I understand what you mean, however we are quite far for a little asymmetric behavior. What I've been getting is also far more than a simple overshoot or ringing. Plus the most important part, is that it must "protect", which means it must when activating, "limit" the dissipation, and not cause more. The big issue is that it acts far more than simply asymmetrically, it's hugely different on the negative side compared to the positive, and it causes big oscillations and big increases in dissipation instead of limiting it.
Like I said, it really misbehaves and causes more problems than it solves.
I was able to make it work somewhat properly on the positive side, with a effect somewhat similar to clipping, which would be ok, but it has a very big impact when acting on the negative side, with really big oscillations and it seems also that when the negative side is enabled to act, it causes a different way to act on the positive side.
I have been trying to isolate the different causes and elements involved, and I was able to make it work somewhat better, on the positive side, by acting on the vas input instead of its output. However there is no way to make this work right on the negative side. So that is why I am thinking the non symmetrical topo of the vas may be the reason, as it doesn't take kindly to be shorted that way.
The limiter should behave better on the elektor based topo, but I haven't yet been able to make the bridged simulation to work right, so I think there must be something wrong with how we "cross-linked" it. Perhaps by taking the divider network from crown as is without tweaking it to the exact need, we made it impossible to work properly as a bridge. I was able to run the low side properly, although it's simplified and by itself it doesn't have a feedback, I verified it was working right anyway. But the grounded bridged version would not work at all in simulations, so this must mean something isn't quite right in the bridging. I want to get a quasi working, so this elektor based version could be tweaked to work properly. In the mean time, since John Ellis's bridge is working beautifully as a grounded bridge, and even as a bridged bridge, although it's not a quasi, we can make that one work, as long as we can make the protection to work right. I started putting the John Ellis bridge into eagle as a schematic, for now, but it's missing proper protections.
We will have a complementary bridge with the John Ellis bridge, and then the quasi with the elektor one.
I simulated a 4 pairs version of the John Ellis bridge, and that works even better and very nicely on a 2ohms load, so that version with 4 pairs will be a nice variant to make for those who want to use it as a bridged bridge. When we get proper protection working right.
There is no avoiding the 2ohms load operation when the bridge is bridged again, with the 8ohms load on the whole thing, so if we don't want to have the protections bothering that mode of operation, we must use a far higher threshold of activation. But with a limiter that doesn't really limit the dissipation, a better protection method must be used.
I think I know why the vi limiter doesn't behave properly, in part, when used on the bridge with the rails flying. The vi limiter can't be referencing to ground, since it's swinging endlessly. And I think that's the problem, the resistors in series with the diodes going to ground.
Plus the action of the limiter makes the vas act strangely as well, and that causes oscillations. The insertion of resistors between the vas and the limiter should put a damper on that. And I've been testing the addition of a cap & resistor in series placed after the sensor resistors, which not only adds a little delay in the limiter's action, which will let the transients pass some more, it also prevents oscillations.
I added a protection transistor on the base of the vas (john ellis' bridge), so it won't behave as wildly when overloaded by the limiter, so this also makes it behave better.
I will change the limiter not to use the diode/resistors going to ground, in favor of the dual slope with zeners, as our esteemed friend kiwanuka describes on fig 28 in his excellent paper.
The limiter transistor starts acting way before the set threashold, but only partly, and the loop gain seems to be sufficient to correct for the little leakage induced by the overly early action. The transistor starts to conduct even before 400mv vbe.
What remains to be solved for the elektor based version is how to make it work properly as a bridge. Each half working properly independently doesn't make it work properly once bridged. The resistive divider network taken as is from a crown schematic will require different values at least.
Plus the action of the limiter makes the vas act strangely as well, and that causes oscillations. The insertion of resistors between the vas and the limiter should put a damper on that. And I've been testing the addition of a cap & resistor in series placed after the sensor resistors, which not only adds a little delay in the limiter's action, which will let the transients pass some more, it also prevents oscillations.
