[anoobis]

It can be difficult to assess acoustic changes over long timescales, especially if the changes are gradual. Mental state (inc. ear fatigue), headphones, audio memory etc. all come into play.

My initial thought was to approach measurement as a signal analysis problem. Leave a source and amp. running for several hundred hours, playing white or pink noise. Send the signal from the HP out to a computer input. Split the recording (in to, say, hourly chunks) then look at the spectra (FFT). The frequency content of each hour should be very similar, so significant changes would be due to changes in the frequency response of the source or amp. Yes, no?

Then I realised this may be over-doing it. Maybe simply removing/reducing audio memory effects would be enough. So, record a sample (noise or music) when the unit is new, then the same sample at intervals. Simply listen to them. I think that the ability to compare the unit at different stages while eliminating 'human effects' could be interesting.

I realise there is the effect of recording and playback but at least these should be constants.

 

[op2003]

Quote:

Originally Posted by anoobis /img/forum/go_quote.gif ... My initial thought was to approach measurement as a signal analysis problem. Leave a source and amp. running for several hundred hours, playing white or pink noise. Split the recording (in to, say, hourly chunks) then look at the spectra (FFT). The frequency content of each hour should be very similar, so significant changes would be due to changes in the frequency response of the source or amp. Yes, no? ...

I would be really interested to see a test like this as it is quite objective. How would you do the spectra analysis?

 

[JaZZ]

The most promising approach would be measuring harmonic distortion in my book. I'm sure burn-in reduces it (or alters its spectrum), at least with some devices.
.

 

[op2003]

Why do you think so?

 

[anoobis]

Quote:

Originally Posted by op2003 /img/forum/go_quote.gif I would be really interested to see a test like this as it is quite objective. How would you do the spectra analysis?

I'm not sure I follow which part you're asking about. However, I did forget to mention I was thinking of a HP out into a computer.

From there, DSP (fast Fourier transform) and that's as far as I got! Eyeball in the first place, take differences. It depends on what sort of differences/effects you think there'll be.

 

[JaZZ]

Quote:

Originally Posted by op2003 /img/forum/go_quote.gif Why do you think so?

I doubt that frequency response varies before and after burn-in (should be near perfect anyway), hence a spectrum analysis won't reveal anything. So what's left?

Even if the reproduction of complex signals and transients changes, it's doubtful if the change will reflect itself in the frequency spectrum.

Of course I could be wrong.
.

 

[op2003]

Quote:

Originally Posted by anoobis /img/forum/go_quote.gif I'm not sure I follow which part you're asking about. However, I did forget to mention I was thinking of a HP out into a computer. From there, DSP (fast Fourier transform) and that's as far as I got! Eyeball in the first place, take differences. It depends on what sort of differences/effects you think there'll be.

That was exactly what i wanted to ask you. If you do a setup like this then you have additional components (PC soundcard at least) which could be subject to burn-in, too. So it would be hard to know, which components burn-in you're actually measuring, if there is a change at all.

Quote:

Originally Posted by JaZZ /img/forum/go_quote.gif I doubt that frequency response varies before and after burn-in (should be near perfect anyway), hence a spectrum analysis won't reveal anything. So what's left? Even if the reproduction of complex signals and transients changes, it's doubtful if the change will reflect itself in the frequency spectrum. Of course I could be wrong. .

Sounds quite logical to me.

Does anybody know, if there has been a scientific approach of measuring burn-in of electronical components at all? I did some quick "google research" and found a quite different definition of burn-in from an industrial point of view here:

"The purpose of the burn-in process [...] is to weed out "infant mortalities"—as seen in the first portion of the well known "bathtub curve" of failure rate versus operational time. These latent, early life failures are attributable to intrinsic gross faults within the bought-in components, assembly errors, and faults induced in components by inappropriate handling (e.g., ESD damage). It should be noted that there are certainly no absolutes in the world of reliability testing, only probability and confidence levels for large populations."

So the purpose of burn-in test, by this definition, is to find failures and not to "mature" the components, as they are obviously expected to operate within certain limits within their whole lifetime.

 

[nick_charles]

Quote:

Originally Posted by anoobis /img/forum/go_quote.gif I'm not sure I follow which part you're asking about. However, I did forget to mention I was thinking of a HP out into a computer. From there, DSP (fast Fourier transform) and that's as far as I got! Eyeball in the first place, take differences. It depends on what sort of differences/effects you think there'll be.

Record the signal using something like Audacity then plot the spectra, export to a text file and load up as external data in Excel from this you can either graph the frequency response or do mathematical comparisons. I used this technique to examine differences between cables. The only thing to remember is you have to be really critical with alignment, trimming and level setting to make sure the samples are comparable.

Quote:

Originally Posted by JaZZ /img/forum/go_quote.gif The most promising approach would be measuring harmonic distortion in my book. I'm sure burn-in reduces it (or alters its spectrum), at least with some devices. .

Quote:

Originally Posted by JaZZ /img/forum/go_quote.gif I doubt that frequency response varies before and after burn-in (should be near perfect anyway), hence a spectrum analysis won't reveal anything. So what's left? Even if the reproduction of complex signals and transients changes, it's doubtful if the change will reflect itself in the frequency spectrum. Of course I could be wrong. .

