Given the dramatically lower higher harmonic figures for the unity gain
simulation vs my real world part with an open loop gain of maybe 30 dB,
I wonder what you would see if the simulated component had more gain?
😎
simulation vs my real world part with an open loop gain of maybe 30 dB,
I wonder what you would see if the simulated component had more gain?
😎
It's not really an apples-to-apples comparison. The intent of what I did was to duplicate/verify Baxandall's work in order to explore some theoretical properties of feedback, starting with the simplest configuration he considered. That configuration has all open-loop distortion components except the second-order identically zero. Yours starts out with open-loop distortion components other than second-order that aren't zero. I would certainly expect the results to be different. In order to model such a case, the B-source would have to be modified to include higher-order distortion. But that wasn't the intent to begin with.
Andy, can you please think about OL circuits? When you will put two or three your 1 * x + a2 * x^2 blocks one after another you will get similar high-order products due to intermodulation fundamental and harmonics. I'm thinking how to introduce the metric to compare OL and CL networks.
dimitri said:
There's lots of possibilities, to be sure. Keith Howard did some interesting things with his AddDistortion utility. If I recall correctly, in its documentation, he had an algorithm for taking a spectrum at a fixed output level, and finding the Taylor series coefficients of the nonlinear transfer characteristic that creates that spectrum at that signal level.
I don't anticipate doing much more with this, as I'm getting ready to start a layout for my project. Your post over in the Otala thread got me thinking about this again, and curiosity got the best of me 🙂.
andy_c said:
I understand that I am asking for something that you are perhaps not
interested in, but to be clear, I was referring to a block with 2nd
order distortion and gain of greater than one. My interest is
simply to compare the simulation to a real circuit with that character.
😎
Andy, sorry to be a cause for this deviation. Please feel free to ask if you need any help (parts, layout, pcb manufacturing, etc, etc)
Nelson Pass said:
as Bob, Andy, (some of) the rest of us have said degeneration appears to act exactly as any other feedback in the sense of Baxendal's harmonic series multiplication
I simmed 3 "Square Law" gm blocks with
no feedback, green
local degeneration, red
+20dB gain block in unity feedback, yellow
output levels close enough to the same: (green fundamental is same level, "under" the red, yellow traces at 1KHz)
green shows only 2nd harmonic
red "local degeneration" shows full harmonic series
yellow shows greater reduction of all harmonics by the added loop gain
as usual LtSpice asc file - just rename w/o .txt
dimitri said:
Thanks a lot, Dimitri. That's very nice of you. No need whatsoever to apologize.
What specific things did you have in mind to look at? I'm in the "kicked-back" mode at the moment, but tomorrow I can look at something.
Hi,
people like M.G.Scroggie and Baxandall shed light on how global feedback works, a process which is much less obvious, at least for me, than the origin of dynamic distortions, as it is easier to suspect that high gain stages are susceptible to overdrive and missing gain, at high frequencies and near cut-off, gives rise to nonlinearity. For the very same reason, GNB is at a loss there, having hard time doing something about distortions of all sorts due to small bandwidth, and the odd high order harmonics and the even higher level of intermodulation products, produced by low biased push-pull output stages, rather attacking the desirable low order harmonics.
There's quite a distance between the idealized virtual world and harsh reality.
people like M.G.Scroggie and Baxandall shed light on how global feedback works, a process which is much less obvious, at least for me, than the origin of dynamic distortions, as it is easier to suspect that high gain stages are susceptible to overdrive and missing gain, at high frequencies and near cut-off, gives rise to nonlinearity. For the very same reason, GNB is at a loss there, having hard time doing something about distortions of all sorts due to small bandwidth, and the odd high order harmonics and the even higher level of intermodulation products, produced by low biased push-pull output stages, rather attacking the desirable low order harmonics.
There's quite a distance between the idealized virtual world and harsh reality.
It's interesting how nobody seems to take note of the error which I pointed out. I'm not silly, I saw an error that Andy C made and I simply pointed it out.
The other thing is that its all very well posting up pretty graphs of distortion, but what relevance is it?
After all nobody listens to square waves or sine waves, so it makes no sense in my view to analyse an amplifier using these.
We really ought to be considering real life music signals.
The other thing is that its all very well posting up pretty graphs of distortion, but what relevance is it?
