Hi Andrew
PTAT = Proportional To Absolute Temperature.
Personally I try to keep the Vbias constant as the temperature increases to keep a a bit of a safety factor.
If it increases a little then I can say "I meant to do that".
I would be concerned if it increased more than PTAT.
Best wishes
David
Last edited:
Thank you.
In many of my earlier amps I played safe with tempco and actually liked to see the bias Vre reduce very slightly as the output stage temperature increased.
I now realise I was wrong to think that way and that holding steady, or increasing slightly, is better for ClassAB performance.
In many of my earlier amps I played safe with tempco and actually liked to see the bias Vre reduce very slightly as the output stage temperature increased.
I now realise I was wrong to think that way and that holding steady, or increasing slightly, is better for ClassAB performance.
Hi Richard,
Thank you for this offer, but my own analyzer will be ready within a month or so, in any event much sooner than that amp.
Cheers, E.
WOW, very good point, it is music to my ears. Thanks.
Ha ha, now you deserve to name it Oliver's Andrew condition. This is an important point!!!
Last edited:
Whenever you are experimenting with bias controllers, it's a very good idea to add a safety circuit across it, eg a Zener, so that if it all goes wrong the current will be under some sort of control. I did this with my Class-A design.
In response to an anxiety expressed earlier, the same amplifier had the whole negative-feeedback bias circuitry moving up and down with the output, much of it bootstrapped from said output to avoid loading the VAS node. Worked and works a treat, but I'd think twice about having a microcontroller whizzing up and down at 20 kHz.
Well the uC can do a peek, make a correction if necessary and then go to sleep for the next 98% of time. With uC's the whole paradigma changes. Not saying it is ideal or easy, but it is a completely different way of looking at the issue. And why would an electrolytic (for instance) across the bias generator wizzing up and down be 'better' or 'un-dangerous' and a wizzing uC not? It's just a blob of plastic really ;-)
Jan
Jan
Last edited:
My 200W CFA mosfet, you said it will not work, use a kind of Cherry compensation, is working and Richard Marsh tested and measured it.I realy think that you have to pay more attention to the CFA.
More of what I know about power amps I've got from your WW articles and latter from your book, but you have to expand your horizont too. It is never to late for curious mind to learn new things.
best wishies
Damir
Last edited:
Yes, Mr. Cordell should have a full chapter on CFA. I have been thinking about why Ostripper put so much effort on CFA. I now understand why. CFA has much higher frequency response.....not that we need the frequency response beyond 20KHz. BUT we need the loop gain at 20KHz ( you called feedback gain). We need a lot of loop gain at 20KHz in order to lower the distortion at 20KHz.
It is well documented in Mr. Cordell's book that the distortion is always a lot higher at 20KHz than 1KHz. Major part if not most of the reason is the loop gain of the VFA falls off. For example, if the -3dB BW is 400KHz, at 20KHz, the loop gain is only 20 or 26dB. That's not a lot of overhead to cancel out distortion. You need higher loop gain of say at least 100 or 40dB to do reasonable cancellation.
To get that in VFA, in order to get loop gain of 40dB, you need the amp to have BW of 2MHz IF it is dominant pole compensation. That's a lot for a big discrete power amp in VFA. The only way around it is to really play with the poles and zeros to get high loop gain at 20KHz, then using 2 poles roll off, then put a zero right before the 0dB crossover to give you back the phase margin.
CFA should be the answer for all these. It is by nature high frequency response and you should get good loop gain overhead at 20KHz. I don't know much about CFA, it is not easy to find a good article to explain how you optimize like how Mr. Cordell step by step lowering the distortion and increase slew rate.
Crossover distortion is a major problem and it happens at low signal level more. You can optimize it, but never eliminate it. In Self's book, it explain that adding more EF stages in parallel makes the kink smoother, which, is moving the distortion harmonics to lower frequency. This you can make use of the feedback to cancel out the distortion. Without loop gain, you are not going to cancel out those distortion even if you push it to lower frequency. Having plenty of loop gain is the ONLY way that I know of to eliminate( well say lower it further) it.
It is well documented in Mr. Cordell's book that the distortion is always a lot higher at 20KHz than 1KHz. Major part if not most of the reason is the loop gain of the VFA falls off. For example, if the -3dB BW is 400KHz, at 20KHz, the loop gain is only 20 or 26dB. That's not a lot of overhead to cancel out distortion. You need higher loop gain of say at least 100 or 40dB to do reasonable cancellation.
To get that in VFA, in order to get loop gain of 40dB, you need the amp to have BW of 2MHz IF it is dominant pole compensation. That's a lot for a big discrete power amp in VFA. The only way around it is to really play with the poles and zeros to get high loop gain at 20KHz, then using 2 poles roll off, then put a zero right before the 0dB crossover to give you back the phase margin.
CFA should be the answer for all these. It is by nature high frequency response and you should get good loop gain overhead at 20KHz. I don't know much about CFA, it is not easy to find a good article to explain how you optimize like how Mr. Cordell step by step lowering the distortion and increase slew rate.
Crossover distortion is a major problem and it happens at low signal level more. You can optimize it, but never eliminate it. In Self's book, it explain that adding more EF stages in parallel makes the kink smoother, which, is moving the distortion harmonics to lower frequency. This you can make use of the feedback to cancel out the distortion. Without loop gain, you are not going to cancel out those distortion even if you push it to lower frequency. Having plenty of loop gain is the ONLY way that I know of to eliminate( well say lower it further) it.
