Amplifier based on 2N3055

The test load was two pairs of dual Delta10A cabs with Selenium HF. They're just top boxes (relatively high tuning), but I was running them full range. On this low a power it's pretty safe.

The other amp is not 3055-based, it is TIP35/36. It's actually 2 pairs - one active and one for the commutators. The 3055's don't have enough Hfe to do that.

"Double Barreled" 3055 is how I would have done it from the beginning, instead of grounded bridge. Voltage stresses and dissipation are EXACTLY the same, as long as you're not in slew limit. I have a pair of double-barreled (cascode) amps that will do about 1200 watts in 2 ohms -and melt the plugs off the extension cord doing it. Those were the last big class AB amps I did.
 
Last edited:
The other amp is not 3055-based, it is TIP35/36. It's actually 2 pairs - one active and one for the commutators. The 3055's don't have enough Hfe to do that.

Interesting project too.

"Double Barreled" 3055 is how I would have done it from the beginning, instead of grounded bridge.

Well, the most interesting part of it is the grounded bridge topo. This will be very cool when we get one running, and then later, perhaps with a 4 pairs version, when we bridge a bridge. I simulated john ellis' bridge in the bridged bridge situation and that works fine. We'll have cool amps done soon.

I think revisiting the leach double barreled on 3055 will be a good future project as well. This one can get quite a bang and also can be bridged.

One 3055 project that I really want to build as well is the one I posted about much earlier, with a quad comp output stage based on bryston's topo. My simulations on that one show superior performance for 3055s.

But one thing at a time, let's get a grounded bridge working first.

One thing that I'm very curious about, is to find out how well that performs when actually measured, compared to the simulations.
 
Here is a state of where we are on that fig 41 limiter.

I switched the protection transistors to the BC547/57C for higher gain, which improved the action of the limiter, reduced their base resistors from 10k to 4k7 and that too improved it a little more, then switched also the vas protection transistors to the same higher gain and increased the 12ohms resistor for the sensing to 27ohms, so that protection is at the threshold of action and that limiter clamps the vas max current to about 22mA, which also improves the action significantly, but we're still not close enough to a full fold back action from the limiter and I am afraid this doesn't keep the dissipation enough within the soa as set by the locus.

The values for the fig 41 are now as follows:

R1 = 12k
R2 = 430
R3 = 147
R4 = 14k
R5 = 4k7
R6 = 1k5
3 zeners: 12V + 12V + 10V = 34V to set the Vref at 1V

The plotted locus does slightly encroach on a part of the critical second breakdown area, but that is on the 50C de-rated DC SOA curve, so it's not too bad. This is the theory/calculations and the reality may be a little different with the real parts. Plus the screened and tested 3055s for higher Vce0 should help this a bit too.

I ran this sim with the output shorted with 1mohms, so that's a frank short! at a drive level on the input of 1.27Vpk, which gives nearly 100W on 4ohms when not shorted, so it's driven almost to clipping.

Attached is a plot of the dissipation only for T21 and I plotted its soa curve/load line, which looks very funky in a big loop, and I drew up a part of the de-rated @ 50 DC soa curve on top of it, which does show how much above the soa it still is. The average power in T21 shows ~115W already, and we can see peaks at about 410W.

The limiter is fully activated, as it should be, according to the values in place.

Please someone verify my values, just in case. But with almost full drive and output shorted, the limiter is definitely active but not effective enough to clamp the dissipation within the soa as it's supposed to.

I didn't have the flyback diodes before for all the other sims, but this time since I'm testing the limiter, I added them to see if they help, and I guess they would probably help on a reactive load and not a short.

I played a bit with the time constant R5/C1 (R68/C13). When I had R5 (R68) at 10k and C1 (C13) at 22n, for a 200us constant, which is below the max of 500us as prescribed by Cordell, the limiter's action showed low levels of oscillations.

The dissipation increases with frequency and a longer time constant dampens oscillations of the limiter, however it also increases the dissipation a lot, so I tried finding a compromise that works across the whole range of 20-20k, and I ended up using a 22n for C1 (C13) (~103us constant), which keeps the oscillations to a bearable minimum while acting towards limiting the dissipation. However as this dissipation info shows, it's too much above soa, and worse at 20k with ~138W average power dissipated in T21 alone.

