You should read your own thread a bit better. Several things that might limit the ultimate distortion performance have been mentioned:
1. Noise floor. I disagree with this, but it has been mentioned a lot.
2. Passive component nonlinearity, if you use passive components to set the gain of the corrected amplifier. For example, voltage coefficient or self heating and temperature coefficient of resistors.
3. Coupling of distorted output stage currents into the input, magnetically or otherwise.
4. Thermocouple effects.
5. Non-ohmic contacts between dissimilar metals.
Marcel,if you have to reduce the FFT bins (measurement bandwidth) to the point that you can read down at some arbitrary low distortion level its going to take time. That time is not compatible with the BW requirements of an audio amplifier i.e. in a practical application the system cant hang around to average the noise floor out to see crud below it.
Secondly, in a real world system, please explain how an EC system can do something useful waaaay below it, or the DUT's, noise floor.
The nonsense here is in the attempts to bypass the physics and magic up specification that just cannot exist on something lashed up on a kitchen sink.
As already mentioned, this is LIGO territory . . .
Secondly, in a real world system, please explain how an EC system can do something useful waaaay below it, or the DUT's, noise floor.
The nonsense here is in the attempts to bypass the physics and magic up specification that just cannot exist on something lashed up on a kitchen sink.
As already mentioned, this is LIGO territory . . .
The SINAD vs pure distortion products is technically correct and we should be more careful to differentiate the two. There is also the practical part of measurement, if nothing more than the generator frequency stability over a 2+ minute run with bin sizes getting into the .1 Hz or less range (and its own distortion products muddying the waters).
Obviously the -290 claim is going to fall apart very quickly when examined there. And, yes, I believe Scott has made mention that the overall gain of LIGO is in the range of 300 dB. 🙂
So that's not to say any party is incorrect, but appearing to be talking past one another. There *is* merit, depending on one's needs to look way past the "BW requirements of (insert DUT here)" for a deep dive into the weak links of an instrument, but we're talking a much more modest -120 to -150 dB range here. 🙂 Marcel's list becomes very important when you're sweating the last details, and I'm fortunate to *not* work in that territory ever! (I would add EM ingress as well, because perfect shielding doesn't exist)
Obviously the -290 claim is going to fall apart very quickly when examined there. And, yes, I believe Scott has made mention that the overall gain of LIGO is in the range of 300 dB. 🙂
So that's not to say any party is incorrect, but appearing to be talking past one another. There *is* merit, depending on one's needs to look way past the "BW requirements of (insert DUT here)" for a deep dive into the weak links of an instrument, but we're talking a much more modest -120 to -150 dB range here. 🙂 Marcel's list becomes very important when you're sweating the last details, and I'm fortunate to *not* work in that territory ever! (I would add EM ingress as well, because perfect shielding doesn't exist)
I went back and checked the measurements - see attached.
With the A-D input shorted with a 50 Ohm BNC plug and a 0.75 Hz FFT bin, the noise is ~ -158 dBV
In the loop back test, with the same FFT bins, its about 20 dB worse, so about -138 dBV.
So my earlier claim of loop back noise floor at -150 dBV was not correct - that's only what you get on the measurement side.
The generator was set to 0dBV as a reference
(BTW - it takes about 50 seconds for the noise grass to settle on this measurement)
With the A-D input shorted with a 50 Ohm BNC plug and a 0.75 Hz FFT bin, the noise is ~ -158 dBV
In the loop back test, with the same FFT bins, its about 20 dB worse, so about -138 dBV.
So my earlier claim of loop back noise floor at -150 dBV was not correct - that's only what you get on the measurement side.
The generator was set to 0dBV as a reference
(BTW - it takes about 50 seconds for the noise grass to settle on this measurement)
Last edited:
Suppose hellokitty123's circuit would indeed produce harmonics with a total RMS value 290 dB below a single-tone desired signal, and a noise floor 130 dB below the signal measured over a 20 kHz bandwidth. By definition, the harmonic distortion would then be -290 dB and the SINAD almost 130 dB.
Proving by measurement that the harmonic distortion is as low as it is would then require a bandwidth of the order of 2E-12 Hz, requiring a measuring time that probably exceeds the lifetime of the amplifier. That is, noise would prehibit direct measurement of the distortion, but I still don't see why it would limit the achievable distortion itself. I certainly don't see what the bandwidth of the amplifier or the critical bandwidth of the human auditory system have to do with it.
Of course all of this is of philosophical interest only, as there are several second-order effects that are likely to limit the achievable distortion to a value well above -290 dB.
I'm still struggling to see how you can measure something below the noise floor or that there is anything happening other than noise down at that level. I would assume signals ride on top of fundamental noise mechanisms - i.e. the signal of interest energy has to be greater than the fundamental noise energy. Do signals exist down at the level of atoms that are bumping into each other and causing Johnson noise?
I've made it pretty clear that I don't understand why people think my extrapolation is a mistake.
90% of the time in my daily life that I take someone word for something I end up finding out they are wrong. It boggles the mind how often people are are wrong about things. Humans are really good at being wrong about everything. The core of it is that people don't wait for all of the information before making an analysis, it is inherently flawed. Presumptuousness is the essence of being wrong.
Being right isn't about knowing everything it's about knowing what you don't know and being able to assess how correct you can be with your current knowledge.
