You must mean "base stoppers" (this is BJT).
What values?
The original superamp didn't have them but his latest version added them, at 10ohms. But I can't get it to stop oscillating in simulations, and base stoppers don't stop anything.
I just made up a sim of just the OPS, without the resistor dividers setting the cascode bias, which I replaced with voltage sources.
I got it biased and working, and at input voltages less than about 35V it works about ok, but above that, it goes haywire, loads of distortion, and oscillations.
Trying to stop that and make it work right, but something is wrong.
What values?
The original superamp didn't have them but his latest version added them, at 10ohms. But I can't get it to stop oscillating in simulations, and base stoppers don't stop anything.
I just made up a sim of just the OPS, without the resistor dividers setting the cascode bias, which I replaced with voltage sources.
I got it biased and working, and at input voltages less than about 35V it works about ok, but above that, it goes haywire, loads of distortion, and oscillations.
Trying to stop that and make it work right, but something is wrong.
Well, oscillations would have nothing to do with how the amp is being used.
4.7 is what I've been using in sims lately, after starting out with the 10ohms that leach has in his latest superamp version.
But still, it just won't work without oscillations, and no amount of compensation can help or stop those.
What I can't understand is why an amp that works well in reality, just can't work right in simulations.
Makes no sense. Unless there is something specific about the cascode topo that calls upon some aspects of the models that cause this. It's true that when switching to different models, the oscillations change a lot, but never go away completely.
We just can't count on compensation and loop gain to correct the defects caused by oscillations. The loop gain can only do so much to compensate, and can't eliminate all distortion. Those pesky oscillations are causing a lot of extra distortion, but how could simple base stoppers in a real amp make it work right, and not in a sim? Beats me!
4.7 is what I've been using in sims lately, after starting out with the 10ohms that leach has in his latest superamp version.
But still, it just won't work without oscillations, and no amount of compensation can help or stop those.
What I can't understand is why an amp that works well in reality, just can't work right in simulations.
Makes no sense. Unless there is something specific about the cascode topo that calls upon some aspects of the models that cause this. It's true that when switching to different models, the oscillations change a lot, but never go away completely.
We just can't count on compensation and loop gain to correct the defects caused by oscillations. The loop gain can only do so much to compensate, and can't eliminate all distortion. Those pesky oscillations are causing a lot of extra distortion, but how could simple base stoppers in a real amp make it work right, and not in a sim? Beats me!
I didn't know about A.Grey, where can we get those?
But yes, I've been trying this out with many models from various sources, including Cordell of course, as well as fairchild, onsemi (modepex. yuk?!?), and others...
The thing is, the simulations works differently with different models, but keeps giving bad results not matter what. Although those same models work just fine in other sims, giving expected "normal" results, in the leach design, they don't give those same results.
Since those amps have been built over and over, for real, by so many people, using many different parts, it MUST be working, so why not in sims?
So it must be something with the models, that have some aspects that aren't bothering in usual circuits, but do in this cascoded topo.
What I'd like to do is use the major advantage of the cascoding to allow using 3055/2955 in a superamp with 60V rails, which would not push the devices beyond the specified limit for Vce0.
I haven't seen any such builds out there, and I'd like to see this done.
One thing I found out when I tried simulating only the output stage from the superamp with the 3055/2955 models at hand, is that they are likely representing the earlier vintage of devices, with a rather low Ft, possibly even the overly low Ft from those early devices at some 800khz.
One thing that hinted at this is the comparison with the MJ15003/4, which when I switched back and forth, showed a huge difference in bandwidth, with a comfortable one for the MJs, even though by today's standard they would be deemed "slow" devices, the 3055s were a whole lot slower, with the rolloff starting a little before 10khz, while it was beyond 1mhz for the MJs.
It would be nice to have tweaked 3055/2955 models to make them correspond to what's made nowdays, which are nearly on par with the MJs.
I did even try out the MJ15015/16 models, but those are modpex from onsemi, and gave the same wild results.
A leach superamp with 3055s would be nice to see working. With 60V rails (under full power), that would be about 200W on 8ohms, and with 3 device sets, it should handle a 4ohms load (2ohms accounting for most reactive loads).
The key is getting those models to work right.
andy_c was the guy with the models, but he only did a few parts - a very sharp person:
MJL3281A and MJL1302A SPICE Models
Why do you want to use such old parts? You don't want to Cascode them rather you
would put them in series as was done in the Double Barrel amp. Or you could just use
about half the rail voltage and run very high current amps bridged. BUT look at the SOA
on those old devices, you'd probably need 2X as many as the best modern devices. Other
old amps with series outputs are the Ampzilla and Tigersaurus.
If you start a new thread about your sim of that amp, I'll try to help. I'm also seeing near
oscillation with the Krell KSA-50 in sim but obviously people have built them with no issues.
I think wiring inductance, especially to the base of the outputs acts as a base stopper, but
I don't think that is the whole story.
