Re: Re: Pros and Cons of THD measurement
Hi syn08,
Thanks! I agree, this distortion is a tough one. The good news is that it shows up on a THD-20 measurement.
I have indeed long been a believer in putting some significant capacitance right at the output stage devices to try to force the currents to flow and resolve locally. Obviously, there is a tradeoff in terms of board size etc. Interestingly, one approach is to put an X capacitor from the positive to the negative rail. To the extent that there is impedance in the supply wires back to the power supply, the X capacitor will try to make those currents sum together and be less distorted, since the sum of the collector currents in a Class B output stage will be largely linear.
I have also resorted to making a test coil of about 25 turns of #30 wire in about a 1/2 - inch diameter to probe around for nasty magnetic fields like this and like mains rectification pulses.
Cheers,
Bob
syn08 said:
Hi syn08,
Thanks! I agree, this distortion is a tough one. The good news is that it shows up on a THD-20 measurement.
I have indeed long been a believer in putting some significant capacitance right at the output stage devices to try to force the currents to flow and resolve locally. Obviously, there is a tradeoff in terms of board size etc. Interestingly, one approach is to put an X capacitor from the positive to the negative rail. To the extent that there is impedance in the supply wires back to the power supply, the X capacitor will try to make those currents sum together and be less distorted, since the sum of the collector currents in a Class B output stage will be largely linear.
I have also resorted to making a test coil of about 25 turns of #30 wire in about a 1/2 - inch diameter to probe around for nasty magnetic fields like this and like mains rectification pulses.
Cheers,
Bob
Re: Re: Re: Re: Re: rail induced distortion
Thanks, good to know. Let me know if and when you have some data from your practice.
Edmond Stuart said:
Thanks, good to know. Let me know if and when you have some data from your practice.
Re: rail induced distortion
That'll take a long time, as I'm still busy with the PMP amp. Then I've to draw the PCB (probably 4 layers) and let's hope I don't get new ideas in the mean time, which makes all the previous work obsolete.
syn08 said:
That'll take a long time, as I'm still busy with the PMP amp. Then I've to draw the PCB (probably 4 layers) and let's hope I don't get new ideas in the mean time, which makes all the previous work obsolete.
SY said:
For those here who have not been following the Blowtorch thread, I thought that I would point out that SY has been kind enough to post the exchange of letters between John Curl et al and myself that occurred about 30 years ago in response to my original article in Audio titled "Another View of TIM".
The original article is on my web site at www.cordellaudio.com.
The contents of the original rebuttal from John et al had been alluded to several times in various threads, but until now was not posted. Since the contents were a bit tied up in controversy, I am glad to see it and my reply posted together (neither was ever published in Audio). Those interested can read both letters and look at the original article and decide for themselves whether or not the substance of my article was incorrect in any area. I think the article was pretty much on target, but I'm interested to know if there are areas where I was wrong and would be happy to discuss.
The TIM controversy was a long time ago, but it was one of the main things that gave negative feedback it undeserved bad rap.
Thanks again to SY for posting the letters.
Cheers,
Bob
Gene Pitts was not shy about the "Let's you and him fight" approach
in editing Audio Magazine back then. A little controversy does wonders
for the circulation.
😎
in editing Audio Magazine back then. A little controversy does wonders
for the circulation.
😎
"Low feedback is good feedback"
"Condemnation without examination is prejudice"
John 😕
But though your slogan is the backbone of good research (I have often used it since I saw it below your posts) .... alas; even that is not as precise as it may appear to be. Nothing is perfect. One actually needs to say:
"Condemnation without proper examination is prejudice."
These previous two posts could not be serious! Yet I notice no smileys. So they must be. We have been round that route now .... how many times?
Bob,
Thanks (also to SY) for those references. Since it all appeared many years ago I took it that the invitation to comment was for THEN. If not, it amazed me that two "sides" could have that much similarity and still provoke such responses, especially from the quadruplets.
It is not necessary to re-open it again. The reply to Bob's question above lies in rereading those (all) good contributions. I fear it does not favour John's super-simplistic remark above, and it explains why - as was also shown in this thread and any other proper dissertation on NFB ever published.
"Condemnation without examination is prejudice"
John 😕
But though your slogan is the backbone of good research (I have often used it since I saw it below your posts) .... alas; even that is not as precise as it may appear to be. Nothing is perfect. One actually needs to say:
"Condemnation without proper examination is prejudice."