I added a protection transistor on the base of the vas (john ellis' bridge), so it won't behave as wildly when overloaded by the limiter, so this also makes it behave better.
I will change the limiter not to use the diode/resistors going to ground, in favor of the dual slope with zeners, as our esteemed friend kiwanuka describes on fig 28 in his excellent paper.
The limiter transistor starts acting way before the set threashold, but only partly, and the loop gain seems to be sufficient to correct for the little leakage induced by the overly early action. The transistor starts to conduct even before 400mv vbe.
What remains to be solved for the elektor based version is how to make it work properly as a bridge. Each half working properly independently doesn't make it work properly once bridged. The resistive divider network taken as is from a crown schematic will require different values at least.
I have been saying this for nearly a decade.
The IV protection MUST pass all valid audio signals to all valid audio loads.
Yes, I know, and you're right on.
The simulations do show that clearly.And this implies that we must have much more soa available than needed for normal use, so the threshold can be placed high above the normal use with that extra headroom required to prevent the early action from messing up the sound.
There is one more thing that the simulations have revealed to me, is that the dissipated power isn't really limited well enough, so although we calculate a threshold for action, counting on that limiter for placing that dissipation limit right there, and in fact, when acting, the dissipation goes up instead of down. There is a limitation action, but it is really much higher than what we set it at with the calculations and with the actual action of the limiter.
Perhaps there are other factors that require investigation, to make the limiter more effective and get the dissipation anticipated.
I DID try to tell them this in not quite the same way about 52 pages ago
Still not to rain on anyone's parade... I still do valve amps too so can hardly complain!
Just for that I'm going to get out my stash of germanium transistors and play with them this weekend
And back to the protection circuits, they're not black and white. They are always shades of gray. You do pretty much not want them to activate at all driving a single pair of speakers. But living with a little action when driving two pair may be ok. And whether it "fully protects" may be time and temperature dependent as well. You want to be able to survive a momentary short or overload - long enough to notice the distortion and correct the problem. But expecting to drive indefinitley into a short in full limit while cooking eggs on the chassis is a bit beyond the call of duty. Transistors just don't blow up the very instant you go outisde rated SOA. It depends for how long, by how much, and how hot it was to begin with.
I would use the zener version that references the rail (not ground) and be done with it. Once the topology is chosen you can finish the pcb and get some made. The answer for exactly where to set the IV limits won't come from simulation. It will come from a prototype.
I would use the zener version that references the rail (not ground) and be done with it. Once the topology is chosen you can finish the pcb and get some made. The answer for exactly where to set the IV limits won't come from simulation. It will come from a prototype.
That's true, I agree, and I have even contemplated adding other types of protections around that to complement the limiters.
I've done many simulations and most of my curiosity and unknowns are now taken care of.
I'll work on the calculations spreadsheet.
There is one thing about the elektor based one that I think will need to be resolved before we consider it ready for prototyping: both sides work independently (in sims), but not when bridged. So there must be something going on that needs to be addressed. I think it has to do with the resistor divider network that does the bridging. As we took it as is from a crown design, there is likely an adjustment to make for this one.
Other than that, the john ellis bridge works like a charm in sims, and all I need to solve on that one is the very nasty behavior of the limiters, which is likely to be fixed in a similar manner as the elektor one. I've seen some improvements already with a few small alterations, such as adding the protection transistor on the vas for example. Even with that, the behavior is wild, with huge dissipation increase when limiting (over 800W peaks per output). I haven't yet tried the zener type limiter, but I suspect the suppression of the diodes/resistors going to ground will solve a good part of the issues.
More work to do, but getting there. We'll do a proto real soon. For both.
Yes! I've been looking at that one too. I need to put this into a spreadsheet though. Some work to do, then comes the fun of prototyping and building.
The manufacturer's SOA plots take account of duration.
That's why we see double the current capability for medium term loading and exceeding 10times the current capability for very short term loading.
