Yeah I guess the DM does do that in a drawn out way. I just find it strange no one ever mentions measuring directly off the mid-point of an inverting amp, seems easier. I already troubleshooted the DM, the error starts at the last opamp, everything before that point works as expected. I simply don't have time to figure why. I have too many other things going on. I need to get it measured professionally anyway.
Only works when gain is exactly -1. And you still have to null it for gain and phase, because the resistors setting the gain (and with that the 'null point') are never exactly the same.
That's why the DM has the tuning pots.
BTW I find it curious that someone with the chops to design a -290dB EC can't fix a simple opamp circuit ... But that's probably just me.
Jan
I'm glad you are still there. There was no reaction from your side to my noise simulations, maybe because it was a bit unclear what I tried to tell.
That's why I have made the simulations again, but now including the noise spectra without EC plus the noise spectra that you showed here -290 dB Distortion?.
I also included a table at the end with noise data extracted from these figures at 10KHz, the frequency that you also used here https://www.diyaudio.com/forums/analog-line-level/338635-290-db-distortion-post5816052.html. These spectra are showing 4 EC steps, whereby each step suppresses the 2nd harmonic by roughly an extra 30dB. So 1/2 of your EC as used for your noise images should suppress ca 60dB at 20Khz. Suppression at 10KHz should be comparable or higher.
However when looking at the table with noise figures @10KHz, the highest noise suppression with 1/2 your EC is from 600nV/rtHz to 68nV/rtHz, or 19dB suppression and not the expected 60dB.
Without noise (and other missing things contributing to THD) in FFT simulations, you can get these very nice figures, almost up to 60dB for the halve set and 150dB error correction for the full set of EC stages that you have in mind. However, the much more reliable noise simulations are showing at the best a noise suppression of 19dB @10KHz, which seems difficult to understand.
So my intention was to stimulate you to find out how these seemingly conflicting results should be regarded. I would be happy to receive a reaction from your side.
Hans
Attachments
Maybe I'm forgetting something but can't I use the dBr function to compensate for the gain of the amp? The mid-point is the distortion relative to the output.
Never said I couldn't fix it, I said don't have time, multiple times. This kind of lack of reading and/or reading comprehension is why I ended my discussion of my circuit in this thread. It's like 10% of my words actually get read. No offense meant but it's frustrating when all I do is repeat myself.
As I said the DM nodes function as expected so probably a connection issue or something silly and small but I don't have a good reason to make time to fix it plus I hate solving these kinds of problems, it's tedious. I have little interest in fixing it at the moment anyway since I've concluded the pointlessness of continuing to discuss my circuit here and the need for third party measurement regardless. My question about measuring an inverting amplifier at the mid-point is mostly academic.
Sorry Hans, the reason I didn't react was because of a compounding frustration from a having to repeat myself continuously. I ended the conversation about my circuit officially speaking in this thread. Don't get me wrong. I highly highly appreciate your effort in trying to help me. I really do. I just needed to take a break and I've been really busy anyway. I would be more than happy to continue the conversation with you in PM if you want. But I'm kind of done in this thread as far as discussing my circuit in particular. But as a response to your concern. You bring up a good point. As I said the circuit is not really a true EC in the normal sense, it's very confusing to me how it would be classified. Also because of the unusual nature and configuration of the circuit, logically figuring out how noise is affected by the EC is also a bit confusing to me. I need to think about it more.
Last edited:
If and I mean IF I had topology that could deliver -290db distortion I would not give up. As all of the worlds amp manufactures would be beating a path to my door!!! ps:- And the good folks on this site!!!!.
Hello,
In the real world in real time -390 dB distortion lies somewhere between paradox and oxymoron. In the real world in real time there in no -390dB distortion only noise. In the imaginary complex world of DSP you can visualize distortion in an ever narrowing FFT window of noise. Somehow some get the mistaken idea that a lower noise floor is reduced noise.
Noise is random, distortion harmonics are steady state. Error Correction feedback does nothing to remove noise.
