Post History

<blockquote>Is it reasonable to use function generator as differential input the way I tried?</blockquote>

In <i>theory</i>, it should have worked, but it's asking for lots of common mode noise.  Even with the output apparently not connected to ground at DC, there can be considerable capacitive coupling to ground, power, and other nasty places.  Unless the function generator is specified to have a truly differential output, I wouldn't try to use it that way.

What you can do is put a small audio transformer between the function generator output and the input of your circuit.

<blockquote>If 2.5V is the only reference for the signal, is that sufficient or does there need to be a direct GND reference somewhere?</blockquote>

There <i>is</i> a GND reference, through the 2.5 V source.  The input as you've shown it is not floating.  Other than general low noise cleanliness (see below), your input bias seems reasonable.

<blockquote>What may be the reasons you can see or think of that the circuit does not behave as expected?</blockquote>

The clipping of positive signals seems strange.  Your suspicion about a TVS being shorted or installed backwards could be right.

I would go back to testing the circuit with the function generator so that you have a well controlled and known input.  Use an audio transformer to guarantee a truly floating differential signal.  Remove the clipping diodes to simplify the circuit and get that working with the function generator.  Then you can put the clipping circuit back and test with a real geophone.

Why TVSs for clipping?  They have rather soft and inaccurate knees.  Why not the more usual clipping diodes to power and ground on each input?  That will guarantee only one diode drop voltage excursion beyond the rails.  With a 5 V TVS, it could be several volts beyond the rails.

I would also address noise in various places more directly.  I don't know what the valid frequency range of geophone signals are, so I'll arbitrarily pick 1 Hz to 1 kHz as example.

Starting with the lines from the geophone, you want to squash high frequency pickup from radio stations and other sources.  That's easy, since they are much higher than the highest signal frequency.

For example, adding 1 k&Omega; followed by 16 nF to ground on each line starts filtering out signals past 10 kHz.  The filters not matching exactly on each side will cause some imbalance, but mostly at frequencies you can filter out later from the single-ended signal.

I would also add a capacitor to ground at the midpoint of the 9090 &Omega; between the two inputs.  If you use 1 k&Omega; on each line as described above, that leaves 7090 &Omega; total between the lines.  That would be split into two 3.5 k&Omega; resistors, with a cap to ground at the center point.  Something like a 20 &micro;F electrolytic cap to ground should keep the center point nicely clean at the frequencies of interest.  Don't use a ceramic, due to possible microphonics.

The pure DC resistance driving the whole input circuit common mode voltage doesn't need to be very low resistance.  It really only needs to be low relative to the diff amp inputs.  10 k&Omega; between the center point and the 2.5 V reference should be fine.  The cap at the center point lowers the impedance at AC, and 10 k&Omega; at DC is plenty low enough.  One downside to this is it will take a few 100 ms for the input voltages to stabilize after startup.  I'll assume that doesn't matter for now.