1G Ohm TIA with large photodiode signal to noise

[MaciejW]

Hi,
Could someone please let me know how to limit noise in 1G Ohm Transimpedance amplifier with few nF photodiode capacitance? I am trying to find working bootstrap circuit.
My goal is to measure night sky brightness with bandwidth of 200Hz. My reasoning is that the bigger the photodiode the bigger the current, but photodiode capacitance and noise as well.
So far with 5 pieces of bpw21r (each has 7mm^2 and 1.2nF) I get 500uV of RMS noise. I would like to get down to 100uV.
I tried double JFET bootstrap from this website, with no improvent https://www.radiolocman.com/shem/schematics.html?di=618081

[Marco]

I don't really understand the two FET circuit.

I understand it's for photovoltaic mode ... but why not just capacitively couple the JFET source follower to the cathode then and avoid the ugliness trying to match current sources?

PS. I assume you are low passing before measuring the noise right?

[moffy]

See: https://www.eevblog.com/forum/projects/help-me-understand-photodiode-amplifier-jfet-bootstrap/msg5070655/#msg5070655

[magic]

A mere 1pF of parasitic capacitance in parallel with 1GΩ already adds a 1st order lowpass at 160Hz, so BW of the circuit is unlikely to be much higher than intended. But, if necessary, the above shows how BW can be readily reduced.

It's important that bootstrapping eliminates most of opamp's voltage noise only to replace it with voltage noise of the JFET source follower, which will now appear across the diode. This helps because good discretes are less noisy than chips, and you can run them at higher bias than a power-conscious IC design does. Make sure to use a low noise JFET at fairly high drain current. If you have two, maybe try connecting them in parallel instead (for lower noise) and adjust bias current (and resulting Vgs) simply by tuning the pulldown resistor.

You could replace the photodiode with an "equivalent" ordinary 1.2nF capacitor to see how much noise comes from the TIA and how much is legitimate noise from the diode.

[iMo]

.. and put your entire stuff into a metal cookie box, inclusive the batteries.. Output via a BNC connector and a shielded coax..
The noise you see may come from outer sources easily, you work with sensitive circuits..

[MaciejW]

I don't really understand the two FET circuit.

I understand it's for photovoltaic mode ... but why not just capacitively couple the JFET source follower to the cathode then and avoid the ugliness trying to match current sources?

PS. I assume you are low passing before measuring the noise right?

yes, 5th order 200Hz low pass filter

See: https://www.eevblog.com/forum/projects/help-me-understand-photodiode-amplifier-jfet-bootstrap/msg5070655/#msg5070655

Thank you, good hint, I found this topic, and I modeled bootstrap in LTspice and unfortunately does not improve things as it is

.. and put your entire stuff into a metal cookie box, inclusive the batteries.. Output via a BNC connector and a shielded coax..
The noise you see may come from outer sources easily, you work with sensitive circuits..

You are right, I did exactly as you advise. I even put metal mesh in front of the photodiode, as it was picking too much 50Hz interference

A mere 1pF of parasitic capacitance in parallel with 1GΩ already adds a 1st order lowpass at 160Hz, so BW of the circuit is unlikely to be much higher than intended. But, if necessary, the above shows how BW can be readily reduced.

It's important that bootstrapping eliminates most of opamp's voltage noise only to replace it with voltage noise of the JFET source follower, which will now appear across the diode. This helps because good discretes are less noisy than chips, and you can run them at higher bias than a power-conscious IC design does. Make sure to use a low noise JFET at fairly high drain current. If you have two, maybe try connecting them in parallel instead (for lower noise) and adjust bias current (and resulting Vgs) simply by tuning the pulldown resistor.

You could replace the photodiode with an "equivalent" ordinary 1.2nF capacitor to see how much noise comes from the TIA and how much is legitimate noise from the diode.

Thank you for so much input! I measured BW and it is around 200Hz. OP amp input, photodiode and feedback resistor are soldered in air.
I replaced photodiode with capacitor and for BPW21R photodiode seems most of the noise comes from TIA, as it did not change.

Interestingly I tried same thing with TEMD5000 photodiode (same surface area, but only 70pF capacitance) and here some of the noise comes from photodiode.

I played with bootstrap circuit like on the picture below, and increasing voltage helps and eliminating one of the resistors helps as well. I am sorry, which resistor is pull down resistor here?

[magic]

I played with bootstrap circuit like on the picture below, and increasing voltage helps and eliminating one of the resistors helps as well. I am sorry, which resistor is pull down resistor here?

I meant R_BIAS on the simpler schematic with only one JFET from the other thread. This resistor sets bias current of the source follower JFET, for low noise the current can't be too low. You could also try temporarily disabling the bootstrap - if this results in lower noise it means that the source follower is noisier than the opamp and needs to be improved.

