dimitri said:
It is medium 😉
In the old days when tube amps were mainstream the rule of the thumb was 12 dB – 20 dB, But that had mainly to do with the frequency dependence of the output tranny and not to get the response peaky, in the high end as well as in the low end.
With SS it can be everything but keep in mind that you need to provide extra loop gain to make overall feedback possible. Whatever you do, it is wise to design each stage of the amp with the best possible linearity by itself.
Cheers 😉
john curl said:
Hi John,
I took data points at 0 dB NFB (i.e., no NFB, defined as by Baxandall as gain without feedback / gain with feedback) 6 dB NFB .... etc.
I did not take a data point at 3 dB NFB.
Sorry, but I'm not sure what you mean by tracking the scale.
Bob
But Dimitri, that is the point I am trying to make! Spice can't be that off, can it? By the way, I hope that you are enjoying your retirement.
john curl said:
Actually, Baxandall himself admits in his article that his formula predicting third-order distortion gives large errors for the high-distortion case where there is little feedback.
Maybe you could explain why this is, John?
Let me explain to everyone WHY I care so much about this. First, I was exposed to the mathematics of the 3'rd harmonic notch with negative feedback, back in the early 1970's. I was amazed, BUT it IS in the equations. Now where is the notch at 3dB feedback? IF SPICE ignores it, then SPICE is next to useless in this application. This is important to me, because I am told that I should be using SPICE on a regular basis, but I must know if it is reliable or not, if I can use it for anything that I am interested in.
john curl said:
John, I don't think I've ever told you that. I've always been of the view that people should use only the tools that they are comfortable with. If you are not comfortable with SPICE, it's not my concern. The solution to that problem is not to use it if that's the case.
But the formula for third-order distortion with feedback is what I was asking about. There are certain assumptions that need to hold for it to be valid. Baxandall himself, in his article, admits these assumptions are violated, causing significant errors in certain portions of his graph. In order to be able to say that a formula proves something, it's necessary that the formula not be used outside the set of assumptions under which it was derived.
This is the situation. Did you know that the 'natural' 3'rd harmonic distortion from an ideal bipolar transistor is expanding? It is only compressing when negative feedback is added, even local feedback. This is an interesting thing to know, because it shows that negative feedback is more complex than we would first assume. IF we are going to model bipolar transistors with different harmonics and different amounts of feedback, then the program better do it accurately or it is a waste of time.
You seem to not understand that I first used computer modeling of electronic circuits over 40 years ago, with mainframe computers, and IBM software. I found problems then that just about wrecked our efforts, and I still get a little suspicious of SPICE, when Bob Pease still hates it. However, many of you have never seen the real equations for a transistor, so for you, SPICE is all that you have known. This is OK for me, but I would prefer to be left alone regarding my own opinion about using it extensively. It really isn't that important to an experienced circuit designer.
You seem to not understand that I first used computer modeling of electronic circuits over 40 years ago, with mainframe computers, and IBM software. I found problems then that just about wrecked our efforts, and I still get a little suspicious of SPICE, when Bob Pease still hates it. However, many of you have never seen the real equations for a transistor, so for you, SPICE is all that you have known. This is OK for me, but I would prefer to be left alone regarding my own opinion about using it extensively. It really isn't that important to an experienced circuit designer.
Hi John,
I think we understand each other now. A "live and let live" viewpoint can go a long way toward preventing conflict and keeping the discussions constructive. If I use a certain tool and it's giving me some bad results, I'd like to understand why, and help fix things if I can. If I don't know whether it's right or not, I'd rather withhold judgment until I can look into it further. Sometimes such investigations can lead to unexpected discoveries and improved understanding. I like doing that sort of thing.
That may be true for some folks, I don't know. If you check out my web pages, you'll see that I've devoted some effort toward providing some much improved models for the OnSemi XJL1302 and 3281 power transistors. It turns out that the models that OnSemi provide are way off from the actual device performance. In order to fix this and create better ones, one has to know... you guessed it... the real equations for a transistor. So in doing that, one is not running away from the fundamental issues, but facing them head on. People who proactively face those issues might be a bit miffed if they are characterized as being someone who blindly trusts the tool. That's where I'm coming from.
I think we understand each other now. A "live and let live" viewpoint can go a long way toward preventing conflict and keeping the discussions constructive. If I use a certain tool and it's giving me some bad results, I'd like to understand why, and help fix things if I can. If I don't know whether it's right or not, I'd rather withhold judgment until I can look into it further. Sometimes such investigations can lead to unexpected discoveries and improved understanding. I like doing that sort of thing.
john curl said:
That may be true for some folks, I don't know. If you check out my web pages, you'll see that I've devoted some effort toward providing some much improved models for the OnSemi XJL1302 and 3281 power transistors. It turns out that the models that OnSemi provide are way off from the actual device performance. In order to fix this and create better ones, one has to know... you guessed it... the real equations for a transistor. So in doing that, one is not running away from the fundamental issues, but facing them head on. People who proactively face those issues might be a bit miffed if they are characterized as being someone who blindly trusts the tool. That's where I'm coming from.
You do realize that SPICE uses all the same fundamental equations regarding transistors that you do when you do a pen and paper analysis? SPICE just happens to also take into account numerous lesser effects that would impractically cumbersome to include in a hand computation.
SPICE is just computer assisted math with an interface optimized for circuit design. It seems just as likely that your analysis is incomplete as it does that SPICE is making things up.
It should be noted that Bob's sim appears to have been done with a complete transistor model, not an idealized one. It is possible that the internal Re of the virtual 2n5501 provides 3db or more of feedback already, which would of course hide the effect you are looking for.
SPICE is just computer assisted math with an interface optimized for circuit design. It seems just as likely that your analysis is incomplete as it does that SPICE is making things up.
It should be noted that Bob's sim appears to have been done with a complete transistor model, not an idealized one. It is possible that the internal Re of the virtual 2n5501 provides 3db or more of feedback already, which would of course hide the effect you are looking for.
Couldn't one of you clever guys with an AP1 or similar just build a circuit and test this? It would check both spice and John's calculations in the real world and be of benefit to us all.
Al,
could you kindly point to the circuit you exactly mean. It is possible to measure distortion like -100dB (-120dB) quite reliably.
Anyway, I doubt it would stop argueing, especially in -160dB/7th harm. order 😉
could you kindly point to the circuit you exactly mean. It is possible to measure distortion like -100dB (-120dB) quite reliably.
Anyway, I doubt it would stop argueing, especially in -160dB/7th harm. order 😉
I stand corrected. The 2N5551 is a very poor part for this example and it is operating at a non favorable current to accurately convey any real information. It is the example, rather than the simultation that is giving the misleading results. The 2N5551 could, in fact, have up to 2 ohms emitter resistance, which is very high for most any part.
PMA said:
I was thinking of the Baxandall stuff that was quoted earlier. Okay, maybe a better challenge, can we build something to test distortion at -160dB?
Bob's simulation use ideal math parts (summator, differentiator etc.) Any real component would bring its transfer function. There is a link to Baxandall's experimental circuit, a question again, parts creating feedback would differ.