Hi Pano,
Yeah, I did that test, too. Think I got to test signals at -40dB relative to the pink noise before it fell apart. It's a bit tricky to set up, involving a digital mixing desk and recording/playing noise/sweeps etc at different levels. I just ran REW sweeps below the pink noise - didn't bother with music etc, but that could be fun to hear.
It went like this:
Play pink noise through the speaker, record it.
Play that same section of pink noise, plus a test signal through the speaker, record that
Null the two, see if the test signal results.
It was interesting that, once again, the mic was picking up things that I plain couldn't hear: sweep & noise at the same level is obvious, and -10dB to -20dB is no problem. Further down, though, and it's easy to end up second-guessing yourself.
Perhaps an interesting listening test would be to identify a song (just by ear), played at some level below the noise.
Chris
Well, that was the result of that particular test on one particular speaker, and all I can say for sure is that the frequency response didn't change, down to inaudibility.
I was really impressed with the engineering on those drivers, Mark, and would have loved to throw a bit of DSP around to bring up the LF, roll in a tweeter, etc. Another time, perhaps, but I suspect my next speakers will include some of those drivers.
Chris
It's entirely possible, but IMO unlikely: the test speaker was a KEF HTS3001SE.
Plastic cone, low-tech motor. Nothing particularly exciting.
I use them for desktop listening, with the ports blocked and lots of EQ applied. They do pretty well: 40Hz-20kHz at sensible levels.
Chris
I'm a little confused here. Are you suggesting using the DUT as a microphone, to pick up other sounds?
Or are you thinking more like an impedance plot?
The latter might give some interesting data at very low signal levels, but there's a serious risk of the driver acting as a microphone and picking up all sorts of acoustic signals.
Chris
There are a few SNRs you might be referring to here, but I'll outline the steps I took:
- Make sure it's really really quiet in the room. I could hear cold water convecting in a nearby radiator once my hearing had adjusted. I was in the furthest corner of the house from the fridge.
- Use a directional mic (10dB+ of rear rejection helps)
- Use a very high-sensitivity mic - more signal feeding into the test system
- Use a very low-noise mic.
The three mic-related points made the Beyerdynamic MC930 a good candidate. Plenty of signal output, very low noise, and good rejection of the outside world.
Finally, I let REW run 8x sweeps for each measurement, to give the software the best chance of eliminating noise.
Chris
Two more
The experiment Chris661 reported in #133 is excellent - the results should be more widely known. I particularly like the idea of using a sweep, for generally decreasing excursion as the frequency rises. One could argue that a more complex signal would "dither" over any "steps" in the response masking their effect.
I had measurement equipment set up for something else, and I couldn't resist having a go: Holm Impulse, taking sweeps at normal, low, and barely-audible levels - as low as I could go given mic and background noise.
The first driver (first plot) is a GRS pt5010 - perhaps regarded as a control for the experiment. There are LR crossovers at 700Hz and 3kHz. The gating is constant.
The lowest measurement is 40dB SPL at 3cm measurement distance, an extremely soft sound at 1m. OK: nothing to see here.
What about a B&C 14NDL76 with a coated, triple-roll surround? This driver is a month old, used at domestic levels. Qms is 8.2 (datasheet).
Measured at ~10cm to catch sound from the whole cone. There's an LR 4th order 700Hz low-pass. A floor reflection shows up at 250Hz, so ignore below ~300Hz. The SPL at the mic is lower due to increased distance, and background noise is not far below either. The green measurement reaches no higher than 30dB SPL at 10cm - i.e., not more than a few nm motion of the cone at 500Hz. Looks fine to me, at any level that may be of concern - sticking would surely show up as a step in the response. Yet, there's nothing to see above -80dB on the plot that's below 10dB SPL at the listening position.
Qms looks unlikely characterize stiction. Even if someone finds a "sticky speaker" that's not enough to suggest that Qms is a useful measure of stiction more generally.
Ken
The experiment Chris661 reported in #133 is excellent - the results should be more widely known. I particularly like the idea of using a sweep, for generally decreasing excursion as the frequency rises. One could argue that a more complex signal would "dither" over any "steps" in the response masking their effect.
