Connect with us

Designing a APD based receiver for use with a TOF laser range finder

Discussion in 'Electronic Design' started by Michael, May 24, 2007.

Scroll to continue with content
  1. Michael

    Michael Guest

    Hi - I'm attempting to design a circuit using an Avalanche Photo Diode
    (APD) to detect pulses sent out by a laser diode, as part of a Time Of
    Flight (TOF) laser range finder. My goal is to get really sharp pulses
    from the APD circuit (ideally with rise times under 100ps, even better
    would be under 10ps). I don't need single photon counting ability -
    but the greater the sensitivity the better.

    I have never worked with an APD before, so I've been reading as much
    as possible about them. My understanding is this: you reverse bias
    them with a very large voltage that is beneath their breakdown
    voltage, normally 100 or more volts. You put a shunt resistor in
    series with the APD, and use that to measure the current flowing
    through it. I'm assuming this would be done with a really high speed
    op-amp. Typical current I believe is in the nano-ampere range. I think
    this is all dark current? When light hits the APD, the current will
    increase for a brief moment, with the magnitude of the added current
    controlled by the magnitude of the bias voltage, the larger the bias
    the larger the current increase. (with the exact relationship shown in
    a graph in the APD's datasheet)

    How am I doing so far?

    I am hoping to use a visible laser diode (for safety, as well as ease
    of debugging). Red seems like a good option as red diodes are so
    common and inexpensive. But either way - visible means that I'll need
    a silicon APD. I can't find many distributors for APDs, unfortunately.
    I found digi-key has a couple:
    Does that pricing seem normal? The cheapest they have is $126.92 in
    single quantities. Are there other distributors or manufacturers I
    should be looking at?

    Also, I've seen some work done with actively cooling the APD to
    decrease the dark current. When is this necessary? From where I've
    seen it done, it looks to only be done when you're trying to count
    single photons, which is not what I'm trying to do.

    Lastly - what is the benefit of having a large active area? It seems
    that price is directly proportional to active area. To me, it seems
    like putting a big lens in front of a APD with a small active area
    would serve the same purpose as using an APD with a large active area,
    but I suspect that I'm missing something.

    Can anybody shed some light on APDs for me?

    Thanks so much!

  2. Guest

    You've missed the point that avalanche photo diodes amplify the
    current generated by a single photon - initially a single hole/
    electron pair - by a process of avalanche multiplication, in which one
    of the charge carriers moving through the lattice picks up enough
    energy to create more hole/elecron pairs.

    Any "dark current" is multiplied in the same way. The diodes break
    down when the reverse voltage across the diode is high enough that
    even the dark current is multiplied up to a current which can creat
    run-away warming in the lattice.
    There are specialised parts for a small market, so they end up
    expensive and not widely available.
    Stabilising the temperature of the APD also stabilises the avanalnche
    gain at a given voltage, which can be helpful.
    No. If your incoming light is well collimated, it is much better to
    use a lens to focus it onto a small area photodiode, with a low dark
    current and a low capacitance.
    Sergio Cova at the Milan Polytechnic has published a number of good
    papers on avalanche photodiodes and single photon avalanche diodes
    over the years - check out Applied Optic and the Review of Scientific

    A search on his name on Google Scholar (
    throws up 713 references - not all of them useful. Adding avalanche
    and photodiode to the required words brngs this back to 34.
  3. John Larkin

    John Larkin Guest

    You should be able to find some cheaper apd's, and probably get one as
    a sample if you write a convincing email. Take a look at a buyer's
    guide, Photonics Spectra or Laser Focus World, and try some emails.
    These things are going for a few dollars in medium quantities, so
    samples shouldn't be a big deal.

    If you're going to be working in background brighter than moonlight,
    an APD may not be worth the trouble; just a fast PIN diode would work
    as well and not need the high voltage.

    If you do use an apd, current limit the supply!

    A large area diode will have a lot of capacitance and be slow.

    What's your laser like? Range? Optics? Resolution?

