AC-coupled low-frequency transimpedance amplifier (TIA)

Hello Maxime,

It certainly appears you have researched some different TIA options for your application. Although circuits 2 and 3 do provide variations on the basic TIA I am concerned that the added circuit complexity might be an issue when the minimum current is on the order of 30 nA. If the circuit must be able to detect small current changes at that current level, then the circuit must be reliable with and input current at a fraction of that; probably single digit nA, or less. Therefore, I tend to stay with the simplest circuit that provides the needed performance and will only migrate to something more complex if necessary. Thus, let's explore circuit 1 first.

I see a couple of issues with the circuit and it is interesting that you were able to get ac sweeps:

• The dc input current (500 uA) from the source flows in the direction that it would force the OPA37 output more negative. The OPA37 is being powered by a single, V+ supply and the output is not able to swing more negative. If the current source is reversed, the output would normally swing more positive.
• The OPA37 minimum common-mode voltage is approximately +4 V above the negative supply rail. It is is set to 0 V in the first circuit, and about +1.5 V in the other two circuits. Normally, the non-inverting input is biased to a common-mode voltage within the linear range.
• Any op amp's output does not swing all the way to zero when powered by a single supply. The actual level to which it swing may be anywhere from tens-of-millivolts to volts, depending on the design. The OPA37 output swing looks like it is about 3 V from the supply rail.
• The OPA37 is not unity gain stable and would likely become unstable in a TIA application.
• There is a very high voltage gain 909 V/V being required of the second stage. That will certainly reduce the attainable bandwidth.
• The OPA37 is a legacy op amp and most of the aforementioned points could be addressed with a modern rail-to-rail op amp. 

I suggest considering the following:

• A modern, low input bias current dual, tail-to-rail op amp such as the OPA2197. See https://www.ti.com/lit/ds/symlink/opa2197.pdf
• Either use a dual polarity supply for V+ and V-, or add a DC common-mode voltage at the non-inverting input so that the output can move more negative as the input current is increased. This will move the output of the lower supply (0 V) rail.
• If possible, increase the transimpedance gain of the TIA so that the voltage gain of the second stage can be reduced and bandwidth increased. Better bandwidth may be achievable by splitting the second stage into two amplifier stages each having lower voltage gain. 
• The 20 Meg feedback resistor in parallel with a 4 pF feedback in the voltage amplifier stage will produce a -3 dB cutoff frequency less and 2 kHz. That resistor and capacitance should be reduced if more bandwidth is needed.

This is a starting place. The circuit can be refined as you get closer to obtaining the output that is needed.

Regards, Thomas

Hello Maxime

Those are good options you show, if you say you can provide any supply, here is a +/-2.5V solution using the dual OPA2397. I was just looking for a lower noise dual that was new, the etrim is not needed. 

This is the input referred current noise here, so pretty low relatively to 30nA signal and the DC servo takes the output of the signal path to 1.5V - it needs to be at -1.5V for the servo amp to work, hence the +/-2.5V supplies, 

I put a place between the DC servo R's for a noise filter cap, here a 1uF is added there to ground, just a little lower here, 

I set the BW with the feedback cap on the 20Mohm Zt stage. You can increase that cap to bandlimit more, 

High gain Zt with servo loop.TSC

Dear Thomas,

Thank you for your broad analysis and structured answer.
Legacy circuitry tends, indeed, to be less «fashion» but more reliable than other proposed designs. Could you please have a look and comment this solution, having capacitors on OP-Amp inputs ?

In the previous (first) schematic, I omit to mention that the DC component was set to negative on Tina, hence the «working» behaviour.
Also, schematics were first issued with ideal OP-amps, then refined with a real device.
OPA37 selection was deliberate but not definitive, in an attempt to deplete a stockpile of them, currently under my hand. I will gladly swap them for newer OPAx197 with decent input bias currents.
Single supply up to 36V might be selected, to allow larger room for first-stage TIA. I will refine models and AC analysis to check both bandwidth, stability and noise propagation.
Even though the 2-stage approach will remain, I require a more efficient high-pass amplifier to allow 0.1Hz signals.

Thank you very much for your help, much appreciated.

Dear Michael,

Thank you for your research and the data provided. All-amp DC-servos were not (so far) considered as I lacked knowledge on them. I will definitely have a look at your simulation file tomorrow and let you know its results.

Thank you for sharing.

Just to go a little further, 

1. I did not actually turn 30nA into a 1V step, but here I did with 33Mohm Zt resistor. 

2. Adjusted the feedback cap down to hold that nominal BW of about 10kHz

3. Offset the supplies to give more output swing on the signal channel but still enough for the required negative bias out of the servo channel -2V ot 3V now, 

4. Increased the gain in the servo summing channel, that requires less negative voltage but incrementally adds more noise from the servo channel

5. got rid of the cap in the servo summing channel, not really helping, 

Here is the DC, the servo output sits at about -0.9V

AC response, to Vout for the current input, 

Transient analysis for a +/-30nA input (most are unipolar inputs though) at 3kHz - amplitude slightly rolled of but looks good, 

And then integrating the output noise, this RMS noise could be reduced with postfiltering if too high. Compare this to an output signal max of 1V or whatever your min detectable needs to be. 4034.High gain Zt with servo loop.TSC

The device I show may not be the best, but suitable for illustration purposes. 

Dear Michael,

Thank you for this extra effort on DC-servo cancellation circuit. Despite very good results in terms on bandwidth and noise generation, it looks that this circuit cannot deal with offset DC currents above 180nA, far from the 500uA component I have to deal with.
Efforts to bring down such high DC component while keeping the large, low-frequency targeted bandwidth seem to lead to impractical Capacitor choice on the servo and significantly lower input resistances.

Yes indeed, I had the feeling I was missing something when I was doing this and that DC correction range is the issue - now, to expand that range without killing the noise? 

Hello Maxime,

Okay, now I understand how you were able to obtain the results that you received for the TINA simulations. Thank you for clarifying how you had set up the negative supply such that the OPA37 circuits were functional.

The AC coupled TIA article that you referenced is interesting and I have no reason to question the author's design or results obtained. It is a clever design and solution for the specific application. I think about my only concern is the potential leakage currents of the large 150 nF capacitors and then to a lesser extent that of the 3.3 nF capacitors. Although the tolerance of the capacitors doesn't appear to be critical, their leakage current characteristics might be since you are interested in AC current changes in the tens of nanoamperes. Certainly using capacitors having high quality, low leakage dielectric should alleviate that concern.

Regards, Thomas

Precision Amplifiers Applications Engineering 

Dear experts,

I proposed the following solution,to be refined later this week. A classic medium-gain (56kV/A) TIA converts all photodiode current (DC+AC) up to near OPAx197 upper rail (+35V), with enough room for accidental over-illumination.
A Sallen-Key filter based on OPAx397 brings a x107 amplification, free of DC component, thanks to symmetric power. Note the 10pF capacitor on R4, improving SNR by 10dB at 1kHz.
A 1V positive DC offset is applied on the output so the useful AC signal stands between +1 and at most, +2V, as requested.

I now have to ensure stability of such circuit.