Originally Posted by Norixone
What I would like to know from Brian is how much the servo design he has implemented in the Rythmik subs can actually compensate for thermal compression.
Although I agree with most of his statements about DSP versus servo, there is a way to compensate for thermal (or memory effect as Brian likes to call it) with DSP and some producers do implement a thermal control to work with the DSP (Revel for one).
Servo is based on remote sensing. So the DCR of the driver is not even in the close-form equation that determines the output. So the compensation is instantaneous and its compensation range is determined by the loop gain. On the other hand, DSP adjustment is at specific time instances or intervals (for instance every second). In addition, it is a "control" meaning it has to know what the resistance of voice coil is before it can control. In the servo, there is no need to know the DCR. So the DSP control loop is: measure -> control -> measure -> control -> measure -> control.... In servo, it is ...............................
Moreover, if the servo used in the Rythmik had all the benefits claimed, I do not see way we would need a limiter switch. The servo circuit could simply act as a limiter as well.
Servo does not know if the driver or amp is at its limit.
BTW, Enrico quoted me for a comment that I made a few years back. I'd like to explain a bit more. The unit-to-unit variations are about T/S parameter variations. Very often we take the T/S parameters off the manufacturer's website and think those apply to all units. But in reality as many have done it before, if someone buys 4 units of the same driver model and measures the T/S paremeters, he actually gets 4 sets of different T/S parameters and each set changes over time and changes between small signal vs large signal. Thilo (TC Sounds) told me a story that he delivered drivers to one of his customers and customer rejected those drivers based on T/S paramets out of spec. It turns out, the difference is due to the fact that they measured at different signal levels. After that, Thilo measure parameters at the signal level specified by the customer.
Now back to the "thermal memory" effect. Here is a plot that I use to study thermal memory in non-servo subs. Here I use "two signal levels" to simplify discussion so that we can see output changes much easier. In real world, the signal level is constantly changing. As you can see the output of the large signal is gradually decreasing due to temperature rise in voice coil in what I call "heat-up" period. After we switch to the small signal, at about 3.1sec, the opposite happens that the output gradually increases, in what I call "cool-down" period. We call it memory effect because it is not repeatable. If we run a waveform like above in a round robin fashion, the output of the large signal is different with each repetition because the cool-down period may not be sufficient to bring it back to room temperature. If we take the correlation of the output variation vs the input signal, it is a delayed correlation as temperatutre change takes time to build up and go away. For instance at the moment when the signal changes from large to small, does it go back to room temperature? No it does not. In fact, at that very moment (3.1S), it has the largest temperature variation. But where is the center of the large signal? It is at 2.8S which is the center of the large signal period. That is why I called it memory effect. It is not just a static thermal compression. The folks with DSP say they solve the problem. But in fact, they misunderstand the problem.