It does look rather strange, but maybe it's the way it's drawn that causes more confusion.
I don't see a resistive divider, so how is this made?
I was just looking at that and the margins along the way. Did you check that part? I am seeing about 26db of open loop gain. That seems odd, as the closed loop is about that much, so there would be nothing to reduce distortion.
If this is the case, it would explain why thd is so high.
I didn't have much luck previously in that sim with the bd139/40, but now it's all quite different, and even the 3055 models are working much better.
The same design can be used with different sets of parts, so those wanting to make a 3055/2955 based amp, should be able to without any troubles.
They do, but it's still quite possible to obtain fairly nice performances from them. I've had many sims done with various topos to try out 3055 amps with good performance and "unusual" topos. The best performances I was able to get from 3055s was with the bryston topo. But I also got rather nice performance as well with a design from John Ellis, a grounded bridge. Unusual as well, but works great.
Sure, we don't need anything stacked beyond the power devices.
Everything can be modern, non stacked, and the last output stage can use the 3055s and it's easily possible to use modern ones there too.
A single design can accommodate all that. It would just be a choice at build time. Perhaps at most a few values to adjust for rail voltages.
The 2N3442 could be an alternative to the 3055s probably with hardly any changes, and with a bit higher rails, since they can handle it easily. They are lower current, so perhaps more device sets would be a good thing to do with those.
I'll look at all this tomorrow, but in the mean time, here is a modpex onsemi model for the 2N3773:
.MODEL 2N3773G NPN ( IS=4.27569e-10 BF=110.833 NF=1.04275 VAF=24.8913 IKF=9.09316 ISE=7.75051e-12 NE=3.34364 BR=2.12581 NR=1.08667 VAR=3.43591 IKR=14.4171 ISC=6.25e-13 NC=3.90625 RB=3.19163 IRB=0.1 RBM=0.1 RE=0.000132792 RC=0.0777476 XTB=0.1 XTI=1.02282 EG=1.05074 CJE=1.49308e-09 VJE=0.890172 MJE=0.698851 TF=3.95e-10 XTF=1.35721 VTF=0.995664 ITF=1 CJC=5e-10 VJC=0.95 MJC=0.404162 XCJC=0.803125 FC=0.8 CJS=0 VJS=0.75 MJS=0.5 TR=1e-07 PTF=0 KF=0 AF=1 mfg=onsemimodpex )
Knowing it's modpex, there is reason to doubt its quality, but they always say they match the specs from datasheets...
Can be tweaked though. If needed.
.MODEL 2N3773G NPN ( IS=4.27569e-10 BF=110.833 NF=1.04275 VAF=24.8913 IKF=9.09316 ISE=7.75051e-12 NE=3.34364 BR=2.12581 NR=1.08667 VAR=3.43591 IKR=14.4171 ISC=6.25e-13 NC=3.90625 RB=3.19163 IRB=0.1 RBM=0.1 RE=0.000132792 RC=0.0777476 XTB=0.1 XTI=1.02282 EG=1.05074 CJE=1.49308e-09 VJE=0.890172 MJE=0.698851 TF=3.95e-10 XTF=1.35721 VTF=0.995664 ITF=1 CJC=5e-10 VJC=0.95 MJC=0.404162 XCJC=0.803125 FC=0.8 CJS=0 VJS=0.75 MJS=0.5 TR=1e-07 PTF=0 KF=0 AF=1 mfg=onsemimodpex )
Knowing it's modpex, there is reason to doubt its quality, but they always say they match the specs from datasheets...
Can be tweaked though. If needed.
I will look more closely later, and compare to the ones I have, but at first glance, the BF seems odd, and VAF probably should be a little higher. Plus the CJC may also be off. And likely others of course...
The 3055 datasheets don't always provide the BC capacitance. I only rarely have seen it mentioned, but I have a mil spec one that states 700p.
And in some models I have the CJC is at 500p. Somewhat close, but not very.
The 50/50 voltage divider in the original simulation is R52 and R54, then follow the BE
junctions for the triple EF to the series outputs. There are tap off points for the VAS
series helper, etc.
junctions for the triple EF to the series outputs. There are tap off points for the VAS
series helper, etc.
I try not to use any modpex models. I do think that I've tried that one in the past and it at
least works but I'm not going to chase down distortion issues with questionable models.
I'd use a Cordell MJ2119X model before anything from Onsemi.
least works but I'm not going to chase down distortion issues with questionable models.