  4. Michael

    Michael Guest

    On May 24, 7:28 pm, wrote:

    Hi Bill -
    So dark current is the only current flowing through the diode when
    reverse biased and not exposed to any light, correct?
    But I think there's more to it than that. One particular project I
    looked at cooled the APD down to, if I'm remembering right, 77K.
    That's cold!!
    I will check him out. However, I should be clear that I'm not looking
    to count single photons, and I think the techniques used for counting
    single photons differs a bit from what I'm attempting to do.


  5. Michael

    Michael Guest

    Hi John - I was hoping you would weigh in. I spent a good deal of time
    reading through the google groups archive of posts related to APDs,
    and your name popped up a good number of times. I'll check those
    resources out.
    My understanding of APDs is that they vastly increase your
    capabilities. Specifically, they'll allow you to do things like
    decrease your transmitted power while maintaining the same sensing
    range as was available with a PIN diode. So I figure I might as well
    start with the more powerful solution, and if that ends up being
    overkill, I can step it down a bit.

    I hope for this device to work in all lighting conditions. My plan is
    to use a diode of a very specific frequency and find a filter that'll
    knock everything else out. To be honest, though, I haven't spent much
    time working on the optics side of things, as I'm an EE, so I'm not as
    familiar with that stuff.
    Which brings up a question: what does a typical power supply for these
    things look like? My understanding is that it doesn't need to be able
    to source much current at all (1ma maybe?) but that it needs to have a
    large voltage - 100-200V typically. The eventual goal for this device
    is for it to be battery powered, probably operated off of a single 5V
    Well - that wouldn't be any good! What use are the really large APDs
    then? Digi-Key lists them as $1500 or so - so surely somebody must
    have a good reason for using them!
    I haven't chosen a laser yet. I thought it'd be best to choose a laser
    to match whatever APD I end up with. My hope is to stay with a visible
    wavelength though, and keep it very low power (5mw or less). I want
    this thing to be very safe. I've been thinking I'd use some laser
    driver chips designed for optical communication to drive the laser.

    My desired range is 0-5 meters. More would be awesome, but I'll
    survive with that. Really, I would be happy with just a couple meters
    of range, but I think this device will be alot more useful if I can
    get a higher range.

    As for optics - I haven't put much thought into it just yet. I thought
    I'd use a collimating lens for the laser's output, and then some
    filter as I mentioned earlier in front of a large lens in front of the

    Regarding resolution - I hope to get a centimeter resolution or
    better. My plan is to use some of the Time to Digital Converter (TDC)
    chips made by Acam. (
  6. Paul Mathews

    Paul Mathews Guest

    Recommended reference: Building ElectroOptical Systems, Phil Hobbs
    High resolution range finding at relatively short ranges is most often
    done with phase measurement rather than pulse echo time measurement.
    You will soon discover why when you attempt to generate and observe
    ultrafast pulses. Echo time for 1 cm in free space is 67 picoseconds.
    Measuring 1 degree of phase at 41 MHz is generally easier than
    measuring 67 ps.
    Paul Mathews
  7. Phil Hobbs

    Phil Hobbs Guest

    The main usefulness of analogue-mode APDs is in the range from the
    practical upper limit of photon counting (say 10-100 MHz average count
    rate, or about 1-10 pW in the visible) to the lowest photocurrent where
    shot-noise limited SNRs are possible with reasonable bandwidths, say
    about 5 uW. Within that range, by adjusting the APD bias, you can get a
    big SNR improvement with an APD, though you may not get to the shot noise.

    APDs slow down at high gains, because it takes awhile for the avalanche
    to build up. If you can get afford to get two matched APDs, and run one
    in the dark, you can temperature-compensate the gain by servoing on the
    amplified dark current.


    Phil Hobbs
  8. Michael

    Michael Guest

    phase difference range finding. I remember talking with a guy from
    MIT's Lincoln Labs about it - and he said that they use TOF range
    finding, and that they'll even watch for things like 2 different
    ranges present at the same position, which often indicates something
    like a car parked beneath a tree. As I recall, their receiver counts
    single photons returned.

    The TDC chips I plan on using should take care of the timing part.
    Generating really crisp pulses for the laser is one of my bigger
    worries, though I think some of the commercially available laser
    driver chips should be able to help me out there.