I thought less limiter action was needed at higher frequencies and expected lower dissipation, but it's the opposite.

We can see peaks of dissipation and oscillations over pretty much the whole range. The shorter time constant helps bring down average dissipation, at the expense of higher peaks and more oscillations.

My question for now is: Do we look only at the averaged out power dissipation or should we consider the peaks?

Also how much oscillation should we tolerate. More oscillations cause more issues and more high frequency dissipation.
 

Attachments

  • test output shorted 10khz.png
    test output shorted 10khz.png
    775.9 KB · Views: 574
  • dissip T21 output shorted prot active 10khz.png
    dissip T21 output shorted prot active 10khz.png
    352 KB · Views: 532
  • SOA-T21-output-shorted-prot-active.png
    SOA-T21-output-shorted-prot-active.png
    202 KB · Views: 468
The way you're going, you might as well have relays shorting out T13 and T14 b/e along with what you've got now. That'll do it. Now I think about it, it'll bang on and off as quick as it's capable of - perhaps that's what's sort of happening with your present circuit with its oscillations, so I suppose all that can be striven for is a low duty cycle repetition rate to keep dissipation down.
 
The way you're going, you might as well have relays shorting out T13 and T14 b/e along with what you've got now. That'll do it. Now I think about it, it'll bang on and off as quick as it's capable of - perhaps that's what's sort of happening with your present circuit with its oscillations, so I suppose all that can be striven for is a low duty cycle repetition rate to keep dissipation down.

Oscillations are "normal" and unavoidable on such limiters, it's their nature of operation. We use a time constant there to dampen those oscillations but we can't eliminate them entirely, or the circuit would no longer work at all.

It's a matter of compromise between what amount of oscillations we keep possible and the dissipation.

On 2EF or 3EF stages, this limiter works a lot more efficiently due to the larger gap to squeeze when acting, but here we only have one junction's worth of gap and the protection transistors have a harder time closing that gap well enough.

When I saw that a higher gain type was more able to squeeze this gap, I thought about using darlingtons there, but then the draw back is that they need twice the Vbe to act, so we'd be back to square one and require a higher value for Re.

If we can't make this limiter act hard enough to keep the dissipation within the soa, then other protection means will be required regardless.

The main issue is with this topology with one junction only. That was this same issue that was giving me a problem making a better bias spreader.

The original elektor schematic with the 2 transistors used as diodes as a bias spreader, with a very low value only resistor in between, was really not so practical.

Using TO92s to couple them thermally wasn't too hard to do, but the bias trimmer resistor needed to be only a few ohms and it was difficult to bring the bias down low enough.

Actually when I used cordell's bd139/140 models, their Vbe being around 400mV only, the 2 bc639/40 were not able to keep the bias current down enough even with ony 1 ohms between them.

I have been thinking about the latch found on crimson's amps, but since we may have some action sometimes on legit signals if speaker impedance drops too low with an overly high reactive load, then we'd have an amp locking up all the time when driven a bit too hard.

I think with 3 pairs of 3055s that making a bridged bridge would require keeping the drive well below max power. I plotted a 4 pairs soa and that looks much better for driving 2ohms to the max. That will be an other incarnation of this amp, but for now let's get a workable 3 pairs model.

The main issue with the detection of overload is with the output shorted, because when the limiter isn't acting on legit signals at max power, the dissipation stays reasonable. It's the limiter's action that causes more dissipation.
 
For the purpose of testing the vi limiter's action on various complex loads and checking how stable the amp is on those types of loads, I picked up a few models of configurations.

The most complex, I got from Rod Elliot's site, for 2 ways with a passive filter. 2 variations of that. Plus an old thread on the forum had what is to be the worst case possibilities for inductive and capacitive, both with 60 degrees phase shift, which I haven't verified, although I changed the 4ohms series resistors for 2ohms, because I want to simulate as close as possible to 4ohms real world load.

And finally I don't remember where that came from, but I also used a single speaker model that is supposed to simulate the real complex load that it presents, with all the caps, resistors and coils to emulate it.

I'm posting what I used for reference.

ESP's 2 examples of 2 ways complex load seem to me to be for 8ohms nominal, and despite their apparent complexity, they don't pose a problem to the amp at all (elektor's quasi), and the amp doesn't break a sweat. Nothing bad going on with phase shift, thd stays very close to the resistive load and there is nowhere near enough phase shift to cause any significant extra dissipation in the outputs, their soa usage stay very well within their limits and the limiter doesn't activate.

