Hi, PMA,
There's something I still dont understand about your drawing. The emitor degeneration is 5ohm, but the load is 100ohm. I assume this is a preamp. It is quite normal values for a preamp.
What is making the loop gain disappears? I think it is not the 5ohm RE. Is it the 741? If we replace with NE5532, the problem disappears?
There's something I still dont understand about your drawing. The emitor degeneration is 5ohm, but the load is 100ohm. I assume this is a preamp. It is quite normal values for a preamp.
What is making the loop gain disappears? I think it is not the 5ohm RE. Is it the 741? If we replace with NE5532, the problem disappears?
OK, fellow designers, let me give a design approach that works for me.
First, I am given a heatsink by the manufacturer. I usually have little or no real choice in what I get. Second, I am told by the client what kind of power output is expected from this chassis and heatsink.
Then I go to work. First, I must select the power supply output voltages to get the proper peak power into 8 ohms. Second, I set the quiescent current so that the total power dissipation at idle does not exceed a 55 degree C temperature on the outside of the heatsink with a nominal 25 degree C environment. Third, I select the emitter resistors to match the quiescent current with a drop of 15-25mV.
Now WHY do I design this way?
First, I want to be as class A as possible.
Second, I want the transistion between class A and class A-B to be as smooth as possible.
I know from experience that most people will listen at moderate levels. This is why class A is so important. Most of the time, they will be listening to a class A amplifer.
When they do use more power, I want the transition to be smooth, almost seamless. I work a lot on this.
And finally, if they want to reproduce an explosion or something, I have virtually unlimited current for a short period of time to cover almost any real load.
First, I am given a heatsink by the manufacturer. I usually have little or no real choice in what I get. Second, I am told by the client what kind of power output is expected from this chassis and heatsink.
Then I go to work. First, I must select the power supply output voltages to get the proper peak power into 8 ohms. Second, I set the quiescent current so that the total power dissipation at idle does not exceed a 55 degree C temperature on the outside of the heatsink with a nominal 25 degree C environment. Third, I select the emitter resistors to match the quiescent current with a drop of 15-25mV.
Now WHY do I design this way?
First, I want to be as class A as possible.
Second, I want the transistion between class A and class A-B to be as smooth as possible.
I know from experience that most people will listen at moderate levels. This is why class A is so important. Most of the time, they will be listening to a class A amplifer.
When they do use more power, I want the transition to be smooth, almost seamless. I work a lot on this.
And finally, if they want to reproduce an explosion or something, I have virtually unlimited current for a short period of time to cover almost any real load.
Re: Re: Class-AB Output Stage Dist vs NFB
Hi Andy,
I'm not really sure whether I would try to explain it that way or not, but I would not rule it out. Sometimes there are multiple ways to explain something that sort of end up being the same when looked at the right way. I think the significant thing here is that this open loop behavior is pretty much what you get with a conventional BJT Class-AB output stage (apart for choice of RE and bias level), and that this is a complex form of distortion.
That NFB behaves in such a predictable, as-advertized way with this nonlinearity is very good news. I literally dropped my jaw when I first saw this plot. I was not expecting this. It is yet another finding that raises questions about the feedback-is-bad position.
Cheers,
Bob
andy_c said:
Hi Andy,
I'm not really sure whether I would try to explain it that way or not, but I would not rule it out. Sometimes there are multiple ways to explain something that sort of end up being the same when looked at the right way. I think the significant thing here is that this open loop behavior is pretty much what you get with a conventional BJT Class-AB output stage (apart for choice of RE and bias level), and that this is a complex form of distortion.
That NFB behaves in such a predictable, as-advertized way with this nonlinearity is very good news. I literally dropped my jaw when I first saw this plot. I was not expecting this. It is yet another finding that raises questions about the feedback-is-bad position.
Cheers,
Bob
lumanauw said:
It is not a preamp, it is an example of strange behaviour of 3rd harmonic vs. loopgain.
The problem of 741 is the open loop gain, the highest trace (green). In case I used 5532, the result would be much better, and even better for OPA627.
Loopgain is a difference between open loop gain and closed loop gain (in dB). In case there is almost no open loop gain at a given frequency, there can be no loopgain. It was my intention to use this example (bad opamp and bad output stage) to emphasize problem. Anyway, some seem not to understand my intention 😉
john curl said:Heavy use of negative feedback promotes UNDERBIASING, not OVERBIASING. After all, if it measures OK, why add extra heat, etc? Wake up and smell the coffee!
Think of a small hill that you have created by G(m) doubling, rather than the open manhole cover (real xover distortion) and adding power to your automobile. You might note with pride that your 250 HP auto didn't even notice the hill, BUT just wait for what happens when you drive over that open manhole in the street.
Cute analogy, John. And I agree that I have always seen overbiasing as being more benign than underbiasing.
But to say that negative feedback promotes under-biasing is at minimum painting with an unfairly broad brush. We are not talking about Walmart amplifiers made in China here, where they might just do that and cut a bunch of other corners as well.
For you to insinuate that all designers who use negative feedback are doing so in order to be able to get away with under-biasing the output stage is plainly ridiculous.
I think that you are getting desperate as yet one more nail is driven into the coffin of the negative-feedback-is-bad school of thinking 🙂.
