I've been trying to do a very low light level project myself the past 2 days with photodiodes and phototransistors. This is for people like myself and the original poster who are pushing light detection without a photomultiplier to the limit (below 0.1 mW/cm^2).
I looked at the first receiver module and its minimum irradiance detection was 0.2 mW/m^2 which is about 10,000 times more (less capable) than what discrete photodiodes and phototransistors can do (maybe they meant cm^2 instead of m^2?). Neither are good for really low light levels according to "Art of Electronics" (1 uA per uW of light page 996), totally not capable of getting near what the human eye can do due to leakage current and noise. He describes using photomultipliers which may be required if your light levels are too low. However, in shining light through my fingers in a well-lit room, I am able to see what my eye can't detect on an oscilliscope (with either PhotoDiode or PhotoTransistor).
Assuming his 1 uA per uW is correct, here is an example: a 5 mm photodiodes and phototransistors have an area of 20 micro m^2. So 1 uW/m^2 (1/1000th of noon sunlight) would generate 20 uA (according to Art of Electr) . [[ 1/1000th of noon sunlight is 1 W/m^2 which is about twice as strong as a 20W incadescent light at 1 meter (6W light output into 12 m^2 surface area of a surrounding sphere). ]]
However, my 880nm phototransistor datasheet indicates 600 uA at 1W/m^2 (0.1 mW/cm^2), wich is 30 times more. This assumes all the light is within the active range of the diode's junction.
Sharp has a much better application note, but it seems to be lacking in explaining which design is best for which situations. Figure 13 is most applicable to what original poster and I need, and figure 10B is very interesting but I don't know what they mean by "improves response". http://physlab.lums.edu.pk/images/1/10/Photodiode_circuit.pdf
When used with an op amp, a phototransistor may not capabale of getting as good of a gain as a photodiode for very low light levels because it to uses a "cheap" method of getting its initial gain (transistor instead of op amp). I suspect a photodiode with a JFET op amp (very low input current) would ultimately provide a higher gain with less noise. In any event, the photodiode or phototransistor with the largest optical receiving area might have the best ability to detect low light levels, but that might also increase noise and leakage by a proportional amount and they are usually the underlying problem. So there is a limit to this type of light detection and ideally efficient phototransistors and photodiodes may ultimately be equally good when used with an op amp, but theoretically I suspect photodiode is a little better. It will be the op amp design that matters and I think figure 13 with a JFET op amp and a well-chosen shunt capacitor across the feedback will be best for phototransistor.
For the dual supply op amp, you can use a "lowish" valued resistor pair (two 1k for 10V Vcc to get 5 mA bias) to split the voltage to create a false ground for the +Vin.
I found R=1M for the feedback resistor much better than R=4.7M. Forrest Mimms in his simple opto book used a 10 M with a parallel 0.002uF and a solar cell instead of a phototransistor or photodiode for "extrememly" low light levels" (maybe a solar cell would be better for your application) It seems all PN junctions seem to operate as a solar cell to some extent, as I've read of using clear-cased small signal diodes to detect light. I'm using a regular 830 nm LED as my "photodiode".
The lens angle of whichever 5mm optical diode you use makes a big difference. +/-10 degrees is roughly 4 times more sensitive than +/-20 degrees....if the light source is coming in from less than +/-10 degrees. If the light source is a big area that is +/-20 degrees in front, then it doesn't matter.
I tested the two circuits below. I could detect 0.3V, 5 ms pulses on the phototransistor's Vo which means 0.3 uA which means 0.05 uW/cm^2 if my reading of the datasheet is correct and if it remained linear (big ifs) all the way down to 0.3uA. Maybe it was 5 uW/cm^2. If 0.05 uW/cm^2 is correct, then the off-the-shelf 830 LED was reading down to 0.5 uW/cm^2. I was shining 10 mW 830 nm light through 1 cm of tissue (my finger). I know that if the light levels I was working with were red, it would have been barely visible. The link below shows using 500 M ohm feedback with a photodiode, indicating much lower light levels. Notice the orientation of their photodiode, which is the same as my LED (backwards from most internet links). I got better results this way.
http://www.optics.arizona.edu/palmer/OPTI400/SuppDocs/pd_char.pdf
What would you be trying to improve? Performance? Cost? Availability?
There are no guarantees on the noise, the offset drift with temperature, the gain tempco. There is a guarantee on the gain error.
The typical noise is not bad for a CMOS-input type. The typical gain tempco is excellent (4ppm/K), but no indication of the drift with time.
Offset voltage drift with temperature (4uV/K typical, no maximum) looks like the biggest error source to me- though if you have a load cell that's zero'd or tared at every measurement like most bathroom scales, it probably doesn't matter much. Same with the initial offset- it's possible to do 50 or 100x better in those departments with a separate op-amp. To do better in the gain tempco department you might have to go to discrete precision resistors and an analog switch, but that would push the price up considerably.
Sounds like you're really tight for space. Do you really need the PGA or could you just use more bits in the ADC? A (real) 20-bit ADC ("24-bit") would give you a resolution of a few uV in a few V, which is probably at or below the noise floor of that op-amp.
Best Answer
To meet the criteria for an oscillator, you need to have positive feedback at a gain of one or greater. Your op amp as configured will act as an integrator at higher frequencies, so you have a 90-degree phase shift (in addition to the 180 degrees of the negative feedback) and with the instrument amp's higher gain at the high end, you have met the criterion for oscillation. The original TI circuit relies on the roll-off of the response of the instrumentation amp (providing lower gain at higher frequency) to stay away from this condition. If you went back to the original instrumentation amp with its R2C2 roll-off network, your circuit would likely be stable but with the lower, 1 Khz frequency response.
You may also want to consider that, because I = c dv/dt, your 50 uA into a 10 uF load will only provide an output voltage slew rate of 5 volts per second. At 1 Mhz or even 100 Khz, any change in current will be primarily charging the large capacitor at this very slow rate, so the voltage response at 1 Mhz will be a few microvolts.