Photodiode transimpedance amplifier for slowly changing signal

In an ordinary TIA circuit, the op amp forces its idea of zero volts AC across the photodiode. This is often a win, but not always.

The difficulty is that it also applies its input noise voltage across the PD, resulting in a noise current of

i_NC = e_NAmp * (2 pi f C).

If you use a chopper amplifier with a 1-Hz noise voltage of (say) 30 nV, this is

i_NC = 8e-17 f amps per root hertz. The 100 meg resistor has a noise current of

i_NR = sqrt(4 kT / R) = 13 fA per root hertz. The two become equal at

f_Cnoise = 13 fA/80 aA = 161 Hz.

At that frequency you've lost 3 dB of SNR. So you can't go very fast. (Your op amp is also liable to oscillate with that big an RC lag in the feedback--you'll want to put in something like 47 pF of feedback capacitance across that 100 M resistor.)

Doing stuff down in the low hertz is hard, though--I'd be thinking about a charge dispensing front end, if I were you, so I could chop the illumination at a minimum of 1 kHz and use a lock-in amplifier. Lock-ins are good medicine for this sort of job.

The other thing is your absolute current. If you're planning to use a single range setting, I'm assuming that your maximum current is less than about 10V/100M = 100 nA, so your minimum current is down around 1 pA. That's very hard to measure at such a low frequency, and your SNR is going to be the pits even if you manage it.

I'd strongly suggest cranking up your light source if you can, or using a photomultiplier if you can't.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs
Principal
ElectroOptical Innovations
55 Orchard Rd
Briarcliff Manor NY 10510
845-480-2058

email: hobbs (atsign) electrooptical (period) net
http://electrooptical.net

Over fifty years? Wow! Best to put a note into your will, to make sure one of your grandkids reads the results. Just kidding ... :-)

The best way would be a synchronous detector. IOW, it only integrates while the light source is pulsed on and otherwise remains completely silent. This avoids integrating up too much in noise or dark-currents. The overall detector bandwidth of the post-filtering which you probably do in the digital domain (DSP, uC or PC) should be very low. 2Hz range I guess because you said that's you maximum.

I see it very often that people provide a super-high gain in the TIA. I suggest reducing that to something more reasonable and less leakage-prone. Say 10M or less. Then follow that TIA with another regular low-drift amplifier.

Provide a few pF from OUT to IN- so it won't oscillate on you. Short path on that cap, meaning it should be closer to the opamp than the resistor.

--
Regards, Joerg

http://www.analogconsultants.com/

"gmail" domain blocked because of excessive spam.
Use another domain or send PM.

One common way to do this is to chop the light source (ie, on/off with a square wave at some frequency) and use a synchronous detector after the TIA and some gain. Lowpass filter or digitally average after the detector, or better yet do the synchronous detection in software.

The synchronous detection thing lets your TIA+amp avoid its 1/F noise region, and lets you have an arbitrarily low net measurement bandwidth.

If your absorption is changing as slowly as you describe, it's optimum to have system bandwidth just enough to track the changes, 2 Hz maybe. Chopping at 1 KHz might be good.

A TIA that uses a feedback capacitor, rather than a resistor, might be interesting. Caps don't have Johnson noise.

John

nd

f t

of

e?

ral

pF

Actually, even a capacitor does have some Johnson noise, if only from its internal series resistance, but John Larkin is perfectly correct in pointing out that this is going to be lot lower than the Johnson noise in a resistor doing the same job.

And you can make it easier to cope with your five decades of range by adjusting the time over which you integrate your signal current. Charge injection from the switch that you use to discharge the capacitor complicates life a bit, particularly for very small currents, and you may end up doing matched integrations - with the light source turned on and off respectively - to reduce this particular error.

And you may need to read up on "charge soak". Capacitors with Teflon dielectric suffer less from charge soak than most, and polypropylene and polystyrene are almost as good.

