There are multiple mechanisms by which powert compression can occur, and minimizing them just about always favors high-efficiency designs. This will not be a textbook-comprehensive look at the topic, and I will make some generalizations along the way.
There are two main types of thermal compression, both originating with voice coil heating, and we’ll come back to them in a minute.
Flux modulation is a compression mechanism which occurs because the permanent magnet’s field strength is modulated - temporarily weakened - by interaction with the magnetic field induced in the voice coil by the amplifier signal. For a given SPL, in general the more powerful the motor (and therefore in general the more efficient the speaker) the less the magnetic flux is modulated. Faraday rings in the motor can reduce flux modulation; alnico motors are inherently relatively immune to flux modulation (and neodymium drivers less so); and field-coil motors (electromagnets) are effectively immune to flux modulation (because the perpetual current from the power supply instantaneously restores the magnetic flux).
Mechanical compression can also occur as the suspension system becomes non-linear at long excursions, and/or the voice coil exceeds its linear excursion limits. Big drivers are less likely to get into mechanical compression than are small drivers, and since big drivers tend to be more efficient, there is again a correlation between efficiency and low compression.
And now for the two types of thermal compression: As the voice coil heats up its resistance rises, and the voice coil in turn heats up the magnet over time, which reduces the magnet’s strength. When the magnet cools back down, its strength returns. Alnico magnets are relatively immune to thermal compression UNLESS they are overheated, and THEN they unfortunately will permanently lose strength. The greater the thermal mass of the voice coil and magnet, the more gradual this heating. And the less wattage required to reach a given SPL, the less heat there is in the first place. Compression due to the magnet heating up and losing strength takes a while to set in, and it takes a while to go away. This topic has been studied because it matters a lot in prosound.
Imo the most interesting and audibly significant thermal compression effect arises from the rapid heating of the voice coil due to high-power music transients. This heating is instantaneous. So a 100-watt peak is (to a crude first approximation) like touching the voice coil with a 100-watt soldering iron that transfers heat instantly. So there will be an instantaneous spike in the voice coil’s temperature and resistance, which dissipates fairly quickly to the surrounding air and motor assembly, but the onset is still much faster than the dissipation. What can happen is this: Since the PEAKS are where that instantaneous heating occurs, it is the PEAKS which are compressed the most. We might call this effect "thermal modulation", to distinguish it from the more long-term "thermal compression" which includes the aforementioned reduction in motor strength due to the magnet heating up over time.
Unlike thermal compression, "thermal modulation" has not been studied because it’s not a critical factor in prosound (where the financing for such studies usually comes from), and because there is insufficient incentive to finance such a study coming from the consumer audio side.
Floyd Toole had this to say on the topic, in a conversation with me on another forum:
"The audibility of power compression in its many variations probably could use some more research to define what is audible and what is tolerable. The magnet heating that you describe is important in pro audio sound reinforcement systems where the loudspeakers are required to work at or close to their design limits for long periods. Such heating and cooling has a very long time constant. This is not the case in most home systems. Although the modification of motor strength through magnet heating is a factor, most of the audible effects are from voice coil heating, which has a much shorter time constant. I just saw a test of a high-end audiophile speaker that in going from an average level of 70 dB (loud conversation, background music) to 90 dB (a moderate crescendo, or foreground rock listening) lost about 4 dB in output over about 3 octaves in the mid-high-frequency range. It became a different loudspeaker at different listening levels." [emphasis Duke’s]
In other words, the speaker Toole measured had a midrange driver which was subject to significant thermal modulation effects, while the woofer and tweeter were not. So 20 dB crescendos were compressed by 4 dB in the midrange region. Yuck!
Years ago Stereophile published an article which supposedly "debunked" the "myth" of thermal modulation. Their measurement procedure was flawed in that they did not sample the voice coil temperature at the instant of peak power, but rather they sampled voice coil temperature at regular time intervals and then averaged the results. So whatever was happening at the peaks was not captured.
Anyway as you can probably see, the less wattage required to reach a given sound pressure level, the less thermal modulation effect on the peaks. And also the greater the thermal mass of the voice coil, the less thermal modulation for a given power level. Therefore in general, big, high-efficiency prosound-style drivers have a significant advantage over small, low-efficiency high-end audio drivers in this area.
If you have ever heard a speaker whose tonal balance is different at different volume levels, this is probably because the thermal modulation effects are different for the various drivers. The designer has probably "voiced" the speaker to sound best at a particular volume level. Any driver will eventually run into thermal (or mechanical or magnetic) compression issues, but big efficient prosound-style drivers will have similar thermal characteristics over a much wider SPL range, so they are more likely to retain the same tonal balance from very low to very high volume levels.
Duke
so-called hardcore technical speaker designer