I'd use a Cordell MJ2119X model before anything from Onsemi.
Last edited:
Modelling Info
I posted this link on here back in 2010:
http://www.intusoft.com/lit/WkwModels.pdf
I skimmed their newsletter back in the 80s and 90s but have never really dug into making or
fixing models, perhaps someday. I've not read that in detail.
Anyone on here used the info in that paper to make models?
I posted this link on here back in 2010:
http://www.intusoft.com/lit/WkwModels.pdf
I skimmed their newsletter back in the 80s and 90s but have never really dug into making or
fixing models, perhaps someday. I've not read that in detail.
Anyone on here used the info in that paper to make models?
Locky_z's curve tracer is impressive and i think has sold at least a few hundreds if not more. I have two and use them extensively for measuring power devices etc. AFAIK he now only sells them as fully built PCB so you need to add case and a PSU. See this link on ebay: locky_z's Intelligent curve tracer PCB(New) | eBay
There are a few threads on this curve tracer on diyaudio and i can safely recommend it for those in search of a dedicated device (that does power devices) and with the funds to buy it ($180 delivered).
It only works on windows.
With respect to the digilent you're right in that its a 'cobbled' together solution using their digilent analog discovery 2 USB scope and waveforms software. Even after patching things together, it (currently) has nowhere near the flexibility and ease of use of the CT above. However, you can also use the box with free 3p software as a very good audio analyzer with differential inputs. The software is waveforms and works on ARM, Linux, Windows and Mac.
The thread i was referring to is: https://forum.digilentinc.com/topic/3556-scripting-example-for-mosfet-vgs-id-matching/
Someone with time time and interest can probably extended this work to measure power BJTs..
If you can live with windoze only platform, then the locky-z would be my bet for the kind of work you want done.
Last edited:
Perhaps, but at least one of them was real, even though it may not be fully reliable. It was then copied and made into the opposite. I've done this a few times when models can't be found for both polarities. Sometimes we have no choice.
Or making up the whole model could be done, from datasheets and measurements, but that's a whole project. Not so simple.
I kept digging into that digilent stuff, and found that earlier versions of their software did have a wider platform support. If they don't make updates for all the platforms any more, we're out of luck.
But the issue I see about digilent is there aren't any cheaper options.
However there are others to look at. For example picoscope:
https://www.picotech.com/oscilloscope/2000/picoscope-2000-overview
And they have more platform choice and cheaper options.
Still even though it seems feasible to use things like this as part of a scheme to do measurements and matching, it is rather cumbersome and it takes a lot of manipulations and wiring.
I would like a ready made system that just plugs in and we can load software and start plugging devices to be tested and then use the stored measurement data to do comparisons and sorting.
Something more customized for the purpose, no extra wiring besides plugging the tested devices in, and no remembering how to put it all together every time.
This is basically the original leach amp.
The original worked in sim, but I could never get really good thd performances.
The same thing seems to be recurring, even though it has excellent margins, way more than required, and enough feedback, it's not enough to counteract the tiny mostly visually invisible oscillations and the spikes to get the thd lower.
Maybe the spikes are commutation artifacts of some kind, perhaps cross conduction or something of that nature...
When running transient sims, take a peek at the signal in its path, and you can see it doesn't look so pretty along the way. Fortunately the global loop manages to iron out most of that stuff, but obviously not enough for really good thd.
Ok, I looked closer at those models, and obviously the 2955 was just made by copying the 3055 to make it into the complementary.
And this is the identical model as the one Jan Didden shared before.
That model has far fewer parameters than most other models I've seen, so it's really simplified and I doubt it could be too representative of the real thing. I would say insufficiently defined to model a device that would be close enough to a real one.
I've tried it before, in other sims, and found it flawed.
One of the models I've been using more recently is one that seems to have come from TI and possibly came from the orcad library. It's the one called 2N3055tx in my posted sim.
.model 2N3055tx NPN ( Is=974.4f Xti=3 Eg=1.11 Vaf=150 Bf=99.49 Ne=1.941 Ise=902.5p Ikf=4.029 Xtb=1.5 Br=2.949 Nc=2 Isc=0 Ikr=0 Rc=.1 Cjc=276p Vjc=.75 Mjc=.3333 Fc=.5 Cje=569.1p Vje=.75 Mje=.3333 Tr=971.7n Tf=39.11n Itf=20 Vtf=10 Xtf=2 Rb=.1 Vceo=60 Icrating=15 mfg=Texas )
It's also rather simplified though, I have other models with more parameters defined. And in this texas model, I tweaked the VAF to 150, because actual measurements on parts made by John Ellis showed it was a more realistic value. More representative of real devices.