  9. John Larkin

    John Larkin Guest

    A dinky 850 nm vcsel driven by an eclips lite gate will give you a few
    milliwatts with an optical risetime of 100 ps or less. The receiver
    and the timing will be a bigger problem.

  10. Michael

    Michael Guest

    Hi Phil - first off, just to be sure: by analogue mode - you are
    referring to analogue mode as opposed to Geiger mode, correct?

    Also - what do you mean by servoing the amplified dark current?


  11. Michael

    Michael Guest

    Hi John - pardon my ignorance - but what is an eclips lite gate?
    Googling for it turns up nothing. Googling for light gate turns up
    some mechanical devices that might be able to cut off an optical
    signal, but I'm not finding any solid sources of information.


  12. Mike Monett

    Mike Monett Guest

    As John mentioned in another post,

    "A dinky 850 nm vcsel driven by an eclips lite gate will give you a
    few milliwatts with an optical risetime of 100 ps or less. The
    receiver and the timing will be a bigger problem."

    Converting the pulse timing to range is a non-trivial problem. TOF
    converters suffer from jitter, so averaging will be needed. This
    will take time, and eventually you hit a barrier where further
    improvement in SNR will simply take too long.

    If you are interested in looking at a newer approach, the Binary
    Sampler allows you to overcome the averaging limit, and it gives
    more accurate results much faster. For example, with very simple
    circuitry, you can obtain greater than 1 ps rms resolution in 1
    second at 1 MHz. Running at 41MHz should give a corresponding

    An ideal timebase method is the heterodyne technique. This has been
    difficult to achieve due to the need for low timing jitter in the
    offset frequency. A regular DDS may give a jitter of 300ps rms or
    more, which is unusable.

    ADI now has the AD9540, which is a very low jitter clock
    synthesizer. I have not had time to try it, but it looks very
    impressive. It offers femtosecond level timing jitter and 48-bit
    frequency tuning word resolution for under $10.00. If these specs
    are true, it would solve the frequency offset problem and make the
    Binary Sampler extremely useful in applications where accurate
    measurements are needed in signals with large timing jitter.

    You can see an early version of the Binary Sampler at

    The concept has been considerably improved since this was posted. If
    you are interested in more information, you can contact me at the
    address shown on the contact page.


    Mike Monett
  13. Guest

    ECLinPS - ON-Semiconductor's recent version of emitter-coupled logic

    You can buffer the emitter-follower outputs with wideband discrete
    transistors if you want a bit more current - I used the 5GHz BFR92
    (NPN) and BFT92 (PNP) some twenty years ago. Farnell still stock them,
    but nowadays they have 10GHz parts and some items in a list that is
    supposed to go up to 45GHz.
  14. colin

    colin Guest


    That binary sampler looked interesting, im trying to average a time interval
    signal down to sub picosecond resolution, but my data's standard deviation
    is nearly 10 nanoseconds. I do however have ~10 million points per day to
    play with.

    It looks to me like it effectivly does the same thing as creating a line of
    best fit with equal number of points above and below, as opposed to line
    with minimum absoulute error or square error. have I got this right ?
    however im not familiar with mathmatical techniques to find such a line of
    best fit, is there any code around for doing such techniques ?

    its just a one off physics measurement experiment im doing. At the moment im
    just doing a fft to find the signal im looking for wich shows up as
    modulation of time interval of <1ps with a period of many hours. I'm playing
    about with stuff like rejecting 10% of the points at the edge of the
    standard deviation. I gues I could just limit the error rather than reject
    points. I havent realy looked into error reduction yet as im still trying to
    reduce the source of the error.

    The noise is from mechanical system so there are many things like short and
    long term drift, and sudden change in offsets, spurious spikes etc.

    However if there is a way to reduce noise in proportion to the number of
    samples rather than the sqrt this would make a big difference.

    As for heterodyne aproach for LIDAR I tried to do this with swept ~2ghz
    modulated laser and an APD with a modulated bias voltage with a freq offset
    or IF of 455khz. the difficulty was keeping ghz VCOs 455khs apart from
    eachother, they always tried to lock. however any phase noise could easily
    be compensated for with a null control path.