I had measurement equipment set up for something else, and I couldn't resist having a go: Holm Impulse, taking sweeps at normal, low, and barely-audible levels - as low as I could go given mic and background noise.
The first driver (first plot) is a GRS pt5010 - perhaps regarded as a control for the experiment. There are LR crossovers at 700Hz and 3kHz. The gating is constant.
The lowest measurement is 40dB SPL at 3cm measurement distance, an extremely soft sound at 1m. OK: nothing to see here.
What about a B&C 14NDL76 with a coated, triple-roll surround? This driver is a month old, used at domestic levels. Qms is 8.2 (datasheet).
Measured at ~10cm to catch sound from the whole cone. There's an LR 4th order 700Hz low-pass. A floor reflection shows up at 250Hz, so ignore below ~300Hz. The SPL at the mic is lower due to increased distance, and background noise is not far below either. The green measurement reaches no higher than 30dB SPL at 10cm - i.e., not more than a few nm motion of the cone at 500Hz. Looks fine to me, at any level that may be of concern - sticking would surely show up as a step in the response. Yet, there's nothing to see above -80dB on the plot that's below 10dB SPL at the listening position.
Qms looks unlikely characterize stiction. Even if someone finds a "sticky speaker" that's not enough to suggest that Qms is a useful measure of stiction more generally.
Ken
Attachments
No not at all - although it's an interesting approach!
Yes, play a signal into the woofer and record that signal taken right at the woofer's terminals. You would get the signal and whatever changes the woofer makes to it.
It happens, for sure. One has to be careful when going impedance sweeps, the driver will always pick up something.
When I did the tests about 6 years ago I simply embedded one piece of music at a low level way down into another piece of music. A recording was made of a song first without, then with the embeded low level info. I was trying to use Bill Waslo's DiffMaker software to extract the difference and find changes to the low level music. Bill was doubtful that it would work on an acoustic recording like that, but I did have modest success. I would take some work and some practice to get it right.
I'd suggest a slight modification, then, and drive the woofer from a current source rather than a voltage source.
A voltage source is a short-circuit that produces voltage, so any signals the woofer is producing would be shorted out.
I guess that so long as the same section of pink noise is played (ie, a pre-recorded section of noise rather than a freshly-generated bit of random noise), then the isolation could be pretty good.
I have quite a lot on at the mo (working 9-5 and coming home to a pile of AKG D12s that need restoratioin), but will give it a try when I get time.
Chris
A voltage source is a short-circuit that produces voltage, so any signals the woofer is producing would be shorted out.
I guess that so long as the same section of pink noise is played (ie, a pre-recorded section of noise rather than a freshly-generated bit of random noise), then the isolation could be pretty good.
I have quite a lot on at the mo (working 9-5 and coming home to a pile of AKG D12s that need restoratioin), but will give it a try when I get time.
Chris
Compliments on your measurement experiment and thx for posting about it. I of course wrote ‘sounding’ because here obviously are perceptive elements at work. Our hearing isn’t linear wrt frequency, neither is it wrt level. So one should hear differences, otherwise some linearity problems of the speaker would be obvious.
Precisely. That makes sense, along with psycho-acoustic preferences (resonances and the like) and our ever-present bias.
Yes, good point. It would be easy enough to try both voltage and current source to compare. A series resistor would somewhat isolate the recording from the amp. Is that good or bad? TDB.
Since it's just woofers that we are testing, I would suggest a 2nd order low pass of approximately 2.5 kHz on the signal. No need to test a woofer higher, and the limited bandwidth would make nulling out the masking signal much easier than full bandwidth.
It might be interesting to evaluate the coil's movements for as long as the cone moves in a pistonic way. Then again, differences resulting in high or low Qms could be originating from more or less damped surrounds and the breakup of the cone involved. I'm inclined to think part of the perceived differences (if they are real) could stem from design choices around the cone rim and surround damping (not forgetting the sound output contribution from the surround itself).