So knowing that, I tweaked most of those models to get about that value.
Plus the BF value also looks suspicious most of the time. Sometimes the CJC looks totally out of whack too! I've seen them with CJC=0 !!! what??@?@
Ok, I put it all in a folder and zipped. Hopefully it has all its bits and pieces and should run as is out of the box.
Made the sim a bit slower, with higher resolution to get a better picture of thd.
We're at 20khz, but still, just about 0.7% thd, and it's on 8ohms resistive only.
3V input sends it way into clipping. It does clip already at 2.6V, and 2.55V is just before clipping.
The thing is, I tried with lower input levels, and we're still at some 0.363% or so with only 2V input.
Not what we would expect from a high end amp.
And bryston don't use degen either...
But there are two main things that were put forth for low tim back then, not only there was this thing about degen, but there was also the idea of lowering open loop gain by having more localized feedback on inner stages, making each stage more linear with less gain, so the global loop has less loop gain for less distortion reduction possibilities. But having less loop gain to lower distortion was compensated, supposedly, by having each stage more linear and thus making less distortion to begin with.
Otala was among the first, if not the first, to promote this concept.
And one big thing was the fact that with more global feedback, the loop wasn't fast enough to come in and correct on fast transients, which overloaded the input stage in the process. I'm not sure this amount of global feedback could make that much difference on the loop's speed to allow correcting on fast transients.
Things have changed quite a bit since those pioneers formulated those ideas.
I think most good amps made today don't attempt that much to reduce each stage's gain and thus reduce the open loop gain. There are compromises, but keeping a fair amount of open loop gain to have sufficient feedback to correct distortion seems to be the way to go I think.
Not sure what's right there. I just looked again at onsemi's datasheet, and there is no capacitance stated, so it's hard to know what it's supposed to be.
Their modpex models have CJC = 1.147E-11 which seems rather tiny to me.
And then cordell's models have CJC=120e-12 and CJC=140e-12, which are even tinier and seem even less realistic. But not having the info on the datasheets keep us in the dark about that.
Of course, if we take the 2SA1381 from cordell, with CJC=8e-12, obviously the MJEs are far above that. But what's right?
If we knew for sure, we could tweak our models and get something more reliable and representative. I wish more people would share about this, so we could get closer to having real good models.
The fact it's only windoze takes me out right off. I'm not going to get a pc or run windoze on the mac just for those things. I must have native software, no emulation, so although it would do a lot of what I'm looking for, it doesn't do enough for my needs.
And there are other missing things it doesn't do and I would want in my own tester: power devices really aren't fully testable, with overly low voltage and currents, and nothing to test for second breakdown.
My goal is a really fully comprehensive dedicated "machine" for more thorough device testing and easier matching. And top important, NO WINDOZE only stuff!
Not me! And it doesn't go far enough.
It does have a lot, but missing more features.
And I'm not going to use windoze, no way! 😀
Switching to the MJL models from cordell only brings down the thd some, but not that much.
I only swapped the models and didn't try adjusting anything else, keeping the sim exactly the same.
That sim needs fixing to get the bias current values again.
I only swapped the models and didn't try adjusting anything else, keeping the sim exactly the same.
That sim needs fixing to get the bias current values again.
In the hope of improving models, I put some data in a spreadsheet, starting with the few 3055 models on hand, so they can be compared and eventually tweaked.
I would really like to have 2 models for the 3055/2955, one representing the original devices from the 60s-70s (and maybe 80s), and the other for the modern newly made devices. When did they starting making those with the newer methods? 80s??
Anyway, once the models are in a sheet, next to each other and their parameters sorted so they can be compared quickly at a glance, we can see how much disparities we're dealing with. Not to mention how different they are in their completeness.
Some values are somewhat in about the same range, but many, important ones, really aren't.
And that's distressing. How could this represent the same part?
Many parameters look just wrong.
Many parameters are missing and most have default values, which are probably not in agreement with the datasheets.
I have been tweaking models for the more representative VAF, at 150 or so, which seems sensible and more in agreement with measurements on actual parts (150-160 for VAF on 3055s).
