Battery-powered chipamp has an intermitted DC-offset issue

[eisnerd]

As suggested by the subject, I have a simple battery-powered chipamp, using two ex-UPS 12V lead-acid batteries and the TDA2050V chip in a non-inverting arrangement, specifically. The DC offset is usually negligable, and there's very little switch on pop, however, on switching off I sometimes get a mid/high pitched sucking/wizzing sound, which I've heard from other amps or powered speakers before. From the look of the driver, while this happens there's a more substantial offset, though not enough to do damage (so far, at least to these speakers).

I have the batteries arranged -12V; 0V; +12V straight to the supply caps & chip, except for a simple DPST rocker switch on the +-12V lines. My suspicion is that the effect I described is caused by the rocker engaging or detaching slightly out of time.

Any suggestions as to switches that may improve this or circuits to guard against this imbalance either at the supply or output? Or perhaps I'm barking up the wrong tree and have another problem or should ignore this.

[otherside]

Schematic at hand?

One thing that your switch will cause the the caps to discharge through the amp. Because for each rail they'll discharge at different rates this could be causing the problem.

I would suggest adding diodes across your switch (just like you do on relays, or regulators) that will discharge your caps back to battery. Alternatively use a DPDT and add discharge resistors on one side, move the battery input on the other side and take the output from the middle.

I assume that the sound only lasts a second or two... or you mean permanent?

Also because with batteries things can be slightly different than PSUs make sure your grounding is correct.

[AndrewT]

Can you measure the voltage across the smoothing caps as they discharge?
Sometimes one side discharges much faster than the other and the charged side then forces the discharged side into reverse polarity.

If this is the case then a pair of diodes solves the problem.

Fit reversed diodes from +ve to zero volts and again from zero volts to -ve. This limits the reverse voltage to <=700mV