The time-domain averaging only works when you have sample-sync'd recording while playback (DA and AD in the same device or even the same chip, running on one single clock), so analog sources and non-sync'ed sources (like two CD-players) are out.
I do it like this (basic procedure) :
- select a snippet of audio, about a second or so, for example a 100k sample block (select a block size and sample rate to get good reduction of mains hum+harmonics)
- loop this 1000++ times in a DAW, ad random dither to the whole length.
- record while playback through the trimmed analog diff and gain stage.
- throw away blocks at the beginning to avoid settling tails.
- condense the recording to one single block of samples, this is the averaging (wrote a piece of software to do that).
This gives a residual with at least 20dB less noise than a simple subtraction. The trimming is somewhat iterative once you get deep below the analog noise floor (no meaningful residual without heavy averaging).
As described, you can also record without the analog diff, sequentially, with and without DUT (or DUT A and DUT B) -- it is important to use the same DA and AD channel, hence not in real time using L/R channels.
It doesnt matter if you average first, then try to trim for lowest residual or vice versa, but averaging first is better because the improved SNR helps a lot to judge the residual.
Haven't played with diffmaker much but IIRC it does not do averaging (to do that it would need some input about the properties of the data it is processing, the block size). It offers resampling to take out some effect of non-synchrous recording by autocorrelation, so some jitter and/or clock drift is taken care of, but I could never get that to work satisfactory.
Diffmaker's trimming does work best with good preprocessing, that is pre-averaged, sample-sync'd and aligned input data.
Those would be one in the same, slight pitch shift would be like SRC which involves resampling. Doing it on a continuous scale (like aligning two plays of an LP) is more involved but the maths are out there.
In either case, KSTR (Klaus?) highlights some of the difficulties of not doing analyses on stationary signals (and the definite need for synchronous sampling).
Not that synchronous sampling doesn't help across the board, e.g. noise. But also what we partially get by using things like distortion amplifiers/nulling before digitization.
Not that synchronous sampling doesn't help across the board, e.g. noise. But also what we partially get by using things like distortion amplifiers/nulling before digitization.
Not saying you do it....... the general thrust here (DIYAudio) is to supply answers and it seems very frustrating to those when they dont get answers. Then the only debate is the interpretation or answere.
It can be frustrating to some to instead ask questions without giving an answere. My hope is that someone will try this or that and give their answer. Or thier interpretation. After all, we already have the authors of those papers JC put up conclusion. Occasionally, a single person will be curious or challenged enough to do the work themselves and talk about it... or write a paper. I encourage them to go further and when they do, new insights happen.
Two or three people here at DIY have taken my questions and done amazing things not available commercially. So, instead of saying its all covered in E101, try it. And, see if anything new - even a different insight - occures.
What I like is to see what new tests or designs can be done. Rather than what has already been done yet again. Push things a bit. Even if it isnt audible but more perfect.
Having said all that, a lot of testing involves some form of averaging or preaveraging-- FFT or other. To me, that is one possible area that moves listening and test data further apart.
THx-RNMarsh
Last edited:
Test are done routinely by smart 'amateurs' (and I use that word as a compliment) that were not possible even 20 years ago because now everyone can afford high performance equipment and software.
But it is all ignored and 30 yr old tests are trotted out on 40 yr old Shibawhatever showing lots of decimal zeros. I tell you a secret: we've moved on from THD numbers. We can do tests with 30 or more simultaneous frequencies that show anything you can think of in a simple test that takes just a few seconds.
People like Hirata thought long and hard to come up with a test that was more revealing than what was possible at the time. Where are the hard thinkers now? Ohh we have a couple, the likes of SW, but they are ridiculed because they make 8-legs. All that happens now is that the old papers are thrown up and the rest of us is told that we are stupid and have no clue. If Hirata could turn in his grave, he would be mistaken for a fan.
It's OK to ask questions, of course. But they should be sensible, based on understanding where we stand, and what kind of steps forward we would like to go. I myself often have no idea what that is, and I then use an advanced technique called 'keeping my mouth shut' .
