Speaker advice redux?

"A speaker is either able to resolve details better than another one or it's not, level isn't going to change that."

That is true in many cases, and would probably be true in all cases if the spectral balance didn't change with volume level. But sometimes that is what happens. Let me give a hypothetical example that illustrates what I'm talking about. I'm going to make some simplifing assumptions here, to keep this hypothetical from getting too complicated:

Suppose we want to design a compact speaker with decent bass, so we choose a beefy little 83 dB efficient 5-inch woofer that has some serious low end going on (honest 40 Hz in about 1/4 cubic feet). Power handling 50 watts RMS, 100 watts "music program", 200 watts peak. And because we want good resolution, we choose a nice little 90 dB efficient 1" dome tweeter, same power handling characteristics as the woofer.

The problem we run into is thermal compression. As a ballpark generalization, a speaker will have about 1 dB of thermal compression at 1/10th its RMS power handling, increasing to about 3 dB at its RMS rated power handling, and going up from there.

The woofer will see far more power than the tweeter, since the tweeter will be padded down to level-match with the woofer. So let's say we "voice" our little speaker to sound correctly balanced at 97 dB, which corresponds to 50 watts input. Our woofer is seeing nearly the full 50 watts, and our padded-down tweeter is seeing a little over 2 watts. So at 97 dB, our woofer is exhibiting about 3 dB of thermal compression (that's why it isn't giving us the full 100 dB we would otherwise expect), and our tweeter isn't exhibiting any thermal compression at all. So instead of being padded down the 7 dB we might expect based on the efficiency difference between the two drivers, we have padded down the tweeter by 10 dB, so that the spectral balance is correct at high sound pressure levels (97 dB ballpark).

So, what happens at low levels, say down at 85 dB or less?

Well, at low levels, neither woofer nor tweeter are seeing enough power to exhibit any significant thermal compression. But remember that we had to pad down the tweeter by an extra 3 dB because we wanted the speaker to sound good at 97 dB. So now, at low levels, the tweeter is 3 dB too soft relative to the woofer! As a result, our killer little speaker sounds dull and lifeless at low levels - you have to hit it with a lot of power before it "wakes up"!

Now this isn't really a change in resolution per se, but it can sound like a change in resolution, because the treble range - which conveys the little details and nuances - is subdued relative to the rest of the spectrum at modest SPLs.

What would happen if we didn't pad down the tweeter by that extra 3 dB? Well it would sound correct at modest levels, but then at high levels it would be too bright, and especially so on peaks. (The correct solution, imo, is to do a better job of matching up the real-world thermal characteristics of woofer and tweeter... but then this hypothetical example would fail to illustrate my point.)

As mentioned early on, I have made some simplifying assumptions here, but the general principle is valid: When you have a large discrepancy in the thermal headroom of the woofer and tweeter, the spectral balance may change significantly as the volume level changes significantly.

Duke
dealer/manufacturer

Sorry I wasn't clear, Bob! Let me try again.

First a bit of background that I left out: Thermal compression is caused by the voice coil heating up which causes its resistance to rise (there are other bad things that happen as well, but this is the main one). As the resistance rises, progressively less wattage goes into moving the voice coil, and progressively more wattage goes into further heating the voice coil, so it's a viscious cycle. Normally thermal compression is negligible until we get up to maybe 1/10th the driver's RMS power handling, then it starts to accelerate rapidly, often reaching 3 dB at full rated power. In the example I gave in my post above, the woofer is subject to significant thermal compression but the tweeter is not.

It might be easier to visualize how this can make the speaker's spectral balance change with level if we start out at low SPL, using the same 83 dB, 50 watt woofer and 90 dB, 50 watt tweeter as before:

Let's say we start out with woofer and tweeter level-matched at 1 watt (83 dB). Since the tweeter is 7 dB more efficient that the woofer, we have padded it down by 7 dB in order to match the levels.

Now if we go up to 50 watts, we would expect both the woofer and tweeter to deliver 100 dB. But the woofer in our example loses 3d B to thermal compression (due to the voice coil getting very hot), so it only gives us 97 dB, while the tweeter is still delivering 100 dB. So the speaker sounds bright at high SPL.

If we wanted to "voice" the speaker to sound correctly balanced at 50 watts, we would want the tweeter to also deliver 97 dB. We do that by padding the tweeter down by an additional 3 dB.

But this additional 3 dB of padding on the tweeter needed to level-match at high SPLs results in the tweeter being -3 dB relative to the woofer at low SPLs, where thermal compression is negligible.