Compare the DC, 1second, 100ms, 10ms, 1ms & 100us SOA plots to see the duration effect.
The manufacturers also gives guidance of de-rating for temperatures exceeding 25°C.
I'm working on the spreadsheet to calculate the limiter's values. Not quite as easy as some may think it is when one isn't really at ease with maths. Plus I have to do simultaneous equation solving in a spreadsheet, so not quite sure how that works, as I may be a casual user of excel, I'm not an expert. I'll figure this out eventually.
One question came to mind though. Thinking about using the TO126 types for both the bias spreader and the pre-drivers (and vas) on the elektor based amp (bd139/140). Since we should thermally couple the bias spreader and pre-drivers, using all TO126 makes this much easier and more efficient thermally than the TO92 types, as the contact surface is much larger and we can use screws to couple them for good.
The question that came to mind is since they have their collectors linked to their metal backside, they must be isolated from each other, which requires a mica between them, and thus adding an extra Rth there. I know the substrate is physically on that metal plate inside and not directly in contact with the plastic front side, however, how much worse or better would the thermal coupling be if we put the front faces instead of the back faces in contact together?
If the front faces are in contact and we use plastic screws to couple them, then with some thermal grease between them, the contact would be much better than with a mica between the metal faces. This could be a good arrangement, as long as there is a good enough thermal link inside the case from the substrate to the front side of the case.
Any ideas on this? I'm just thinking ahead about the physical assembly.
One question came to mind though. Thinking about using the TO126 types for both the bias spreader and the pre-drivers (and vas) on the elektor based amp (bd139/140). Since we should thermally couple the bias spreader and pre-drivers, using all TO126 makes this much easier and more efficient thermally than the TO92 types, as the contact surface is much larger and we can use screws to couple them for good.
The question that came to mind is since they have their collectors linked to their metal backside, they must be isolated from each other, which requires a mica between them, and thus adding an extra Rth there. I know the substrate is physically on that metal plate inside and not directly in contact with the plastic front side, however, how much worse or better would the thermal coupling be if we put the front faces instead of the back faces in contact together?
If the front faces are in contact and we use plastic screws to couple them, then with some thermal grease between them, the contact would be much better than with a mica between the metal faces. This could be a good arrangement, as long as there is a good enough thermal link inside the case from the substrate to the front side of the case.
Any ideas on this? I'm just thinking ahead about the physical assembly.
There is one other thing that I've been experiencing, which may be due to the models, but could also be due simply to how the bd139/40 are. Their Vbe seems to be lower than the common 0.6V, at somewhere around 0.4V, so that was why biasing it with the simple elektor original spreader didn't work so well and the bias trimmer had to be just above zero ohms to reach a low enough bias.
Since we switch to the bd139/40 as pre-drivers and also to the bias spreader, we need 2 of them also on the spreader so they can be coupled by pairs with the pre-drivers.
But the spreader needs to "squeeze" hard enough to bring the total spread between the bases to a little less than 800mV total, so the bias can be adjusted low enough.
I've tried various models, and It seems to me those bd139/40 models don't seem bad, and they just may reflect how the real ones are. I'd like confirmation about, this, as this does make a big difference.
Since we switch to the bd139/40 as pre-drivers and also to the bias spreader, we need 2 of them also on the spreader so they can be coupled by pairs with the pre-drivers.
But the spreader needs to "squeeze" hard enough to bring the total spread between the bases to a little less than 800mV total, so the bias can be adjusted low enough.
I've tried various models, and It seems to me those bd139/40 models don't seem bad, and they just may reflect how the real ones are. I'd like confirmation about, this, as this does make a big difference.
