I'm quite surprised at this discovery. Have you come to any explanation that confirms why the THD is multiplied ten fold for separate (cooler) chips?
Then the voltage might be too high? Ironically, it is the builders who aren't trying to push for power voltage, who usually get more power. That can apply when using 4 ohm speakers, or if bridged with lighter load speakers.
I ran a quick check with my old LM3886T on Tabor's black alloy, a support design similar to Philfr's, my old rusty Stancor 20-0-20vac 4A center tap, a simple power circuit with 470u//10,000u (in AN1192), and paralleled 8 ohm speakers stacked vertically, with the top one upside down so the tweeters would be close together (I didn't have a proper double woofer speaker handy at the moment). Plenty good enough for a quick concept check.
What happened was a placid amp (cool running, pretty tone) and almost twice the usable power. Well, it is good to have nice surprises.
So, I came out with this:
Mod86 with 4 ohm speakers, +-20vac transformer
Bridged Mod86 with 8 ohm speakers, +-20vac transformer
Bridged Mod86 with 4 ohm speakers, +-15vac transformer
Bridged Par86 with 4 ohm speakers +-22vac transformer
Those are not typical transformer choices with the LM3886, but is it correct?
Oh right! Even though it works if the power voltage is low enough, it isn't useful to do it. With the efficiency difference (the majority of the power into the heatsink instead of the speaker), then power to the speaker is similar or less (irritated mr spike) than if it wasn't bridged. Sorry, I hadn't considered enough variables. Thanks!
However,
These two bridge options still look pretty good for big amp sound:
Bridged Mod86 with 8 ohm speakers, +-20vac transformer
Bridged Par86 with 4 ohm speakers +-22vac transformer
The bridged Par86 should probably be monoblocs.
I do still figure that there are some people in need of a bridge amp. But, lets do continue to make sure that we deliver that when appropriate. Thanks again!
It has little to do with temperature, actually. The main culprit is the mismatch in offset voltage.
I doubt the two dice in an LM4780 are matched. That would add cost. But they may come from adjacent spots on the wafer, hence, have a beter chance of matching than two ICs picked at random from the same batch. This is a guess, but based on my experience with the LM4780 vs two LM3886es, that makes sense. There will always be a difference in the input offset voltage between two LM3886es (and between the two halves of an LM4780. In my experience, based on rather few samples, the difference in offset voltage is considerably less for the two halves of an LM4780 than for two LM3886es from the same tube.
The mismatch in offset voltage is what causes two LM3886es in parallel to "fight". The mismatch in offset sets up a standing current in the ballast resistors. Worst case, the mismatch is 20 mV. Most use Rballast = 0.1 Ω, so you'll have 20 mV across 0.2 Ω -> 1 A of standing current. Now the good news is that the ICs will operate in Class A for most of the cycle, however, the bad news is that they'll hit their current limits rather quickly and basically provide little benefit over a single LM3886. On top of that the THD degrades by roughly 20 dB. That's probably why people tend to shy away from the BPA200 build.
There are a couple of solutions to this: One is to use ICs that are better matched and run at the same temperature (Vos tends to depend on temperature). This minimizes the standing current. Another solution is what Jeff Rowland (sp?) used in his 6xLM3886 amp. Notice the blue multi-turn trimpots? I bet those are for offset adjustment.
Tom
Last edited:
I can see that temperature has little to nothing to do you what you found, otherwise one might expect the dual 3886 with the cooler operating capability to perform better.
You have found the opposite.
Input offset voltage difference is possible/likely.
I have made comment on that in numerous posts about paralleled 3886. National give no specific warning against differing input offsets causing an operational problem.
I have suggested always DC blocking EACH input to prevent unwanted interaction between the input offsets of the channels.
You have found the opposite.
Input offset voltage difference is possible/likely.
I have made comment on that in numerous posts about paralleled 3886. National give no specific warning against differing input offsets causing an operational problem.
I have suggested always DC blocking EACH input to prevent unwanted interaction between the input offsets of the channels.
Unless the time constants are truly precision matched this will only make the cross current problem worse, for LF signals.
