advertisement


How do the 'non-subjectivists' choose their hi-fi systems?

I've heard enough - it's clear to me that certain individuals here who like to portray themselves as experts are actually anything but.
 
Since a non-linear system can be literally anything, I'm assuming we're talking about reasonably well-behaved audio devices without unstable states or other crazy features. Also, we don't have to actually model the non-linearities as long as we can show that they are too small to be audible.

That's fairy snuff in the main. I'm an (ex-)academic so nit picking is a habit. :) Nevertheless, remember that some people were surprised by TID for example. And an unwary designer can be caught out by a 'wrong' combination of feedback, nonlinearity, and some state memory cause by some capacitors, say. Shouldn't happen with sensible designs, though. But it means people need a pesky injuneer to know and check. Also note the way DSD/SACD was nearly caught out by this.
 
Have we got the answer yet ? 48 pages and a lot of waffle ... a simple question that could easily be answered without all the thread crapping, so can we get to the point please :rolleyes:.
 
By device I mean anything with an input and an output. It could be an amplifier. Or a cable.

The transfer function is the mathematical relationship between the signal entering the input and what comes out the other end. It is an expression of everything the device does to the signal. It doesn't matter whether the signal is a sine wave, a frequency sweep, or a piece of music.
Thank you for the response. I infer that the transfer function is assumed not to be frequency dependent, or that it remains constant over time. Is this a valid assumption? Might the function vary, for instance if the device is operating outside its design envelope (eg, clipping, or perhaps a power supply insufficient to maintain output under an extreme demand), might this affect the transfer function? Is it absolutely consistent over the frequency range of the unit, or is it sufficient that it remains within an acceptable range?
 
That's fairy snuff in the main. I'm an (ex-)academic so nit picking is a habit. :) Nevertheless, remember that some people were surprised by TID for example. And an unwary designer can be caught out by a 'wrong' combination of feedback, nonlinearity, and some state memory cause by some capacitors, say. Shouldn't happen with sensible designs, though. But it means people need a pesky injuneer to know and check.
And how would you go about doing this checking, with test signals or with music?

Also note the way DSD/SACD was nearly caught out by this.
DSD is extremely non-linear, and deliberately so, with entire books written on the topic.
 
Thank you for the response. I infer that the transfer function is assumed not to be frequency dependent, or that it remains constant over time. Is this a valid assumption? Might the function vary, for instance if the device is operating outside its design envelope (eg, clipping, or perhaps a power supply insufficient to maintain output under an extreme demand), might this affect the transfer function? Is it absolutely consistent over the frequency range of the unit, or is it sufficient that it remains within an acceptable range?
Frequency dependency would be included in the transfer function. I think we all assume that our audio gear isn't changing over time, disregarding any warm-up period. Clipping and power supply sag can be modelled, but it's simpler if we stay within the design parameters. It's not terribly interesting exactly how things go wrong so long as we can prevent them going wrong in the first place (that is, by using a sufficiently large amp).
 
Frequency dependency would be included in the transfer function. I think we all assume that our audio gear isn't changing over time, disregarding any warm-up period. Clipping and power supply sag can be modelled, but it's simpler if we stay within the design parameters. It's not terribly interesting exactly how things go wrong so long as we can prevent them going wrong in the first place (that is, by using a sufficiently large amp).
Fine, that’s what I thought you’d suggest, given what you posted in your explanation.Thanks. But sometimes music takes devices outside their design parameters, and we don’t always notice, nor might we always anticipate this before purchase. I suspect this is one of those things that emerges over the sort of extended listening I prefer.
 
And how would you go about doing this checking, with test signals or with music?

I wonder if Jim is referring to this:

zFbf0tG.png
 
Science isn't quite so volatile. Reading what you wrote, one gets the impression that scientific theories are like the spring collection of a fashion house, constantly and unpredictably changing with the whims of the designers. .
You're missing my point again. I'm not saying that is the case, I am saying that is a common perception, so trust is lost. You can explain the details of scientific research and conclusions until you are blue in the face, but if trust is lost then it means nothing.
 
Again, your main point isn't clear, so I'll seek to resolve it:

Can you perhaps point to two or three AES journal papers or conference reports which support your contention and that it is need for all tests for all audio purposes? We can then examine this in the relevant audio context, not simply as a page on stats plucked from a wiki, or on the basis of your assertions. FWIW my understanding from having read various audio comparison tests is that the methods vary according to the purpose of the test, etc, as appropriate- as I've pointed out in the past.

