Parallel LM3886 problem

[shmulik]

as a BPA300 group buy veteran, I'm still struggling with problems I would like get rid of but don't know how.
to state the problem in simple terms, using the default components values, when a square wave of ~400 Hz is played at the input, a ringing effect is watched with an oscilloscope at the output, at a single PA150 board solely connected to the PS.
installing larger NFB capacitor (15pF instead of the recommended 8pF) smooths the ringing but also kills the frequency response. both the ringing and the worse frequency response are audible.
having another board (a clone bought on ebay before PA saved the day) and some spare parts, I was determined to track the problem without ruining the existing completed boards.
Unlike the first attempt that was in success-oriented mode (meaning everything was built and installed into chassis before measurements taken place), I now placed the LMs without installing the output resistors and without DC offset canceling trimmers (direct connection to ground on non-inverting input). guess what - I had 22, 26 and 40 mV dc on the three outputs BUT the square wave was perfect. I installed the trimmers (100K - lesson learned from the previous attemt with 50K) and brought the offset to 0 on all. that required a lot of back and forth resistance changes since there is coupling between LMs (why?).
now I had zero offset but ringing on edges of the square wave. the amplitude is influenced by the amount of offset correction applied - more correction, larger ringing amplitude.
in despair I shuffled some wires and turned the circuit into a non-inverting one. installing a feedback capacitor to ground I managed 12, 15 and 30 mV offset and nice square wave.

How can I have get rid of the dc offset without compromising response? I have all components from the same batch (same batch number) but obviously not matched. is there a way to mach without soldering them into the circuit?
Need to mention also that the FB resistors are 200k and 10k 0.1%.

[Mark Allen]

This is an interesting problem and I have yet to look at my boards with a square wave... but I will try to do that this weekend and get back on it. I have to say they are still some of the best sounding boards I;ve ever listened to though so I suspect mine are not suffering from the ringing problem. I always thought this project should have included a servo on each chip and this may be the answer to the problem, especially if I find that mine are doing the same thing. I believe it could be built on a small board and then wired in with the small board mounted piggyback on the bottom of the BPA-150. If I find any ringing I'll post scope pics and we can compare... Also, as far as my offsets go they were similar to yours and all my chips were also from the same batch... I was also able to get it nulled out by going back and forth but I also noticed some slight drift.

Mark

[shmulik]

to my ears, the ringing manifested itself most in double bass plunking in high quality jazz recordings. where I was accustomed to hear the attack related to plunking when the bass takes lead no I hear somewhat boomy sound. in some sense I would expect edge enhancement, like similar effect on images but maybe our auditory sensation differs in that from our visual system.

[Mark Allen]

YEP! Mine is also ringing. I think we have some work to do here. First one is 1khz 10 watts, second one 10khz 10 watts. Both into 8 ohm resistive load.

[peranders]

How does your insignal look like? It's very important to compare input and output. Are the probes adjusted properly?

[shmulik]

in my case the input signal looks like perfect square wave, taken from a proper signal generator. moreover, if the trimmers are shorted, meaning non-inverting input goes right to ground, the output of each LM has no ringing (done with no output power resistors). I felt somewhat alone in this until now. I am much more relieved now that other have witnessed the phenomenon and better skilled people will be looking for a solution.

[Conrad Hoffman]

Be careful with square wave tests. If you don't limit the rise and fall times of the input, all sorts of unexpected things can happen. Think about the slew rate of the input signal, then convert that to frequency. I think you'll find you're up in the RF region. See if limiting the slew rate a bit gives a huge improvement in the output. If that's the case it may not be important. If it really is affecting the audio, you might try fooling with the output network values. (I don't have this circuit, so can't be specific, but most amps like about .1uF in series with an inductance, and the value of that inductance needs to be above a minimum value- there was a huge thread on this some months (years?) back.)

[Mark Allen]

P.A., Yes input square wave is more or less perfect. Scope connection was direct hook up across the load resistor as I normally do. I do have an input capacitor of 5uf. So I didn't show low frequency response which does exhibit some roll off. I disagree this is a rise and fall time problem in and of itself. I'm thinking something else in the circuit is causing(inducing)this to happen This is the only amp I have here out of more than half a dozen that looks like this. The rest of them reproduce more or less perfect square waves.

I have the Lundahl input transformers to bridge these into mono and it gets rid of the input cap. This is the next step I'm trying... beyond that a servo on each 3886 may be the only way out of this mess.

[shmulik]

Originally Posted by Mark Allen

Scope connection was direct hook up across the load resistor as I normally do.

I have the Lundahl input transformers to bridge these into mono and it gets rid of the input cap. This is the next step I'm trying... beyond that a servo on each 3886 may be the only way out of this mess.

Mark, from the above I gather you measure the bridged configuration - is your scope input floating? otherwise you cannot measure across the load resistor of the bridged amp since you cannot connect the ground to any of the terminals.

[Mark Allen]

No, I'm not bridged as of yet but I'm wondering if bridging the boards might not cancel out the peoblem...