[optoelectronics] how to protect a photodetector to avoid its saturation

[EugenioN]

Hi all!

I'm designing a measuring device for detecting the presence of some polluting elements mixed in a stream of dust. I need to illuminate the dust under test ( DUT  ) with lots of energy in the near infrared region for a brief interval of time (~100us), then activate a photodetector to sense if the dust contains polluting elements that emit back some energy in the same near infrared region.
The DUT can travel fast (1-5m/s) on a conveyor belt and I assume I need to place the emitter and the detector close to each other to minimize the influence of DUT speed.

My problem is that the detector, a BPV11F phototransistor, is deeply saturated when the IR led is on, and it takes ages (~500u) to recover. During this time, the infinitesimally small light emitted back fades out, and I'm left with no signal at all.

The phototransistor is kept biased with emitter at -12V and collector fed to the inverting input of a fet opamp connected as a transimpedance amplifier with 47k feedback resistor. I used a TL072 because I have many of them in my drawers and there is plenty of slew rate.
The limit is the phototransistor itself, somehow blinded by the short emitted light. I've tried to help the phototransistor with resistors from base to emitter to drain the excess charges (220k, 1Meg, 2Meg), with slight improvement in speed but also a drop of gain, so no real help.
Also tried to short the base to emitter with an electronic SPST switch (DG449) when light is on, but this is not enough to shutter the phototransistor: when I open the short circuit from base to emitter the signal goes all the way up to full scale in a couple of microseconds with minor improvement in saturation recovery speed...

Decreasing the excitation led current will improve the phototransistor saturation recovery time, but also does not allow the glowing particles to be activated enough to be detected.

Mechanical shutters are a no go, way too slow and with a limited lifespan.
Any hints?

[RoGeorge]

Put the transmitter and the receiver on the same side, such that the receiver can only see the reflected light.

If that is not enough, maybe use a brushless motor to rotate a disc with many slots in it, as a mechanical shutter.  If the speed of a rotating disk is not enough, try a rotating mirror with many faces and placed at a distance from the receiver.

[S. Petrukhin]

Maybe to short-circuit the collector-emitter while the LED is on?

[MasterT]

I'd think about "power control loop" - when LED driver is adjusted based on receiver signal strength. Same concepr used on mobile network, cell phone saves battery and prevents jamming for neighbours

[S. Petrukhin]

We need a bright flash and then registration of the residual (fluorescent) glow, as far as I understand.
Are you trying to control the quality of grain?

Similar is used in our photo separators, it is expensive and complex equipment with a several Canon SLR cameras.

But I looked at the grain flow through these cameras (can monitor the picture in real time) and noticed that parasites and damage in the UV spectrum stand out very noticeably.
Maybe you should think about changing the pulse mode and controlling the live image?

[ejeffrey]

The standard approach when a photo transistor is too slow or has too little dynamic range is to use a photodiode.  They don't have internal gain so you will need an external low noise amplifier with a fair bit of gain.

About how much light are you trying to detect? 

Can you separate the signal from the excitation by wavelength?  An optical interference filter might be able to reject the excitation light from the detector l.

[Circuitauger]

Like ejeffrey stated photodiodes are what you want, and if you have the budget for it these guys are very convenient https://www.thorlabs.com/newgrouppage9.cfm?objectgroup_id=3257

[EugenioN]

[ Specified attachment is not available ]

The standard approach when a photo transistor is too slow or has too little dynamic range is to use a photodiode.  They don't have internal gain so you will need an external low noise amplifier with a fair bit of gain.

About how much light are you trying to detect? 

Can you separate the signal from the excitation by wavelength?  An optical interference filter might be able to reject the excitation light from the detector l.

I switched to a photodiode: it's a PIN-5D with spectral sensitivity quite well matched to the fluorescent response of the particles, biased at -12V to minimize capacitance and, hopefully maximize speed. Led diode should emit at least 1800mW/sr (it's a  SHF4545 binned at 450mW/sr@100mA minimum, driven with 350mA pulse). This is the basic detector circuit:



Rf=50kOhm, tl072 opamp with +-12V dual supply.

I'm trying to understand the response speed characteristics and how to improve it, thus I placed the emitting led and the photodiode facing to each other with less than 1mm of gap between lenses: led is on for 200us and diode takes ~350us to fully recover from saturation  after led goes off .
Yellow trace is opamp output, green is the voltage at inverting node (inverted for better display results).