Feedback low level harmonics plus the noise that is always there and the THD + Noise will not go down but increase. Out of phase noise added to noise is even more noise.
Just for fun see the attached plot of APx555 1.2M sized FFT window with 10 averages. What happened to the noise did it disappear?
Thanks DT
In the real world in real time -390 dB distortion lies somewhere between paradox and oxymoron. In the real world in real time there in no -390dB distortion only noise. In the imaginary complex world of DSP you can visualize distortion in an ever narrowing FFT window of noise. Somehow some get the mistaken idea that a lower noise floor is reduced noise.
Noise is random, distortion harmonics are steady state. Error Correction feedback does nothing to remove noise.
Feedback low level harmonics plus the noise that is always there and the THD + Noise will not go down but increase. Out of phase noise added to noise is even more noise.
Just for fun see the attached plot of APx555 1.2M sized FFT window with 10 averages. What happened to the noise did it disappear?
Thanks DT
Attachments
Last edited:
Oh yes you can, see the head of the 2nd harmonic sticking his head out of the weeds at -165dBV. The thing is you would never see it without the 1.2M sample size and 10 times averaging, otherwise the second and even 3rd harmonic would both be swamped by the noise. Real world and real time the harmonics are completely masked at the low levels in the plot.
Thanks DT
Thanks DT
Last edited:
Of course not, dBr is just a way to scale the measurement. Example: changing a size measurement from millimeters to tenths of an inch changes the numbers but doesn't change the object size!
And why would it be the midpoint? If I have a circuit with an input of -1V and an output of +2.345V, *somewhere* there is a point where the magnitude is zero. I don't understand why that should be the midpoint, whatever that is. And you can also bet your sweet behind that zero phase is not the same point as the zero level!
This is exactly the reason that the DM has controls for tuning both level and phase! You really should read the article to understand the background. It will be very hard to sell a revolutionary EC if you miss the basics.
Jan
Sorry, this is plain nonsense.
Where's the fun? Nobody expects noise to disappear when averaging. Only the distribution curve of the noise narrows down, giving a flatter spectrum at the average level, making it easier to see the distortion products popping out. Narrowing the Bin Width by using a larger FFT does lower the noise energy per bin, resulting in a spectrum with a lower noise level, but all the noise is still there spread over more bins, nothing disappeared.
Hans
It's like the length of the population. Some people are 150cm tall, others are 190cm tall, but if the average of the whole population comes out to 171cm, that doesn't mean that all of a sudden everyone is 171cm tall!
Noise is bipolar so it you average it long enough the average comes out to zero. But the noise is still there!
Its the difference between what it really is and the way you measure it.
Jan
Noise is bipolar so it you average it long enough the average comes out to zero. But the noise is still there!
Its the difference between what it really is and the way you measure it.
Jan
Last edited:
Assuming the paper reflects correctly what he did, I think it is.
If you measure two ropes, one is 123 feet, the other is 234 feet, is it reasonable to state that the two ropes together are 357 feet long, even if you haven't measured them as a single rope? I think it is.
Jan
If you measure two ropes, one is 123 feet, the other is 234 feet, is it reasonable to state that the two ropes together are 357 feet long, even if you haven't measured them as a single rope? I think it is.
Jan
Last edited:
Oh, for some reason I thought the dBr function would multiply the results based on the relative position to 0dbv. So dBr only level shifts the view or am I still wrong?
dBr is for relative measurements. It moves the entire scale to make it easier to read. Less math in the head. If you move the measurement to the nearest reference line, odB as an example, then you only have to glance at the measurement instead of calculating it. dBc is an example of a kind of dBr and so is THD for that matter. It's relative to the fundamental. But dBr does not change the scale. dBr can only be used for relative measures because you have changed the absolute measure. Best not to use it for extrapolation unless you really know what you are doing. When we use dBr the unit of the measure become meaning less because it is no longer referenced. So dBr is always in dB not dBV for example unless the math normalizes it.
Hello All,
Noise per bin times the number of bins equals the Total Noise.
