• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required as is 20 years of participation in forums (not all true). Come here to have fun, be ready to be teased and not take online life too seriously. We now measure and review equipment for free! Click here for details.

Dummy question: amplifiers measurement

Joined
Nov 17, 2018
Messages
4
Likes
2
#1
Hi everyone!
Hope you'll excuse my english, first of all! It's not my idiom, so I'll use translator...
I would like to know why there is no mention of transient intermodulation distortion, memory distortion or thermal distortion concerning amplifiers measurement, and if those measurements would be relevant?
I also wanted to know if it was possible and relevant to measure amplifiers on transient signals (not burst), and on complex loads (like true loudspeakers)?

As you've probably understood, I'm not 100% objectivist, and I think there are differences in sound rendering between amplifiers, that the "classic" measures do not show... I'm I wrong? Thank you for enlighten me!

Best regards from France
 

DonH56

Technical Expert
Technical Expert
Patreon Donor
Joined
Mar 15, 2016
Messages
3,400
Likes
4,489
Location
Monument, CO
#2
Hand-waving loose guesses:

TIM is mostly a solved problem for many years. Better technology provided greater internal bandwidth and allowed improvements in the feedback path.

I do not know what you mean by "memory distortion". Hysteresis and component PIM (passive intermodulation distortion) is again well below audibility for most modern designs.

Thermal distortion, or bias shifts due to thermal issues (*), are very real and remain with us today though more as a consequence of poor design and layout practice, or perhaps just cost-saving. Again the effects are inaudible unless very bad. Severe distortion should show up in the longer steady-state tests.

We have asked Amir about doing more transient tests and more realistic loads. It takes more time, a different set of equipment in the case of transient tests, and a good speaker load is not a cheap thing for high-power amplifiers so thus far they have not happened (though Amir has shown some transient tests when he can).

What sort of non-burst transient signals did you have in mind?


(*) Edited to clarify that thermal distortion and thermal bias shifts are not the same thing although bias shifts can change distortion and noise levels among other things.
 
Last edited:
Joined
Nov 17, 2018
Messages
4
Likes
2
#3
Good evening, DonH56 (it's night here)
Thank you for your answers.
If you say that TIM is no longer a problem, it's one less ...
Regarding memory distortion or memory effect, I read that it was caused by dielectrics as capacitors, insulators, etc ...
If indeed transient measures can be put in place, super, maybe something will come out of it.
For the transient signals, they should be like musical signals, extended in frequency, with a lot of dynamics and harmonics, which make it possible to measure the slew rate or the rise time on all the audio band, on a realistic load.
In fact, I have the impression that we measure practically in "static" and not on the extent of the spectrum to be reproduced, but I am wrong maybe ...

Regards,
 

DonH56

Technical Expert
Technical Expert
Patreon Donor
Joined
Mar 15, 2016
Messages
3,400
Likes
4,489
Location
Monument, CO
#4
Memory effect = hysteresis and is for the most part a non-issue for audio (aside from marketing). A real effect but inaudible, at least when decent components are used (as virtually everyone does).

Nothing is really "static" because the signals are time-varying. Amir also performs multitone testing that emulates music so look at those plots in his tests. That is a particularly stressful test for many amplifiers and is a very good indicator of what music will look like. He also does frequency sweeps spanning the full audio band and power sweeps that vary amplitude.

The typical "transient" test signal is a square wave and he does not use those (requires different gear, different test setup). I and others would like to see square wave testing to check the stability of the amplifiers and look for any abnormal setting response but in practice it is unlikely you would see anything that the tests already performed do not cover, albeit with a resistive load. One thing that might not be too hard is to add a LC across the test load to introduce some reactance but again it takes more time and effort to do that. I am not sure I have read any review in years that found significant stability problems with an amp save a few "garage shop" designs.
 

amirm

Founder/Admin
Staff Member
CFO (Chief Fun Officer)
Joined
Feb 13, 2016
Messages
22,563
Likes
34,610
Location
Seattle Area
#5
In fact, I have the impression that we measure practically in "static" and not on the extent of the spectrum to be reproduced, but I am wrong maybe ...
All measurements are dynamic in that even a sine wave is swinging from negative to positive voltage. I also run 32-tone tests where the signal is changing a lot more:

1575325740472.png


I usually show this in frequency domain so it looks more orderly:
1575325765217.png


But what the amplifier sees is the top graph in blue.
 

amirm

Founder/Admin
Staff Member
CFO (Chief Fun Officer)
Joined
Feb 13, 2016
Messages
22,563
Likes
34,610
Location
Seattle Area
#6
If indeed transient measures can be put in place, super, maybe something will come out of it.
Let's step back and think about what a transient signal is. A textbook, ideal transient will go from zero voltage in one instant, to maximum in another. If we decompose this signal into a combination of frequencies required to create it, we get:



In other words, to get that single transient in time domain (i.e. what you/amplifier sees), you need to have infinite number of frequencies added together. By definition then, that signal will take infinite energy to create!

