High speed Transimpedance amplifiers

[max_torque]

I have a photodiode circuit that receives ~100ns long pulses.  I need to amplify the receiving photodiode as much as is practically possible to be able to maintain the link with poor optical coupling.  The edges of my signal have rise/fall times in the order of 10ns.  Has anyone direct experience of a suitable op amp to act as the amplifier, and also, i guess that amp, with it's high speed nature, will be tricky to prototype with (ie breadboard etc)?

(assume i'm going to be looking for an IC with it's -3db point up in the low GHz?)

I have a circuit working with an AD8031 opamp (80MHz, 30v/us) but have obviously had to increase the pulse width to around 1us to get sufficient gain.

[Zero999]

The NE592?

[max_torque]

NE592 looks like an useful little amp, but it's too slow at just 120Mhz b/w i suspect.  It also has quite a low input impedance that is going to make hooking it to a photodiode more difficult

[Alex Nikitin]

http://www.linear.com/product/LTC6268 and it's family.

Cheers

Alex

[Zero999]

NE592 looks like an useful little amp, but it's too slow at just 120Mhz b/w i suspect.  It also has quite a low input impedance that is going to make hooking it to a photodiode more difficult

I can understand your issue with the bandwidth being too low but the low impedance shouldn't be a problem. In a transimpedance amplifier, the input impedance to very close to zero, due to negative feedback.

[max_torque]

LTC6268 looks promising, and also LT1226 as well

(wow, there are LOTS and LOTS of op amps available these days!    )

[bson]

OPA836?  It's rail-to-rail output with fast recovery (60ns) on the input.  http://www.ti.com/product/opa836/datasheet

In general, if you need more gain you can always cascade a dual unit...  But 10MHz 20dB isn't exactly a tall order, especially for a dual op amp.  This will bring 30mV to 3V.  For 100-200MHz (more what I think of as high speed) I'd go for a 1-2GHz CFB amp, like THS3202, but these aren't R2R and tend to go bonkers when they get too close to rail so need clamped inputs.  10MHz should pose little problems and you can easily clamp the input if you find a nice non-R2R op amp you like to use.

[tggzzz]

There are some subtleties in the design and implementation of wideband and/or low noise TIAs that aren't immediately apparent. TIAs for optoelectronics are commercially important and have been well-studied and documented, so it would be pointless to quarter-duplicate that here.

There is a good introduction in The Art of Electronics 3 pp537-555. Do yourself a favour and get that book. It isn't cheap, but it will save you a lot of time on this and other topics. For example, it lists a range of opamps (and transistors etc) that will cut down the search space and time [1].

If you want more depth after that, then optoelectronics books and application notes (e.g. lasers, comms) would be my next port of call.

[1] “Wow. Chapter 5 details every circuit artifact that I’ve encountered in the past 30 years in a thorough, pragmatic, and straightforward way. My only ‘twinge’ is that it discloses and explains (in glorious graphical detail and with real part numbers) many topics that I thought were my personal trade secrets. I love the plots. I know that it must take an enormous effort to collate all of the device characteristics. It’s worth the effort. The way the data is presented allows the reader to get terrific perspective on a lot of landscape in a single view. Nice work.” — John Willison, founder, Stanford Research Systems

[nfmax]

A good reference for this application is: Jerald Graeme, Photodiode Amplifiers - Op Amp Solutions, McGraw Hill, Boston 1996, ISBN 0-07-024247-X.

For the ultimate noise performance, 100ns pulses is somewhere near the limit of Si technology, as the input capacitance/transconductance ratio is more or less fixed and becomes the limiting factor. You might also look at using a GaAs HEMT input device followed by a Si amplifier, though this is a lot trickier. It depends how critical noise is to you. You should be using a FET input device, choose one with low input capacitance. Output swing isn't critical as bandwidth requirements will limit the feedback resistor value - its easier to add a second stage of voltage amplification.