-- Bill Sloman, Nijmegen

d

By five decades do you mean that you want to see changes in the absorption of one part in 100,000? Say 10 uV's out of a 1 volt signal.

f ?

al

Reducing the bandwidth is easy, add some C in parallel with the feedback R. But that may not be what you want. As the 'experts' have pointed out you should think about lock-in detection, perhaps raising the bandwidth... More light if you can get it. (a kHz with 100 Meg ohm is not so easy, a few pF are going to ruin your day.)

George H.

F

With a 100 megohm Rf, you can achieve low bandwidth simply by achieving stability in your op-amp circuit.

My knee-jerk response to this would be to use a TIA preamp with a lower Rf, followed by at least one regular op-amp stage doing ordinary voltage amplification. I'd do a thorough noise analysis to determine just how to balance my gains for the lowest noise at the ADC. I may or may not AC couple the analog electronics to eliminate problems with photodiode dark currents or bias at the preamp throwing the whole thing out of range at the ADC.

I _would_ take advantage of the pulsed light source to chopper-stabilize the whole thing in software. Take ADC readings with the light sources off, take readings with the light source on, then subtract the two. This gets rid of bias (until the ADC starts saturating), and will go a long way to pounding down 1/f noise, to boot.

Note that nearly _everyone_ so far has suggested this chopper- stabilization, just using different language. This is what the 'lock-in amplifier' that has been suggested will do (only you're taking ADC bias into account this way, instead of worrying about bias in your analog mixers and your ADC), and is exactly, or very nearly, what John Larkin has suggested.

You probably want to use one of those nifty 24-bit sigma-delta ADCs to get the precision you need, but if you do you need to be careful about lining up your measurement with your illumination, so that the effects of the illumination change has time to settle in the amplifier chain and the ADC. If, for some reason, you really, really, had to use a 16-bit ADC, then oversample like mad, average, and hope that the noise averaging makes up for all the ADC deficiencies*.

• ADCs these days always come with "marketing bits", or at least more bits than they seem to be able to sign up to providing solid data for. The each bit of extra resolution does help, but not by a whole bit's worth of accuracy for the bottom-most ones.

--
http://www.wescottdesign.com

Hello all,

A more sophisticated way is called lock-in. Modulating the light source with a frequency above the 1/f and demodulate it in software with a bandpass with high Q. So your noiseband is limmeted in both directions.

Marte

Thanks for your input Phil. It feels kind of a paradox that the low hertz region makes the measurements hard. One would easily think that the higher the frequency the greater the challenge. My first experiments indicate an initial photo current of around 12nA. The desired dynamic range is somewhere between 3 and 4 decades (rather than five as I first stated) which still gives a minimum current in the pA-region. I can increase the LED (=light source) current with a factor of around 10 to 20 but that doesn?t change the situation dramatically, right. It shortens the already short lifetime of my LED however. Your suggestion of a lock-in amplifier sounds really exciting but it?s a completely new and unknown technology for me. Feels like I have a lot of homework to do?

My original design idea was to use a normal TIA (low noise and bias current amp) with perhaps a diode bootstrap (unsure if it would give me any benefit). Then sample the signal with a high resolution delta sigma ADC using a large OSR giving good noise reduction. I?ll guess that I need to do the feedback with a resistor tee in order to reach 100meg (or above). Are there any drawbacks with this?

/Quist

--------------------------------------- Posted through

Lock-in is just another word for synchronous detection + lowpass filtering.

John

bandwidth of

But it sounds more fancy :-)

Got any snow yet?

--
Regards, Joerg

http://www.analogconsultants.com/

"gmail" domain blocked because of excessive spam.
Use another domain or send PM.

The easiest range is between 1 and 100 kHz. Below that, you get into power line pickup, 1/f noise, thermal drift, and so on. Above that, especially with low light levels, you have to work harder to keep your signal-to-noise ratio.

If you pulse the LEDs, you can get higher instantaneous photocurrent. That's a win, because the detected electrical power goes as the square of the optical power.

Do read up on lock-ins. You don't necessarily have to buy one, but for a relative beginner it's a lot easier if you do. Do you have an equipment budget for this, or a reasonably well equipped laboratory to use?

NoNoNoNoNo. You can use a tee network to get a bit of voltage gain, but whatever you do, don't try using it to reduce your feedback resistor, or your SNR will go down the toilet. Keep the voltage drop across the feedback resistor as high as you can while getting enough bandwidth.