Harmonic distortion wll show up in frequency spectra if you use a single frequency signal. Bung a 1Khz signal through the system and record it then do a frequency analysis on it and record the energy level at each frequency, Audacity lets you do this, then load it up into Excel and graph it, the fundamental will show up as the dominant tone and any harmonics will also show up as lesser peaks. Rinse and repeat...

 

[ericj]

My two bits: A year ago i built a Bijou all-tube headphone amp. I used NOS tubes. The instructions state that it needs about 50 hours before it settles in.

I periodically measured component temperatures and voltages at specific reference points.

Temperatures and voltages did in fact change over time. The rate of change tapered off after 40 hours, and eventually the temperatures and voltages settled out pretty much exactly where the designer said they would.

So, empirically speaking, I'm going to say that i believe that tubes at least do have this behavior, and this is probably where the concept of 'burn in' in audio gear comes from. The variance in voltages would have a measurable impact on the frequency response, for sure.

Whether the tubes are behaving that way because the getters are soaking up the last bits of gas in the tubes or because the heat causes physical changes in the metal, or both, is up for debate.

 

[anoobis]

Quote:

Originally Posted by JaZZ /img/forum/go_quote.gif I doubt that frequency response varies before and after burn-in (should be near perfect anyway), hence a spectrum analysis won't reveal anything. So what's left? .

I'm simply going off anecdotal evidence that states things like the bass quantity increases. This surely would change the frequency response? I'm also not suggesting that frequency analysis would/could measure every type of change.

By recording the signals, you could still measure things like impulse and step responses though (with suitable input signals).

Quote:

Originally Posted by op2003 /img/forum/go_quote.gif That was exactly what i wanted to ask you. If you do a setup like this then you have additional components (PC soundcard at least) which could be subject to burn-in, too. So it would be hard to know, which components burn-in you're actually measuring, if there is a change at all.

I acknowledged the fact that how you record the signal (PC, microphone, oscilloscope etc.) will play a part but are you not presupposing these are new components? More to the point, how are you going to do any measurement without any other equipment?


To an extent, both of the above quotes are reasons for simply listening to the recorded samples. When it comes down to it, you will be listening to the components. The original question posed is, has the response of the system changed significantly over time? You should be able to ascertain if it has changed by listening to the samples, if not exactly how it has changed. And the beauty of the former is that the playback equipment is irrelevant.

Quote:

Originally Posted by nick_charles /img/forum/go_quote.gif Record the signal using something like Audacity then plot the spectra, export to a text file and load up as external data in Excel from this you can either graph the frequency response or do mathematical comparisons. I used this technique to examine differences between cables. The only thing to remember is you have to be really critical with alignment, trimming and level setting to make sure the samples are comparable.

I didn't get in to exactly what I would look for because it depends to some extent on what the initial evaluation shows. I wasn't saying I don't have a clue what I'd do!

Decent suggestion on Audacity but Excel? Hell no! Admittedly I've not used any recent versions but the words 'unweildy', 'slow' and 'unsuitable' spring to mind...

Ultimately you use what you have access to though.


As ericj was saying, there are plausible physical explanations for what could change and why. The two questions are i) what changes can be measured? and ii) how (if at all) do they affect the sound?

 

[nick_charles]

Quote:

Originally Posted by anoobis /img/forum/go_quote.gif I didn't get in to exactly what I would look for because it depends to some extent on what the initial evaluation shows. I wasn't saying I don't have a clue what I'd do! Decent suggestion on Audacity but Excel? Hell no! Admittedly I've not used any recent versions but the words 'unweildy', 'slow' and 'unsuitable' spring to mind...

Okay, but (apologies for sounding patronising) I repeat the need for care in making your samples. A small difference due to a lack of consistency can give you very misleading results. I thought I had found a 0.6db difference between cables only to discover that one of my 20 samples was 0.5 seconds longer than it should have been.

Sounds like a fun project.

 

[anoobis]

No need to apologise, you're simply offering advice. Good point about the sample length; 0.5s is a lot of samples in 44kHz audio.

 

[op2003]

Quote:

Originally Posted by anoobis /img/forum/go_quote.gif ... I acknowledged the fact that how you record the signal (PC, microphone, oscilloscope etc.) will play a part but are you not presupposing these are new components? More to the point, how are you going to do any measurement without any other equipment?

Normally you use calibrated equipment for scientific measurement.

Quote:

Originally Posted by anoobis /img/forum/go_quote.gif To an extent, both of the above quotes are reasons for simply listening to the recorded samples. When it comes down to it, you will be listening to the components. The original question posed is, has the response of the system changed significantly over time? You should be able to ascertain if it has changed by listening to the samples, if not exactly how it has changed. And the beauty of the former is that the playback equipment is irrelevant.

For a fun project this is probably the way to go, for scientific results this method won't work. You have to keep in mind that your own ability to hear changes from day to day, so this test would be highly subjective. You won't be able to tell, if the changes you are hearing are because of you or your system.

So again, for a fun project you can do it like this, for serious results there is probably no other way but going with calibrated equipment under controlled and repeatable conditions (environment).

 

[ph0rk]

Caps eventually die, so I could believe they had a "setting" period. I'd be surprised if it was longer than a few seconds.

A similar angle, is "warming up" gear - for solid state equipment at least, if it sound different cold I'd think it was a bad design. (Tubes are a different story, of course - one of the many reasons I prefer not to play with them).