Then I tried both supposedly worst case loads with 60 degrees capacitive and inductive, and those don't pose any problem either, the dissipation in the outputs is no threat at all.

I can see there is a phase shift, which for the capacitive load is hardly noticeable at the lowest frequencies and it increases with frequency, while the amp's output current and power transistor's dissipation is also very low at the lowest end and much higher at the high end (20k) although less than 15W worth of dissipation in T21...

The phase margin on 4 ohms resistive is at almost 70 degrees and the gain margin a little too low at a bit over 1.54db.

The capacitive load decreases the phase margin down to a bit over 61.8 degrees and it increases the gain margin to a little more than 2db. There are no oscillations at all on that load and the thd goes down as well, actually cut almost in half at 20khz.

Things don't go quite that way on the inductive load, as the phase margin increases a bunch to nearly 79 degrees, while the gain margin drops to negative at some -1.33db, and there are some very high frequency oscillations superimposed on the sine signal, and curiously, the thd also drops some 200ppm lower than on the 4ohms resistive load, as with the capacitive load.

I am thinking perhaps those oscillations happen because there is no more gain margin, being negative and that's enough to allow them. So I will have to tweak some more the compensation scheme for more gain margin, hopefully to stabilize this better.

I haven't yet been able to do simulations on lower impedance loads, as those are as they are and I don't know what I should tweak to change realistically their impedance, perhaps down to some 3 or 2 ohms. This will be needed to test the limiter.

All the tests done so far show me that on a full short, the limiter doesn't suffice and it should be supplemented by something else more efficient and perhaps as radical as causing something to shut down somewhat quickly, to prevent damage, as the dissipation in the ouputs while the limiter is active on a short is much too high and quite above the soa.

Although I haven't yet tested the limiter on complex low impedance loads, I think it probably will be sufficient and should keep the outputs in their soa. I will have to confirm that.

What I'd like to do now is tweak the complex loads to make them 4ohms nominal instead of 8, and somewhat badly behaved with impedance dropping to say 2 or 3 ohms, which should at least some of the time activate the limiter, which is the while point.

Then perhaps try to have those load also present a 2ohms nominal, with their bad behavior accordingly, which should trigger the limiter pretty much all the time at full power.

When I put in place the new calculated values for the limiter (M.K.'s fig 41), I found although it wasn't really activating, that it was a tiny bit turned on anyway, from having a threshold close above the full power (see soa and locus), and doing so, it was adding a little too much extra thd, which I cured by increasing the time constant from about 103us to 470us. The lowish values calculated for the sensing resistors (R2=430 & R3=147) also caused unwelcomed loading on the pre-drivers and caused significant extra thd (more than 200ppm more), so I went back to the calculations and managed to make that higher (R2=953 & R3=310), with a different breakpoint threshold as well, slightly negative this time at -3V.
Then the thd increase from the addition of the limiter was minimized.

It's funny that the more complex and reactive loads are actually making the thd go down...

I'd like some pointers on tweaking the speaker simulated loads for the lower nominal impedance, keeping them as realistic as possible. And I'm going to attempt getting more gain margin, hopefully to cure the hf oscillations on the inductive loads.
 

Attachments

  • Screen Shot 2014-04-29 at 11.23.51 PM.png
    Screen Shot 2014-04-29 at 11.23.51 PM.png
    11.9 KB · Views: 86
  • Screen Shot 2014-04-29 at 11.23.59 PM.png
    Screen Shot 2014-04-29 at 11.23.59 PM.png
    20.8 KB · Views: 82
  • Screen Shot 2014-04-29 at 11.23.42 PM.png
    Screen Shot 2014-04-29 at 11.23.42 PM.png
    30.2 KB · Views: 399
  • Screen Shot 2014-04-29 at 11.23.35 PM.png
    Screen Shot 2014-04-29 at 11.23.35 PM.png
    33.3 KB · Views: 434
Last edited:
I have tested the fig 41 limiter on both the elektor quasi and the john ellis complementary version of the bridge.

It works roughly the same and won't be much help in case of a short, however on overly difficult reactive loads, it does ok and as long as the impedance doesn't drop too low, that is close to a full short, the limiting action should suffice.

