experiencing stability problems in my simple mic preamp

[Dominique]

Hello!

I'm experiencing stability problems with my simple microphone preamp. The preamp is meant to be very very simple and small. I tried several dual op-amps (unlike my circuit the schematic is single channel for easier review) and they all got the same stability problem but to different extents. It only appears on one channel (on the circuit's picture it's the right channel): the output lowers it's volume and distorts, while there are sounds of several frequencies appearing. Touching the circuit or loading the mic with a loud signal can make the operation normal (on some op-amps (OPA2228 is most stable here)), but the problem may come back after some seconds.

What could be the reason? I tried those op-amps: OPA2228, OPA2227, MC33078, LT1124, LT1126, Max412, RC4580
The gain can be set by selecting R2 via jumper for a gain of about 28, 80 and 180. (on the upside where there's a blue shunt). The circuit was tested in an unshielded case. Oh, and yes, most resistors are SMD and soldered below the board.

Any input will be much appreciated!
Cheers!
Dominique

[Dominique]

here's the schematic (just one audio channel!)

[amplifierguru]

Looks OK from the schematic Dominique. The CL gain as shown is 46dB (or x200).

You could try a small C say 100pF from +in of the chip to ground or up to 100pF across the feedback R (the 25K). This would provide some out of band rolloff.

Greg

[mikeks]

Closed-loop gain is pretty high...which suggests high loop gain has nothing to do with the problem....

Increase rail decoupling caps to 47uF shunted by 100nF...connected as close as supply pins as possible...

[jcx]

i would look at joints, endcaps, resistor bodies... under high magnification and verify correct connectiivy ( and the important lack of adjacent shorted paths) with ohmmeter

hand soldered smt parts can be damaged in several visually subtle ways, plated terminations detach, ceramic bodies crack, excess flux chars and forms shorts/leaky paths

high z inputs can drift and give intermitent operation

input rf filter cap per greg's suggestion may help stability but 100 nF is too large for the feedback cap - requires unity gain stability, extends response to frequencies where your proto's impedance may not be so well defined, parasitic C should be fine at the high gains used, at most add a few pF directly across the op amp pins

[amplifierguru]

My mistake, thanks jcx.

The 100nF suggestion should have been 100pF, in both cases. Hell they're not even close on the keyboard.

Cheers,
Greg

[mastertech]

pcb may be fine but you have long wires that could cause problem


"Increase rail decoupling caps to 47uF shunted by 100nF...connected as close as supply pins as possible..."

good point

cheers

[jcx]

my mistake, looks like your post was pF, not nF, but even 100 pF has the same problems as feedback C

[amplifierguru]

When I have designed mic preamps (and I used to design mixing desks back 30 years for PA, theatre, and broadcast) I have always used a feedback rolloff to bring the large gains back to 2-5 times the minimum stable closed loop gain. This ensures any stray out of band pickup is not passed on to subsequent stages/amplifiers. It worked well.

In this case 100pF //25K will rolloff from 63KHz and -1dB at half that. Ideally then a, say,470ohm in series with the feedback 100pF would limit minimum gain to +14db (x5) and contain the hf output stage loading to 600ohms. A larger than 470, say, 1K would limit to above 1K with >10K ext load.

Cheers,
Greg

[jcx]

at high freq the feedback C forms a divider with the - in terminal C, typically single digit pF, so a high freq gain of 5 will limit feedback C from a fraction of a pF to just a few pF

at Av=180 bandwidth is only 44 KHz with 8 MHz op amps like the 227 or 2134 - genuinely faster (possibly decompensated) op amps or composite amps are really needed at this gain level to permit active filter feedback rather than gbw to define band edge

a low 1st order roll off can have audible consequences (0.1 dB amplitude response variation can be detectable double blind)- the in band response can be flattened by going to a higher order butterworth filter alignment