If you increase the number of bins there is a corresponding reduction in noise per bin, the total noise remains the same. Jan, the average man is still 5 foot 10 inches tall.
Thing is with an increase in the number of bins the apparent “Noise Floor” of the FFT plot is reduced. A lower noise floor allows you to see the low level harmonics.
For the sake of HANS POLAK the Error Correction technology we are speaking of here adds back in 180 degree out of phase harmonics and, if you will allow me, 180 degree out of phase noise. Because the harmonics are periodic they will cancel. Because the noise is not periodic it will not cancel, it sums causing an increase in noise. The net effect is a decrease in THD and a net increase in Noise. If the starting levels of harmonics are low the overall effect of EC may be an increase in the THD+N percentage.
Thanks DT
Noise per bin times the number of bins equals the Total Noise.
If you increase the number of bins there is a corresponding reduction in noise per bin, the total noise remains the same. Jan, the average man is still 5 foot 10 inches tall.
Thing is with an increase in the number of bins the apparent “Noise Floor” of the FFT plot is reduced. A lower noise floor allows you to see the low level harmonics.
For the sake of HANS POLAK the Error Correction technology we are speaking of here adds back in 180 degree out of phase harmonics and, if you will allow me, 180 degree out of phase noise. Because the harmonics are periodic they will cancel. Because the noise is not periodic it will not cancel, it sums causing an increase in noise. The net effect is a decrease in THD and a net increase in Noise. If the starting levels of harmonics are low the overall effect of EC may be an increase in the THD+N percentage.
Thanks DT
Come on! How could changing a scale of a graph change the actual voltage??
Jan
So how does the EC know what the harmonics are, and how large they have to be cancelled? This doesn't sounds encouraging.
However, if it means that the difference between Vin and Vout is properly scaled and added/fed back, that could work and is a very familiar way to do it.
Jan
fiddle with the pot while watching the plot
Jan,
The EC is not knowing, there is nothing AI or new here, just inverted output fed back to the input with the level adjusted with a potentiometer.
You know, fiddle with the pot while watching the plot on the analyzer.
Thanks DT
Jan,
The EC is not knowing, there is nothing AI or new here, just inverted output fed back to the input with the level adjusted with a potentiometer.
You know, fiddle with the pot while watching the plot on the analyzer.
Thanks DT
Energy of uncorrelated signals adds up. Only correlated signals can be subtracted.
Two signals 180 degrees out of phase are fully correlated - that is, if you know one, you know the other with certainty. You can add them and get a null. That is how error correction of any kind works. In reality, null wil not be perfect - there will be some residual noise.
If you have two separate error correction mechanisms each leaving -150dB of residual noise, and the noise from the two sources is uncorrelated, it will add up to -147dB. If one mechanism reduces the error from the other, you still get -150dB residual from the first one.
Two signals 180 degrees out of phase are fully correlated - that is, if you know one, you know the other with certainty. You can add them and get a null. That is how error correction of any kind works. In reality, null wil not be perfect - there will be some residual noise.
If you have two separate error correction mechanisms each leaving -150dB of residual noise, and the noise from the two sources is uncorrelated, it will add up to -147dB. If one mechanism reduces the error from the other, you still get -150dB residual from the first one.
Last edited:
No. An inverting amplifier strives to make the mid-point (i.e. its own input) exactly whatever it needs as input in order to produce the required output. Anything other than 0V is input; it includes an inverted and filtered form of the output distortion.hellokitty123 said:
If you struggle with this simple concept I wonder how you made any EC, let alone one with anything like your wild claims.
Repeating things which are untrue or confused does not make them true and clear. Looking from our side, we cannot easily tell between someone who will not tell us what he is doing and someone who cannot tell us what he is doing because he doesn't know hmself. You think we don't understand; we are sure we understand only too well.hellokitty123 said:
We are still patiently waiting for you to begin to discuss your circuit. Thus far you have told us nothing about it, apart from some amusing claims about what you think it can do.
- Status
- Not open for further replies.