[moffy]

If you have a 1G ohm feedback resistor then your noise is dominated by the input bias current noise which for the AD8610 is 5fA/sqrt(Hz) @ 1kHz which gives 5uV/sqrt(Hz) for the given feedback resistor and that is not including the other noise sources such as the FETs etc. Try the opamp with just the 1G resistor connected as feedback, with nothing else connected to the input and measure your output noise, use batteries for your supply. This will give you a base reading for your noise.

[magic]

That still doesn't explain 500μV RMS, unless it gets much worse at low frequencies. Which it could, so is there a noise vs frequency plot in the datasheet?

BTW, Johnson noise is 4μV/rtHz.

[moffy]

That still doesn't explain 500μV RMS, unless it gets much worse at low frequencies. Which it could, so is there a noise vs frequency plot in the datasheet?

BTW, Johnson noise is 4μV/rtHz.

Good point, according to: https://www.analog.com/en/resources/analog-dialogue/raqs/raq-issue-174.html
the input current noise should be fairly flat up to 1kHz, so expected noise should be around sqrt(5*5 + 4*4)*sqrt(200) uV = 91uV RMS for a 200Hz BW, anything above that for just the opamp with the 1G feedback resistor is contamination of some sort, hence the need for a baseline measurement.

[MaciejW]

If you have a 1G ohm feedback resistor then your noise is dominated by the input bias current noise which for the AD8610 is 5fA/sqrt(Hz) @ 1kHz which gives 5uV/sqrt(Hz) for the given feedback resistor and that is not including the other noise sources such as the FETs etc. Try the opamp with just the 1G resistor connected as feedback, with nothing else connected to the input and measure your output noise, use batteries for your supply. This will give you a base reading for your noise.

I use AD8655, and bateries and I get 110uV RMS noise without photodiode https://www.analog.com/media/en/technical-documentation/data-sheets/AD8655_8656.pdf

That still doesn't explain 500μV RMS, unless it gets much worse at low frequencies. Which it could, so is there a noise vs frequency plot in the datasheet?

BTW, Johnson noise is 4μV/rtHz.

From what I understand the problem is that op amp noise amplification =  (Feedback impedance) / (photodiode impedance). And the amplification increases with frequency, as photodiode impedance goes quickly down with frequency, as it has big capacitance.

@magic low noise JFET was crucial advice, thank you. Simulation shows I got down to 76uV RMS, with 7 x bpw21r photodiodes. This should be good enough for me.
I wonder If I can replace IFN147 with JFETs used in audio, as they are cheaper: BF862 or JFE150

I do not understand fuction of second JFET from circuit below, but it makes noise lower on simulation (I get 86uV RMS with only JFET1)



And here is classic vesion:

[magic]

Noise of JFET1 varies with its drain current, which is determined by the source load (JFET2 and R15). I doubt that adding JFET2 decreased simulated noise - more likely, you inadvertently increased JFET1 bias current at the same time.

The point of JFET2 is to act as a current source matched to JFET1. If you set R15=0 then JFET1 Vgs=0 and it conducts full Idss, which is also conducted by JFET1 causing JFET1 Vgs to become zero as well. This ensures true zero bias across the diode. Similar thing happens if you add equal resistors in series with both JFETs. This trick requires them to be matched to each other and at the same temperature - ideally a monolithic pair, but if experimenting with discretes one could start with ensuring that they are mounted closely together and have same power dissipation (same current and same drain voltage).

One could also use a simple resistor as a current sink for JFET1, but then it's harder to ensure Vgs=0, particularly if temperature of the circuit is to vary significantly during operation.

[MaciejW]

@magic low noise JFET was crucial advice, thank you. Simulation shows I got down to 76uV RMS, with 7 x bpw21r photodiodes. This should be good enough for me.
I wonder If I can replace IFN147 with JFETs used in audio, as they are cheaper: BF862 or JFE150

JFE150 helped, but not as much as simulation showed, probably I put something wrong in it.

I experimented with 3 photodiodes: bpw21r, centronic osd100 and temd5000. Bootstraping decreased noise only for BPW21R. (is high dark resistance necessary?)
Anyway circuits with centronic osd100 and temd5000 had higher SNR than BPW21R.
The goal was to measure how much city night sky does blink - sodium street lamp blink with double main frequency, so lighted poluted sky should as well.
I think I managed to measure it, in my spot it is only 0.25% of total night sky brightness. Simple photodiode with TIA and 200Hz low pass filter did the job.
Probably I will leave this project at this stage as I would like to start project with SPAD.