Bob
Bob, even your original amplifier is UNDERBIASED. Why did you not chose 1/2A like I would have? Could it be because it measured OK anyway? Does 'hiding' distortion really get rid of it? Never worked for me.
john curl said:Heavy use of negative feedback promotes UNDERBIASING, not OVERBIASING. After all, if it measures OK, why add extra heat, etc? Wake up and smell the coffee!
John, none of my power amplifiers has had less than 350mA bias current!!
john curl said:
Throwing the baby out with the bathwater again, John. Stop exaggerating. It adds nothing to the technical merits of the discussion.
Yes, I did overbias and use larger than normal RE on purpose to show the effect clearly.
And, Yes, I could run the sim again with an optimum bias and lower RE. I would expect the behavior to be the same, namely that NFB monotonically reduces the distortion products; but I admit, one never knows, and I should probably do the exercise.
One of the key things here is that the NFB predictably reduces the distortion in an over-biased design (or a design with the same absolute bias, but larger RE). Thus, the feedback allows us, if we choose, to more freely over-bias a design (getting further away from the risk of underbias in a thermal mis-tracking situation).
Cheers,
Bob
PMA said:
Hi Pavel,
Yes, I was thinking about doing something along these lines also. Since it is time consuming though, I'd like to have the simulation try to represent some kind of real-world situation, but with many real-world complications removed so that (hopefully) some general trends can be seen.
How about something like this. How about a Leach amp output stage with 2+2 BJTs, driven by a distortionless integrator, and a 6 Ohm load on the output stage. But instead of the less common 0.33 Ohm RE of the Leach, maybe use 0.22 Ohms, with Self's optimum bias for this resistor value. The integrator time constant could be varied so that the unity loop gain crossover would go from, say, 100 kHz (for very low FB) to maybe 2 MHz (for an aggressive high FB design).
I think John likes to do his harmonic distortion measurements at 5 kHz, so maybe choose that as the test frequency. Does that sound reasonable?
john curl said:
Hi John,
Thanks. This sounds pretty reasonable to me.
Have you ever done a dynamic bias stability test on your amplifier like Charles described several posts ago in, I think, the BJT-MOSFET thread? For example, monitor the total output stage bias, and make note of it at times during the test when there is no signal being applied to the amplifier. Turn the amplifier on with no signal and let it totally stabilize. Then run it with signal at, say, 1/3 rated power into 8 ohms for 1/2 hour, then kill signal and plot bias current vs time.
Bob
john curl said:
Are you talking about my MOSFET power amplifier with error correction?
If so, it was biased at 150 mA, but as you should know, there is no real defined "under-bias" for a MOSFET. It tends to be the more the merrier. And, indeed, I explicitly stated in the article that a big reason for using the error correction was to mitigate the transconductance droop of the MOSFETs.
The EC was not "hiding" any distortion, it was greatly reducing it locally in the output stage. I suppose if a person is phobic about NFB, they would be just as phobic about EC, but there is little I can do about that. The 0.001% THD-20 at all power levels up to full power was a very good figure.
Cheers,
Bob
It is underbiased because it is in class A-B too much of the time. I started with .5A in 1967. I have tried to go up from there with my later designs. Of course, only my BEST efforts. I do have to allow for cheaper, more compromised amps, that cost a lot less. I can only do the best that I can with those amps, and make a smooth transition.
Now let's see what negative feedback does with 50ma idle current? That is pretty typical.
Now let's see what negative feedback does with 50ma idle current? That is pretty typical.
High biased non-NFB output stage
Colleagues fellowdesigners,
I would like to share with you measurements made on my real-life power amplifier. High biased, non-global NFB, local NFB CFP diamond buffer based output stage, biased at 1A.
the 1st image is for 1W/8ohm, pure class A
http://web.telecom.cz/macura/pma1_1w_8o.gif
the 2nd is for 17W/8ohm, just after transition to AB
http://web.telecom.cz/macura/pma1_17w_8o.gif
Not bad results, are they?
P.S. Noise raise at higher freq. is converter noise shaping, not amplifier noise.
Colleagues fellowdesigners,
I would like to share with you measurements made on my real-life power amplifier. High biased, non-global NFB, local NFB CFP diamond buffer based output stage, biased at 1A.
the 1st image is for 1W/8ohm, pure class A
http://web.telecom.cz/macura/pma1_1w_8o.gif
the 2nd is for 17W/8ohm, just after transition to AB
http://web.telecom.cz/macura/pma1_17w_8o.gif
Not bad results, are they?
P.S. Noise raise at higher freq. is converter noise shaping, not amplifier noise.
john curl said:
John,
Are you serious? YOU should care. If your bias is walking all over the place, then you are sometimes under-biased and sometimes over-biased, and sometimes correctly biased.
Anyway, let's not lose sight of the point of this current very interesting discussion. It is about negative feedback and under what conditions its application makes things worse or better. It appears that the Baxandall results, while valid in the context of what they looked at, may have been a bit of a red herring in the bigger picture. It is especially significant that the application of negative feedback around a Class-AB output stage (that is not severely under-biased) improves things, rather than introducing new spectra or enhancing existing spectra under certain conditions.
Cheers,
Bob
Folks, thanks for your positive posts in relation to my queries. As a "build it, and watch it go bang" kind of guy rather than tweaking things on a sim. program, I find these difficult concepts easier to understand with real world examples. Keep up the good work, I might actually be learning something, ( I suspect this 'cos my head hurts when I think about it!). 🙂