One tip is not to trust any of the models, I would use all models from Bob Cordell and just
sub similar parts to the small signal types. Note that you can use devices that don't meet
the breakdown voltage since that is not modeled in SPICE. BD139/140 is fine for med power
for example.
Last edited:
That's unfortunate when someone gets overly sensitive in open discussions. It's all in good fun and constructive...
Really? not Forums?
(like Cactii for Cactus)Anyway, is there some place to look? I didn't find anything searching...
Got those. If this was what andrew was referring to, then I already tried them.
That argument keeps coming all the time, but there are people in some parts of the world who are looking for such builds, not only because those old parts are cheap for them, but also because they can be easily found there.
And there are people, like me, who have old stock of such parts, and want to put them to work. I have dozens of those puppies on hand, and it'd be nice to see them pulling their weight, and if possible in ways they were never thought they could long ago.
I think despite their ancient status, they can still be used today, and they're still popular and manufactured.
Plus it's a good challenge to try to make those old things do more than they were meant to back then. With their main limit being their overly low Vce0, finding ways to use them with more voltage is a nice trick.
Now if you're going into the details of semantics... The leach superamp is all cascoded, and as far as I know, those arrangements at the output stage, altough in series, is also called cascode (common base).
I've been working on that, and a good way to push them to their limits is a grounded bridge, and that was the subject on other threads, still ongoing project.
Doing this is an entirely different project. I was thinking the superleach would be a good thing with 3055s to see those 60V devices work in a setup with rails at 60V. Actually because of the extra overhead from the cascoding, I suspect the SOA would not be violated (not much) even if the rails were pushed to some 65V to compensate.
And then that superamp can also be bridged, so that would make a beefy amp made of devices that were never meant to deliver that much.
I've been looking at the SOA, and just now was running such sims, and a 3055 based superamp with 3 sets would do just fine for 4ohms load.
I'd have to see what happens with the protections under a dead short, but in normal usage, they do just fine.
That's nice, maybe later. I'll see how it goes with the models.
Could be a similar issue, with models needing tweaking to be more like the real thing. There is no good reason for a simulation to fail when the real builds work just fine.
Could be. Layout can make a difference.
A good one! And I always have doubts, and try to figure out how to fix them.
I already do use those, and more.
Sure. And I already do often. But always keeping in mind the real thing's limits.
I used Cordell's and Fairchild's in this superamp, and nothing good came out of it.
Those models are among those making it totally unstable, although I have used those same models a lot in many other sims, and they always worked just fine there, just not right at all in the superamp. Beats me!
I was doing a sim lately on a grounded bridge designed by John Ellis, that works quite nicely, and we had been developing it for use with 3055s as well, which worked very nicely with good performance, despite the old devices. I just could not get the protections to work as intended. However, what I've been trying lately is to scale it down a little with BD139/140 for the outputs, with 4 pairs of outputs (per side of the bridge), and rails at 30V, to be used as a headphone amp, which could be capable of driving the whole range of possible headphone impedance (16ohms to 600ohms).
Not so easy to do, with such a wide impedance range, but it works and there I am having the same issue again making the protections work properly.
I haven't yet tried the protections on the superamp, but will try that later on.
I just would like it to work properly with the 3055s first. The 3055s are a prime example of the need to cascode to get over their low Vce0 limit, beyond that, with today's devices, there is no need to complicate that much, so cascoding is no longer needed, and it's rather tricky to get working it would seem anyway...
Andrew,
in pure latin language it is 'fora' (plural neutrum), in english it is as usual
' forums'.
BR
The intent of a cascode is to hold the Vce of the main transistor constant and usually low
in order to minimize the Miller effect.
This is done by having the second transistor act as an emitter follower to the collector of
the first transistor with a reference on the base. The second transistor in in a common
base configuration.
A series connected output stage is just that with the intent being to split the Vce in half
between the two transistors. So in this case the base is driven by a resistor divider of
half of the total split.
You need to use slow models for those old devices, not sure where you might get good
ones. The Leach often used a small amount of Cdom compensation, upping it should
make it stable, then when you build it measure the margins and reduce it a reasonable
amount. You might try other better compensation methods. TMC for example.
Anyway, good luck to you.
in order to minimize the Miller effect.
This is done by having the second transistor act as an emitter follower to the collector of
the first transistor with a reference on the base. The second transistor in in a common
base configuration.
A series connected output stage is just that with the intent being to split the Vce in half
between the two transistors. So in this case the base is driven by a resistor divider of
half of the total split.
You need to use slow models for those old devices, not sure where you might get good
ones. The Leach often used a small amount of Cdom compensation, upping it should
make it stable, then when you build it measure the margins and reduce it a reasonable
amount. You might try other better compensation methods. TMC for example.
Anyway, good luck to you.
Hi spookydd,
When I built member Pingrs version of the Leach amp, I noticed it has an extra slot for a small 10pf caps in the pcb.[could not remember though if it was intended for the drivers or outputs]. I was told that it was an extra cap to prevent oscillation if it ever occurs. It was version 4.5 though but you may want to ping member Pingrs [Brian Pingres] he had more info about the add'l compensation cap.