The Vbe's are more than 0.4V. Maybe at microamp current levels it's that low, but at 10mA it's more like 0.6. I use them for temp sensors quite a bit - and not just for the bias spreaders. The Fairchild BD139's are completely encapsulated - no metal tab showing. Heat going *out* is affected quite a bit (more themal rise per watt than a tabbed device), but heat going *in* happens quickly. Put your finger on it and you can watch the Vbe drop pretty fast- stabilizing in less than a second. Even a high Rth device like a TO-92 will do that - because the heat source is external and not dissipation-dependent. I mount them directly to output transistor cases (TO-264's, TO218 tabs, TO-3's) for all sorts of stuff - bias spreaders, sensors for turning on fans, muting the inputs on overtemp, etc.
Actually I have been using mostly cordell's models for that, and I didn't have many choices, but assumed his models are trustworthy.
However I dug out a couple of phillips models and those are not doing that, with a vbe in the same position (pre-drivers) around 650mv instead of less than 400 for cordell's. Wierd!
I switched to those models to see how that works out. I have them on the pre-drivers, bias spreader and vas (enhanced with bc550/60c).
It works at a very different operating point with those different vbe voltages, the trimmer adjustment for the bias would have to cover any such large differences in case something like this happens in the real world.
The issue now is that this amp has high thd at 20k. This could be due to the fact that I am now running the lpts with their ccs at about 6mA. I really don't see the real advantage of running the lpts so heavy, all that to earn some high slew rate. I think that increases noise a lot, but I suspect it does make them go out of their linear range, despite what some say. I would like to be certain about this, but I feel like it's not a plus to run them that high. Maybe 3 or 4mA tops.
I'm doing those sims on the elektor based one, just about the way we had it somewhat recently, with ccs but no mirrors, and the enhanced vas.
The thd @ 20k that I'm getting right now is above 0.6%, with rather small caps on the compensation, using the bridged tmc as described by cordell and only a 2p2 on the feedback, with the values in the tmc around 3-4pf. I would expect better performance.
It's not the 3055s causing distortion, at least not the transistors themselves, but rather the topology. I'm not even sure going complementary would bring much more improvements.
The only way to get lower thd that I found was to add complexity, either making the lpts as sziklai pairs, or loading them with mirrors. I even had tried the enhanced vas with the cascode as cordell describes it, but that doesn't bring much compared to the extra complexity. I think the sziklai pair would probably bring the most improvement, because with the degenerated bases and emitters on the lpts, the open loop gain is only around 70db. Not enough for a more effective feedback reduced thd.
I've had open loop gains over 125db with some topos, which really helps bringing thd down, at the expense of greated complexity and more difficulty to keep it stable.
I returned to our less improved schematic and tried a few things that would give us some fair performance without too much complexity.
What I don't understand is how elektor was claiming less than 0.02% thd at all frequencies, while no sims can get even close to that with the schematic as is was.
There is one other thing that came up during all those simulations: the fact that too much dissipation can happen in the emitter resistors on the ouput, the ones we sense for the limiters.
It's easy to see real quick that very beefy types would need to be used, if the bridge ends up loaded with 4 ohms, each side seeing 2.
With the goal of having the threshold at 27A for the limiters action, we can easily see that some serious peak dissipation can happen in those 0.22ohms resistors, so I reduced them to 0.1, which is quite sufficient to trigger the limiters because they still cause a good amount of voltage drop for that. And since there are other such resistors on each output as well, we cause a good amount of drop there too, although a bit less because they only see 1/3rd the current.
If we allow for the psu to sag some, plus the drops in 2 resistors in series in the output emitters, plus the vce drops in those outputs, we lose 5 or 6 volts. That's a fair amount of headroom lost. So I think using 0.1ohms is enough. We wouldn't go any lower, so the sensor still has enough to work with, and the others in the emitters can still do their job of equalizing currents. All transistors should be matched the best possible.
The choice for those power resistors isn't going to be so simple, if this amp is to be built in places where choices are limited.
It's easy to see real quick that very beefy types would need to be used, if the bridge ends up loaded with 4 ohms, each side seeing 2.