Or is your point that pro audio engineers, etc, never get this right?
Isn't it your turn to demonstrate that the said AES papers are following standard DOE protocol?

You write as if you are familiar with the research and the methodology.

It should be easy for you to show how the basic issue is dealt with and what "the workings" are. I offered a link to a standard DOE protocol to refresh your memory, since in your posts in this thread you appear to need it.

So far I see a lot of slow motion hand waving and eye rolling.
 
Fine, that’s what I thought you’d suggest, given what you posted in your explanation.Thanks. But sometimes music takes devices outside their design parameters, and we don’t always notice, nor might we always anticipate this before purchase. I suspect this is one of those things that emerges over the sort of extended listening I prefer.
It is trivially easy to determine the point where an amp starts clipping. If it doesn't clip with the highest input level your source can output (also easy to determine), then you're good.
 
It is trivially easy to determine the point where an amp starts clipping. If it doesn't clip with the highest input level your source can output (also easy to determine), then you're good.
Just like it's "trivial" to determine when a metal yields in an Instron test. Except when it's not - when a material has an ill-defined elastic behavior, suffers from creep, etc.

It is my understanding that some design, like SETs have some of this complex behavior.
 
And how would you go about doing this checking, with test signals or with music?


DSD is extremely non-linear, and deliberately so, with entire books written on the topic.


Agreed. However it actually came to market before some of the problems became clear. The best-know symptom being that the first players had a claimed 100kHz audio bandwidth, but this was swiftly rolled back to 50kHz because of the ultrasonic noise problems the wider bandwidth caused. They also discovered late on that allowing the audio to go above about -3dBFS could easily 'lock up' the demondulator. (Their term for the 'DAC'.) So had to limit the levels modulated (their term for the 'ADC'.)
 
I wonder if Jim is referring to this:

zFbf0tG.png

That was when he noticed it and got known for it, yes. But at least some engineers had known about it for years. As a result I was surprised by the fuss.

But yes. the problem is that the *rate of change* of an input may alter the nominal transfer function. And this can in principle happen at various points in a system. Add feedback to that and some nonlinearity and you can, quite literally, get chaotic (in the mathematical sense) behaviour. In non-chaotic cases the result is that the transfer function becomes a differential nonlinear equation, which can be fun to handle. :) Chaos is another breed of animal.

In some ways its an analog parallel with the problems with 1 bit high order processes.

Well designed kit produced by someone who know about the above, and then used within the intended limits, should be fine. But in other cases...
 
That was when he noticed it and got known for it, yes. But at least some engineers had known about it for years. As a result I was surprised by the fuss.

But yes. the problem is that the *rate of change* of an input may alter the nominal transfer function. And this can in principle happen at various points in a system. Add feedback to that and some nonlinearity and you can, quite literally, get chaotic (in the mathematical sense) behaviour. In non-chaotic cases the result is that the transfer function becomes a differential nonlinear equation, which can be fun to handle. :) Chaos is another breed of animal.

In some ways its an analog parallel with the problems with 1 bit high order processes.

Well designed kit produced by someone who know about the above, and then used within the intended limits, should be fine. But in other cases...
Thanks Jim, if I understand correctly (not, by any means, a given...) this is the sort of potential problem I was trying to articulate in post #945 above.

If so, then this makes mansr's assertions about the predictability and knowability of system behaviour by being able to know the transfer function, somewhat more complicated in the real world?
 
Agreed. However it actually came to market before some of the problems became clear. The best-know symptom being that the first players had a claimed 100kHz audio bandwidth, but this was swiftly rolled back to 50kHz because of the ultrasonic noise problems the wider bandwidth caused. They also discovered late on that allowing the audio to go above about -3dBFS could easily 'lock up' the demondulator. (Their term for the 'DAC'.) So had to limit the levels modulated (their term for the 'ADC'.)
The ultrasonic noise should have been no surprise. It's impossible to make a modulator without it. 1-bit sigma-delta DACs had been around for quite some time too. The only new thing was using that signal in the distribution format. On the second part, you are almost right. It's the modulator (ADC/encoder) that goes crazy. The DAC is just a plain old linear low-pass filter.
 
Thanks Jim, if I understand correctly (not, by any means, a given...) this is the sort of potential problem I was trying to articulate in post #945 above.

If so, then this makes mansr's assertions about the predictability and knowability of system behaviour by being able to know the transfer function, somewhat more complicated in the real world?
If you exceed the bandwidth limit or maximum signal level of an amp, bad things can happen. Don't do that.
 


advertisement


Back
Top