So a true impulse cannot exist in real-life since all real devices have limited bandwidth.

In the case of audio, we have severe, very severe bandwidth limitations. CD quality audio for example can't have any components above just 22 kHz. That is a way, way short of infinite!

Turns out we can create impulse signals -- as "illegal" as they can be -- in digital domain. In the case of CD audio, we can simply have a PCM value representing zero and in the next sample, the maximum value. Such a signal will have huge bandwidth although not infinite. The moment you attempt to play back such a signal though, the filter in the DAC will truncate everything past 22 kHz. The result is that you now have a very slow rising signal from zero to max. Same thing happens with LP sources by the way as they too have limited bandwidth.

People performing impulse testing for audio don't take into the above issues and use a generator with megahertz of bandwidth and a scope with hundreds of megahertz of bandwidth to analyze the signal. As a result, the test fixture has no association with reality of how it is used. You can get pathological outcomes this way that don't at all occur in real life audio signals.

Conclusion
Impulse testing subjects an amplifier to hugely wide bandwidth signal way, way past what an audio source can create, or you can hear in real life music. Or capture for that matter. The results then can be quite misleading with respect to assessing audible fidelity of an amplifier. Designers can utilize impulse response to verify stability of what they have built and other similar uses. But that is a different task than measuring fidelity which is what I do.
 

RayDunzl

Major Contributor
Central Scrutinizer
Joined
Mar 9, 2016
Messages
7,693
Likes
4,518
Location
Riverview FL
#7

restorer-john

Major Contributor
Joined
Mar 1, 2018
Messages
3,063
Likes
5,481
Location
Gold Coast, Queensland, Australia
#8
Transients in the real world, recorded at CD quality
Transient sound series in the real world recorded at CD quality:

1575327980889.png


Zoomed in a little

1575328060567.png


More:

1575328098811.png


More:

1575328140684.png


individual samples:

1575328190672.png


These are full swings within two samples.

Spectrum

1575328410742.png
 

RayDunzl

Major Contributor
Central Scrutinizer
Joined
Mar 9, 2016
Messages
7,693
Likes
4,518
Location
Riverview FL
#9

restorer-john

Major Contributor
Joined
Mar 1, 2018
Messages
3,063
Likes
5,481
Location
Gold Coast, Queensland, Australia
#10
What's the source?
Arguably the best, most consistent natural fast transient/impulse sound I know of. You can buy them and I can buy them on the other side of the world, and they will sound exactly the same. They cost virtually nothing.

(download the file and change the extension to .wav from .pdf and you will have your answer. :)

Recorded with a little Tascam sitting on my kitchen countertop.
 

Attachments

RayDunzl

Major Contributor
Central Scrutinizer
Joined
Mar 9, 2016
Messages
7,693
Likes
4,518
Location
Riverview FL
#11
Ok, I picked the first transient that didn't appear to be clipped.

The waves occupy at least (about) four samples by my quick scan, so around 10kHz maximum.

Resampled to 768kHz (my pseudo reconstruction filter) doesn't show anomalies

Sound is a something, probably spherical, maybe a ball bearing, dropped onto and bouncing from a hard surface.

1575330798855.png
 

restorer-john

Major Contributor
Joined
Mar 1, 2018
Messages
3,063
Likes
5,481
Location
Gold Coast, Queensland, Australia
#13
It's a ping pong ball dropped on the marble kitchen counter top.

Individual samples need to be filtered to represent what we hear out of a DAC.
Don't forget, the audio sample above has already been through the recorder's A/D front end filter before this capture. I can up the sample rate to 96KHz and record the same ping pong ball on my bench and see what we see.

Here it is. Not much difference except a little more HF extension.

1575333254783.png


Zoomed right in.

1575333334196.png


1575333158760.png


So this is a natural sound, not at all unusual and heard around the world every day by millions of people playing ping pong. Multiple 10KHz transient impulses.

Common natural sounds like this justify testing preamplifiers and amplifiers for large amplitude, high frequency transient behaviour. 10KHz fixed number of cycles toneburst repeated at regular intervals, with examination of the rise time, behaviour at full amplitude, and recovery/overshoot.