Yellow trace is opamp output, green voltage is led current (through a sensing resistor); the drop is caused by a relatively large supply impedance and a non existent capacitor; I repurposed and old photodetector circuit to evaluate the whole thing.


If I partialize the quantity of light reaching the photodiode, its speed will increase dramatically and almost "copy" the excitation current, so it does not seems a source related problem. 

[tggzzz]

By "protection" it looks like you don't mean "avoiding destruction" but do mean "avoiding the consequences of receiver saturation".

TAoE x-chapters secion 4x.3.7 is about a transimpedance amplifier with a dynamic range of 10⁷, from 10^-10A to 10^-3A.

Very good book: "Building Electro-Optical Systems: Making it all Work". The author, Phil Hobbs, is referred to several times in TAoE and TAoE x-chapters. https://electrooptical.net/building-electrooptical-systems

[EugenioN]

About how much light are you trying to detect? 
Can you separate the signal from the excitation by wavelength?  An optical interference filter might be able to reject the excitation light from the detector l.

I can't measure the emitted light, but I suspect its at least -40dB with respect to the excitation.
Thank you for the signal separation: light response should be 50nm away from excitation, maybe an optical filter with a sharp enough notch can attenuate. I'll take a look at Edmund catalogue. 

[EugenioN]

By "protection" it looks like you don't mean "avoiding destruction" but do mean "avoiding the consequences of receiver saturation".

Yes!

TAoE x-chapters secion 4x.3.7 is about a transimpedance amplifier with a dynamic range of 10⁷, from 10^-10A to 10^-3A.

Wow, super comprehensive chapter! I will read it today and try to test some updates.
At a first glance my opamp selection was quite wrong: TL072 is too jellybean with limited slew rate, and I can also increase feedback resistor. Also very interesting the bootsrapping technique.

EDIT:
Running the math, TL072 has GBW of ~5MHz and Cin ~3pF. PIN-5D biased at -12V should be ~15pF, all this adds up to a ball park figure of 20pF with pcb traces and wires. So frc,in=150kHz. No compensation capacitor is present: I definitely can see ringing and underdamped behavior, but this should be in favor of speed. All this considered, time constant is ~1us, so a steady state condition should be reached in a ten-ish of microseconds.
From waveforms I see a tenfold worse condition: am I missing something? 

[ejeffrey]

If I remember my semiconductor physics (which is questionable) PIN diodes have a much worse recovery from saturation than PN didoes.  It's the same issue as electrical reverse recovery: the saturation pulse delivers a lot of charges to the I layer and it takes a long time to sweep them out.  A PN diode would work  better.

You are also operating at the edge of silicon diodes.  Your LED is centered at 950 and you said the signal is red shifted 50 nm?  950 nm is the peak reapomsicity of most silicon photodiodes and they start dropping quickly between 1000-1050.  You might also end up more temperature sensitive because of that.  If the shift is only 50 nm you should be fine but if the signal extends much past 1000 nm you might consider InGaAs photodiodes which are noisier and more expensive and more delicate but have better responsoivity in this region.

[ejeffrey]

Also response time of silicon photodiodes increases rapidly at long wavelength, see for example:

https://www.thorlabs.com/newgrouppage9.cfm?objectgroup_id=14218

For 1000 nm, the thorlabs det100a (also a PIN detector) has a 20 us rise time.  That is less than the 300 us recovery time you are seeing, but saturation and a 40 dB lower signal than background will make that worse.

[EugenioN]

Also response time of silicon photodiodes increases rapidly at long wavelength, see for example:

https://www.thorlabs.com/newgrouppage9.cfm?objectgroup_id=14218

For 1000 nm, the thorlabs det100a (also a PIN detector) has a 20 us rise time.  That is less than the 300 us recovery time you are seeing, but saturation and a 40 dB lower signal than background will make that worse.

Good to know, thank you!
Also rise time with PIN-5D and TL072 transimpedance is way less than 10us; it's the falling time that upsets me a lot 

[Kleinstein]

With a good PIN diode the response time can get lower quite a bit with bias at the diode.  Good sensitivity to the long wavelength end (e.g. > 1000 nm) requires thick and pure (long minority carrier lifetime) material. The advantage is more with a PIN diode as the I layer can result in a larger depletion region and less bulk area.
There can be after effects depending on the purity of the material. So it may be worth looking a more scientific grade photodiode as oppossed to a low cost more comsumer part.