These previous two posts could not be serious! Yet I notice no smileys. So they must be. We have been round that route now .... how many times?
Bob,
Thanks (also to SY) for those references. Since it all appeared many years ago I took it that the invitation to comment was for THEN. If not, it amazed me that two "sides" could have that much similarity and still provoke such responses, especially from the quadruplets.
It is not necessary to re-open it again. The reply to Bob's question above lies in rereading those (all) good contributions. I fear it does not favour John's super-simplistic remark above, and it explains why - as was also shown in this thread and any other proper dissertation on NFB ever published.
😀 😀 😀 😀
But you are a very good jerk, John 🙂
I would not try, I might very likely succeed! (I would not reverse that challenge, John; you might not succeed! 😱 )
But you are a very good jerk, John 🙂
I would not try, I might very likely succeed! (I would not reverse that challenge, John; you might not succeed! 😱 )
Actually, I am a very knowledgeable and experienced jerk. I know better than most, have read more than most, and have made more successful designs than many here.
john curl said:
Hi John,
All of what you have said here is true, except that you are not a jerk. Although we diagree on some things, I personally over the years have learned a tremendous amount from you. It is not necessarily that you have taiught me directly, but you have challenged me to think hard and work hard to understand the origins of our disagreements, and at times to try and prove you wrong when I thought you were wrong. I can obviously say the same thing about Matti Otala.
My question "why" was actually a serious one. I'd really like to know why, knowing all that you know now, you believe that NFB is bad. I would like to know what that basis is, so that if there is a technical element, it can be discussed and examined. For example, do you believe that NFB is bad because of some known, measurable thing it does to the signal, or is it due to something that you speculate about, but which we cannot, or have not, been able to measure? Do we just chalk it up to the X factor, or is there some testable theory behing it?
If you tell me that in your own experience it just sounds better to have little or no negative feedback, but you have no idea why, that is an OK answer (but one that is difficult to pursue further).
I think that the answer to my question "why" gets to the core of this thread.
Thanks,
Bob
Bob,
few days ago I was exploring hard a power amplifier designed by a colleague of mine, for transient response and stability at various amplitudes and loads. The amp had very low and nice THD spectrum. It was quite complex topology, heavy NFB design.
When measuring the amp in time domain, with input step with rise time of 10ns, and whole range of amplitudes, with reduced RC filter at the input, I found interesting behaviour. The transient response is not a smooth function, but its derivative differs quite a lot for different amplitudes, in different parts of the step response curve. This, IMHO, depends on instant operating points (and charging/discharging of nonlin. junction capacitances) of the devices used. This phenomenon is masked, when we use input RC filter. However, I do not find this behaviour correct, and I blame it to complex circuit with high NFB. What would you think, if I may ask you?
Regards,
Pavel
few days ago I was exploring hard a power amplifier designed by a colleague of mine, for transient response and stability at various amplitudes and loads. The amp had very low and nice THD spectrum. It was quite complex topology, heavy NFB design.
When measuring the amp in time domain, with input step with rise time of 10ns, and whole range of amplitudes, with reduced RC filter at the input, I found interesting behaviour. The transient response is not a smooth function, but its derivative differs quite a lot for different amplitudes, in different parts of the step response curve. This, IMHO, depends on instant operating points (and charging/discharging of nonlin. junction capacitances) of the devices used. This phenomenon is masked, when we use input RC filter. However, I do not find this behaviour correct, and I blame it to complex circuit with high NFB. What would you think, if I may ask you?
