Younger brother of the LM3886 - LM2876 in parallel

It's the matching between the resistors that matter. They need to be tightly matched. They do not need to be of high absolute tolerance. 0.10 vs 0.12 Ω really doesn't matter (much). But if you buy a handful of ±0.1 % tolerance resistors you know the resistors will match to within 0.1 % for an average pair and 0.2 % worst case. A pair of ±5 % resistors could be mismatched by as much as 10 %.

Knowing that it's the matching that matters changes things. It would be fairly easy to set up a resistor bridge and find sets of resistors that matched. If you have enough of those ±5 % tolerance resistors, you could possibly create matched pairs or matched quads.

Resistors are fairly cheap and time is expensive. At least my time is. So personally I would just buy some resistors of known, tight tolerance. You can find 0.1 Ω, 3 W, ±0.1 % without much fuss.

Another thing to keep in mind is that you, ideally, want resistors with a low temperature coefficient. Recall that, unless you get fancy, the resistors will be outside the feedback loop, so they will contribute distortion to the amplifier output if they have significant temperature coefficient. So if you go resistor shopping see if you can find some ±50 ppm/ºC ones instead of the ±hundreds of ppm/ºC commonly available for power resistors. If you care about distortion anyway.

I said that 0.10 vs 0.12 Ω does not matter much. Higher resistance -> higher dissipated power. That's basically the tradeoff. So I'd go with 0.1 Ω.

Tom
 
I’ve updated the project to use four LM2876 (or LM3886) chips in parallel. This configuration splits the output current across four amplifiers, so for a 4-ohm speaker delivering 80W, each chip handles only about 1.58 A peak (total peak current ≈ 6.33 A, with a 25.3 V peak output).

I’m using 0.47 Ω ballast resistors for current sharing, as they offer the best safety margin in terms of thermal balance and mismatch protection. In the worst-case offset scenario (±10 mV), the maximum circulating current is only about 21 mA. By comparison, using 0.1 Ω resistors would increase this to 100 mA, which raises distortion and the risk of thermal runaway.

If you’re using a DC servo, then 0.1 Ω could work well for improved efficiency and damping. But without a servo, I recommend sticking to 0.47 Ω, or at minimum 0.22 Ω, to maintain reliability and thermal stability.

For the input buffer, I tested multiple op-amps and found the OPA1656 to perform best. It runs at a gain of 1.5 and includes a 22 Ω series output resistor (R_iso) to ensure stability and isolate capacitive loading.

Measures 0.001% thd+n into 4 ohms load at 1khz and sounds absolutely amazing!
ICs get barely warm even at full power out!
 
  • Like
Reactions: Sensation45
If you’re using a DC servo, then 0.1 Ω could work well for improved efficiency and damping. But without a servo, I recommend sticking to 0.47 Ω, or at minimum 0.22 Ω, to maintain reliability and thermal stability.
Your thinking works if each LM2876/LM3886 has its own DC servo. There's no difference if the DC servo is global, i.e., works on all four power amps at once.

I haven't had any issue with 0.1 Ω with 2-3 LM3886 in parallel, but I also run the LM3886 with unity gain at DC (i.e., a capacitor in the grounded leg of the feedback divider). This ensures a maximum DC offset of 10 mV, guaranteed. Typical is 2 mV. 2 mV will result in 20 mA of error current through the ballast resistors. That seems manageable.

Tom
 
  • Like
Reactions: pbilous
4.26 mA of current for typical offset 2mV using 0.47 ohms ballast resistors

Overkill but in practice ICs do not heat up like they were when I used 0.1 ohms.

Jeff Rowland design used 0.5 ohms.
 

Attachments

  • IMG_5531.jpeg
    IMG_5531.jpeg
    225.9 KB · Views: 5