It's not always so easy to make certain measurements, and it takes a minimum of equipment to do so.
But if we go by the datasheets and perhaps a mix of measurements and data from the datasheets, this can help come to a consensus and get more functional and realistic models.
In the case of the 3055, there aren't recognized sources for known good models, so that needs to be done.
Among the 3055 models in that spreadsheet, there is a large disparity for BF, and they pretty much all look wrong and overstated. The datasheets give the 3055 the hfe of 20-70, so all of the values in the models look rather optimistic.
I would really like to have 2 models for the 3055/2955, one representing the original devices from the 60s-70s (and maybe 80s), and the other for the modern newly made devices. When did they starting making those with the newer methods? 80s??
Anyway, once the models are in a sheet, next to each other and their parameters sorted so they can be compared quickly at a glance, we can see how much disparities we're dealing with. Not to mention how different they are in their completeness.
Some values are somewhat in about the same range, but many, important ones, really aren't.
And that's distressing. How could this represent the same part?
Many parameters look just wrong.
Many parameters are missing and most have default values, which are probably not in agreement with the datasheets.
I have been tweaking models for the more representative VAF, at 150 or so, which seems sensible and more in agreement with measurements on actual parts (150-160 for VAF on 3055s).
It's not always so easy to make certain measurements, and it takes a minimum of equipment to do so.
But if we go by the datasheets and perhaps a mix of measurements and data from the datasheets, this can help come to a consensus and get more functional and realistic models.
In the case of the 3055, there aren't recognized sources for known good models, so that needs to be done.
Among the 3055 models in that spreadsheet, there is a large disparity for BF, and they pretty much all look wrong and overstated. The datasheets give the 3055 the hfe of 20-70, so all of the values in the models look rather optimistic.
When you're doing something that requires models you don't already have and aren't in some libraries of known working models, you don't have much choice.
You can hunt down models and try them out. Unfortunately most of the time they're not going to be up to par.
However, although it's a lot of work, it's still feasible to take the best models you find and tweak them. At some point, if sharing them and they get tested by enough people, we may arrive at something workable.
That's what I'm trying to do right now with this 3055 issue. We do have models, but most seem odd or just plain bad.
However, although it's hard to trust those modpex ones, they seem to be the most complete, often having the largest number of defined parameters.
So they may have bad parameters, and for some it seems rather obvious.
And with a little effort and collaboration, we should be able to fix them.
I'm just doing this right now for the 3055, because I've been doing a lot of sims with those, and the 2955 obviously is going to have to be fixed along with it.
More models not commonly encountered could be worked on afterwards.
I'm documenting some more that spreadsheet that I made for the 3055. Adding comments on the names of the parameters, so they can be readily identified and we know what they are without having to dig for info.
One thing I just found out while doing that is that some parameters have been included but they are at the default value, so rather useless to mention.
I'll post the more complete sheet when it's done.
Are you aware that the 2N3055 was the first high power SI transistor? And I'm sure you've
read that there are many different versions. Why go to so much trouble with such out
dated technology? Many of them were made for Power supply regulators - DC.
I was hoping your answer when I asked which ones you had was 1990s Motorola, because
I have their Power data books and it is a decent part.
I have no idea what you have there.
Here's an idea, those very slow RCA parts were used in many old amps, Dyna ST-80/120,
several british amps, NAD amps, guitar amps that often oscillate and fail when newer
parts are used. You could sell your older parts as replacements, probably at a premium
price. I'd find a better home for them.
Then if any of your parts have an Ft over 2MHz use those.
What brand are your 2N3773s, don't recall if you mentioned it?
I know that I, and others have asked this, but the first time I asked I assumed that you
had at least 2 MHz parts and knew better than to try to build an amp out of parts intended
for a DC power supply. I've never seen data sheets for several of the manufacturers
parts that you mention.
I have about 24 Bendix TR-24 transistors that look to me like a
house number 2N3055, they were used in amplifiers so I know they are not too slow. I
was going to build a quasi Tigasaurus with them around 1970 built a prototype of just the
output stage that worked but then decided to just build the real thing. It was a Quasi with
series connected parts. I measured their Vce at around 100V.
I've been meaning to mention that if you determine that a bunch are higher speed parts
you can measure their actual Vce, you'll probably find it to be 90 or higher and then
70V rails are possible.
read that there are many different versions. Why go to so much trouble with such out
dated technology? Many of them were made for Power supply regulators - DC.
