key aspect is to analyze a broad swath of music to actually determine the necessary peak to average ratios and the required duty cycle. We might have a situation where you have an easily produced result on test tones, but you will never see it in a recording or in a sound reinforcement setting.
I would also like to see if dynamics change in a current drive vs. a voltage drive system.
Finally, I would like to see tests with at least a 2 way loudspeaker, one where the transducers have equal power handling specs, and another where the power handling specs are vastly different.
I don’t ask for much!
I remember a test, not that I believe in them so much
It was a 2way speaker with TAD woofers and Fostex tweeter
Supposedly they suffered from compression at low levels, but really "came to life" at louder levels 😕 well, I could think of a flawed xo, but other than that I dont know
Earl, no worry, people will believe you more than me, and thats fine 😉
how they made those Altecs in the old days, does make one wonder
It was a 2way speaker with TAD woofers and Fostex tweeter
Supposedly they suffered from compression at low levels, but really "came to life" at louder levels 😕 well, I could think of a flawed xo, but other than that I dont know
Earl, no worry, people will believe you more than me, and thats fine 😉
how they made those Altecs in the old days, does make one wonder
Hi,
The reason I feel an adaptive algorithm is a necessity, is that the algorithm would be modeling the path from amp in to mic out, and it’s the time varying changes in this path's characteristic that we are looking to measure. No averaging required, which makes it agile in the time domain, able to hunt out quick changes in performance.
This is the root echo canceller design problem, in a nut shell. It’s not what really makes a good ecan (that's the non linear processing wrapped around this), however much can be gleaned from that field, and applied here (hint pointing towards a literature search). I agree that the MLS model could be (and should be) tightly constrained, and pretrained on a reference frequency response. One non-intuitive “gotcha” is that, when the signal envelope drops, the band edge frequency content becomes low in power, as the spectrum is waited to lower levels there. You then start running into lower resolution calculations in these frequency bins, and the “adapter” can mistake this for a change in the characteristic of the path being measured. It’s very important that the signal envelope, signal spectrum, measurement levels, and system-under test background noise (electrical, acoustic, round off error etc) be such that this effect doesn’t occur. Another classic echo canceller issue.
I agree that the ear’s ability to distinguish changes in time is dependant on the frequency of interest. This has been studied in the audition field, as “temporal integration”. The papers I’ve read on the topic show it is definitely frequency dependant. I used it to effect here:
http://patft.uspto.gov/netacgi/nph-...d=PTXT&s1=6,178,162&OS=6,178,162&RS=6,178,162
My temporal integration references are at home, but many of the books sold through the ASA will have applicable data.
I might have a speech file available, but not one for music. This is a great discussion, but near the limits of the personal cycles I have available to explore this. As you know, good audio jobs are hen’s teeth and my involvement in this and time available are squarely at the “hobbyist” level of commitment, for the present time.
The reason I feel an adaptive algorithm is a necessity, is that the algorithm would be modeling the path from amp in to mic out, and it’s the time varying changes in this path's characteristic that we are looking to measure. No averaging required, which makes it agile in the time domain, able to hunt out quick changes in performance.
This is the root echo canceller design problem, in a nut shell. It’s not what really makes a good ecan (that's the non linear processing wrapped around this), however much can be gleaned from that field, and applied here (hint pointing towards a literature search). I agree that the MLS model could be (and should be) tightly constrained, and pretrained on a reference frequency response. One non-intuitive “gotcha” is that, when the signal envelope drops, the band edge frequency content becomes low in power, as the spectrum is waited to lower levels there. You then start running into lower resolution calculations in these frequency bins, and the “adapter” can mistake this for a change in the characteristic of the path being measured. It’s very important that the signal envelope, signal spectrum, measurement levels, and system-under test background noise (electrical, acoustic, round off error etc) be such that this effect doesn’t occur. Another classic echo canceller issue.
I agree that the ear’s ability to distinguish changes in time is dependant on the frequency of interest. This has been studied in the audition field, as “temporal integration”. The papers I’ve read on the topic show it is definitely frequency dependant. I used it to effect here:
http://patft.uspto.gov/netacgi/nph-...d=PTXT&s1=6,178,162&OS=6,178,162&RS=6,178,162
My temporal integration references are at home, but many of the books sold through the ASA will have applicable data.
I might have a speech file available, but not one for music. This is a great discussion, but near the limits of the personal cycles I have available to explore this. As you know, good audio jobs are hen’s teeth and my involvement in this and time available are squarely at the “hobbyist” level of commitment, for the present time.