Low damping surrounds don't damp the transversal sound wave in the cone that well. The reflected wave from the frame could be stronger, all dependent on the elasticity, mass and damping in the surround. Bottom line, not only the spider but also the surround have to be carefully balanced wrt those qualities to optimize the resonant behavior of the cone. Striving for a low Rms / high Qms would limit one's options. The only argument there would be higher efficiency really. There's no news for driver designers in this all I guess. I actually would be more interested in the perceptions than in the mechanics on this. What differences do people hear, what circumstances and are the differences reproducible?
Low damping surrounds don't damp the transversal sound wave in the cone that well. The reflected wave from the frame could be stronger, all dependent on the elasticity, mass and damping in the surround. Bottom line, not only the spider but also the surround have to be carefully balanced wrt those qualities to optimize the resonant behavior of the cone. Striving for a low Rms / high Qms would limit one's options. The only argument there would be higher efficiency really. There's no news for driver designers in this all I guess. I actually would be more interested in the perceptions than in the mechanics on this. What differences do people hear, what circumstances and are the differences reproducible?
Q vs loss
I'm hesitant to interject where there may be strongly-held opinion, but I believe Qms is unlikely to be a reliable guide to high-frequency behavior of a loudspeaker.
People like speakers that happen to have high (or low) Qms, and that's fine, of course: I'd not argue with preference. It might even be that within a given range of speakers, the preference scales with Qms. That does not mean that Qms can be used to predict the sound of another speaker more generally.
The following overview of some of the different types of damping may be helpful. It represents a modern understanding developed over the last several decades in engineering and material physics.
Damping in Structural Dynamics: Theory and Sources | COMSOL Blog
This was the first hit in a search, I have no connection to the author or companies involved.
The problem comes from the definition of the quality factor Q of a resonance:
Q is defined to be 2 pi times the energy stored in the resonance divided by the energy lost per cycle of the resonant frequency. As a consequence, it has no well-defined meaning other than at the resonant frequency. It used to be thought that all damping was viscous, and Q at one frequency could be used to predict loss at another, but I do not believe that's a good guide given the kinds and range of materials used in loudspeakers.
A more informative quantity which is defined at all frequencies is the damping loss, aka the loss angle, or often tan delta in electronics - often given the symbol phi(f) - a function of frequency (f).
For real materials phi is often a nontrivial function of frequency; at some frequency like 10 times fs, it can differ from phi(fs) by easily a factor three; or in other cases, it could take almost the same value. That's a big spread, as Qms values are not very spread out in the first place.
Thinking of cone-edge resonances (often at 10 to 50 times fs). You could find two speakers with the same Qms (pick 5), and would know that the loss phi(fs) =0.2 for both. If one of the speakers has most of the loss in the surround, but the other has more of its loss in the spider, you'd have to figure this into the argument along with Qms. So in this case you might expect the speakers to sound very different even though Qms is the same.
I couldn't find two speakers that make a tight comparison, with Qms =5, but the following come close enough to make the point:
B&C 12NDL76 Qms 4.2, double roll surround, single spider
B&C 12NDL88 Qms 5.0, triple roll surround, double spider
Double spiders tend to have considerably more loss than single spiders, yet the Q of the 12NDL88 is higher, but then so is the cone mass.
If someone wants damped or undamped cone resonances, isn't it better to look for direct measurements of the break-up resonances, rather than a proxy that is not necessarily reliable?
Ken
I'm hesitant to interject where there may be strongly-held opinion, but I believe Qms is unlikely to be a reliable guide to high-frequency behavior of a loudspeaker.
People like speakers that happen to have high (or low) Qms, and that's fine, of course: I'd not argue with preference. It might even be that within a given range of speakers, the preference scales with Qms. That does not mean that Qms can be used to predict the sound of another speaker more generally.
The following overview of some of the different types of damping may be helpful. It represents a modern understanding developed over the last several decades in engineering and material physics.
Damping in Structural Dynamics: Theory and Sources | COMSOL Blog
This was the first hit in a search, I have no connection to the author or companies involved.