After all nobody listens to square waves or sine waves, so it makes no sense in my view to analyse an amplifier using these.
We really ought to be considering real life music signals.
Dear Professor Smith;
people invent repeatable complex signals in order to test amplifiers because real musical signals are not well made for easy measurements.
Pretty graphs of distortions illustrate what I said before many times: either don't use feedback at all, or use it as more as applicable. Low feedback turns errors that our perception filters out into those that are audible even when are smaller and sound nasty.
As a professional mathematician you should know well that if to multiply a function by itself the resulting function will have higher order. That means, 2 stages with low order distortions will give higher order of distortions. And that means if to add a negative feedback you are multiplying the transfer function by fraction of itself multiplied by fraction of itself... and so on, that means even more distortions. And it does not matter, is it a "local" feedback around a single stage (like an emitter follower), or is it a global feedback over 2 or 3 stages, the result will be straightened transfer function, i.e. the function that on the graph looks close to linear one, but has higher order.
people invent repeatable complex signals in order to test amplifiers because real musical signals are not well made for easy measurements.
Pretty graphs of distortions illustrate what I said before many times: either don't use feedback at all, or use it as more as applicable. Low feedback turns errors that our perception filters out into those that are audible even when are smaller and sound nasty.
As a professional mathematician you should know well that if to multiply a function by itself the resulting function will have higher order. That means, 2 stages with low order distortions will give higher order of distortions. And that means if to add a negative feedback you are multiplying the transfer function by fraction of itself multiplied by fraction of itself... and so on, that means even more distortions. And it does not matter, is it a "local" feedback around a single stage (like an emitter follower), or is it a global feedback over 2 or 3 stages, the result will be straightened transfer function, i.e. the function that on the graph looks close to linear one, but has higher order.
andy_c said:
Andy,
I know it wasn't your intention but thanks for this link as it leads to a whole slew of utilities that can help to understand what's going on in our amps, our ears and, most importantly, our brains 😉
jd
Professor smith said:
Lumba Ogir said:
Gentlemen ??!
Firstly, it is well known that a fellow called Fourier showed that any periodic signal (thus including just about all real life music!) could be broken down into a fundamental and number of harmonics of particular frequency and amplitude. Further, it is known that the audio frequency region goes (nominally) from 20Hz - 20kHz. Thus limited, it is fairly easy to test for every aspect of audio equipment using Fourier analysis, i.e. a number of sine waves. Yes, Professor - we DO listen to sine waves, in the sense that whatever music you fancy, can be represented by said analyses, unless you want to contest the Fourier Theorem. Each one by itself - true, such sine waves would certainly not sound very intersting; but then nobody suggested that such a break-down is for listening to. But it does make for very easy analysis of sound equipment by the designer.
Next, the limits under which hearing operates has been fairly well researched. Thus, it is simple for the audio analyst to test, and when his equipment satisfies the conditions, to state that all music, however wonderfully intricate, will reach the hearing faculty unadulterated.
In that sense, Lumba Ogir, I fear, respectfully, that you are mistaken. It has everything to do with the sound. You might be able to point to a flaw by listening, if the part played is of the right composition to reveal the same. But will that enable you or the engineer to go to the appropriate part of the system to exactly correct the alleged flaw? I fear not; only measuring and analysis will.
To conclude, a square wave is simply a mathematical representation of an audio band full of signal, which enables the engineer at a glance to judge whether everything is in order. Any deviation shows him what to go look for in greater detail.
These are basics and rather off-topic so I will not expand - but you did ask the questions or made the remarks. If Andy C made mistakes it should be in order to point them out respectfully, but your statements are somewhat ironic judging by your own miscomprehensions - again respectfully, naturally.
Jan and Johan,
Thanks for your kind words. I seem to be experiencing some "ignore list leakage" here, though (through quotes). LOL!
Dimitri,
Did you come up with a scenario you'd like me to try out?
Thanks for your kind words. I seem to be experiencing some "ignore list leakage" here, though (through quotes). LOL!
Dimitri,
Did you come up with a scenario you'd like me to try out?
Thanks Bob. 🙂
Edit: I recall it was your idea originally to duplicate the Baxandall graphs with SPICE (in the earlier part of this thread).
Edit: I recall it was your idea originally to duplicate the Baxandall graphs with SPICE (in the earlier part of this thread).