Last edited:
I am not even thinking about the development stage, I am worry about in the production stage. In order for servo to work, you need individual servo for each transistor. If one of the servo fails and it can get stuck at max current, that transistor is going to burn very quick. The servo fail, you have a good % of probability of driving the transistor to hog all the current and it will smoke. The more circuit you have, the more possibility something fails and it's not going to be pretty.
Yes, I would never consider hanging circuits, power supplies on the sensitive VAS note. This might sound good on paper, but in real life, you can create so much problems having a big circuit in the VAS and bias spreader. No offense, from the layout I've seen, I won't go there.
Another potential issue is uP run at some frequency, this will create noise. Don't think the frequency is too high to create problem, the IM can move down to audio frequency.
Last edited:
I agree that having a uC whizzing up and down at 20 KHz is not a smart thing to do. However, using a controller to modify the response of a 1st order Vbe spreader seems to be an elegant way to do things.
Quite true. Some american PA-amps from the eighties died sometimes during power-up due to the fact that some caps had to be charged and thus put an extra inrush current through the Vbe multiplier. I fixed that with one or two LEDs parallel to the VBE-multiplier.
Hi
What law of physics would prevent an Output Inclusive Compensation scheme?
Any feedback loop must be covered by the rules worked out by Nyquist more than 80 years back.
It is not as easy to apply this for some loops as for the main loop but the laws are the same, no need to put them into a different class.
The discussions in this forum have lead to better techniques to analyse the less simple loops.
I expect Richard will reply that he has actually done this stuff and Damir has mentioned his own amp.
So just to mention a few threads that can be easily found, where discussion helped me learn.
Toni "ASTX" has a very nice amp indeed with partial Output Inclusive compensation, while Paul "MC99UK" has built some stable prototypes with complex, somewhat similar schemes.
Finally, Mathias "Matze" has built a multiple nested Output Inclusive Compensation amplifier that is the most extreme example of the theory I have seen.
Not very practical but it does demonstrate that the theory works.
Best wishes
David
Last edited:
Absolutely nothing!
If you build your amp wid dis supa Virginal stuff and it doesn't release the Holy Smoke .. you will become one of the "my devices are hand-selected by Virgins after washing in liquid Unobtainium and the ONE amp I've built is OK" brigade ... and the design uninteresting.
Much more interesting if you do release the Holy Smoke .. preferably after 12 mths heavy use ... ie its a marginal case rather than absolute BS. 😀
__________________________
Thanks Guru Zan for answering Self on 'pure Cherry'. As several forum members have successfully used 'pure Cherry' to obtain supa THD20k with simple circuits .. and there's Prof. Cherry himself and even a couple of commercial designs .. perhaps Doug needs to do a bit of revision on the Laws of Physics .. and also refresh himself with Bode etc.
Last edited:
@ bwaslo & kgrlee
OK, well thanx for looking @ it & explaining. Just throwing out some ideas, that might have worked. ! Sometimes unconventional ideas do provide new solutions 😉
OK, well thanx for looking @ it & explaining. Just throwing out some ideas, that might have worked. ! Sometimes unconventional ideas do provide new solutions 😉
No I don't deserve that.WOW, very good point, it is music to my ears. Thanks.
Ha ha, now you deserve to name it Oliver's Andrew condition. This is an important point!!!
I just research my projects properly and have learned to sort the wheat from the chaff, when the science is not too complicated.
Well, I did not do any research whether you are the first, but this is an important point that really make sense. Go do some research to see anyone wrote about this, if not, I think you can do a few measurement to prove your assertion and submit a paper to HIFI journal or magazine.
I am serious and I really mean this. Don't think this is too simple to worth anything. Even Mr. Cordell quote your post to me. I am not one easily impressed, I meant what I said about Class XD and I meant what I said about this. I've seen paper published with less. This is much more than some people dream up with some little nifty circuits.
Ha ha, everyone knows VT=kT/q, Vbe=VT ln(Ic/Is), but how come nobody think about this here that I read?
If Mr. Cordell use this idea, you deserve a foot note.
Last edited:
But why would it then? What would 'create so much problems'? I concede that intuitively it may appear 'not good' but is there any basis for that?
It's not difficult to prevent this. After all, the uC doesn't have to run all the time. You can run it at very slow speed and put it to sleep 99.xxx% of the time anyway. Not a practical issue
Jan
No, I don't agree this is virgins something. This is engineering decision. I know plenty about guaranty by design, design for reliability and all. What I suggested is sound according to theory. AND it's cheaper than all this monkeying around with all the servos. AND I get lower distortion with more pairs.
If you can do it with one servo without touching the bias spreader, it might worth looking into. But I cannot see it can be done unless you use servo on each and every transistor. You decrease reliability with that much more circuit.
Not to mention if one of the servo fails, there is a good possibility that it hung up in making the transistor draw max current and you really going to have fire work. Doing all this to prevent current hogging is NOT good engineering. I think you guys do it just for the fun of it and it is not helping any new comers.
Indeed Jan.
Further, the whole MCU circuit is bootstrapped off the VAS midpoint so there really is not an issue with noise.
In most MCUs today, you can set an interrupt based on an A-D value, so the MCU would be sleeping most of the time - active for a few microseconds every few seconds.