If it's still unclear, let me know and I'll try again.

I first became aware of thermal compression when listening to a well-respected three-way speaker (which cost more than my car) that sounded overly laid-back at at low and medium levels, magnificent at fairly high levels, and forward and harsh at very high levels. In this case, the woofer and tweeter were suffering from thermal compression, but the midrange driver was not (it was the "showpiece" of the system). Since then I've noticed the effects of thermal compression on spectral balance in more speakers than I can remember. It's most noticeable when a speaker gets bright or harsh or too forward-sounding at high SPL, and is most common when a high-efficiency tweeter is paired with a low-efficiency woofer.

Duke

Vapor1, I don't think I've ever seen published Klippel data on thermal compression. An explanatory graph on Klippel's website implies that thermal compression in the neighborhood of 3-4 dB is normal:

http://www.klippel.de/measurements/nonlinear-distortion/compression-of-fundamental-components.html (first graph on the page)

If pushed hard enough, even the top tier drivers you refer to will exhibit significant thermal compression (assuming something doesn't melt first - which I guess would be permanent total compression!). But they may be rated conservatively enough that thermal compression is negligible at their rated input power.

Earl Geddes on the subject, from the text "Transducers", page 241: "Thermal effects in transducers do not generate distortion byproducts, but they do distort the frequency response - i.e. they cause severe linear distortion." He then goes on to discuss both short-term and long-term thermal effects, the latter including the effect of magnet heating on the BL curve.

Here is some further clarification from Earl on short-term thermal effects, from a DIY audio thread: "Awhile back I did some calculations of thermal rise in a VC. It happens almost instantaneously. This means that the VC resistance is being modulated at a very fast rate. The magnet heating and other effects take much much longer. But I became interested in the effect of temperture rise of the voice coil since this happens very fast.... The temperature problems occur almost instantaneously with the large signals even if we don't hear them as loud... These thermal issues may occur even at low levels if the signal's dynamic range is sufficient."

[ubergeek warning] There are two primary heating events at play in thermal compression: That of the voice coil, and that of the motor structure. We find evidence of very rapid voice coil heating in the generation of subharmonics at low frequencies and high power levels. Non-linear theory says that subharmonic generation is impossible in a time-invariant system. So there must be a time-variant factor, and that could well be voice coil heating on a timescale comparable to the period of a low frequency signal. Which is pretty darn fast. Geddes is again my source. [/ubergeek warning]

EighteenSound, a high-end Italian prosound driver manufacturer, is among the few companies that publish thermal compression data on their drivers, and my ballpark estimates of 1 dB @ 1/10th the rated power and 3 db @ the rated power are largely based on their published data.

From JBL's FAQ page: "Some speakers may exhibit 3 to 6 dB of power compression." Tom Danley published a paper a while back showing power compression of drivers operated at their rated power to be generally in the 3-6 dB range. (I'm guilty of using the terms "thermal compression" and "power compression" interchangeably; thermal compression is the primary component of power compression.)

On the other hand, in 2006 Keith Howard published an article in Stereophile which seemed to disprove the significance of thermal compression based on his measurements. (Earl disputes Keith's findings and technique, which did not include either a direct measurement of the voice coil temperature or of the frequency response.) So anyway there are two sides to the issue, but I think the more professionally qualified opinions fall on the side of thermal compression being a potentially significant issue, even if it's not an issue in all cases.

Two conventional but very effective defenses against thermal compression are high efficiency and large voice coil diameter. High efficiency = less wattage needed for a given SPL = less heat, and large diameter voice coil = more thermal mass = greater thermal inertia, and more surface area for better cooling. Of course using multiple drivers helps in both areas.

That being said, it is quite possible for a well-designed (top tier) speaker with a smaller voice coil diameter to incorporate features that give it very good thermal characteristics. I believe that you use Acoustic Elegance woofers in some of your models, and they are superb in that regard (and in many others). That drastically extended pole piece wicks away heat right where it's needed the most - at the forward edge of the voice coil (which is where voice coils tend to burn out because of insufficient local heat sinking).

Duke

Hi Bob,

Good catch! I mis-used (and mis-spelled) the term "vicious cycle"; it only becomes such when a human enters the loop and keeps turning the volume control up higher and higher because he's not hearing the volume level increase as much as he expected.

Duke

My apologies to Margot also... sometimes the nerd in me takes over, and this time it ran rampant.

Duke