I was hoping your answer when I asked which ones you had was 1990s Motorola, because
I have their Power data books and it is a decent part.
I have no idea what you have there.
Here's an idea, those very slow RCA parts were used in many old amps, Dyna ST-80/120,
several british amps, NAD amps, guitar amps that often oscillate and fail when newer
parts are used. You could sell your older parts as replacements, probably at a premium
price. I'd find a better home for them.
Then if any of your parts have an Ft over 2MHz use those.
What brand are your 2N3773s, don't recall if you mentioned it?
I know that I, and others have asked this, but the first time I asked I assumed that you
had at least 2 MHz parts and knew better than to try to build an amp out of parts intended
for a DC power supply. I've never seen data sheets for several of the manufacturers
parts that you mention.
I have about 24 Bendix TR-24 transistors that look to me like a
house number 2N3055, they were used in amplifiers so I know they are not too slow. I
was going to build a quasi Tigasaurus with them around 1970 built a prototype of just the
output stage that worked but then decided to just build the real thing. It was a Quasi with
series connected parts. I measured their Vce at around 100V.
I've been meaning to mention that if you determine that a bunch are higher speed parts
you can measure their actual Vce, you'll probably find it to be 90 or higher and then
70V rails are possible.
Last edited:
I'm going to go on the assumption that if you build with those old parts, that you'll use at
least the 2 MHz + Ft versions only. I'm fairly certain that if a part is labelled MJ2955 it is
intended to meet the Motorola now Onsemi spec, can't say the same for all 2N3055s:
http://www.onsemi.com/pub/Collateral/2N3055-D.PDF
least the 2 MHz + Ft versions only. I'm fairly certain that if a part is labelled MJ2955 it is
intended to meet the Motorola now Onsemi spec, can't say the same for all 2N3055s:
http://www.onsemi.com/pub/Collateral/2N3055-D.PDF
Last edited:
Not too surprising.
Because there are people who today can find them more easily in some parts of the world than they can find more modern parts, and in those regions, they are also cheap. Plus there are people like me who have old stock and want to do something with them.
Plus it's also an interesting challenge to get those old things to do things that weren't done back then when they were popular, and push them to new limits.
I've had quite a few requests from others who wanted me to share the work results on such 3055 based amps. I'm still working on several such projects, and fact is, there are several threads where those things draw interest.
Especially the 800khz versions, not so good for audio.
For some of it, I'm not so sure myself.
I tried again taking photos, but they always come out blurry and we can't make out what's on the cases, so I had an idea, I used a scanner, and that worked much better. The cases not in contact with the glass are blurred, but who cares, since the top of the case are clear enough to read the stuff...
Posting a few shots of a part of the stash. Some are of unknown origin, and some have date codes I don't understand.
Maybe, but I would have to know who needs them, when they need them.
But what would be the fun in that? 🙂
Depending on when they started making the newer ones, I may have some relatively young enough to be so.
I haven't yet found my batch of MJ2955s, but I suspect they may be from ST. That is after they merged, so no longer SGS. I do have some SGS, some thomson, some Sescosem...
I didn't mention because I didn't know. I couldn't remember, plus I forgot where they were stored. I have parts stored in 3 places (geographically) and many aren't readily accessible right there, but I searched today at one of those places and found my batch, and oh surprise! I don't even know who made them. Look at the posted photo. Can't tell.
But in any case, all of them are from an era before it became common to counterfeit them. So even though the 3773s don't seem to have a manufacturer's seal on them, I doubt greatly that they would be counterfeits. But then again, you never know for sure...
I know they improved their manufacturing process and although they pretty much kept the specs the same in datasheets, the parts perform better than stated, and the second breakdown is usually much higher than the original 60V.
That is why many have built such amps with higher rails than they should and had no troubles. And among our 3055 projects, we've systematically been aiming for 70V rails (+-35V).
I must still have an old amp with 3055 lying around, from the mid seventies, a quasi, which should still be functional. It never gave any troubles and got a lot of use back then, even with a very flimsy heatsinking...
In that small sample from my stash in the photos, there are 2N3055E and 2N3055S, and I never knew what the E and S were all about.
There are a few 2N3055H, and with any chance, those could be the real H with the actually announced 100V Vce0 in their datasheet.
The ones from the mid 80s could be potentially the newer types, maybe.