Photodiode amp with switchable sensitivity

I would like some comments on a photodiode amp I am designing. I would like to make it with several different switchable sensitivities. The basic design is a LTC6241 op amp with the PIN photodiode connected in photovoltaic mode (transimpedance amplifier configuration). The photodiode goes to a virtual ground at the (-) input of the amp along with a feedback resistor. The size of this feedback resistor determines the sensitivity. What I would like to do is switch in several different feedback resistors using a JFET switch. As some of these resistors might be in the multi-megohms, I want to be very careful about leakage current. The BFT46 is a very small JFET with a cut-off drain current of less than 0.5nA and a gate leakage of less than 0.2nA. Since the Source of the FET goes to the (-) input of the op amp, which is held at virtual ground, the gate only needs to be driven to either 0v or -5v to switch on or off the feedback resistor connected to its drain.

I have never seen a photodiode transimpedance amp configured this way, so I suspect that something is wrong with the approach. Can anyone comment on the approach or tell me what is likely to cause me trouble if I do things this way? For example, should I take great pains to LP filter the gate drive to these FETs? I am not concerned about behavior during the transition between different gains. It is OK if switching gains causes a temporary spike due to capacitively coupled gate drive getting to the summing junction. But after a few msec I need things to operate as if only a single feedback resistor was connected.

The application is for timing reference for a prop balancer for small planes. I plan to shine an artificial light source on the front of the prop and pick up the shadow of the prop from behind the prop with this photodiode.

Rather than attempt something as tricky as you propose, why don't you better do a single-gain transimpedance circuit, followed by a second opamp with switcheable gain?

Because as you correctly mention, you are talking multi-megohm feedback resistors, and any component or board leakage will cause all sorts of drift.
I remember a circuit with such high valued feedback resistors, and changes in the air's relative humidity would make the readings to drift.

As high sensitivity may not be required I think, I would also go for the 2 stage approach. Of course a lower value feed back resistor will introduce more current noise, but you avoid many parasitic problems and you can use a real world feedback capacitor to get a descent response.

If you run into 10MOhm or more feedback resistors, and your circuit needs to be fast (fast rise/fall time, etc), your feedback capacitance (across the resistor) becomes so small that capacitance from resistor to ambient is likely more. In addition the higher your feedback resistor, the more GBW you need to get same speed (of course depending on diode and stray capacitances).

A series of smaller resistors to get many MOhms can behave as a ladder RC network resulting in stability problems. There are methods to compensate for this, but if you can live without them I would do it.

With a switched, sampling CTIA your gain is just the
pulse width. That's eminently variable, you could use
a uC pin and code to set the integration time (= gain,
for slow signals at least) arbitrarily. I think that the
timescale of propeller shadow movement is "slow" in
relation to microsecond-range integration times and
you could be looking for coarse time "slots" if the
difference in timescales is multiple decades.

AC-coupling the sensor would remove what often
limits simple photosensor designs - the baseline
ambient light. Making the source modulated would
let you go even further, use a frequency detector
behind a bandpass filter (perhaps a tuned load on
the amp, rejecting HF and LF right up front) to get
your timing signal. Again this would need detect
time to be << event time. But if you can discard all
the non-event-related input components of the
photocurrent, early, then you probably can just set
& forget the gain.

@dick_freebird: If he takes a directional photo diode (narrow half power beam width, SFH213?) into a tube with some baffles in it to avoid internal reflection, and he can keep it aimed at an IR or Red narrow beam LED, he should have sufficient diode current to get a descent S/N ratio. I assume this is stationary setup.

I think only direct sunlight that bounces of the propellor towards his detector may be a show stopper, but that can be avoided I think. He can also do some sanity checks in SW as well, as I assume the balancing process is slow.

Thanks for all the responses. Yes, I think sunlight reflecting from the back of the prop would be a problem. That will have to be addressed with user instructions not to position the plane so that is a problem. As for response time, I want to be able to time a 2400 RPM prop at 24" from the axis at a resolution of 0.5". That comes out to 83 usec. Not terribly fast, but not slow either. The reason I need so much resolution is that I intend to have two photo detectors offset slightly so that they can be digitized and fed to a quadrature encoder. This will allow me to track prop passes in both directions. The reason both directions are important is that I don't intend to mark a reference blade ahead of time (like competing systems do with a bit of reflective tape). Instead the system will pick a reference blade at random during the testing. Then when the engine is stopped, the system will tell the user where to add weight with respect to the blade that just went by the detector. In the process of stopping the engine it is possible for the prop to back up slightly at the very end. With quadrature detection I can track these backward transitions properly. By the way, this also means that I have to be DC-coupled. Sorry, Dick, A-C coupling would have been a nice idea, but it a luxury I cannot use. Yes, the balancing process is slow and iterative. At each iteration some weight is added or removed. But it is of great value to reduce the number of iterations by making more accurate measurements. Starting up an airplane engine is a big deal, safety-wise and with respect to wear and tear. You don't want to do it more than absolutely necessary.

However I am beginning to think that schmitt trigger was right at the beginning. Perhaps I can ensure enough illumination that a wide range of sensitivities is unnecessary. So maybe just a second stage switchable gain is best. I will have to do some tests to see.

A bit off topic: When you have sufficient processing power (and a fast ADC+mux), why not using an analog input signal (so not the Schmitttrigger)? You can do everything in SW afterwards and instruct the user to change something as you will "see" when the received signal is not as expected.

Something in between is that you determine the average DC level (slow ADC) and display this to the user to help alignment of receiver and transmitter.

Fully clear that you want to keep the number of start/stop cycles as low as possible.