I have the same ping pong sounds somewhere on a vinyl test record, I've never looked at the waveform, but it could be fun. :)
 
Last edited:

amirm

Founder/Admin
Staff Member
CFO (Chief Fun Officer)
Joined
Feb 13, 2016
Messages
22,563
Likes
34,610
Location
Seattle Area
#14
Don't forget, the audio sample above has already been through the recorder's A/D front end filter before this capture.
That is only the case after a DAC filters that signal. In digital domain, that filtering is missing so it looks like you have step functions. Those step functions will not and cannot be there or you will be violating Nyquist. Here is a graph that shows that difference well:



You are showing the bottom samples which seems to show (to our eyes) that there are step functions. But post filtering we get that smooth sine wave.
 

scott wurcer

Senior Member
Audio Luminary
Technical Expert
Joined
Apr 24, 2019
Messages
481
Likes
835
#15
In other words, to get that single transient in time domain (i.e. what you/amplifier sees), you need to have infinite number of frequencies added together. By definition then, that signal will take infinite energy to create!

.
This is nonsense or at least a very poor way of looking at things. As I said earlier a single sample of amplitude A and zero everywhere else is a perfectly legal result in the limit of an ideal sampler and ideal brickwall anti-aliasing filter, it is a mater of theory vs limits of physical realization. Infinite energy does not figure into it.
 

restorer-john

Major Contributor
Joined
Mar 1, 2018
Messages
3,063
Likes
5,481
Location
Gold Coast, Queensland, Australia
#16
The analog signal (the sound) is picked up by the mic, goes through the preamp stage and an aliasing filter before hitting the A/D converter. That means only signals below the LPF are digitized in the first place. We all know that. The quantized values are the individual samples the A/D converter produced for valid audible frequencies during that capture.

Count the samples in a cycle (any one you like), look at the sampling rate and determine the frequency. Nobody is saying the steps come out the analog output- certainly not me. Again, we have naturally occurring high level, 10KHz transients repeated at a regular intervals in this recording of a humble ping pong ball.

Common natural sounds like this justify testing preamplifiers and amplifiers for large amplitude, high frequency transient behaviour. 10KHz fixed number of cycles tonebursts/square waves repeated at regular intervals, with examination of the rise time, behaviour at full amplitude, and recovery/overshoot.
It was done for decades. Why fight it? And the notion that it is somehow not-representative of "real-world" sounds and content can be easily disproved.
 

amirm

Founder/Admin
Staff Member
CFO (Chief Fun Officer)
Joined
Feb 13, 2016
Messages
22,563
Likes
34,610
Location
Seattle Area
#17
Count the samples in a cycle (any one you like), look at the sampling rate and determine the frequency. Nobody is saying the steps come out the analog output- certainly not me.
But you showed a graph that represents that. I am saying that is a visual representation that doesn't show what happens to an analog signal.
 

amirm

Founder/Admin
Staff Member
CFO (Chief Fun Officer)
Joined
Feb 13, 2016
Messages
22,563
Likes
34,610
Location
Seattle Area
#18
This is nonsense or at least a very poor way of looking at things. As I said earlier a single sample of amplitude A and zero everywhere else is a perfectly legal result in the limit of an ideal sampler and ideal brickwall anti-aliasing filter, it is a mater of theory vs limits of physical realization. Infinite energy does not figure into it.
It does in this scenario where I am trying to explain what an impulse signal is. The fact that it is infinite in bandwidth is what makes it useful to characterize a system in theory. That same power makes it useless to represent an amplifier for audio application.
 

restorer-john

Major Contributor
Joined
Mar 1, 2018
Messages
3,063
Likes
5,481
Location
Gold Coast, Queensland, Australia
#19
But you showed a graph that represents that. I am saying that is a visual representation that doesn't show what happens to an analog signal.
This makes no sense Amir. The spectrogram shows the energy centered around 10KHz does it not? The samples and the period show the dominant cycles in the transients to be centered around 10KHz.

Here's what any of my amplifiers 'see' with the analog output of the world's first CD player, the Sony CDP-101 playing a single sample impulse at 0dB. (no linear phase filters here)

Are you really trying to tell me my amplifer doesn't really get to 'see' a half cycle 22.05KHz impulse, followed by rapidly decaying ringing?

1575341260126.jpeg


Would you like to see the output of the entire chain, CD player, preamplifer and power amplifier with that single sample? (maybe tomorrow after I've dug out the old girl (the Sony CDP-101)) We both know what we are going to see...
 
Last edited:

pma

Active Member
Joined
Feb 23, 2019
Messages
220
Likes
302
Location
Prague
#20
This is nonsense or at least a very poor way of looking at things. As I said earlier a single sample of amplitude A and zero everywhere else is a perfectly legal result in the limit of an ideal sampler and ideal brickwall anti-aliasing filter, it is a mater of theory vs limits of physical realization. Infinite energy does not figure into it.
Agreed, Scott.

1575359190791.png
 
Thread starter Similar threads Forum Replies Date
A DIY Audio Forum 32
DonH56 Testing Area 0
DonH56 Testing Area 0

Similar threads

Top Bottom