G.Kleinschmidt said:
I've given the Sanken parts the skip in favour of the non-ThermalTrack version of those devices for my latest design for nearly all the reasons you just listed.
Cheers,
Glen
Glad to hear it, Glen.
Bob
Re: S-K Filter Simulation and Monte Carlo
Thanks, I had not thought of that, and I'll try it.
WTR to the converse case, at least that corresponds to a real
behaviour. Oscillation will often break out in only a portion
of the operating swing.
😎
analog_guy said:
Thanks, I had not thought of that, and I'll try it.
WTR to the converse case, at least that corresponds to a real
behaviour. Oscillation will often break out in only a portion
of the operating swing.
😎
Re: S-K Filter Simulation and Monte Carlo
Hmmm... I gave it a good kick, and it still doesn't oscillate.
😎
analog_guy said:
Hmmm... I gave it a good kick, and it still doesn't oscillate.
😎
later...
I set it to a much longer observation period, kicked it again,
and it very slowly lurched into oscillation after a rather long
time.
😎
I set it to a much longer observation period, kicked it again,
and it very slowly lurched into oscillation after a rather long
time.
😎
john curl said:Yes, Nelson. That would be greatly appreciated!
Send me your shipping address.
😎
Hi Tim_x,Nelson
Alternatively, you should match the transistors in practice! I've pointed out in previous threads that I match transistors wherever they are needed. This means driver pairs, output (parallel) pairs and diff input pairs, current mirrors, etc. Matching small signal devices is almost as easy as falling off a log, since several purchased at once will almost certainly have the same date code, and maybe from the same wafer (if you ar lucky) or at least the same lot run.
I have in the past used a series of models for individual devices, but with suffixes, so that they represent different gain spreads (and when plotted will give a Gaussian distribution). But it's hard work as model parameters are in some cases correlated, and each variation has to be determined by some knowledge about what causes the variations.
And of course, one reason lots of feedback is used is to minimise the spreads in transistor/passive component variation.
I've also used the kickstart method to simulate a crystal oscillator. Ever tried? the huge equivalent inductance and small capacitance really takes some momentum to get going (i.e. it never starts without being kicked). I just used a voltage source with a single pulse (in this case 500V/1uS).
Nelson's point about models never being right is correct in that some effects aren't modelled well in the older generations of simulation. I can believe that single precision simulators got things badly wrong, but 64 bit precision, which is now standard, should be good for quite a low % distortion. I agree though that some effects such as quasi-saturation resistance, which the gummel-poon/Level 3 type models don't have, and breakdown effects (which some models don't have) can make simulations optimistic. Thermal effects are only just being thought about, and no doubt soon we can plug in our thermal insulators, heatsinks and watch for temperature effects (it might be quicker to build the amp and measure). But if you want your 1 ppm amplifier, I still think that the simulator needs to say it is possible with the tools available. Then you check.
cheers
John
Alternatively, you should match the transistors in practice! I've pointed out in previous threads that I match transistors wherever they are needed. This means driver pairs, output (parallel) pairs and diff input pairs, current mirrors, etc. Matching small signal devices is almost as easy as falling off a log, since several purchased at once will almost certainly have the same date code, and maybe from the same wafer (if you ar lucky) or at least the same lot run.
I have in the past used a series of models for individual devices, but with suffixes, so that they represent different gain spreads (and when plotted will give a Gaussian distribution). But it's hard work as model parameters are in some cases correlated, and each variation has to be determined by some knowledge about what causes the variations.
And of course, one reason lots of feedback is used is to minimise the spreads in transistor/passive component variation.
I've also used the kickstart method to simulate a crystal oscillator. Ever tried? the huge equivalent inductance and small capacitance really takes some momentum to get going (i.e. it never starts without being kicked). I just used a voltage source with a single pulse (in this case 500V/1uS).
Nelson's point about models never being right is correct in that some effects aren't modelled well in the older generations of simulation. I can believe that single precision simulators got things badly wrong, but 64 bit precision, which is now standard, should be good for quite a low % distortion. I agree though that some effects such as quasi-saturation resistance, which the gummel-poon/Level 3 type models don't have, and breakdown effects (which some models don't have) can make simulations optimistic. Thermal effects are only just being thought about, and no doubt soon we can plug in our thermal insulators, heatsinks and watch for temperature effects (it might be quicker to build the amp and measure). But if you want your 1 ppm amplifier, I still think that the simulator needs to say it is possible with the tools available. Then you check.
cheers
John
Hi John,
We are in complete agreement on the subject of transistor matching. Experience bears this out.
-Chris
We are in complete agreement on the subject of transistor matching. Experience bears this out.
-Chris
Interestingly, some of the best results I've seen have not been
cases of matching per se but from swapping parts in and
out and tweaking current and resistance values around each
one. If you do it with a distortion analyzer, you can often find the
more narrow "sweeter spot" than with matching alone.
😎
cases of matching per se but from swapping parts in and
out and tweaking current and resistance values around each
one. If you do it with a distortion analyzer, you can often find the
more narrow "sweeter spot" than with matching alone.
😎
Nelson Pass said:
Perzackly. The distortion analyser is already looking at the second order effect (rather than the first order). Useful for those of us who can measure distortion, but perhaps less useful for everyone else. Mind you, I've often found that such distortion minima (in valves, not semiconductors) tend merely to minimise H2, and possibly increase higher harmonics, making them less useful. There's just no free lunch.
I'll go away now and leave you to semiconductors...
Hi Nelson,
I used to just match on beta, but I recently built a "diff pair" tester. It's just a current source that you plug in your hopefuls and look for a collector (or drain) current balance. It mimics a diff pair without any feedback. That gives me very tight matches.
-Chris
Hi Nelson,
I find it interesting to experiment to see what make a real difference. I'm sure you do the same on the bench. This does as you do, but with less bother and no soldering iron. 😉
-Chris
There's the rub. This little circuit comes with sockets. My other ones don't.
I find it interesting to experiment to see what make a real difference. I'm sure you do the same on the bench. This does as you do, but with less bother and no soldering iron. 😉
-Chris
Bob Cordell said:
Hi John,
I'm still interested in learning your experience and opinion on the OnSemi ThermalTrak transistors.
Cheers,
Bob
Nelson Pass said:
Hi Nelson,
Have you used LTspice, and if so, what has been you experience of Microcap versus LTspice?
What do you do for MOSFET models?
Thanks!
Bob
Are you referring to me, Bob? You should contact Charles Hansen. He is the going expert on ON semi at the moment. I don't need these devices in my current designs at this time, but for a smaller amp, they might be perfect. Why? Quiz on Friday.
john curl said:
Hi John,
Yes, I was referring to you. Now I'm really curious with regard to your quiz question about why these devices might be perfect in [only] a smaller amplifier.
I'm guessing that you are referring to the TO-264 package for the OnSemi parts as opposed to the double-screw package of the Sanken parts. I'm guessing that you like the larger metal surface of the latter to obtain a smaller theta CS. Am I right?
If I'm right, that would seem to make sense in amp designs where the choice of the number of paralleled output pairs was dominated by average junction temperature considerations as opposed to concerns about short-term SOA or high-current beta or ft drop-off. Am I right, or have I missed something.
Thanks,
Bob
john curl said:
Hi John,
I believe the "A" version of the devices is essentially the same, but without the ThermnalTrak diodes. I was mainly referring to the BJT characteristics of the OnSemi RET devices versus the Sanken devices.
Cheers,
Bob