Originally Posted by Steve1981
I'm a bit curious, is there anything in the measurement suite by Josh Ricci that would illustrate these differences you're referring to? As it stands, he does fairly comprehensive measurements including FR & THD at various drive levels, CEA 2010 testing, and waterfall plots/spectrograms. From what I can gather based on the measurements he did of an FV15HP, while it is a fantastic sub for the money, I don't really see anything in the measurements that indicate "ohh yeah, that's a servo sub, so it's much better than an Epik Empire or SVS PB13U". If not, have you talked to him and others about adding a measurement that would better illustrate the advantages to people like me that like fancy graphs?
Steve, have you ever played with oscilloscope? When you input the sine wave or square wave, or anything repetitive like that, the scope can easily capture and display it. Now if you feed the music signal to it, it shows something like random signal. Is it really random? (This give you another direction to explore, how to differentiate two spectral plots and reconstruct them back to the original signal? You can't. If you can't, what information is lost?) So the first problem engineers run into is how to select a convenient signal test for them. Can't they just use music signal, instead of sine waves? This leads to the first problem. How creative are we with test methods. Some methods are century old. A lot of them are out of convenience when the equipments are primitive. But they stick. So people use them till today. That is one aspect of the issue. To address the problem, there are tests being added all the time. But no matter what they do, they are always limited by repeatibility and feasibility. In the end the tests are not really that different from one another and no new information is revealed. So we seem to be stuck in the dilemma that we can hear the difference, but there is no clear tests to demonstrate that. But for subwoofer at least, the problem is that the same response cannot be repeated over time, which is called memory effect.
Now, if you read any university textbook of the course "Signal and System" (in Electrical Engineering), it defines the linear system as "time-invariant" system and linear property: F(ax+by)=a*F(x)+b*F(y), scaling and superimposition. The first part "time invariant" is convenient forgotten in all discussion on this forum. The system who cannot maintain time invariant characteristic is either random or with memory effect (meaning the current state depends on the history of the states). In the test done by Ricci, there is a plot "LONG TERM POWER COMPRESSION). Almost everyone read it wrong. Actually, all designers with all their life designing nonservo subs read them wrong because they don't have alternative but live with power compression. It is indication of a memory effect. When you can compare all these plots with DB increments, the thermal memory effect is hidden there. Our subs perform far better than others until you hit the "ceiling" by amplifier power, ie clipping. If you overlay these plots, you will see the portion above 40hz is almost identical, whereas all nonservo subs has drop around 60hz to 100hz as SPL level goes up. The cause of that drop is because voice coil temperature has gone up during each sweep. Most people read this plot as static plots, meaning if the signal level drop from 105db to 90db again, it will have the same plot as 90db plot. This is never true for nonservo sub. To demonstrate that, I asked Ricci to repeat the baseline 90db plot again and compare that to the first 90db plot. It is best done with no time left between those two "consecutive plots". But there is a difficulty here because it takes time for the computer to crunch the number after a sweep so during that time, the voice coil has cooled down quite a bit. Some may ask the discrepancy below say 20hz for one port, that is due to compression from high air velocity of ports and that has no memory effect. Ilkka has tested our sealed version kit and it that plots it shows perfectly identical plots until it hits amp clipping. Secondly, if we can completely remove the computer crunch time and do two sweep back to back, one will find the 105db plots is not completely "repeatable". Each additional sweep increases the voice coil temperature and produces a slightly different plot.
To further demonstrate the idea, we have some illustration of memory effect on our website:
and its spectral plot is
Look at how wide the 60hz base frequency is. It is indication that the signal is no longer fixed at 60hz. Does that translate to blurry sound?
One the other hand, our servo subs:
and spectral plot:
This is just a beginning of a demonstration. Now chop these waveform into chunks and space them, what kind of waveform it will produce from the sub? With the same rate of decrease in output as above? Of course not, because between two chunks of waveform, now the voice coil can cool down again and the rate of cooling down depends how far the chunks are spaced. It almost like create a micro rhythm within these steady state signals. If you play signal with real rhythm, it creates a micro rhythm on top of music rhythm. Confusing? Probably more to our ears than our brain
Now in the space between chunks, let us fill in with small level signal. That is exactly what I have covered in another thread with this plot:
To the right side of this plot is a repeat from the left again. What it will do is as the temperature cools down, it will do the opposite of compression, it will do "expansion" over time until it hit the large signal again. The signal strength of the onset of that large signal is almost unpredicable as it depends on the voice coil temperature. So in the real signal we just do this type of yo-yo all the time and we cannot capture it directly with all the "conventional" tests. One needs to be creative and think outside the box to come up with tests that can expose this. Ricci cannot do these because his test is far field measurement, it has a lot of background noise in there. His hands are tied by the so-called "standard". But I don't have that limitation.
This is just the demonstration of memory effect. Another part of the puzzle is how do diffentiate the distortion coming from mechanical source from the electronics source. That is another very complicated procedure. But mechanical distortion tends to be higher order. So one really needs to pay attention to higher order distortions.
Hope the above answers your question.