With the goal of having the threshold at 27A for the limiters action, we can easily see that some serious peak dissipation can happen in those 0.22ohms resistors, so I reduced them to 0.1, which is quite sufficient to trigger the limiters because they still cause a good amount of voltage drop for that. And since there are other such resistors on each output as well, we cause a good amount of drop there too, although a bit less because they only see 1/3rd the current.
If we allow for the psu to sag some, plus the drops in 2 resistors in series in the output emitters, plus the vce drops in those outputs, we lose 5 or 6 volts. That's a fair amount of headroom lost. So I think using 0.1ohms is enough. We wouldn't go any lower, so the sensor still has enough to work with, and the others in the emitters can still do their job of equalizing currents. All transistors should be matched the best possible.
The choice for those power resistors isn't going to be so simple, if this amp is to be built in places where choices are limited.
Mounting the sot-BJTs reverse should be no issue as long as there is no significant power dissipation inside them.
Providing this thermal flow is at minimum, and additional thermal resistance does not cause significant temperature gradients.
Anyway, keep in mind that nylon srews are not designed for high temperatures.
But you could use metal srews and insert the mica on the backside, just to isolate the srew.
Providing this thermal flow is at minimum, and additional thermal resistance does not cause significant temperature gradients.
Anyway, keep in mind that nylon srews are not designed for high temperatures.
But you could use metal srews and insert the mica on the backside, just to isolate the srew.
The bias spreader dissipates only 3 or 4mw in each, so that shouldn't cause any heat to speak of. The pre-drivers dissipate less than 30mw, with an Ic around 700-800uA, so they probably will heat up a tiny bit, but nothing big, and since we're coupling the pre-drivers to the bias spreader and not to the main heatsink, they should hardly receive any heat feedback from the main heatsink. I will layout the pcb with some distance away from the sink, so this can't interfere.
The bias spreader is supposed to only "see" (sense) the pre-drivers. So the heat flow would only be minor and from the pre-drivers to the spreader. Let's hope this scheme works as it should.
With less than 30mw in one and less than 5mw in the other in each coupled pair, I really doubt we need to worry about much heat there.
That could be done, and the diyers who will be building them will have the choices. I would only use plastic screws. There is no heat to speak of, and there is no need to a very strong torque to keep them coupled.
The good thing is not having to mount them on the main sink. We'll only have the drivers with the output on the sinks.
I'm still working on the vi limiter and looking at the worst conditions in case of a short or near short, the dissipation in the output's emitter resistors are a concern. I'm not too worried about each transistor's emitter resistor, as their value is already low and they only take 1/3rd of the total output current. The resistors used for the limiter's sensing can be subjected to quite a load when a short happens, and with the limiter acting and the short maintained, they have a long term dissipation to handle, unless there is some other protection that will act and stop it.
I had already reduced the 0.22 to 0.1, as we don't really need that much voltage there to sense. Keeping the goal of limiting at 27A total, we have a 2.7V drop on that resistor, which is far more than needed to trigger the limiter, especially since we're going to use the arrangement with the zeners (kiwanuka fig 41). The limiter's action should cause some fold back, as I understand it, but the simulations have not shown me that, and the actual dissipation in the outputs is higher with the limiter acting than without it at all.
Anyway, assuming a near short on the output, perhaps with some length of cable there, with connections along the way, we don't really have 0 ohms on the amp's output, especially with the output L//R thingy in series there, which might be a few 10s or perhaps close to 100 mohms. Sticking with a max current of 27A, we're way in overload for a 10W 0.1ohms resistor. Those are rather tough and can take some punishment for a while, but not forever. So eventually a resistor could throw the towel and there goes the protection sensor...
For any short duration shorts, the overloads shouldn't be too much of a cause for concern, however a longer one would cause issues in the resistors probably before the transistors, depending on how actually effective the limiter is in limiting the dissipation.
I was just looking at the MPC types of power resistors, and they do have values below 0.1ohms in the 5W MPC75 models, so for example using 0.05ohms instead of 0.1 would cut in half the fault condition dissipation, all while also reducing the drop during normal use, for a lower overhead and even a tiny bit less distortion.
