I was thinking of observing 2 or three stationary signal for IM while at the same time the amp is handling a 4th low frequency signal, wildly varying in amplitude. FFT with long (seconds) averaging of course. This to vary and draw more juice out of the amp - like when the 16ft pedal comes on during an organ piece or when the double basses come in in the beginning of the 3rd movement in Shostakovich 8th (Petrenko / RLP / Naxos). Like small waves carried by large ones.
//
//
To me that looks rather contradictory - the live musical performance is non-stationary but the recording, being a signal is. An ECG signal isn't a signal because its not produced by a stationary process even though its not a recording of something that's already happened.
So then it looks like the word 'stationary' in this context is practically useless, no?
So then it looks like the word 'stationary' in this context is practically useless, no?
It is wrong and without finesse in the first place, medium, noise and signal mixed up in an unholy mess, senseless mathematical terminology. We have DC and all the rest is AC, it is stupid to complicate things unnecessarily, no? The subject is of trifling importance in engineering anyway.
You have learnt your lessons thoroughly, but your knowledge has little relevance to the audio field.
Now you change your mind about thermal distortion
For easy to analysis, input of amplifier should be time invariant. This is how engineering work. We modeled the system for easy to analysis and usually it work almost 100% accurate. But nothing in the world is time invariant. They all changing every time, they are aging. Even living things will be die.
So is the whole time (in)varient discussion, isn't it? All red herrings?
You could also use "steady state".
A deterministic signal is said to be stationary if it can be written as a discrete sum of cosine waves or exponentials, i.e. as a sum of elements which have constant instantaneous amplitude and instantaneous frequency. In the random case, a signal {x
} is said to be wide-sense stationary (or stationary up to the second order) if its variance is independent of time.So abraxalito you're using the wrong definition, as said above a stationary signal is periodic. Now, one could set the song on repeat to make it periodic and nitpick that "a discrete sum of cosine" is still legit if you have a billion of them, in which case it would fit the definition of "stationary", but that's splitting hairs.
Andrea>
When trying to measure or understand something, you need an experiment and a model that are as simple as possible to be usable. But if it is too simple, maybe it doesn't model or allow to measure the thing you're trying to study. For example, a multimeter can't answer the question "does the amp have sticky clipping?"
If the experiment is more complex than necessary (for example, using music instead of a simple test signal) then when a result changes, the number of potential causes to explain why it changed is larger, so it's more difficult to know what happened.
For example, I fixed an old Marantz. I didn't test it with music, I tested it with a multimeter. The +24V power supply was +5V, all the current sources in the power amp were off, and it had no output stage bias. That narrowed it down pretty quickly to a particular series of fusible resistors which were the common denominator in these circuits, and they all had failed open. Looking at the output with a scope gave different information, which was much less effective at narrowing it down.
OK, so more on "stationary". Since stationary signals have infinite length, they don't exist outside of math. But... if the signal is much longer than all the time constants in the amp, then we can assume it has settled to steady state conditions, so it will be a good approximation of a stationary signal. So in the context of test and measurement, the definition of what is a good approximation of a stationary signal depends on the time constants of the device under test.
Since most testing is done with resistive loads, the typical test ignores all the time constants in the load (state of flux in inductive load, loudspeaker state variables, speed of moving mass, air pressure in enclosure, airspeed in vent). And signals usually have long enough at constant power that the temperature of output devices and the bias loop have time to settle. So the usual tests use a good approximation of a stationary signal from the point of view of the DUT.
My point is to
1) Do non-stationary tests, for example how does the distortion change on a low power signal right after a high power transient, which of course implies a way to measure distortion that doesn't use FFT over a long time window
2) Not use only zero-centered signals which miss the interesting parts of the crossover when amplitude goes down
3) Also extract the small-signal parameters at each operating point, not just at zero voltage/current output. That's what you do in SPICE when you simulate phase margin versus output current, to make sure it is stable over the whole operating envelope, for example.
These guys already do similar tests, otherwise it wouldn't work. Also you can't put a DSP in the feedback loop of a class-AB amp, so no nonlinear control algos.
Ah, okay, right!
Looking at harmonics on a FFT, it's difficult to know where they come from.
But looking at the transfer function, it's a lot easier...
It's like an output stage wingspan diagram (gm vs current) makes problems obvious, whereas a FFT does not. But wingspan is usually done at DC which is not very relevant. It's much more interesting when done with AC.
Of course it should be use measurement. You misunderstand.
In engineering you should make right model and choose what kind of measurement. In sound perception, THD or FFT measurement at one frequency and one power level is not enough. It is can not describe of thermal distortion for example.
Peufeu, i like your point of view. Many engineer assume their engineering knowledge as dogma. It is only the truth to describe the reality, but they can not describe the reality 100% accurate. It is depend how we modeled the reality.
- Status
- Not open for further replies.