[bobaruni]

As others have said, there is a fine art to TIA design but it doesn't have to be that hard.
The main limiting factors will be the photo diode capacitance, op amp capacitance and how much gain you want as the gain resistor determines the pole frequency in combination with the capacitance.
Then there's noise and stability to consider but both can be calculated here: http://www.jensign.com/transimpedanceamp/

About op amps, OPA355 family work well with a supply of 5V or less but are a bit noisy and probably will be only marginally better than what you are using now.
With split power supplies, it's hard to beat OPA659 in a TIA unless you do some trickery using a cascode at the diode junction to eliminate the effect of the capacitance.
For high speed, always apply as much negative bias to the diode to eliminate the capacitance.
And you must use a low input bias op amp FET or CMOS, not bipolar input.

What diode are you using and in what configuration?
What supply do you have to play with?
Is noise important?
Does it need to be linear or is this for a data link or similar?
How much TIA gain do you need (what is  the value of the feedback resistor)?

EDIT: Alex Nikitin previously suggested an LTC6268 and this looks like a really low capacitance op amp (the lowest so far), worth considering.
And OPA657 is 1.6Ghz bandwidth but may not work as well due to higher capacitance and it's also not unity gain stable which might prevent it's use depending on your configuration.
 

[grouchobyte]

I have a photodiode circuit that receives ~100ns long pulses.  I need to amplify the receiving photodiode as much as is practically possible to be able to maintain the link with poor optical coupling.  The edges of my signal have rise/fall times in the order of 10ns.  Has anyone direct experience of a suitable op amp to act as the amplifier, and also, i guess that amp, with it's high speed nature, will be tricky to prototype with (ie breadboard etc)?

(assume i'm going to be looking for an IC with it's -3db point up in the low GHz?)

I have a circuit working with an AD8031 opamp (80MHz, 30v/us) but have obviously had to increase the pulse width to around 1us to get sufficient gain.

First things first....

1. Optical input signal range?...ie: optical power and PD responsivity (A/W)
2. Photodiode selected Si, Ge, InGaAs....? Active area and Cd
3. Wavelength?
4. Photodiode type? PIN, APD, etc
5. Do you need gain in the photdiode to achieve desired SNR. The TIA wont provide a free lunch WRTT
6. Recovery needs. Fast edge rates are one thing recovery from a large optical input can generate thermal tails
7. Clamp requirements?
8. Output voltage range requirements
9. Noise and thus SNR ........very important
10. TIA gain

There are many smart ways to do a fast TIA and I can recommend many great opamps,  articles and practical topologies based on 30 years of designing this stuff in the fiberoptic design field. Many good sugesstions already posted. Read bobaruni's post. He gets it....

@grouchobyte

[Kleinstein]

At a high speed OPs might not be the best choise. The noise is often lower with discrete JFETs or BJTs. At the high speed BJTs can become attractive again, as input capacitance of FETs can be a problem. Also FETs do have quite some current noise at higher frequency.

Even a simpe BJT in base configuation could be an option.

The NE592 is not that slow: the 120 MHz bandwith with gain - this is not GBW, like usually noted with OPs. So GBW can be well into the GHz range.

For the S/N ration the shot noise could be a limiting factor - so there is a potentialy significant noise source that is essentially unavoidable. Once the amplifier noise is well below that limit, there is not much to gain any more. The power level of the signal decides if a really low noise amplifier is really needed.

[grouchobyte]

At a high speed OPs might not be the best choise. The noise is often lower with discrete JFETs or BJTs. At the high speed BJTs can become attractive again, as input capacitance of FETs can be a problem. Also FETs do have quite some current noise at higher frequency.

Even a simpe BJT in base configuation could be an option.

The NE592 is not that slow: the 120 MHz bandwith with gain - this is not GBW, like usually noted with OPs. So GBW can be well into the GHz range.

For the S/N ration the shot noise could be a limiting factor - so there is a potentialy significant noise source that is essentially unavoidable. Once the amplifier noise is well below that limit, there is not much to gain any more. The power level of the signal decides if a really low noise amplifier is really needed.