This was just an example. Can you provide transient response or describe what you think is a problem in soft dome and where the hard dome is better except breakup region? Do you mean only decay time or/and levels below -30dB?
People usually say "Soft cones/domes hides music details" - I would ask which details? Low level? Have you ever seen impulse/CSD discussion on drivers below -30dB?
Greets,
Jack
DDF said:
Well I am not convinced that an adaptive algorithm is a necessity. I am not even convinced that it is desirable. At least not in my case. Spectral estimation techniques have their limits just like everything else, but you are right they are adapt at tracking changes in the time domain.
I understand the situation very well having done Active Noise Cancellation for several years.
You see in my situation I have to gate out the room reflections. Hence, if I have to use the FFT to gate out and aquire a clean impulse response, the "bugs" are already in the cake. Of course you will say "no problem" with adaptive processing because that is what echo-cancellers do, but even if the adaptive approach is the best method - and perhaps it is - thats more work than I have time to spend.
I can get very clean impulses about every second or so, but with my situation this is about the limit.
So, for now, I have to live with whats readily available.
mbutzkies said:
I did not change my mind about the perception of nonlinear distortion in loudspeakers. In fact, I alluded to the fact that mechanical systems cannot have high orders of nonlinearity because of the forces that would be required. I hypothesized - at the time - that this would make nonlinearities in loudspeakers relatively inaudible. The Gedlee Metric could never be very large for loudspeakers because of the nature of the nonlinearities in them.
Then B&C and Lidia and I tested the perception of nonlinearity in compression drivers (see AES paper from 2006). No one was able to hear distortion as high as 25% THD, - in this test nonlinear distortion was completely inaudible. This result confirmed my suspicions.
My results on thermal compression do not agree with the Stereophile study either.
Current drive would reduce the thermal ompression because it is mostly a change in impedance and a current drive would be immune to this.
not sure what you mean by breakdown
If you are talking about low xo point to get down distortion, I dont do that ... as others are beginning to realize too, you get better sound with higher xo point
I admit that distortion gets higher, but sound is really more natural ... maybe better phase behaviour is one benefit, anyway what I experience is the better virtues of a good tubeamp
Did I just hear Geddes say that you cannot hear fairly high levels of nonlinear distortion ... anyway, high xo point lay less strain on the tweeter ... maybe its about less thermal compression ?
If you are talking about low xo point to get down distortion, I dont do that ... as others are beginning to realize too, you get better sound with higher xo point
I admit that distortion gets higher, but sound is really more natural ... maybe better phase behaviour is one benefit, anyway what I experience is the better virtues of a good tubeamp
Did I just hear Geddes say that you cannot hear fairly high levels of nonlinear distortion ... anyway, high xo point lay less strain on the tweeter ... maybe its about less thermal compression ?
"As for thermal compression, the JBL tests seem to contradict the Stereophile test. By the way, I trust JBL way more than Stereophile."
Look at the differences and the mindset of the testers. Why did they do the tests and what did they want to prove??
The JBL test was done as part of the introduction to the vented gap technology against competitors drivers and the previous SFG motors. They dropped in 300 watts of pink noise from 50hz-500hz, which is half their new drivers power rating by the way. They then measured the output over time to measure the compression and see how quickly the compression occurred.
It's kind of obvious why they tested the drivers this way as their new design was a clear winner.
The Stereophille test was done with how much power?? They were running the test using a 60 watt amp with the speakers measuring 100-104db @ 1 meter. The base sensitivity on the speaker system used was 86Db @ 1 meter. So a 14-18dB over 1 watt and only over a very narrow frequency change. Look at the frequency spectrum used. The crossover is at 3K so how much power is there going into the tweeter??
These tests are apples and oranges. In the text it states that the tweeter only receiver 3% of the power the mid bass unit did. So what's 3% of 60 watts??
I trust JBL's testing as well.
Rob🙂
Look at the differences and the mindset of the testers. Why did they do the tests and what did they want to prove??
The JBL test was done as part of the introduction to the vented gap technology against competitors drivers and the previous SFG motors. They dropped in 300 watts of pink noise from 50hz-500hz, which is half their new drivers power rating by the way. They then measured the output over time to measure the compression and see how quickly the compression occurred.
It's kind of obvious why they tested the drivers this way as their new design was a clear winner.