Hope it helps!
Sa muli,
Albert
The Superamp with lots of part count and a not so simple schematic would be easy to get wrong in simulation (build too) so my advise is to check connections very carefully.
Go to Leach site in the original (not superamp) amp, there is LTspice file. I tried it and it worked. You can start with this and slowly add the cascodes, test every cascode pair added and save then continue until output transistors.
Go to Leach site in the original (not superamp) amp, there is LTspice file. I tried it and it worked. You can start with this and slowly add the cascodes, test every cascode pair added and save then continue until output transistors.
Right! I had not examined the superamp output stage enough before to realize it's not really a cascode and the devices are simply "stacked" to share the voltage.
Now I see it, and only the front end and vas are cascoded in the superamp, and the rest of it is just stacked (totem pole?).
But this works out, so the devices are only exposed to half the voltage, so it fills the purpose of being able to use lower Vce0 devices where they would normally not be suitable. And that's where the superamp topo comes in, for making possible the use of 3055s with rails at 60 or 65V.
Well, not too slow, or the bandwidth can get limited.
I've been comparing using it with 3055s and the MJ15003/4, which are much faster, although definitely not among the deemed "fast" types, the difference is huge. My early tests showed the 3055s caused the rolloff to start near about 10k, while the MJs rolled off much higher at some 1Mhz or so, or a bit above. I even noticed a small bump, a slight gain peak above nominal, before the rolloff at the high end with the MJs.
This to me hints at the 3055s models on hand being much slower, and not resembling the newer manufactured versions, which are almost as fast as the MJs.
I've been playing with that to try to keep it stable, but it's not enough to curb those pesky oscillations occurring in the output stages. We may arrive at minimizing the effect overall of those oscillations, with the closed loop, but the distortion remains too high because the output stage continues oscillating.
We can't get a fully stable and low distortion amp if the oscillations aren't stopped.
My main problem with the simulation has been mostly at the high end of the spectrum and at higher power levels.
With a small enough signal, it's possible to have the amp work without oscillations, but that's no fix.
I've been having it stable below 5k, with nice and low distortions, even at full power, but with the frequency increase, the oscillations start.
And that's what I've been trying recently, and since I don't have a good understanding of Leach's compensation scheme to be able to calculate values, I didn't dare playing with it, so I only touched the Cdom and tried adding caps in some places in the output stage, hoping to put an end to those oscillations.
TMC has worked much better than his compensation to keep it stable, although the phase and gain are margins obtained are far less than with the Leach compensation.
Obviously his compensation method makes it easier to make it super stable with extravagant phase and gain margins. I've had it work with nearly 90 degrees of phase and a little more than 30db of gain margins. Far more than really necessary.
But although that "should" be making for a stable amp, it wasn't, with the output stage oscillating a lot.
Cool, and I think 60V rails is perfectly safe. Actually my early SOA measurements show that with 60V rails, the Vce never gets quite to 60V, with the various losses, overhead, plus a real amp would see its rails sag when loaded under full power, so the rails could be pushed a little higher and still be within the intended Vce0 limits of the devices.
Then for those who have what it takes to test the actual Vce0 true breakdown voltage, it can then be determined if the devices can sustain more than the 60V, which is often the case, and then the rails could actually be even higher. But then the dissipation could become a problem, with the SOA being used more fully.
The key is to have the protections work right to keep it safe. That's something I plan to look at, when I can get it to stop oscillating.
It's no real big deal, as more sets of devices can be added and that would push up the available SOA some more.
My superamp sims have been with 3 sets of 3055/2955. The MJs can do the same easily with 2 sets.
I've tried more caps all over the place, and once the Leach compensation has been removed and replace by TMC, plus the sprinkled caps here and there in the output stage, as long as models that work are used, it can work.
Right now I need to figure out what's causing the 3055 models to work so badly with it. Not just them, using BD139/40 for vas and predrivers also cause problems, and those models are from Cordell's lib. I the KSA1381/KSC3503 from fairchild also cause big issues, even though they've been fine in other sims.
I just got it stable now even at full power and at 20khz, which was where the oscillations were the worst and unstoppable, and that was using the 2SA1381/2SC3503 for the vas/cascode and the pre-drivers from Cordell's, and the MJE15032/3 from Cordell's don't work right in this sim, but the MJE15030/1 from an other source did work, as long as no 3055/2955 or similar are used, it works if the power devices are the MJL21193/4 from Cordell's.
Using the models that come with the published sim for the superamp doesn't work too well either. There are oscillations often, and they also can't be stopped.
With that sim with the original models, if the outputs are switched to Cordell's MJL21193/4, then it can be made to work, although the distortion is quite high. And that's leaving intact Leach's compensation, and only 2 sets of outputs as the superamp was designed.
What I'm trying to do is make it work right with 3055/2955, with 3 sets of outputs, and bring the rails at 60-65V. If this can work properly, it would make a nice 3055 amp. Bridge it and it's a screamer 3055 amp.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.