Regards,
Pavel
10 ns edges implies significant levels of 10s of MHz frequency components
few audio power amplifiers have loop gain crossovers beyond a few MHz
basic feedback theory: negative feedback improves performance only at frequencies where the loop gain is large - at low frequencies
beyond the loop gain intercept the global feedback actually becomes positive feedback (but lower than unity so oscillations don't occur) and positive feedback increases distortions vs the open loop response
audio amp inputs should have low pass filtering to prevent high levels of high frequencies from causing internal distortion and consequently audio frequency IMD products
some robustness to RF/EMI detection and IMD generation can be had by degenerating bjt input diff pairs or using fet diff pairs - but keeping large levels of out of band signals out of the amp is still necessary
removing/reducing input filtering to inject a fast test signal for stability assessment is valid, but the amplitude should be low to stay with in "small signal" limits for linear stability - response to large signals and clipping may also be explored to look for large signal oscillations/nonlinear stability
but I don't see why the response of the amp with non-audio/out of band signals causing multiple stages to saturate should be considered relevant to audio reproduction if the properly filtered input doesn't have the high level/high frequency distortion exciting components
few audio power amplifiers have loop gain crossovers beyond a few MHz
basic feedback theory: negative feedback improves performance only at frequencies where the loop gain is large - at low frequencies
beyond the loop gain intercept the global feedback actually becomes positive feedback (but lower than unity so oscillations don't occur) and positive feedback increases distortions vs the open loop response
audio amp inputs should have low pass filtering to prevent high levels of high frequencies from causing internal distortion and consequently audio frequency IMD products
some robustness to RF/EMI detection and IMD generation can be had by degenerating bjt input diff pairs or using fet diff pairs - but keeping large levels of out of band signals out of the amp is still necessary
removing/reducing input filtering to inject a fast test signal for stability assessment is valid, but the amplitude should be low to stay with in "small signal" limits for linear stability - response to large signals and clipping may also be explored to look for large signal oscillations/nonlinear stability
but I don't see why the response of the amp with non-audio/out of band signals causing multiple stages to saturate should be considered relevant to audio reproduction if the properly filtered input doesn't have the high level/high frequency distortion exciting components
My design philosophy is different - my amps MUST work with input step of very short rise time, in ns. Then the input RC is applied.
jcx said:
That's very true, but I once encountered a strange phenomenon at high frequency sharp edges input. At the workbench my not-yet-so-well-stabilized amplifier started behave like a divide-by-two circuit. This was observed only at a certain frequency range, that's not covered by basic feedback theory.
Regards
Adam
PMA said:
Hi Pavel,
Off-hand, it sounds like the amplifier is entering the slew rate limiting region, although other strange things could be happening as a result of the extremely fast edges, such as output stage common mode conduction.
I would ask the following questions:
What is the rated power of the amplifier into 8 ohms?
At what power level and into what load were these observations made?
Would these observations be the same if the amplifier was run into no-load?
What is the measured slew rate of the amplifier?
What is the full-power THD-20?
With regard to operating without an input filter, even a good amplifier with adequate slew rate (say, 50V per microsecond or more for a 100w/8-ohm amplifier) may go into slew rate limiting with extremely fast input edges, and this is analogous to running the amplifier into clipping.
If the amplifier has adequate slew rate, and an input filter is used that is just enough to keep the demanded slew rate less than that which the amplifier is capable of, this is a better test.
What you are seeing is not necessarily a fault of NFB per se, although it could be the fault of improper design of the NFB.
Cheers,
Bob
Bob Cordell said:
Hi Bob,
I had the amp in my workshop for no more than 2 hours, so the number of records collected would be limited.
The amp gave 180W into 4 ohms. The output voltage swing with 4 ohms load quite same as without load.
I monitored the whole range of output amplitudes, starting at some 1V. At about 18Vp, the step response shape, that
was RC-like for voltages less or equal 18Vp, started to be slew-rate limited and started to exhibit the behaviour described earlier.
Yes, the observations were similar for no load.
SR measured was 28V/us. The amp was almost unstable to 3.3uF, but handled 390nF with damped oscillations.
I do not know THD 20, I am enclosing THD 1, which was almost same for all power levels and almost copied distortion of my workshop audio generator.
There was an input filter 1k + 100pF (100ns tau), which was insufficient for this amp.
----------------
SIMULATION RESULT
The colleague had the amp simulated in MC in his notebook. I played with simulation some 10 minutes and found quite same step responses as measured by oscilloscope. I played a bit with compensations (in sim only) and removed overshoots and increased SR to 58V/us (in sim). The simulated SR with original compensation was same as measured, 28V/us.
Even with modified compensations, the step response was not a smooth function. Overhoot was suppressed and SR increased to 58V/us.
The topology of the amp was unusual, combination of CFB and VFB. Three stages, differential VAS with 1 half cascoded, VAS buffer and 2EF output stage.
Regards,
Pavel
Attachments
Just thinking out loud...
I wonder if phase correcting the NFB would improve the sound?
yes, i know the phase error is very small.
I wonder if phase correcting the NFB would improve the sound?
yes, i know the phase error is very small.