    I managed to see resolutions down to less than a millimeter with no signal
    proceesing. I hope to revisit this project and apply some DSP like the above
    too wich might help acheive my goal of detecting 1um change, idealy it would
    be able to replace a mechanical dial indicator guage, advantage being non
    contact of a rotating shaft for example. maybe DDS of 1ghz wil be possible
    by then.

    im not sure about how well it would be possible to account for multiple
    targets on the same path.

    However I also saw an interesting method of determining absolute distance
    using laser interference, by modulating the laser frequency and doing
    correlation on the change in the resulting interference signal.
    claimed resolution of the wavelength of light over distances >10M. it doesnt
    even need a fast detector.

    Colin =^.^=
  15. John Larkin

    John Larkin Guest

    For certain values of "newer", as 45 years maybe.

    Yes. I argued with MM over this for some number of years, and gave up.
    You saw it right away.

    Life doesn't allow that. MM would have a Nobel by now if it did.
    A good laser interferometer could do what you want. But if you want to
    use tof, be aware that it takes extreme measures to get a signal chain
    like yours down to 1 ps per degree C drift. The tof chip you suggest
    using is going to be far, far worse, as will the laser+driver, the
    pin/apd amplifier, and whatever comparator you use.

    What's the physics here? Could you use an incremental, as opposed to
    absolute, position measurement system? Could you use some other
    distance measuring scheme, capacitance maybe?

    I just don't think you can hold < 1 ps for any usable amount of time
    using tof as described.

  16. Phil Hobbs

    Phil Hobbs Guest

    You run both diodes from the same bias supply, and servo the voltage to
    keep the amplified dark current of the dark diode constant. That keeps
    the multiplication gain pretty well constant too. Protection circuits
    are required to avoid blowing up expensive APDs.


    Phil Hobbs
  17. colin

    colin Guest

    hmm i see.
    dam. how sure are you ? it seemed to claim that it did, im not convinced
    either way yet.
    I was going to test it on my test data just need to make a different stat
    analysis algorithm.
    I was realy hopefull :/

    im not the right person to enjoy doing a proveable mathmatical analasys to
    find how the noise reduction equates to the number of samples with this so
    called binary aproach. any mathmatical types here ? its probably already
    been done somewhere anyway, not that I would know exactly where to look for

    I gues a test program would suffice however.
    for me its 2 seperate projects, the <1ps timing is not from a distance
    although they do have similar needs for precise timing wich is why i
    mentioned both of them.

    trying to hold the <1ps for any length of time ... hell yeah tel me about it
    fortunatly I can null out any offset error every few seconds.
    however ive found if the table tilts a bit it gives ~1ns offset error wich
    isnt so easy to null out.

    My LIDAR is for general purpose for machining/robotic sensing so realy needs
    to be absolute.
    The corelating interferometer is as acurate as an ordinary inteferometer,
    but does absolute too.

    Colin =^.^=
  18. colin

    colin Guest

    ive had a play about with this binary sampling but I cant get it to do any
    better than normal averaging,
    ive introduced a filter in software wich mimics the comparator,
    although it reduces the noise the signal almost disapears too,
    unless I increase the slew rate to many times what is needed and then the
    noise just gets through again.
    ive also added slew rate limit detection etc.

    maybe it is dependant on the signal and noise ?
    my test noise is just from a weighted random number generator.
    my test signal is just a sinewave many times lower than the noise.

    there might be other ways I can play around with it but the only effect its
    had so far is to reduce the recovered SNR.

    Colin =^.^=
  19. John Larkin

    John Larkin Guest

    "Binary sampling" [1] has bad statistical properties. It servoes on
    the median of the signal, not on the average; so it really screws up
    if the noise is not perfectly symmetric. It looks sort of like
    delta-sigma, but d-s sums the signal and the feedback into the
    integrator, and the integrator feeds the comparator, so d-s does
    indeed servo on the average. BS sums the integrator output plus the
    signal into the comparator, which changes things.


    [1] called "slideback sampling" in the 1964 GE Transistor Manual.
Ask a Question
Want to reply to this thread or ask your own question?
You'll need to choose a username for the site, which only take a couple of moments (here). After that, you can post your question and our members will help you out.
Electronics Point Logo
Continue to site
Quote of the day