If there are shorts, they may not be near enough zero ohms, and they may not last very long, so this may be fine. If we want a real short protection, then something beyond the limiter would be required.

I've been running a lot of sims with various loads, and the most difficult ones aren't even the highly reactive, it's the short and with a short, there isn't much reactance left, mostly the zobel and output network, so nothing much to cause phase shift.

Now I've tried again to make the elektor based one to work in the bridged configuration, and I think something isn't right. It won't work that way and I think despite all our discussions to arrive at a bridging somewhat like crown's, we're not quite doing it right.

I tried keeping each side separate and test them side by side. With a feedback on the low side like on the high side, both side work just fine, but not when bridged.

I thought perhaps the TMC compensation could be an issue, not being certain where the tap from the output should really be coming from, so after comparing all the compensation methods and seeing how TPC performs quite well in this amp, I used that in the bridge configuration, because it refers neither to ground nor the output, only the rails, which seems ok to me in a bridge with flying rails.

With the TPC compensation, the thd went down to somewhere a bit above 200ppm, which is far better than plain miller when we started this.

What I observe when trying to give the low side the types of signals that I think it should see in the bridge, is an open loop behavior. We can tap the ground for feedback for the low side, so it should take that from the high side, but it's an opposite phase signal, so not a negative feedback. And this may be where there is a need for more reflections.

It's only a matter of getting the bridge to work now, as I think I have the limiter working properly (mostly. not on shorts)

Until we get this bridge working right, I will further the work on john ellis' bridge, as that one works and the limiter is the same on it.
 
The best sound with 2N3055's? Probably that really old 1970's-era Nelson Pass single ended class A that put out a whopping 20 watts. Takes 8 per channel and a big heat sink. It originally specified 2N5878's but 3055's or 3772's can be used. I literally threw one together one afternoon and the sound was incredible. Not terribly loud, but damn good. Better than any two-transistor Quasi class B you will ever find using these old devices.

This particular project aims to get more watts out of 3055's than you're really supposed to. If that's what you mean by 'best', you'll have to wait a while till Spooky gets it going.
 
I'm tempted to rebuild my Marantz 2220 power amp. It's pretty basic, runs from a single 60v supply. Bought the receiver off eBay, someone blew one channel, removed the power switch, etc. I've replaced a lot of components on one channel, including the outputs which were shorted C-E and replaced with 2n3055s. Quasi-comp. I've been reading the soundwest site, and I'm tempted to use MOSFETs instead of 3055s. FETs dont need VI limiter and are inherently more robust. I like that part for sure. The downside is that bipolars provide more power for a given rail voltage, so I'm not crazy about FETs for this reason. Any suggestions?
 
I'm sorry I haven't had time to continue the work on our project for some time. I will get back to this, hopefully soon, depending on my activity.
We got close to a point of making a prototype, so this should come together soon.
I am concerned that I was unable to get the elektor based one to function in simulation once bridged. Each side works fine individually in simulation, even the low side with its simpler structure. I had tried many things to make this work when bridged, but it's unstable.
More work is needed to get the pcb done for the john ellis version, which works great in simulations. Other than that, I had finished pcb design for the elektor based proto in 2 pcbs, so this could be tested already. I am just wary of testing that since it won't work as is in simulations. I don't have a bunch of test equipment, so I am too limited to properly test the proto.
 
Testing a simple proto

I have the single sided dual board proto board designed for the elektor based grounded bridge, without the protection circuits, to keep it simple for testing.

Is there anyone who has the needed stuff to test this?

I can post some outputs from eagle, or perhaps the gerbers, whatever is needed to get this made for testing.

If someone wants to try this out, keep in mind there may be some tweaking needed to make it work when bridged, as it didn't work right in simulations, so a minimum of know how is needed about this.

This version hasn't got most of the improvements we've tried making and still uses the TO-92 bias spreader to couple with the pre-drivers.

I'm posting the schematics and what the boards look like.

I'm mostly concerned about making the bridge work. Once we know the concept works properly, then we can look at the improved versions.
 

Attachments

  • 2N3055 amp.pdf
    151.2 KB · Views: 350
  • 2N3055 amp high side.pdf
    36.3 KB · Views: 227
  • 2N3055 amp low side.pdf
    35.4 KB · Views: 192
Maplin did a 2n3055/mj2955 bipolar amplifier.
Despite the transistor running on a slightly too high B+/B- it worked very well.
It ran on my mobile disco for a good few years.