Plus, the best way to eliminate any additional current offset mismatch is to have neglegible pin to pin resistance, I've used stoppers in the 1k range max.
The biggest problem in parallel 4780 builds was a huge(!) cross current spike during start up. When supply voltages ramp up the two dice don't start to operate synchronously. Causes a lot of confusion on the bench when I used a current limited bench supply.
Check your math
I skipped a step.
--> Worst case is one LM3886 with -10 mV, the other, +10 mV --> 20 mV.
The gain of the LM3886 section is +11 V/V, so 220 mV is the worst case offset.
220 mV across 2*0.1 Ω --> 1.1 A in standing current.
Tom
Where do you get that the LM3886 will work better at cold? I don't recall any statements to that effect in the data sheet, but I haven't committed every bit of the .pdf to memory. Am I missing something?
The BPA200 has a local DC servo for each LM3886. That minimizes the issues with offset voltage.
That will do absolutely nothing to address the issue with the input offset voltage of the LM3886, though. The input offsets don't interacts at the input of the amp. They're in series with each input and grounded through the source output impedance.
Tom
Could be one of the causes of failure.
Does not explain the ten fold increase in distortion measured by Tom.
I'm not sure what you're trying to say. I was responding to your statement that AC coupling the input to the LM3886 would minimize the input offset. That statement is false.
AC coupling the input to the LM3886 will prevent offsets from whatever drives the LM3886 from making it to the LM3886, but will do nothing for the input offset voltage inherent to the LM3886.
Tom
So the LM3886 has a minimum gain of 10 V/V (20 dB) for stability. Sadly, this means its input offset will be amplified by 10x. This is what leads to the 100 mV of output voltage offset and 1 A of standing current that I calculated earlier.
Increasing the feedback factor to 1 at DC by adding a capacitor to ground, such as the one circled below, will cause the LM3886 to have a gain of 1 V/V at DC, hence, lower the max. output offset to 10 mV and reduce the standing current by 10x.
This is the exact reason why you see four bright green towering electrolytic caps on the Parallel-86 board.
Reducing the gain at DC does improve the DC offset on the output of the amp.
AC coupling the input does NOT improve the DC offset on the output of the amp. In case of two LM3886es in parallel, an input offset coming from the source would apply equally to both LM3886es. This would result in a DC voltage on the output of the amp, but would not cause a mismatch in offset voltage between the two LM3886es, hence, not cause an increase in the standing current.
If you're looking at just driving two LM3886es in parallel, you can do even better with a DC servo on each LM3886, but in a composite amplifier design this approach won't work. I'll leave it as an exercise for the reader to figure out why.
Yeah, well. Even a blind squirrel finds a nut every now and then.
Andrew has his moments of glory. Extracting them from the terse and at times cryptic communication can be a bit challenging, though.Tom
Last edited:
Talking about challenges, try reverse engineering you PCB designs.
Why bother? I have more boards available.
Thanks, by the way. I put a lot of effort into those layouts. They're a product of thought, understanding, simulations, and measurements.
Tom
Hi Tom,
I tried to use you BOM to order components for my Modulus-86, but two of the listed capacitors are currently out of stock at Mouser.ca. They are available at Digikey.ca but Digikey doesn't sell THAT1200. I am trying to source components from only one supplier to save shipping costs. Could you specify Mouse P/N for replacement capacitors that I can use to guarantee Mod-86's stella performance?
The two caps that are currently out of stock at Mouser are:
22uf Mouser P/N 647-UPM1H220MED1TD
100pf NP0 Mouser P/N 810-FK28C0G1H101J
Many thanks!
pakultra
Toronto
I tried to use you BOM to order components for my Modulus-86, but two of the listed capacitors are currently out of stock at Mouser.ca. They are available at Digikey.ca but Digikey doesn't sell THAT1200. I am trying to source components from only one supplier to save shipping costs. Could you specify Mouse P/N for replacement capacitors that I can use to guarantee Mod-86's stella performance?
The two caps that are currently out of stock at Mouser are:
22uf Mouser P/N 647-UPM1H220MED1TD
100pf NP0 Mouser P/N 810-FK28C0G1H101J
Many thanks!
pakultra
Toronto