There are bootsrapping and cascode buffering techniques and all kinds of other ways to optimize performance after getting the basics defined. Phil Hobbs is a great resource and is on the forum from time to time. He is an IBM fellow, i believe and he clearly has a good handle on TIA behavior. I have read several papers authored by him and I consider him brilliant on the subject. Perhaps he can chime in on this..,..Phil?

Here's a classic Phil Hobbs paper

http://electrooptical.net/www/frontends/frontends.pdf

That said, the OP has some work to do and we are not going to help him unless he clarifies his requirements a little more..

@grouchobyte

[max_torque]

At the moment, unfortunately, things are rather poorly defined, especially the signal return power i can expect to have to amplify!  The TX signal gets bounced off a target, and the reflectivity, shape and orientation of that target all affect the optical power of the returned signal!  So until i have accurate data for what to expect i'm "designing for worst case" and going to stick in as much amplification as i can reasonably do ;-)

So, at this point, i just wanted suggestions as to a suitable opamp, something to get me started, and that can be investigated and optimised as the project progresses   

[grouchobyte]

At the moment, unfortunately, things are rather poorly defined, especially the signal return power i can expect to have to amplify!  The TX signal gets bounced off a target, and the reflectivity, shape and orientation of that target all affect the optical power of the returned signal!  So until i have accurate data for what to expect i'm "designing for worst case" and going to stick in as much amplification as i can reasonably do ;-)

So, at this point, i just wanted suggestions as to a suitable opamp, something to get me started, and that can be investigated and optimised as the project progresses   

In that case do the following

1. Start with a PIN photodiode/detector. If that doesn't get you the SNR you need than use an APD
2. Find an opamp with the following attributes
    a. Low bias current and low noise (pA per root Hz)
    b. Unity gain stable....good phase margin at 0db so it wont oscillate
    c. Large output swing
    d. Low input capacitance
    e. Low input offset ( both cuurent and voltage)
    f. High GBW product

Pretty easy, right? Not really, but for a first order whack to get somethig going you are gonna need something pretty awesome since you are shooting for all the SNR you can get without widening your pulsewidth. Remember, all your performace is in the photodiode and the TIA. Anything you do post the TiA will get you nothing more. You want to be detector-noise limited. If you need clarification on what I mean by that...just ask

Oh, and watch your layout. Make it tight.....no meandering connections crock clips and whatnot

Good luck

@grouchobyte

[Marco]

Cheap APDs have been popping up on ebay/alixpress/taobao BTW.

[tggzzz]

The TX signal gets bounced off a target, and the reflectivity, shape and orientation of that target all affect the optical power of the returned signal!

Sounds like a LIDAR. Will any ambient light enter the receiver in addition to the signal's light?

[grouchobyte]

The TX signal gets bounced off a target, and the reflectivity, shape and orientation of that target all affect the optical power of the returned signal!

Sounds like a LIDAR. Will any ambient light enter the receiver in addition to the signal's light?

If it is LIDAR, and it sure seems like it is, he will need to do some DC and/or synchronous modulation/demodulation or coherent subtraction and DSP target discrimination? Ambient light can play havoc on the returned optical signal as well, depending on its makeup. Moreover,  reflections can really be challenging just like they are in the microwave domain, with mixing, homodyning and range-gating circuit considerations

After the OPs initial tests, I think he will quickly realize ( if he hasn't already) that the TIA challenge was the low hanging fruit on this tree. 


@grouchobyte

[tggzzz]

The TX signal gets bounced off a target, and the reflectivity, shape and orientation of that target all affect the optical power of the returned signal!

Sounds like a LIDAR. Will any ambient light enter the receiver in addition to the signal's light?

If it is LIDAR, and it sure seems like it is, he will need to do some DC and/or synchronous modulation/demodulation or coherent subtraction and DSP target discrimination? Ambient light can play havoc on the returned optical signal as well, depending on its makeup. Moreover,  reflections can really be challenging just like they are in the microwave domain, with mixing, homodyning and range-gating circuit considerations

After the OPs initial tests, I think he will quickly realize ( if he hasn't already) that the TIA challenge was the low hanging fruit on this tree. 

