On (03 Feb 97) Bob Lawrence wrote to Roy Mcneill...



 RM> I suspect the "real fault" was poor layout.



 BL> It's never layout. These things take off around 1-10 MHz and

 BL> layout problems don't begin till the tens of Mhz. My guess is

 BL> that a large proportion of *all* those converters oscillated,

 BL> depending on the actual transistors fitted... a design fault.



 RM> It's a fault I haven't seen before, so I wouldn't call it a

 RM> "large" proportion



 BL>   He, he. I'm not fazed by that. I've designed sets with inherent

 BL> faults that no tech *ever* found. A tech's approach is to keep things

 BL> working; an engineer's approach is to make things fail.



 BL>   With a power supply, dc/dc converter, or even an audio power amp,

 BL> the load itself is the main variable. They can work perfectly until

 BL> you get one particular funny load... but I still count that a design

 BL> fault. The engineer should have found it.



The load in question was a resistor, with about 40 turns at about

3" radius (a hot water system heater resistor). Added capacitance

(all kinds) didn't alter the instability.



 RM> So any Darlington pair is a potential HF oscillator?



 BL>   Not Darlingtons so much, which is the reason for their popularity.

 BL> It depends on the configuration. Adding a funny emitter circuit can

 BL> set them off.



More info? This converter had a current limiter looking at one of

the 3055 emitter resistors, but that was one of the first things I

disabled.



I'll say it again: the unstable circuit was a TIP41 driving three

3055s with emitter resistors. Adding a 10nF cap from the TIPs

collector to its base killed the oscillation. The opamp that drove

the TIP through a 470 ohm wasn't the culprit. Where was the

positive feedback coming from?



 I wonder if a cap across the 470 ohm resistor between the

opamp output and the TIP41 base would have done any good?



 RM> I've seen amplifiers of all sorts take off, and the problem is

 RM> either poor design (read: layout)



 BL>   I see. I've just told you that it is *never* (read: never) layout,

 BL> based on extensive actual design experience of these things and you

 BL> choose to contradict me. Okay



Poor layout includes poor ground layout. Poor ground layout can

cause hf instability. Ask any digital designer. Very poor ground

layout can cause audio instability.



Point taken, though; the grounding on this converter looked ok.



 BL>   I suppose you've confirmed this by fixing an oscillating amp or

 BL> DC-DC converter by changing the layout, and then changed the layout of

 BL> a good one to make it oscillate? Pig's arse, you have!



Yeah, and I've fixed floods by asking God to change the weather

pattern. Get real.





 BL>   Service techs just slap on a 0.01 and hope for the best - if you

 BL> can't fix it, modify it - and never find the real reason, never

 BL> understand the actual process, and accept wild guesses (read:layout)

 BL> as actual fact. You don't have to do any more. Most of the time, all

 BL> you guys have to do is fix something that an engineer designed right,

 BL> and when you find a bad one you use standard cures... slap a 0.01

 BL> across the collector base (0.01 is bit high, btw).



 BL>   Unfortunately, design engineers can't work like that. We have all

 BL> the time in the world to get it right. I offered you a little of that

 BL> experience but you know better.



 BL>   Pig's arse, you do.



I recall you suggesting a cause. I don't recall you suggesting a

cure. Real helpful.



 RM> I'm quite aware of the risks of sticking a low freq pole

 RM> partway down a feedback amp. In this case, however, it worked.

 RM> QED.



 BL>   If you understand it so well, why use such a ridiculously high

 BL> value? 1000pF would usually do the trick c-b, or 0.01uF b-e (at power

 BL> frequencies). All you have to do is move the already-existing pole,

 BL> and 20-times is usually enough.



1n probably would have worked, but I was running out of patience.

To do it right, I should have found the lowest value that could

keep it just stable, then stick in one about 10 times bigger for a

safety margin.







--- PPoint 1.88