LM3886 driving long cable instability

[mobyd]

I have a situation where I need to send 5 channels of audio about 35 metres to the speakers. The multichannel amp uses LM3886s. All went well in the workshop but when I included the cable, lots of white noise and heating of amps (though they don't sound distorted). The cable I'm using is 100 core telephone cable connected as 10 core x 10 strands. I know this isn't ideal but that's what there is. I'm assuming HF instability. The situation is no good for test equipment. The amps have both LR and LC zobels. Any ideas where to apply bandaids ?.
M

[AndrewT]

RC zobel right next to the output pin feeding directly into the PCB decoupling ground (Power Ground).
L//R of the Thiele Network on the route to the speaker output terminals.

RC Zobel across the speaker output terminals.

This can make the amplifier insensitive to any reactive loading on the output.

What gain are you using?

[Michael Bean]

Sounds like that too much cable capacitance. For a run of 35 meters you would probably get better results with 14 or even 12 gauge two conductor zip or extension cord wire.

Mike

[godfrey]

The best solution is an RC Zobel at the loudspeaker end of the cable. Ideally the resistance should match the characteristic impedance of the cable, but anywhere in the ballpark is good.

It sounds like a low-impedance arrangement so I would guess 10R or 22R in series with 0.1uF would do the trick. For normal figure-of-eight cable, 100R would be about right.

[infinia]

Sounds like cross coupling problems with all outputs so closely packed on a single bulk cable. you could decouple each amp output better with a series resistor or experiment with selecting different numbers of twisted pairs. Try around 1 ohm or so. grounding and RF input filters may help here as well. images of the test build can't hurt for better advice.

[jcx]

chip amps are decompensated, have minimum gain requirements, gain higher than the minimum spec improves stability

also chip amps are slow, the loop gain intercept is in the 100s of KHz, so the series L may have to be higher in value than you expect - what are you using for the L? - ferrite can saturate reducing L

look at the Zobel R too after oscillation - not many are not sized to survive sustained full amplitude oscillation - they may be open now

PS pin bypass, possibly Zobel damping may be needed if the oscillation frequency is readily visible at the chip amp pins do to poor PS impedance, long wires


the phone cable will never be low C, but different wiring arrangements may be better or worse on crosstalk - I think you should use the pairs as pairs, +,- for a single speaker in one pair, then parallel pairs for each speaker

of course miswiring so amps are shorted together will cause problems too

[infinia]

Originally Posted by jcx

also chip amps are slow, the loop gain intercept is in the 100s of KHz, so the series L may have to be higher in value than you expect - what are you using for the L? - ferrite can saturate reducing L

look at the Zobel R too after oscillation - not many are not sized to survive sustained full amplitude oscillation - they may be open now

actually the opposite, for NS the UG LBW is beyond several MHz, quite a bit faster than most discrete designs by at least 2 times.
something else besides down-rite oscillations is going on , cause the zobel woulds tend to smoke awhile 1st before actually going open. the output L would tend decouple things higher up, and is why I suspect cross modulation, even ferrites once saturated can work OK here. The OP could test things in steps to help trouble shoot and give more clues. of course a scope and square wave testing could put to bed most concerns folks have ie stability.

[jcx]

can't read the datasheet? or don't understand GBW relation to loop gain, intercept?

from the datasheet:

GBW is 2 MHz min, 8 Mhz typ

min gain is 10 with hedging:

"The LM3886 is designed to be stable when operated at a
closed-loop gain of 10 or greater, but..."

"Normally the gain is set between 20 and 200;..."


then I would derate them for the min GBW of 2 MHz - ie by 4x


so a LM3886 with Av +20 could have loop gain intercept as low as 100 KHz

the feedback causes impedance peaking at the intercept when the phase margin is low, Cload typically degrades phase margin more which is the reason for a decouping L

the L reactance needs to be larger than the C load by the gain intercept frequency to effectively decouple it

the OP's cable construction probably has ~ 10x the parasitic load C of conventional loudspeaker cables for which the stability was tested with the given output L - so a quick estimate would be to use 10x higher L = ~ 7 uH vs the datasheet example circuit 0.7 uH


some ferrite core L are hopelessly optimistic in giving "max I" as the value that gives 20% L from saturation (that's "reduced to 20%" not "by 20%")
you have to look up the material curves, derate by large factors when using ferrite
on distortion grounds you probably don't want ferrite series L changing by even 10% with operating current

[infinia]

my point being is that NS chip amps are not slow! It's normal for monolithic op-amps using internal dominate pole compensation to merely speak of unity gain BW, every other BW that really matters is circuit dependent, ie closed loop, small signal etc. I doubt the use ferrite's in this app, but even if they were saturated the parasitics would be helpful, ferrite beads are a good example off the materials lossy attenuation after saturation.

[mobyd]

OK, I've attached a piccy of the beast in question. The central amp block was recycled out of a domestic 12 channel distributed sound system, and the designer seems to have followed the NS applications circuit (pg.5) to the letter - L and C zobels identical etc. The only mod I have done is to up Rf series from 20k to 40k to double the gain to 40 as my little SD card players only put out 1.5v p/p. I have discovered that I can kill the hiss (which appears to be broadband white noise with no apparent dominant frequency) by putting ferrite rings (the big orange donuts) on the speaker output lines.
Should I be having a lot more turns on those chokes ?.
Thanks for all your suggestions.
M