The problem comes from the definition of the quality factor Q of a resonance:
Q is defined to be 2 pi times the energy stored in the resonance divided by the energy lost per cycle of the resonant frequency. As a consequence, it has no well-defined meaning other than at the resonant frequency. It used to be thought that all damping was viscous, and Q at one frequency could be used to predict loss at another, but I do not believe that's a good guide given the kinds and range of materials used in loudspeakers.
A more informative quantity which is defined at all frequencies is the damping loss, aka the loss angle, or often tan delta in electronics - often given the symbol phi(f) - a function of frequency (f).
For real materials phi is often a nontrivial function of frequency; at some frequency like 10 times fs, it can differ from phi(fs) by easily a factor three; or in other cases, it could take almost the same value. That's a big spread, as Qms values are not very spread out in the first place.
Thinking of cone-edge resonances (often at 10 to 50 times fs). You could find two speakers with the same Qms (pick 5), and would know that the loss phi(fs) =0.2 for both. If one of the speakers has most of the loss in the surround, but the other has more of its loss in the spider, you'd have to figure this into the argument along with Qms. So in this case you might expect the speakers to sound very different even though Qms is the same.
I couldn't find two speakers that make a tight comparison, with Qms =5, but the following come close enough to make the point:
B&C 12NDL76 Qms 4.2, double roll surround, single spider
B&C 12NDL88 Qms 5.0, triple roll surround, double spider
Double spiders tend to have considerably more loss than single spiders, yet the Q of the 12NDL88 is higher, but then so is the cone mass.
If someone wants damped or undamped cone resonances, isn't it better to look for direct measurements of the break-up resonances, rather than a proxy that is not necessarily reliable?
Ken
Thank you Ken - You have stated some thoughts / concepts that have been mulling about in my head for a while regarding this thread. ... said it better than I could have.
Q is meaningful at the driver Fs. At higher frequencies the numeric value of Qms may not mean much.
Any dynamic model that assumes ideal visco-elastic behavior is only going to take us so far. Spring rates are not linear unless you are working with an actual engineered spring. More importantly, Damping is not viscous unless it is purposefully designed that way, such as an automobile shock absorber. In real world dynamic systems, damping is a combination of viscous behavior and frequency-dependent friction, and it must be empirically characterized ... i.e. not modelled.
My speculation is that at driver resonance, the system behaves close enough to an ideal visco-elastic system that the T/S model works, and therefore it is appropriate to employ Qes, Qts, and Qms to characterize the system. Cone resonances and surround resonances - these are more complicated, and Qms may be not be helpful in understanding those resonances.
jim
Q is meaningful at the driver Fs. At higher frequencies the numeric value of Qms may not mean much.
Any dynamic model that assumes ideal visco-elastic behavior is only going to take us so far. Spring rates are not linear unless you are working with an actual engineered spring. More importantly, Damping is not viscous unless it is purposefully designed that way, such as an automobile shock absorber. In real world dynamic systems, damping is a combination of viscous behavior and frequency-dependent friction, and it must be empirically characterized ... i.e. not modelled.
My speculation is that at driver resonance, the system behaves close enough to an ideal visco-elastic system that the T/S model works, and therefore it is appropriate to employ Qes, Qts, and Qms to characterize the system. Cone resonances and surround resonances - these are more complicated, and Qms may be not be helpful in understanding those resonances.
jim
Last edited:
Friction is speed dependent. And for direct radiators with sizes about 3 times smaller than the wavelength cone speed stays fairly constant (it goes down when the radiation impedance reaches a constant level). So a first order iteration would be that the friction-induced losses don't vary that much over the so-called resonant range of the transducer. But I agree modeling is a bit hazardous. And like Ken states, things are measurable. One has to believe measurements though...
Yes and no.
Focusing on one (global) parameter and forgetting others certainly doesn't bring you to the win. If a loudspeaker would be a F1 race car, would putting all your attention to the engine give you the championship? I personally become more and more annoyed when people simplify things too much and then start ranting/preaching or whatever.
Focusing on one (global) parameter and forgetting others certainly doesn't bring you to the win. If a loudspeaker would be a F1 race car, would putting all your attention to the engine give you the championship? I personally become more and more annoyed when people simplify things too much and then start ranting/preaching or whatever.