Proton Precession Magnetometer - Part 4. Amplifier Fail!

The amplifier circuit is reasonably simple and came together quite well and I made a power supply in a plastic food storage container o hold 4 6V Lantern batteries which serve as a bi-polar power supply. It sat neatly in a grounded aluminium case and I was confident it was connected correctly. 

I did a quick calculation based on the likey local magnetic field strength as to what capacitors should be switched on via the dip switch:

I even found that my local supermarket was selling "highly distilled" water as a health drink so I got some and was ready to go. The first time I did a run I was very excited to see a strong signal.

Doing an FFT analysis (via scipy.signal.periodogram) showed a number of promising peaks in rougly the right place.

Perhaps a little on the high side but in the right area for sure. Then I stopped to think a bit. The signal was remarkably clean, even before filtering. There was no sign of the 50Hz noise others have seen. And when I looked more closely it did look a bit more like it was "ringing". Was I actually seeing what I thought I was? I did a number of runs with different conditions (longer delay, longer data collection) and was getting similar results. The I decided to try it without the water sample in place. The results, to say the least, were disappointing:

Similar results were obtained when the polarisation coil was unconnected and the data collection run with no polarisation. If the sensor coil is disconnected completely from the amplifier, only broad-spectrum noise is recorded.

So what's going on? I went back to re-read the appropriate chapter of "Signals from the Sub-Atomic World".  I also downloaded the Gerber files for their PCB and examined their photos of the amplifier. From this I realised that I had not full appreciated the importance of some of the precautions against oscillation. I thought that simply having a ground plane, an insulated case and a (more or less) linear signal path would do. But I see now that there's some important things I overlooked (admittedly it would have been useful to show the decoupling caps on the schematic, but they are only part of the problem). I will not redesign the PCB to have:

• Full surface mount. Their design has no through-hole components and even the few unavoidably through-hole components (an op-amp and the audio transformer) are mounted in an SMD style.
• Proper decoupling caps on the positive and negative power suppliesas close as possible to each amplifier IC. 
• Very short power traces on the PCB (their design has power cables attached to several places on the board right near where they are needed)
• A full ground plane, uninterrupted by traces, on the underside of the board and substantial ground planes on the top. Also thick ground and signal traces.
• A smaller, more compact board.
• No test points - these made the signal path non-linear and were not useful anyway as any testing can be done directly on the ICs.

Version 2.0 is being designed. I have enough parts to do a second board, although I will try and recover the 12-way DIP switch as that's hard to obtain. Hopefully a better design will address the oscillation issues I'm observing. One positive is that at least it appears the ADC sampling at ca. 16,0000 samples/s is sufficient to capture the signal of ca. 3000 Hz. In order to improve the accuracy of the sampling I set up an external DS3231 Real time clock and used the 32KHz signal with a simple interrupt on the arduino and a counter. This gives me an accurate timebase for the sampling.