LOL, probably true, but I learned on this amp and am somewhat determined to make it work!
Actually, just pure curiousity - I want to know why they blow up!
bias the output stage driver Q with some standing current?
if I'm looking at the right sim you have way too low bias in the cfp input/driver Q they should be higher bias for speed, low impedance to control the output Q ~ 50-100mA isn't too much, reduce both R4,13 and increase spreader V until at least Q5,6 have quiescent current
I also moved the compensation C back to the VAS out, giving "input stage inclusive" if for some reason the normal Miller Cdom is being avoided
if I'm looking at the right sim you have way too low bias in the cfp input/driver Q they should be higher bias for speed, low impedance to control the output Q ~ 50-100mA isn't too much, reduce both R4,13 and increase spreader V until at least Q5,6 have quiescent current
I also moved the compensation C back to the VAS out, giving "input stage inclusive" if for some reason the normal Miller Cdom is being avoided
I've been trying to simulate the original design so far and I did suggest somewhere lowering those resistors to reduce the cross conduction issue. The original design had about 8mA bias in the drivers. I do agree that 50 to 100 mA would be best, however .1A with 35V across the drivers is 3.5 W. The original drivers are rated for 10W at 25 deg C ambient but I believe that 3.5W is too much once derating is taken into consideration. I'd probably substitute MJE15032/33 with a decent heat sink and then go for .1A in the drivers.
I considered Miller inclusive but again, I'm looking for solutions to the other issues first.
I considered Miller inclusive but again, I'm looking for solutions to the other issues first.
the trouble
most of the trouble comes when the amp is driven into clipping. What if we could forestall most of the ill effects? Enclosed schematic changed the compensation (not a big deal), but adds two diodes (1N4148 PIV maybe a bit low). If either output transistor saturates, it steals away base drive from the output stage. This stops the stage from hard saturation, and makes the result much more civilized. Note that for the 3 V input case, which causes overdrive, we don't get crazy big reverse biases, and also avoid simultaneous conduction.
most of the trouble comes when the amp is driven into clipping. What if we could forestall most of the ill effects? Enclosed schematic changed the compensation (not a big deal), but adds two diodes (1N4148 PIV maybe a bit low). If either output transistor saturates, it steals away base drive from the output stage. This stops the stage from hard saturation, and makes the result much more civilized. Note that for the 3 V input case, which causes overdrive, we don't get crazy big reverse biases, and also avoid simultaneous conduction.
I see no reason to think there is any reason to respect the orginal design parts selection, bias points - the topology isn't the best but could be made to work adequately
it is really hard to fix a feedback amplifier when the output has substantial deadzone from inadequate bias - I would just write off the orginal as fundamentally flawed in this respect
a few W even in TO-126 is no problem with a PCB mount heatsink
it is really hard to fix a feedback amplifier when the output has substantial deadzone from inadequate bias - I would just write off the orginal as fundamentally flawed in this respect
a few W even in TO-126 is no problem with a PCB mount heatsink
The only reason to "respect" the original design was to figure out why it blew up, but again, I don't expect to get an answer with flawed SPICE models so given time I will compare to real hardware at some point. If/when I am convinced that this can be made to work, then I'll consider changes in more detail, but we generally are thinking along the same lines. The originals drivers were an odd ball TO-5 with integral heat sink. Yes, I do think that a TO-126 or 220 will do it.
Nice work! I experimented, A LOT, with diodes around the output and drivers, outputs to drivers - which did not work, but never went as far back as to the VAS. Your solution does seem to work and I would use low capacitance diodes in order to keep the loading on the VAS low. If people take a look at this, plot the new diode currents, they are high in the milliamp range in hard clipping, and micro amp leakage before clipping.
Simple, I like it!
I've got to look at it a bit more since it seems to me that the diodes should go forward biased due to the gain in the output stage which would make it not work. I have to look at it a bit more to make sense out of it.
Edit: OK of course you are going to the input of the VAS and current source which are always close to the rail so they only clamp when the output is close to the rail.
Last edited:
Hi Pete,
I came up with an idea which I think is a reasonable solution to the problem of the reverse bias of the Vbe of the driver transistors.
The idea is to create a virtual ground independent of power supply ground that is not bidirectional to supply current to R12 and R2.
This circuit satisfies the requirements with out loss of gain in the OPS and without clamping. It's not an efficient way of doing this and it adds a bit of complexity but it does rectify the problem with the reverse bias of the drivers Vbe. The circuit values are not tweaked but it does demonstrate the principle.
Four zeners were used for no other reason than a lack of availability in LT's library.
Cheers,
David.
I came up with an idea which I think is a reasonable solution to the problem of the reverse bias of the Vbe of the driver transistors.
The idea is to create a virtual ground independent of power supply ground that is not bidirectional to supply current to R12 and R2.
This circuit satisfies the requirements with out loss of gain in the OPS and without clamping. It's not an efficient way of doing this and it adds a bit of complexity but it does rectify the problem with the reverse bias of the drivers Vbe. The circuit values are not tweaked but it does demonstrate the principle.
Four zeners were used for no other reason than a lack of availability in LT's library.
Cheers,
David.
Attachments
Even simpler?
David,
I ran a sim to confirm...your mod makes the output stage (say from the base of Q5 to the output) have a gain of 1, rather than a gain of 2. In that way, the driver stage can't really overdrive the output stage, and the gross clipping issue goes away.
Interestingly, if you remove R7, D4-D7, and R29, just leaving R12 and R2 connected (but not grounded), you get pretty much the same result.
