Didn't we agree that the distortion can never be corrected to be lower than the noise so performance potential doesn't matter in any sense?
Last edited:
Nothing to do. I already know I can reach near arbitrary levels of performance but if the noise can't be corrected then it's pointless to go beyond it. If the potential performance cannot be extrapolated then there's nothing to be gained and nothing to show. As far as I'm concerned I'll just make sure whatever amp I make never goes lower than -150db under any condition within spec which is painfully easy with this circuit. Nothing more to do here I guess.
-290 dB noise floor into 1 ohm at LHe temperatures and .0000001 hz bandwidth I think gets you there. 😉
I suspect that any junction of different metals will give more than -290 dB nonlinearity
I also think this is one of the most successful troll fests I have ever seen
I also think this is one of the most successful troll fests I have ever seen
I'm as skeptical as you are about claims of audible distortion below -80 dB, but I don't believe your arguments are very solid. Distortion causes peaks in the spectrum, so if you want to compare them to noise, you would have to compare them to noise in one critical bandwidth of the ears of the listener, rather than to the noise over the entire audio band.
Regarding Google, I've worked as an integrated circuit designer for the past couple of decades, but you will find very little evidence for that with search engines. What you might find are some hobby articles, some in Jan Didden's former bookazine and two in Electronics World, some papers about very low power analogue MOS circuits from when I was still at university and a very small number of patent applications. The reason is quite simple: the company I work for publishes as little as possible because it wants to keep its know-how to itself. When publishing something is inevitable, it is usually done by the people involved in management, marketing or application engineering rather than by the designers.
Prove you can actually get your amplifier's nonlinearities (or measured distortion spectra) to -150db and you'll have basis for good employment. Heck, just get the noise floor to -150 on an audio frequency speaker or headphone amp and the same is true. You keep repeating that the -150db number is trivial to achieve, and people who have made very successful careers doing this keep mentioning that it is, in fact, quite difficult...
So when you get your EC all built up, I personally would like to see some measurements that correlate that kind of performance. (Not even asking to see the schematic of your circuit, just the testing procedure, parameters and results.)
Simple. People measure THD. People hear IMD. The IMD level is usually higher than the THD, and often audible. That isn't even getting into the fact that not all THD sounds the same, or is equally audible depending on its harmonic composition.
Last edited:
To expand on that idea, people measure HD rather than IM because it's easier to get higher resolution, and generally, you can't have IM without HD, so measuring one will make implications about the other.
Along those lines, people don't listen to pure sine waves either, they listen to complex signals that do not have stationary power levels or static frequency components. I suspect that measurements with these sources will result in far worse distortion numbers than the HD results from the same circuitry, but again, it's very difficult to do measurements using actual music, so instead, we use ultra pure sine waves, and worry about levels that would probably not be audible taken alone.
The final idea that I use to justify scraping into the parts per billion range of distortion is to get a better idea of how the circuit is nonlinear. Does the nonlinearity increase at 6dB per octave, as would be expected by open loop gain rolloff, or does it increase by 12dB per octave, implying that maybe there's an additional frequency dependent mechanism, such as a finite circuit current charging a capacitance? Digging into realms that might be 'inaudible' can possibly help to understand how the circuit works, and whether a particular fault can be made to go away, or to understand what the tradeoffs are related to that fault.
Along those lines, people don't listen to pure sine waves either, they listen to complex signals that do not have stationary power levels or static frequency components. I suspect that measurements with these sources will result in far worse distortion numbers than the HD results from the same circuitry, but again, it's very difficult to do measurements using actual music, so instead, we use ultra pure sine waves, and worry about levels that would probably not be audible taken alone.
The final idea that I use to justify scraping into the parts per billion range of distortion is to get a better idea of how the circuit is nonlinear. Does the nonlinearity increase at 6dB per octave, as would be expected by open loop gain rolloff, or does it increase by 12dB per octave, implying that maybe there's an additional frequency dependent mechanism, such as a finite circuit current charging a capacitance? Digging into realms that might be 'inaudible' can possibly help to understand how the circuit works, and whether a particular fault can be made to go away, or to understand what the tradeoffs are related to that fault.
He has a real talent for sucking in otherwise intelligent people, he's done it before, I will give him that. What have we learnt, that total harmonic distortion isn't a very good measure of audible performance.....🙄
I dont know why I am responding to this.
Well, I do, it's because I have had two Russian River Damnations, which is en excellent (strong) Belgian style beer made in California.
For this to work means the EC circuit is capable of -290 dB accuracy. It does not follow that a circuit that can reduce large distortion D to D-X dB (iow by factor -X dB) can reduce distortion smaller than D by (say) 120 dB to D-X-120 dB. The EC circuit will have has its own nonlinearity, which may be in the noise when dealing with a near clipping signal, thus giving the claimed distortion reduction, but the claim that it (it == the EC circuit) works "proportionally" as well at lower levels is a claim that it's more linear than a solder joint. In other words the ec itself is better than -290 dB thd. If you make that from semiconductors, you need to throw many virgins into the volcano, no not volcano, I mean doping furnace, to get that level of matching in the devices that perform the detection of errors in whatever circuit is doing the error detection. It is beyond absurd. It's "not even wrong".