[Circuitauger]

As a sanity check I'd make sure your emission source LED is optically behaving as you think it is, so while typically emission is thought to be lock step with the LED's current there can be slew.

So I'm wondering if what you're seeing on your TIA isn't just a squared up version of an otherwise shark fin emission profile off your LED.

There's a post about an LEDs opto-electrical profile here, the last comment of https://electronics.stackexchange.com/questions/86717/what-is-the-latency-of-an-led ... which I suppose would only account for a 1uS or so, but possible there's some other sorta capacitance allowing the LED to be on in-spite of the apparent lack of current. 
A known good pulsed laser source would also server as a sanity check against the photo detector.

[ConKbot]

About how much light are you trying to detect? 
Can you separate the signal from the excitation by wavelength?  An optical interference filter might be able to reject the excitation light from the detector l.

I can't measure the emitted light, but I suspect its at least -40dB with respect to the excitation.
Thank you for the signal separation: light response should be 50nm away from excitation, maybe an optical filter with a sharp enough notch can attenuate. I'll take a look at Edmund catalogue. 

Filtering would definitely be a good path to investigate. Thor Labs has high pass filters that transition from 90% transmission to >OD5 in 25ish nm, available in 50nm increments. And bandpass filters that are even narrower. ( https://www.thorlabs.com/newgrouppage9.cfm?objectgroup_id=16235 ) If one works well with your fluorescence emission, then that should improve your dynamic range requirements a lot.

[mikeselectricstuff]

Definitely use a photodiode for this application. Phototransistors are great for higher light levels and lower speeds, but not good for a wide range of signal levels at speed.

[LaserSteve]

Phil Hobbs takes an alternate route from the classical PD amplifier design math by Grame. He bootstraps the PD with a superbeta PNP for speed. He also shows some methods for optical  noise canceling with a second PD.. Those PDFs are free on the web site mentioned above. All of this is done with fiendishly clever additions of transistors ( usually diff pairs)to the opamp front end.  This, and neutral density filters or adjustable crossed polarizers on your PD may help with your problems. Although his sources are noisy lasers,  you may find his methods of use. Read his book, too.

Thorlabs sells some PDs with very small area for reduced capacitance.

Dichroic optical band pass filters in front of your PD may give you good selectivity for your signal. Also a Thorlabs or Newport item.

Steve

[EugenioN]

Friday update: after some fiddling around the main areas that should affect *turn off* speed (photodiode capacitance and operating point and feedback capacitance) I think I have a clue about the offending one.
It seems to me that the opamp is well overdriven because photodiode current when fully illuminated is in the tens of milliamps and with 47k-100k feedback resistor there is no hope to keep the opamp inputs at virtual short circuit.

I've added a diode-connected transistor to keep the inverting input a Vbe away from the non inverting input, with a way bigger improvement in turn off speed. Yellow trace is opamp output, blue trace is emitting led *voltage* for reference and green trace is opamp inverting input:


The turn on speed is three order of magnitude faster than turn off and only limited by opamp slew rate as expected, so the real question should have been: how to avoid the deep saturation of the opamp while keeping the high gain I need? Any chance with a better behaved opamp than the jellybean ones?

[Doctorandus_P]

Maybe put a Zener (TVS) diode over the feedback resistor. This raises the photo diode current once the Zener starts conducting and therefore clamps the opamp output before it goes into saturation. But it shifts the problem to the speed of the Zener diode. TVS diodes are designed to be quick, but some do have a weird turn-off behavior after triggering.

[Terry Bites]

Keep the sensor gain down. Keep the collector resistance low. Add gain later.

The Miller capacitance in the phototransistor can be an issue, try cascoding to neutralise it.

I agree, use a photodiode instead.
Many low cost PDs have rise and fall times in the 10nS area. eg an SFH203 has rise and fall times ~5nS
Use a servoed PD amplifier. It will reject the time integrated value of the illumnator IR.
That lets you apply high gain to fast events without getting blinded.

Aslo see https://tools.analog.com/en/photodiode/