Connect with us

Photodiode wich is fast enough to detect +50Mhz analog (sinus) signal??

Discussion in 'Electronic Design' started by Yannick, Jul 19, 2004.

Scroll to continue with content
  1. Yannick

    Yannick Guest


    I am developing a laser distance meter for my Master thesis. I have
    problems with finding a photodiode where the junction capactitance is
    low enough to detect frequenties higher then 50Mhz(laser diode is
    amplitude modulated with these frequentie). The problem is also with
    the amplifier after the photodiode, wich amplifier is best for this
    purpose? (transconductance amplifier i suppose) i am now doing it with
    discrete components (low noise transistors as common emitter or common
    basis but still the input impedance is to high to detect such high
    frequenties with my current photodiode)

    note that the distance is calculated with phase meausurement detection
    and not with time of flight measurement of pulses.

    any help is very appreciated!

  2. John Larkin

    John Larkin Guest

    What's your optical wavelength? 50 MHz isn't especially fast for a PIN
    or APD diode detector. An integrated PIN-TIA would be convenient (but
    be careful that the AGC doesn't cause phase shift errors) or you could
    use a tuned RF amplifier in the front-end to optimize matching and

  3. Yannick wrote...
    You can take encouragement from the GHz response the optical
    networking guys are getting. The secret to high speed is a
    small detector size. This also means you must tightly focus
    the received light. If you then locate the amplifier right
    near the detector, you'll have low total capacitance, which
    is the 2nd secret. The 3rd secret is a high-bandwidth opamp,
    natch, of course, what'cha expect anyway? The 4th secret is
    a low-value transresistance. The 5th secret is a high light
    level for good SNR, but perhaps you have a signal-processing
    solution to this requirement. 50MHz, no problem.

    - Win

    (email: use hill_at_rowland-dot-org for now)
  4. colin

    colin Guest

    at 50mhz the capacitance of any available pd wil be quite low impedance at
    50mhz, only way to get it up is to tune it with an inductor.
    once you got it tuned at a high impedance its best to use a dual gate mosfet
    to make use of the high impedance. watch out for anything that can change
    the capacitance or inductance with signal strength, temp etc as this will
    alter the phase.

    what acuracy, resolution and limits of distance are you hoping to measure ?

    im doing a similar thing to see what is posible. limiting problems seem to
    be .. the detector amplifier picking up rf interference from the laser
    driver, phase change with amplitude cuased by varying distortion in amp
    stages in conjunction with ac coupled filters (enourmous variation in signal
    strength) . jitter cuased by suden and long term drift in capacitors and
    other interference etc, and thermal drift of most things and minute amounts
    of light finding a short path back to the detector. oh and not to mention
    johnson noise of the estimated 100k input impedance and pd and amplifier

    id be realy interested to see how you do. ive got down to about 15 mm
    resolution, any more is realy swamped by drift/jitter wich im trying to get
    rid of, although the digital resolution will cope with ridiculosly small

    Colin =^.^=
  5. John Larkin

    John Larkin Guest

    Pin pd's are available with sub-pF capacitances and 60 GHz response.
    If you load a pin diode with a transimpedance amplifier, the amp's
    effective input impedance is very low, so the diode capacitance is
    mostly zapped. Agreed, for a narrowband signal, tuning is a good idea,
    and it supresses dc (daylight) response.

  6. Bill Sloman

    Bill Sloman Guest

    Tuning can be a bit risky if you are interested in measuring phase
    shifts - the phase shift through a high-Q filter near resonance
    changes rapidly with frequency, so any small change in the the
    reactive component of any of the filter elements can produce a much
    larger change in phase shift near resonance.

    Low pass filtering after synchronous detection (ideally, both in-phase
    and quadrature) might be better in this context.
  7. Yannick

    Yannick Guest

    at 50mhz the capacitance of any available pd wil be quite low impedance at
    so i have to put an inductor parallel with the photodiode!?
    i was first thinking of using a low load impedance so u still can
    measure signals at these frequenties with relative higher junction
    capacitance, not?
    f = 1/(2*pi*Rload * Cj)
    soo if i get it right, u use a photodiode in parralell with an
    inductor too increase the impedance in function of the frequentie and
    then u use a high Rload to have a relative high voltage with the very
    low current through the photodiode and this voltage u put on the gate
    of the mosfet. then u use the drain current to make a voltage(with a
    resistor:)) where u do phase measurement with the original oscillator
    sinus wave, right?
    distance maximum 10m, acuracy as best as possible, i wanted to do
    better then 10mm I have made an algoritm wich calculate the distance
    when many samples of phase measurements (between 0 and 180 degrees)
    are done by different (increasing)frequenties,soo i solved the
    ambuguity problem (sorry for my bad english) and it also minimze phase
    errors because i compare the ideal triangle wave for every distance
    with the measured wave for the distance that has to be calculated, the
    best match is the measured distance. it's a kind of fourier
    transformation but with triangle waves (because phase measurement
    withing intervals of 0 and 180 degrees of increasing frequenties are
    triangles in function of the frequentie).
    indeed that problem i also have, i am now seperate transmitter and
    receiver completely , this will solve it for now:)

    phase change with amplitude caused by varying distortion in amp
    thats already good, how do you measure phase difference? i am gonne
    use the ad8035 from analog devices. The dual gate mosfet u are using,
    can you give the type nr. ?