Jan
But it is all ignored and 30 yr old tests are trotted out on 40 yr old Shibawhatever showing lots of decimal zeros. I tell you a secret: we've moved on from THD numbers. We can do tests with 30 or more simultaneous frequencies that show anything you can think of in a simple test that takes just a few seconds.
People like Hirata thought long and hard to come up with a test that was more revealing than what was possible at the time. Where are the hard thinkers now? Ohh we have a couple, the likes of SW, but they are ridiculed because they make 8-legs. All that happens now is that the old papers are thrown up and the rest of us is told that we are stupid and have no clue. If Hirata could turn in his grave, he would be mistaken for a fan.
It's OK to ask questions, of course. But they should be sensible, based on understanding where we stand, and what kind of steps forward we would like to go. I myself often have no idea what that is, and I then use an advanced technique called 'keeping my mouth shut' .
Jan
Last edited:
My problem with the recent links remains the same as it has been for years. Not a single reference outside a tiny cadre of people (usually Brits, please don't take that the wrong way). This is not how one presents something that is supposedly also of general electrical engineering interest, it's known as due diligence.
Some of the circuits supposedly patented in the late 90's were well known for decades. The tight feedback input stages were used by Fairchild and even made one of the IEEE historical collections that I got in the 70's. The cross-coupled mirror load was also old hat.
The fuzzy distortion and granularity of noise stuff is just nonsense.
All know I start broad and then narrow it down depending on response.
Let me ask if there is an interest here in what could be the factors which lead to audible changes? And which tests help determine or predict it? Certainly it isnt just distortion or freq response only.
Do other tests help better to determine or predict how it will sound? Which ones or new tests will close the gap on measured data and how it sounds? What are they? Do these old tests revisited offer any insights to that question?
So, I am looking at parasitics and their possible affects in different areas of apps. Or interfacing issues. And, system wide results.
but, nothing wrong with figuring out how to design and make the lowest noise amp than anyone else. Or pushing some other parameter to new levels. Even if it isnt audible. It's all interesting to me.
THx-RNMarsh
Let me ask if there is an interest here in what could be the factors which lead to audible changes? And which tests help determine or predict it? Certainly it isnt just distortion or freq response only.
Do other tests help better to determine or predict how it will sound? Which ones or new tests will close the gap on measured data and how it sounds? What are they? Do these old tests revisited offer any insights to that question?
So, I am looking at parasitics and their possible affects in different areas of apps. Or interfacing issues. And, system wide results.
but, nothing wrong with figuring out how to design and make the lowest noise amp than anyone else. Or pushing some other parameter to new levels. Even if it isnt audible. It's all interesting to me.
THx-RNMarsh
Last edited:
PMA, I agree with you that most 'reasonably engineered' amplifiers might not show much with the Hirata test, BUT I have measured one glaring problem with one of my early op amp prototypes, when I added an AC balance pot. I was hoping to discuss it in detail, once people got the Hirata test in their perception.
What is most important is that the Hirata test CAN show problems not shown by Harmonic or IM tests. It is not the only test to show something not seen with thd, etc tests, but it is available in print, so I put it up as an example of WHY thd or IM (no matter how many tones) is not necessarily enough. Back in the early 80's I had access to even more material by Dr. Hirata, including measurements of some common commercial amps at the time. This was a more detailed, and revealing of even some DC coupled amps, as well as Dolby, etc.
I don't expect anybody, including myself, to completely understand Dr. Hirata's papers, but it concludes accurately.
What is most important is that the Hirata test CAN show problems not shown by Harmonic or IM tests. It is not the only test to show something not seen with thd, etc tests, but it is available in print, so I put it up as an example of WHY thd or IM (no matter how many tones) is not necessarily enough. Back in the early 80's I had access to even more material by Dr. Hirata, including measurements of some common commercial amps at the time. This was a more detailed, and revealing of even some DC coupled amps, as well as Dolby, etc.
I don't expect anybody, including myself, to completely understand Dr. Hirata's papers, but it concludes accurately.