As far as I'm concerned I have a lot more data than I've shown and people keep thinking of the circuit in question in terms of typically used and known circuit configurations based upon the word "EC" which as I've said doesn't apply because the circuit is strange and doesn't use typical methodologies to achieve its goal.
So basically I'm taking an innocent until proven guilty approach.
Which is all the more reason why I'm asking for advice on how best to approach measuring this thing.
I don't, I've said this many times, I was content on that fact until people talked about how Samuel Gronner's circuit was measured to -180db, now clearly this is below the noise floor yet it was measured. So Apparently that is not entirely true and my results were still potentially valid. In either case clearly this shows there can be more I can do to measure it than simply giving up and saying that the noise floor is going to stop me from measuring it.
Which is part of the point that I was making, I've done tests from all different angles and scenarios as I've said toward the begining of the thread and the result is always the same, the same amount of improvement.
Which is why you guys are going to have to give me more information than just "it won't work" for me to take your word on it.
Well as I said I am only using a simple example circuit using an LT1364 at the moment, with long wires and not even any bypass caps.
The tests I did a year-ish ago were on a much more sophisticated design using low noise parts, some of which were on a PCB.
However the scope itself only levels out at a little more than -130db even with out connecting the probe to the analyzer.
At least on DBV anyway.
Indeed, there is a lot of information and data that I haven't shared. But that is just the nature of the situation. I'm keeping my mouth shut on the details of operation.
That's why I keep saying that I'm willing to do whatever tests people want me to try, just not involving a schematic.
I'm still waiting on a response from that local engineer I was recommended to.
In case you guys haven't noticed I'm not going to respond to useless hate comments and unconstructive responses.
Also in case you haven't noticed at this point, discussing details about what the circuit can and can't do further is pointless since I haven't shared a schematic.
You're skepticism is correct, but I will maintain my position until proven wrong. I haven't heard any compelling evidence that my design will not function on a low distorting amplifier other than the noise floor argument.
Again, I'm looking for testing methology advice, that's it.
I get it, you all hate me and think I'm an idiot, that's fine.
It was not measured. Samuel clearly stated in his paper, which you should read, that his AP SYS2722 cannot reliably measure below -120dB. So he used the noise gain method to artificially increase the distortion, measure it and then correct for the noise gain.
Several here have suggested that this would be a good way to test your amp but you seem to disregard yet keep asking for 'a measurement method'. I can understand why some think you are a troll (I don't).
Jan
Did you tell Dick Sequerra that -290dB number and/or about this thread? If you did I doubt it a reply will be forthcoming.
Do you know who this man is?
Jan
It's called (auto)correlation. Noise is averaged out, signal remains, see the lock in amplifier principle http://electronics.physics.helsinki.fi/wp-content/uploads/2011/02/lockin_rob_web.pdf. FFT synchronous averaging and a digital lock in amplifier are providing the same end result (can be mathematically proven).
Last edited:
There are at least four other second-order effects that may limit how low you can go, see post 261. The problem is that you need to quantify what distortion you will get including such effects, and I haven't a clue how to do that unless the distortion is high enough to be measured.
I did use the resistor test method, I've shown the results of this already.
However my original method used the same concept. I made the main amplifier distort greatly in order to have more measurement headroom, but people largely ignored those results.
https://www.diyaudio.com/forums/analog-line-level/338635-290-db-distortion-6.html#post5810225
https://www.diyaudio.com/forums/analog-line-level/338635-290-db-distortion-7.html#post5810310
Why is Gronner's result more valid than mine?
At what levels of distortion do these second order effects take effect?
Last edited:
Yes, he actually measured it by known techniques. Not trying to be mean spirited but you already admitted you didn't understand the concept of noise bandwidth of an FFT bin and correction for the windowing function.
Perhaps, but the result is the result, if you can explain to me exactly why my methodology is somehow incorrect that would be great. I mean, I used the resistor method that gronner used as well and came up with the same result. The resistor method only allowed me about -30db extra headroom and I only required the most basic level of my EC to make the distortion dissapear behind the noise floor. My method gave me around -90db headroom and required a more sophisticated version of my EC to make it go below the noise floor, although it was still only at 40% of the full circuit. If I did the resistor method side by side with my method you would find the same improvement while using my most basic version of EC on both methods. I've shown this in earlier screenshots. I see no evidence, logical or otherwise that my method is somehow invalid.
Last edited:
Bonsai, think of a harmonic component exactly a sub-multiple of the sampling frequency the FFT bin is a complex vector that always points to the same place. The FFT of gaussian noise taken as a complex vector (not as a magnitude) has 0 mean so averaging approaches the limit of nothing but the harmonic content. If you saw the NIST paper on battery chemistry noise you will see that you can even remove the noise of the instrument from the noise of the stimulus.
The averaging has the usual square root of n problem so going from -120dB to -290dB is obviously a real matter of time.
Last edited:
So you are saying with enough time I can average out the noise if I leave my QA401 running long enough?
This is the result of the resistor method
-290 dB Distortion?
This is the result of my method using the same version of the EC that I used on the resistor method. I used the lowest averaging settings here though because I was only trying to make a quick point.
-290 dB Distortion?
This is a result with my method at 12.1v p-p using a more sophisticated version of my circuit
-290 dB Distortion?
This is the result with my method at 12.9v p.p with yet another even more sophisticated version of my circuit.
-290 dB Distortion?
Which I further expanded on here
-290 dB Distortion?
- Status
- Not open for further replies.