One thing that could also be done, since even with a 50mohms and 27A going through it, we're still above 36W of dissipation in case of short, there is the MPC725 model which is 2 times 5W and they have the 2 x 0.1ohms model. That would be used by paralleling the 2 halves to make a single 10W 50mohms power res. 10W is better than 5W to handle an overload of over 36W, so they should handle it much longer.
With 50mohms, at 27A, we still have 1.35V of drop, which is enough to trigger the vi limiter.
I've seen other types of more expensive power resistors in TO220 or similar cases, with much higher dissipation possible, that can even take a heatsink for yet more dissipation. This is quite a bit more expensive, but then there would be no risk of overload induced breakage.
Does this make sense? Or am I a little overprotective?
We need to keep in mind that while we are planning a bridge for 8ohms loads, each bridge side seeing 4ohms, there will be some who would load the bridge with 4ohms. We're aiming at allowing such low loads, although the max power might be a bit lower than theoretical then.
Because of the limiter's transistors wanting to start acting a little bit far before the threshold we calculate, we have no choice but to place the threshold as far above the normal usage as we can, so even a 2ohms load on an amp side would be allowed (mostly). I think it would be good to supplement the limiter with some other type of protection for the longer short durations.
I had already reduced the 0.22 to 0.1, as we don't really need that much voltage there to sense. Keeping the goal of limiting at 27A total, we have a 2.7V drop on that resistor, which is far more than needed to trigger the limiter, especially since we're going to use the arrangement with the zeners (kiwanuka fig 41). The limiter's action should cause some fold back, as I understand it, but the simulations have not shown me that, and the actual dissipation in the outputs is higher with the limiter acting than without it at all.
Anyway, assuming a near short on the output, perhaps with some length of cable there, with connections along the way, we don't really have 0 ohms on the amp's output, especially with the output L//R thingy in series there, which might be a few 10s or perhaps close to 100 mohms. Sticking with a max current of 27A, we're way in overload for a 10W 0.1ohms resistor. Those are rather tough and can take some punishment for a while, but not forever. So eventually a resistor could throw the towel and there goes the protection sensor...
For any short duration shorts, the overloads shouldn't be too much of a cause for concern, however a longer one would cause issues in the resistors probably before the transistors, depending on how actually effective the limiter is in limiting the dissipation.
I was just looking at the MPC types of power resistors, and they do have values below 0.1ohms in the 5W MPC75 models, so for example using 0.05ohms instead of 0.1 would cut in half the fault condition dissipation, all while also reducing the drop during normal use, for a lower overhead and even a tiny bit less distortion.
One thing that could also be done, since even with a 50mohms and 27A going through it, we're still above 36W of dissipation in case of short, there is the MPC725 model which is 2 times 5W and they have the 2 x 0.1ohms model. That would be used by paralleling the 2 halves to make a single 10W 50mohms power res. 10W is better than 5W to handle an overload of over 36W, so they should handle it much longer.
With 50mohms, at 27A, we still have 1.35V of drop, which is enough to trigger the vi limiter.
I've seen other types of more expensive power resistors in TO220 or similar cases, with much higher dissipation possible, that can even take a heatsink for yet more dissipation. This is quite a bit more expensive, but then there would be no risk of overload induced breakage.
Does this make sense? Or am I a little overprotective?
We need to keep in mind that while we are planning a bridge for 8ohms loads, each bridge side seeing 4ohms, there will be some who would load the bridge with 4ohms. We're aiming at allowing such low loads, although the max power might be a bit lower than theoretical then.
Because of the limiter's transistors wanting to start acting a little bit far before the threshold we calculate, we have no choice but to place the threshold as far above the normal usage as we can, so even a 2ohms load on an amp side would be allowed (mostly). I think it would be good to supplement the limiter with some other type of protection for the longer short durations.