So if I understand correctly, you want to be able to "read" a 10 mV variation on top of a 1.9V signal?
If that is the case then I would suggest two separate stages. The first will be a photodiode amplifier (page 9 is the most standard of circuits). This will help to get the current from your photodiode translated into voltage.
The second stage will be an instrumentatation amplifier, such as the INA family from Texas Instruments (the best but also can be expensive). This will help to remove your "common mode" signal, which in this case is the 1.9 V. You can also add gain in to the instrumentation amp or alternately add a simple op amp in a non-inverting configuration at the end to help gain your signal up to the necessary 5 V.
I'm not saying it'll be perfect, but I think that's a good start.
As a final note, I like David's idea above about the clamps, even though those can cause some measurement errors at the A/D converter. What is more important though is if you can swing it, try a better op amp than the 741. Those are common but the specs are terrible. The 3 or 4 mV of offset voltage at the input terminals could really mess up a small signal like you're trying to measure.
~Chris Gammell
This is more a set of comments than an answer per se, but too long to fit in a comment.
Signal from oscillators: 8Vpp 1~20KHz with an offset of ~10V with a new 9V battery.
So the issue is how to couple this to another stage which can amplify it, but at the same time set an appropriate input voltage DC offset suitable to the next stage.
The obvious solution would be to use any amplifier design with reasonably high input impedance, and AC couple to it via a capacitor, so for example a cap from OSC1 to R8.
"The main problem is, on Q1 base, where the signals meet, there's no signal." Whatever voltage signal is at Q1 base will be quite small because the impedance at Q1 base will be small compared to the 1 Meg input resistors. (Especially for frequencies above the knee of the R5-C7 highpass filter.)
So the voltages at Q1 base may well be only 1/100 or 1/1000 of the signals into R8 and R9. In any case what you are more concerned with is the AC currents through R8 and R9 (and thence into Q1-base).
And probably also of concern is the DC voltage at Q1-base -- is it in a sensible range to bias Q1 to operate in it's active range, say with 3 to 4 V DC at Q1 collector? Since you have a 100k collector resistor on Q1, that suggests you are expecting a DC Ic of around 0.03mA to 0.04mA, and thus a DC voltage of rather precisely 0.03V-0.04V across R5 (and not, for example, 0.08V), but there's nothing to set a suitable voltage on Q1-base to make that happen so far as I can see.
Finally, what is the role of C9, 10nF? In parallel with R11 that appears to create a filter that will attenuate output above 160Hz or so, working to considerably suppress the signals in your range of interest, 1 kHz-20kHz.
It's difficult to say anything about what you wrote after "My mission: be able to make its output signal usable" because you don't show a schematic of what your did and it's hard to guess.
FWIW, if you feed an AC audio signal via a capacitor into a voltage follower (which has a high impedance input, hence shouldn't disrupt the source of the signal), you are going to get an output voltage that follows the input voltage. That's assuming you've set the DC level at the follower input to something reasonable. There's not much that can go wrong there, so we need to see exactly what you did that might have cause this to fail.
Bottom line, it looks like your challenge here may be simply understanding how amplifiers work (either op amps or with discrete transistors) and how to satisfy their input requirements for signal voltage or current, impedance, and DC bias (aka offset). Perhaps reading up on that topic might allow you to navigate more satisfactorily?
Best Answer
Roughly, your quote means that a buffers main purpose is to provide current sink/source capacity for a given output, translating to, for example, less susceptibility to noise and more tolerance to various loads.
Regarging your worries on affecting the voltage and fluctuations:
simulate this circuit – Schematic created using CircuitLab
As per Atmega328 datasheet, page 244, the Analog Input will be happy with an input impedance of 10kOhm or less.
Which means that if you manage to keep R1 and R2 in the few kOhm range you will be fine.
Note that the buffer cannot isolate the battery nor the resistive divider from fluctuations caused by the rest of your circuit, it will only "reproduce them" with higher current drive capability. So, if the load of the divider(Vbat/(R1 + R2)) is low enough to not affect the total load (which probably is the case), the buffer will be of little use.
In summary, make R1+R2 too large and they'll be susceptible to noise and charge the sampling caps slower. Make them too small and they might start loading your system a bit. (Although this second scenario is much less likely)