Last edited:
Now that would be interesting, to see how the Hirata test revealed a problem (audible or not as RNM said), that other tests/measurements did not reveal, what you changed to fix it, and test results on the revised amplifier.
As Jan said, it's not very interesting to put up an old paper without relating it to actual design decisions. I would find it far more interesting to see how Hirata test results would drive design changes.
As Jan said, it's not very interesting to put up an old paper without relating it to actual design decisions. I would find it far more interesting to see how Hirata test results would drive design changes.
Speaking of 40 year old papers, I was re-reading the Holman 1975 AES paper 'New Factors in Phonograph Preamplifier Design' as part of looking around for some references on something and noted he has a section in that on square wave testing. I suspect he was on a work limit as he skates over some of the analysis but he does show a test that some well regarded amplifiers of the time failed miserably on, and gives us a circuit that passes this test. The interesting part is that he suggests FFT would be a better way to do this, which is exactly what we can do now with the computing power resting at our finger tips. He does also give a passing mention to transient distortion whilst noting the test is beyond any realistic music programme material.
But for me the interesting part was that there was a letter to the editor from Reg Williamson from 1973 wireless world on MM cartridge loading and the fact that people were getting the basic electrical equivalent model of the generator wrong. 2018 and people are STILL getting it wrong. The internet has failed to enlighten mankind in its race to the bottom
But for me the interesting part was that there was a letter to the editor from Reg Williamson from 1973 wireless world on MM cartridge loading and the fact that people were getting the basic electrical equivalent model of the generator wrong. 2018 and people are STILL getting it wrong. The internet has failed to enlighten mankind in its race to the bottom
So, I am all for trying to gain linearity, but I have to ask how much more valuable do you guys using the Hirata's test, find it, over using very good transistor modeling and thermal balancing? Have you been able to improve your linearity farther? (there is measuring, and there is doing something with it - separate things rather frequently)
I doubt it, in fact that can't be true. Not wasting my time with this stuff anymore the boulder will be at the bottom of the hill again next week. Yes, a large signal square wave how profound would show up in any simulator or these days a simple laptop soundcard FFT.
EDIT - I see even then Mr. THX was sensitive enough to simply state that square wave tests correlated "perfectly" (his words) with the so called asymmetric signals without editorializing.
The best of all - YouTube
Last edited:
Tom Holman's paper is a useful read, in fact you will find it as Reference #6 of my 'Omitted Factors in Audio Design' article that I have put up here.
In fact, the AES paper on TIM Measurement in 1976, improved on Tom Holman's test by adding an extra harmonic frequency that is not closely mathematically related to the square wave frequencies, in order to bring out the added distortions further, rather than having many of them fall back on the input frequencies, as it would with the Holman Square Wave. We debated when making the test in 1976, whether to use 20KHz or 30KHz as the added 'rolloff' frequency when we developed the Sine-Square TIM distortion test. We chose 30KHz, partly because it correlated to Tom Holman's original test, but also because it gives a more realistic effective bandwidth for moving coil cartridge frequency response.
In fact, the AES paper on TIM Measurement in 1976, improved on Tom Holman's test by adding an extra harmonic frequency that is not closely mathematically related to the square wave frequencies, in order to bring out the added distortions further, rather than having many of them fall back on the input frequencies, as it would with the Holman Square Wave. We debated when making the test in 1976, whether to use 20KHz or 30KHz as the added 'rolloff' frequency when we developed the Sine-Square TIM distortion test. We chose 30KHz, partly because it correlated to Tom Holman's original test, but also because it gives a more realistic effective bandwidth for moving coil cartridge frequency response.
Are we back to the physically impossible in-harmonic frequencies yet? You know Gilligan's Island never ended either.
Scott, have you worked out what happens when you distort a square wave. Where do you think the 3rd, and other odd harmonics land? Somewhere where you can easily find them?
For those who are interested in what I just said, please read the first two pages of the TIM paper that I put up here.
For those who are interested in what I just said, please read the first two pages of the TIM paper that I put up here.
Last edited:
- Status
- Not open for further replies.