Just so; hence my question

Many of the techniques used in high speed (fibre optic) comms systems are to minimise the effects of dark current, thermal noise and shot noise. Such techniques, e.g. APDs, are unlikely to be relevant in LIDAR systems.

[grouchobyte]

APDs= PITA

Trust me, avoid them if you can. Especially if you are doing LIDAR, you certainly dont want to use an APD.
I have plenty of design experience with APDs in the front ends in OTDRs for over a decade. In OTDRs you have no choice:  use them or be un-competitive.


Tggzzz is right and I concur with his sage comments in this thread.

@grouchobyte

[Marco]

APDs= PITA

Why exactly? AD500-8-TO52S1's are currently relatively cheaply available on ebay and co. Capacitance isn't much different than the SFH-2701 you'd probably be using otherwise, what makes it harder to use?

PS. these seem to be old stock from the previous incarnation of First Sensor.

[tggzzz]

APDs= PITA

Why exactly? AD500-8-TO52S1's are currently relatively cheaply available on ebay and co. Capacitance isn't much different than the SFH-2701 you'd probably be using otherwise, what makes it harder to use?

You have to actively maintain the optimum operating conditions in the presence of significant sensitivity to temperature and voltage changes. Ambient light complicates that, plus removes APD's advantages.

(When I last looked at them mumble years ago, I would also have needed to cool them so they could count individual photons. Such cooling was difficult even with Peltier devices.)

[grouchobyte]

Tggzzz:

Your experience is dead-on vis-a-vis APDs.
All anyone has to do is look and the gain curve, slope and tempco and if that doesn't scare you enough, the fact that they are sourced from some offshore packager for cheap is like asking for trouble.

Another interesting technical fact you might not have heard about and it is not typically spec'd: (One would only know about this if you have actually used them in a design). Photons that land outside of the devices active area are carriers that get trapped and turned into current much slower then active area carriers. The result is a slow tail or sluggish step response. This phenomenon is worse in APDs then in any other device like say a PIN and it varies with material, wavelength, power level, spot size, temperature, gain and vendor. Add that  to the other design factors and you have one hell of a design challenge. Sorry Marco.

That is what I mean by PITA

@grouchobyte

[nfmax]

If this is a LIDAR application, then I would raise the issue of stray light. Any light that gets to your photodiodes other than the pulse signal reflected from the target will cause trouble. Even if you use (as you should) some form of synchronous detection, stray light will add its contribution to the shot noise, and in many cases this can completely overwhelm all other noise sources (as well as risking saturating the TIA). You should look at using optical filtering to remove light outside the operating waveband - but remember that narrow-band filters are designed to work with normal incidence light. You may have difficulty matching the source & filter wavelengths for very narrow bandwidths, such that one or other may need to be temperature-tuned. The receiver optics & housing need to be designed to allow this. Also take great care when designing your internal and external baffling.

Despite the warnings from @tggzzz and @grouchobyte (and I have personally encountered this issues they raise), APD's can be made to work well. But they are 'difficult', and generally most applicable in situations where the light level is very low, and you are struggling for every photon, even if you aren't actually trying to count them.

@tggzzz: have you come across SiPMT's? Basically an array of Si APDs operating in Geiger mode and wired to a common output. They can be used for photon-counting, but seem to be restricted to shorter wavelengths so not much use to us here.

[chris_leyson]

Hi guys, interesting topic with interesting comments especially from grouchobyte and tggzzz, I guess I won't be using those Perkin Elmer APDs then, don't know the part number, can't find them at the moment. I was looking at developing a LIDAR range finder for use up to say 1 meter, modulate a 25mW or 30mW 660nm or 690nm laser diode at around 300MHz or 350MHz and determine distance by measuring the phase of the returned signal. To keep out ambient light I would use a small band pass filter, 10nm or so wide, in front of the photo detector. Optically it would be something like "A twofold modulation frequency laser range finder" available on CiteSeerX. The laser modulation is the easy bit but what about the detector, PIN diode maybe followed by a MMIC ?