I recently reworked the circuit for my own needs.
I replaced the 2n3055/mj2955 with MJ15003/4 and it works very well.
It did need 10R base stopper resistors so it didn't oscillate.

An externally hosted image should be here but it was not working when we last tested it.
 
The Crimson PCBs intended as 4 Ohm drivers are designed around a load of 2.7 Ohms plus any value of series reactance. Our latest PCBs sense the overcurrent in the VAS and as the limiters operate a diode pump charges a capacitor. In typically 0.5 to 1 second the 0.4 volt theshold is reached for the thyristor connected pair which reduces all the bias constant current sources to zero. The amp is thus latched into a standby zero current state, with hardly any transient. Removing the power resets the system after a few minutes. We now build in clip indicators that are pulse stetching; any protection circuit operation shows up very clearly and its time to back off the volume before the tweeters fry.

So if I get this right, then it means that as soon as the limiter detects the very first tiny bit of overload, it will remove the drive and latch on in that "position", meaning no more sound, the show stops right there.

If this is the case, then it's not the type of protection that I'd be looking into. I would want the amp to protect itself, but not to stop working at the very first sign of an overload.
 
Would anyone be able to test the proto for the 3055 grounded bridge amp?
The dual single sided design from post #594 above was made for that purpose. It needs to be done by someone who knows enough to make this work, because although the simulations of each side were working individually, it was having troubles once the bridge was made.
Or perhaps someone can take that simulation to figure out what's wrong that prevents proper operation as a bridge.
We're so close to a functional design, and the pcb design for the whole thing has been done. It only needs to be tested, made to work, and we have a workable diy amp to make.
 
Hi spookydd
I haven't had a chance to look at your circuit in detail but I recall that to make a differential amplifier work the signal has to be truly differential - "ground" has to be ungrounded or the distortion cancellation does not work and each amp behaves as a single. If so then a single-to-differential preamp might solve this.
I don't know if this is your problem or if it is another issue - perhaps oscillating or something when the two are joined, which will need better frequency compensation on each side.
I'll take a look when I have a chance but it won't be just yet as I have a big list of things to do.
 
Hi spookydd
I haven't had a chance to look at your circuit in detail but I recall that to make a differential amplifier work the signal has to be truly differential - "ground" has to be ungrounded or the distortion cancellation does not work and each amp behaves as a single. If so then a single-to-differential preamp might solve this.
I don't know if this is your problem or if it is another issue - perhaps oscillating or something when the two are joined, which will need better frequency compensation on each side.
I'll take a look when I have a chance but it won't be just yet as I have a big list of things to do.

Hi john,
The one having the issue in the sims when bridged is the elektor based one that we have all contributed on modifying for use as a grounded bridge.

Your design worked beautifully and is worthy of getting a pcb design done.
I haven't been able to keep working on the hobby for some time, and I'm not quite ready yet to get back into it as I should. Hopefully soon. So I was stalled on the elektor based version, for which I had done a single board pcb design, knowing it wasn't ready for prime time yet, but hoping we could solve the issue of the bridging not working as expected, without too many changes.
I had done a separate pcb design with 2 boards, without the protections, to make it a simple prototype to test it and find whatever possible faults and make it work.
Once this is done, then all there is left to do is finalize the protection calculations and it's ready for pcb finalization.
There is interest out there for those amps. I received a few request recently about sharing the design, but it's not to be considered fully operational yet.
It would be nice to finish it and I think many diyers will want to build it, including myself.
If we can finish that one, then I'll go on working on the pcb for your design, which I would also like to build.
I think they are quite decent amps, considering the old trusty 3055 is used.

The main difference between the elektor based one and your design was that the elektor based was a quasi, and your design is complementary, and quite nicely symetric too.

I think there are many diyers awaiting a ready-to-build design.

Let's get those old 3055 out of their dungeons...

For your complementary design, I think it's worth buying the sets from onsemi, which are far better matched now than decades ago. The 2N3055/MJ2955 pair.
I simulated it with the MJ15015/16 pair as well and it also worked quite nicely, so for those who feel they want to squeeze even more power out of that puppy, they can bump up the rails some more, perhaps add one more pair, and of course recalculate the protections.