100M resistors aren't that hard to get. Your SNR will be limited by the photocurrent shot noise as long as you keep the voltage across the feedback resistor at least 50 mV. (You can work that out from the shot noise and Johnson noise current formulas--set the two contributions equal, and solve for (I_photo*R_F).

If your library has a copy of Mark Johnson's _Photodetection_and_Measurement_, reading it will help a lot with this. Johnson is an elementary-level book on what to do if you have a photodiode in one hand and a BNC connector in the other and aren't sure what to put in between.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs
Principal
ElectroOptical Innovations
55 Orchard Rd
Briarcliff Manor NY 10510
845-480-2058

email: hobbs (atsign) electrooptical (period) net
http://electrooptical.net

d e

to

is

n

Check out the Linear Technology "no latency" 24-bit part

As Tim Westcott points out it's actually a 20-bit part with 4-bits of what is essentially random noise tacked on to fill up the low order byte, but that pretty much describes every 24-bit A/D converter. The "no latency" feature does look as if it might be useful, not that I've ever used one.

-- Bill Sloman, Nijmegen

Even before you get to the data sheet, you find that "no latency" means "one sample interval at 15Hz". It sounds like a regular 60Hz part that's been crippled to only allow one of four samples. (All the 24-bit S-D ADCs that I've looked at use CIC filters, which have finite settling time -- what do you want to bet that the settling time in question is three or four samples at 60Hz?)

Twenty honest bits should give you six honest decimal digits -- 1ppm is a pretty rough spec to follow for the electronics leading up to the ADC, so it's probably not going to be the limiting factor in your system accuracy.

--

Tim Wescott
Wescott Design Services
http://www.wescottdesign.com

Do you need to implement control loops in software?
"Applied Control Theory for Embedded Systems" was written for you.
See details at http://www.wescottdesign.com/actfes/actfes.html

It just a couple of minutes ago started raining hard. Some people are predicting snow on Market Street tomorrow. I'm keeping a camera handy.

Driving back from Truckee yesterday, there was snow as low as Colfax.

Sugar Bowl got 10 feet of snow out of the last storm. Then the sun came out. It was glorious.

John

d s

nce

y

al

. g r

ge

ow to

AC

rk

t

ze

es

=A0This

k-in

s

of

the

DC,

t

The data sheet doesn't read as if that was the way they got their "no latency"

and figure 37 at the end of the data sheet shows a "Correlated Double Sampling" circuit that does seem to exploit the "no latency" feature in the way that I had in mind.

The application note

isn't all that helpful, but the original press release for the range, back in May 2001

does seem to claim that the whole LTC2400 ragne can only give you one valid output per four 60Hz cycles when set up to maximise 60Hz rejection, so it isn't the LTC2415 which is crippled, but all the LTC2400-series sigma-delta A/D converters.

Maybe not in yours. National standards labs around the world do seem to want to do better, and - for some applications at least - the technology is available if you are prepared to make the effort.

-- Bill Sloman, Nijmegem

y of

e it

th of

oice?

everal

just saw an interesting example of synchronous detection :)

-Lasse

cy of

se it

bandwidth of

hoice?

several

h e .

It's easier for me to spell.

That snycro-- word is hard for me.

George H.

Our snow almost all melted a few weeks ago. :^( Spring is pretty much mud around here.

Geo

Hi Bill,

As long as you don't use any digital filtering this may be right. But ig you modulate the signal an use digital bandpass after sampling you will be thankful about these "4 bit noise generators" Understanding the lock-in is a wonderfull thing. You don't have to buy these lock-in amps if you know what to do.

Marte

C

V,

noise

ual at

t.

e

ut

a 1

NR

g

ents

the

The

or of

s a

f

e?

rent

r to

t e t

Hmmh, is this book any good: Building Electro-Optical Systems

I never had the need to design any of this low level optical interface circuitry, but on a lark I picked up "Photodiode Amplifiers: OP AMP Solutions" at a local used book store. Damn interesting reading.