The Stereophille test was done with how much power?? They were running the test using a 60 watt amp with the speakers measuring 100-104db @ 1 meter. The base sensitivity on the speaker system used was 86Db @ 1 meter. So a 14-18dB over 1 watt and only over a very narrow frequency change. Look at the frequency spectrum used. The crossover is at 3K so how much power is there going into the tweeter??
These tests are apples and oranges. In the text it states that the tweeter only receiver 3% of the power the mid bass unit did. So what's 3% of 60 watts??
I trust JBL's testing as well.
Rob🙂
Keith Howard did the tests for Stereophile - I know Keith and he is a good researcher. I'm just saying that I got different results - could be many things. What I did not like about Keiths approach is that he did not measure the frequency response directly, he only surmised the VC temp from the current draw. This is not a direct measure of the total problem.
On the other hand, its clear that in my test I used a signal with much more HF content than it would see in the real world. BUT, I did compare two speakers under the exact same test and they were vastly different. The comparison is fair if not the absolute numbers.
From what I am gleaning thus far, I think that the issue has more to do with the time constants of measurement.
On the other hand, its clear that in my test I used a signal with much more HF content than it would see in the real world. BUT, I did compare two speakers under the exact same test and they were vastly different. The comparison is fair if not the absolute numbers.
From what I am gleaning thus far, I think that the issue has more to do with the time constants of measurement.
Opps I didn't mean to slight Mr Howard. My point was they really were done under a different set of conditions. One test is an all out assault on the drivers the other much less demanding. Under those circumstances I don't think it's all that surprising the conclusions reached were not the same.
Rob🙂
Rob🙂
It's easyer to look at two sets of data and compare while also listening. I think this woud be what gedlee calls "closing the loop" But just data alone, although I cannnot see the scaling on the charts due to resolution limit, as far as hidhing detail, look at the rising section at the root of the so called impulse, my guess is that curvy region is an indicator of lost of detail. In this impulse graph, they must be using an impulse long enough to let the low frequency tail show instead of a unit impulse. If this is an impulse generated from MLS signals, then this kind of shape at the root rising edge is a sure indicator of less detail.jzagaja said:
I think this is consistent with my impression on Dynaudio speakers, and I have listened to a few of their high priced series. Which details? It really needs to be explained side by side while listening.
tinnitus wrote:
I have often experienced that speakers come to life at high SPLs if they are lacking low bass response (hint: Fletcher - Munson).
This has IMO nothing to do with power compression.
Regards
Charles
I have often experienced that speakers come to life at high SPLs if they are lacking low bass response (hint: Fletcher - Munson).
This has IMO nothing to do with power compression.
Regards
Charles
Impulse rising region - I've only seen Herr Manger who looked in this part but I couldn't repeat his experiment from Holoprofile booklet. Please look on my Jordan JXR6 high resolution step responses . Herr Manger claims that initial curve should be constant with a broad angle of incidence but of course rising time can be different.
Now you are saying that soft/hard dome differs here. It is very hard to tell because:
1) They differ in bandwidth, rising time, in linear region of operation
2) Published impulses have too small horizontal resolution for discussion
phase_accurate said:
The subject is dynamics in loudspeakers ... power compression is only one aspect ... and I did state that xo might be an issue
What you say , if I understand correctly, is that lack of sub lowend response can lead to compressed sound at low levels ... sounds logical🙂 maybe due to phase issues and lack of resolution ... at low levels phase may be exstremely important, as phase and some noise issues are not attenuated
Since the amplitude and time scales are not comparible, no meaningful comparison can be made, but if you draw a line along the rising edge until it intersects the time axis, then draw a line from the peak of the pulse perpendicularly down to the time axis, then look at the ratio of distance to the 0 point and see what ratios you get.jzagaja said:
phase_accurate
I think Soongsc have in mind only the rising part so that you can imagine it as a set of power curves (y=x^a) and tangential lines at maximum amplitude. Isn't it the ratio a slope order? Electrostatics and compression drivers have a->1 - you see a gentle curve. All in all it is difficult because:
1) Measurements contains pre ringing from aliasing filters
2) You need >500k samples per second AD
I think Soongsc have in mind only the rising part so that you can imagine it as a set of power curves (y=x^a) and tangential lines at maximum amplitude. Isn't it the ratio a slope order? Electrostatics and compression drivers have a->1 - you see a gentle curve. All in all it is difficult because:
1) Measurements contains pre ringing from aliasing filters
2) You need >500k samples per second AD
- Status
- Not open for further replies.