Make a 2:1 resistive divider with -290 dB nonlinearity. And measure it. Good luck.
SMH
c
Well, I do, it's because I have had two Russian River Damnations, which is en excellent (strong) Belgian style beer made in California.
For this to work means the EC circuit is capable of -290 dB accuracy. It does not follow that a circuit that can reduce large distortion D to D-X dB (iow by factor -X dB) can reduce distortion smaller than D by (say) 120 dB to D-X-120 dB. The EC circuit will have has its own nonlinearity, which may be in the noise when dealing with a near clipping signal, thus giving the claimed distortion reduction, but the claim that it (it == the EC circuit) works "proportionally" as well at lower levels is a claim that it's more linear than a solder joint. In other words the ec itself is better than -290 dB thd. If you make that from semiconductors, you need to throw many virgins into the volcano, no not volcano, I mean doping furnace, to get that level of matching in the devices that perform the detection of errors in whatever circuit is doing the error detection. It is beyond absurd. It's "not even wrong".
Make a 2:1 resistive divider with -290 dB nonlinearity. And measure it. Good luck.
SMH
c
Not necessarily. Yes, the noise floor is your ultimate practical limit, but I don’t think we’ve really proven if your EC works at all or this is entirely a measurement artifact. You still seem to think -150 dB is easy and it’s not. The Benchmark doesn’t reach that level, so what does? If you’ve measured that it’s most likely a mistake with the setup.
It takes a ridiculous amount of feedback and conditional stability to achieve this for the composite op-amp described in the paper.
Last edited:
Bill Wazlo's DiffMaker comes to mind, a track of music with a whole symphonic orchestra buried at -60dB. Most listeners simply do not hear that.
Jan
A buried symphony is completely uncorrelated with what the brain is focused on listening to. Human perception is a very complex thing, not linear in the way engineers are trained to model everything. That most listeners don't notice the symphony doesn't necessarily say anything about what people may notice when they listen for an aberration in the sound of music they are focused on.
That’s your theory, anyway. Could probably be tested via fMRI or EEG but those with the equipment have better things to do.
As a somewhat reliable data point, I can offer that an Audio Precision APx-555 test set will show a loopback residual figure of around -155dB for 2nd harmonic and -160dB for 3rd harmonic, at a sweet spot (at least for my unit) of 2kHz with a stimulus level of +6dBV and a 200kΩ load.
I have assembled some very simple balanced inverter PCBs with RF filtering and DC servos on PCBs using AD797 and LM4562 which challenge that residual, showing all of the confusion of a low distortion DUT interfering with the residual of the test set, showing essentially the same level of 3HD or slightly more, but with 2HD levels that may be the result of constructive or destructive summation of the generator or DUT spurs.
So, I can show you small signal amplifiers that, when driving a light load, behave very politely, probably to the -150dBc level. However, I won't say that it's "trivial" to get one of these amplifiers to correct an amplifier that's driving a lot of current into a load - that's actually slightly tough, even with loads around 50-100Ω, much less an 8Ω speaker.
So yes, speaking to the thread in general, I say "bring on the proof" - I'd love to learn something I don't know, but at the same time, I can say that there are few surprises down here in the residual of quality test sets, commercial, or hand-assembled. The very nature of some forms of distortion requires a discipline to interpreting the numbers, given the usual problem of cancellation or reinforcement of the DUT numbers by the analyzer residual.
Ok...maybe i'm just expressing it in a too strong manner...I can accept defeat only when stronger arguments show off, but this is actually my way of reasoning :
If people measure very low thd in laboratory to guess what's happening in real world it might be that there's no unanimity in the scientific world about how to get the real numbers from the basic thd, imd tests.
I saw that many people proposed various ways to test audio amplifiers over the time , claiming better approach to the real world behavior.
So why don't the scientific world get together and make the Ultimate Test a must ? It's you, the specialists, the professionals who should come with a simulation model that works 99% as in the real world and come with a testing technique that can hint in a very scientific way to the real world behavior of an audio amp and we won't comment on it anymore being stripped of any scientific authority.
But we have a schizophrenia digging deep in the audio world that tells us that we should consider the lowest or highest numbers to have a correlation to the real world while some reputable scientists come back to tell that it doesn't matter...that hunting numbers is just the 80's and 90's negative feedback fetish .
So its still us, the simple technicians and users who are called to make a difference and the "big names in the industry" will ever loose against me , a musician or even a deaf person, because i can point to another "big name in the industry" telling the opposite and escape the measuring inquisition.
You know the Beethoven case! Every time a guy pretend that a musician is the best because he has perfect hearing somebody will tell that Beethoven was deaf when composing his masterpieces as if it weren't the talent and memory who made the real working class in our brains doing wonders for the upper class!
Last edited:
Thanks for that Monte, is of interest to me. I see similar behaviour on my SYS2722 albeit at 15dB higher level ...
As to the possibility of measuring -150dBc when delivering some current in a load that is not particularly linear (a headphone for instance), you can only measure it, if at all, at the amplifier at the feedback pick-off point. Measuring something like that at the load is futile in the extreme.
Jan
- Status
- Not open for further replies.