So, it seems another way to stop the overdrive problems is to drop the output stage gain from 2 to 1. It puts a bit more feedback in the output stage, which could lead to better control of quiescent current, and maybe less distortion in the output stage. I might worry a bit output stability in the output stage, but it's probably pretty manageable.
David,
I ran a sim to confirm...your mod makes the output stage (say from the base of Q5 to the output) have a gain of 1, rather than a gain of 2. In that way, the driver stage can't really overdrive the output stage, and the gross clipping issue goes away.
Interestingly, if you remove R7, D4-D7, and R29, just leaving R12 and R2 connected (but not grounded), you get pretty much the same result.
So, it seems another way to stop the overdrive problems is to drop the output stage gain from 2 to 1. It puts a bit more feedback in the output stage, which could lead to better control of quiescent current, and maybe less distortion in the output stage. I might worry a bit output stability in the output stage, but it's probably pretty manageable.
Following that thought...
I haven't simmed it to confirm, but perhaps just raising the values of R12 and R2, and leaving the ground, might make a nice way to drop the gain so overdrive isn't so nasty, but still leave a bit of the benefit of a little extra gain in the output stage.
OK...I just simmed it...R12=R2=500 is quite civilized with a 3 volt peak input.
It looks like that's yet another way to mitigate the overdrive issues...
It may take some more simulations and though to figure out which of these approaches is optimal...
I haven't simmed it to confirm, but perhaps just raising the values of R12 and R2, and leaving the ground, might make a nice way to drop the gain so overdrive isn't so nasty, but still leave a bit of the benefit of a little extra gain in the output stage.
OK...I just simmed it...R12=R2=500 is quite civilized with a 3 volt peak input.
It looks like that's yet another way to mitigate the overdrive issues...
It may take some more simulations and though to figure out which of these approaches is optimal...
For whatever it's worth, I just subbed MJ21193/4s in a 207A and it runs just fine; yet faster transistors might be problematic, haven't tried those. I believe I'll need substitutes for the old RCA 40409/10 drivers and VAS stage devices. I'm going to replace all resistors with quality film caps and update the boards to the B version, and give them a new listen. They just didn't have the air and power that my Leach amp does, as suggested by the higher order distortion products as power increased from a few watts.
Increasing the current through R7, R29, D4 - D7 by decreasing the value of R7 and R29 to 100 ohms and decreasing the value of R2 and R12 to 10 ohm, restores the gain of the OPS to ~2 and the the drivers reverse Vbe is still well within safe limits. So it's not just a matter of lowering the OPS gain.
It's not necessary to remove R7, R29, D4 - D7 to do a comparison, just ground the node between D5 and D7.
David.
Taking the hint...
David,
I took your hint, and perhaps simplified it a bit...Both your mod and this simplification thereof remove the excessive reverse bias at (modest to severe) clipping as an issue even with the output stage gain restored to near two. However, it does once again allow some cross-conduction when over-driven, which could be damaging.
Perhaps owing to the cross-conduction when over-driven, I'd still vote for either some kind of clipping, or reduced gain in the output stage...
Comments invited...
Dan
David,
I took your hint, and perhaps simplified it a bit...Both your mod and this simplification thereof remove the excessive reverse bias at (modest to severe) clipping as an issue even with the output stage gain restored to near two. However, it does once again allow some cross-conduction when over-driven, which could be damaging.
Perhaps owing to the cross-conduction when over-driven, I'd still vote for either some kind of clipping, or reduced gain in the output stage...
Comments invited...
Dan
Nice work gentlemen, just wondering any reason to be moving on from the diode fix, or just exploring other alternatives? I expect that a 1N914B should work fine there, rated for 75 at 5uA reverse leakage and 100V at 100 uA leakage, capacitance is less than 5 pF so I doubt a lower capacitance diode is even needed.
Possible fixes
the possible fixes are:
1. diode clamps
2. lower gain in the output stage
It's hard to say which one is better without a number of sims, and probably a number of experiments.
On the diode clamp side, I'd be a bit lconcernd about the 1N914, just because with plus and minus 42 volt supplies, there will be nearly 84 volts across the diode when you push the amp to clipping...at that voltage, their leakage is getting a bit of concern, maybe especially at temperature...maybe 2 in series would alleviate my concerns to some degree, or perhaps just a low capacitance diode with higher PIV rating.
the possible fixes are:
1. diode clamps
2. lower gain in the output stage
It's hard to say which one is better without a number of sims, and probably a number of experiments.
On the diode clamp side, I'd be a bit lconcernd about the 1N914, just because with plus and minus 42 volt supplies, there will be nearly 84 volts across the diode when you push the amp to clipping...at that voltage, their leakage is getting a bit of concern, maybe especially at temperature...maybe 2 in series would alleviate my concerns to some degree, or perhaps just a low capacitance diode with higher PIV rating.
I consider the gain in the output stage to be a key feature, not sure I'd consider anything less than 1.5 worth doing. What is interesting is that it allows for easily regulating the front end, using cascodes, etc. without needed higher supply voltages.
At clipping it only comes to about 40, not 42 and we are splitting hairs here, 80 is not much more than 75 which is a worst case number so do you think the leakage would be much more than 5 uA - I seriously doubt it especially with statistical considerations. If you are really worried the BAV21 is probably a good choice, just about 2 cents a piece:
Product Folder - BAV20 - High Voltage General Purpose Diode, Fairchild Semiconductor - Global Leader in Power Optimization
Last edited:
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.