    thanks a lot for the help so far,

  8. Yannick

    Yannick Guest

    i heard about it:)

    The secret to high speed is a
    why? because the resistance is low due short paths??

    The 3rd secret is a high-bandwidth opamp,
    ofcourse, do you have a type nr. single suply?

    The 4th secret is
    50Mhz was to begin with, i want to go much higher as this will improve
    resolution.but i first have to get it to work with 50Mhz

    thanks for your help

  9. Yannick

    Yannick Guest

    optical wavelengt is 650nm. I fount these avalance photodiodes but
    these are quite expensive.(150 euro)

    thanks for the help,

  10. colin

    colin Guest


    even 1pf has 3k impedance, wich for a signal of only a few nanoamps does
    make for poor sn ratio. reducing the input impedance makes the situation
    worse generaly, signal goes up with R johnson noise only goes up with root
    R, although this is far from being the only noise source.

    i had a good rf transistor as a comon emiter config and just thought of it
    as a curent multiplier stage ie pd curent multiplied by hfe (100) and fed
    into the colector load. but the noise is far less form the tuned dual gate
    yes this is a problem for me i think, the frequency is precise but
    variations in capacitances i think are the problem, i think even the pd
    curent afects this - not sure, obviously temp does, signal strength too.
    maybe it wld be possible to arange several lc elements so it has a wider
    passband with minimal phase variance ? wld be tricky, maybe easier at
    microwave freq where you cld use pcb tracks as elements .

    at the moment cycle to cycle jitter and short term suden changes too is my
    problem, longer term drift isnt to much of an issue as if its predictable i
    can subtract it, although i might think about resorting to maintaining a
    constant temp to ensure temp changes dont afect it. i put a fan close but
    the interference from the fan cuased more of a problem. ( i have to remember
    to turn of my monitor when i make measurments)

    i think its a trade off between noise and phase variance and for me atm im
    not sure wich is cuasing the most problem. im no longer sure what to try
    next ... i think problem is ive got quite a few stages to fiddle with. its
    probably down to needing better capacitors, maybe even my 0.1 uf decoupling
    capacitors need to be hi stability types ? .. ive even resorted to stacking
    up several SM tantalums in order to get rid of supply noise, soldering one
    on top of another ... wich made quite a diference, i can still see 1mv of
    ripple at 32khz tho but i think i can ignore this, although im looking for
    changes of 10ns in that 32khz signal, jitter from this part seems to be less
    than that now.

    im wondering if noise iteslf can be considered in the form of varying pahse
    shift in components somehow ??
    wouldnt you still need some stage of amplification before a detector ? i
    have a single stage then fed into a mixer, its not synchronous as i use an
    IF to efectivly multiuply length of the phase angle and you still need to
    amp the signal quite a bit and at some stage you need to limit the signal
    before you finaly measure it as it cld easilay vary by 100db.

    tbh i think the limiter is cuasing my problems as the sinewave input to it
    doesnt seem to show as much jitter as the output although its hard to tell
    just by looking at it on a scope.

    maybe synchrounous detection wld be a lot easier. actualy mine looks a lot
    like a fm receiver, it just uses a phase discriminator instead of the freq
    discriminator ive even used a sa605 fm ic. you wld have to seperate the
    signal well from the reference i wld gues or it cld swamp it, or at least
    cuase significant error.

    could you somehow have an optical mixer ? maybe not at 50mhz but vanes in
    front of 2 detectors ...

    interestingly my first atempt was to use the light delay as part of the
    feedback in an oscilator so the frequency would depend on the distance,
    however the change in signal strenght cuased more of an efect on frequency.

    this seemed a very simple idea many years ago when i started it, but now its
    grown out of proportion, im not the sort to ever give up tho even tho its
    driving me mad now ... originaly i thought it would be a good idea to use it
    to replace a simple vernier dial gauge, so you could use it on revolving
    shafts etc or while you were actualy cuting it, but this was before id
    calculated the efects of noise and returned signal strengths etc etc ...

    Colin =^.^=
  11. colin

    colin Guest

    the current in the inductor is 180 out of phase with the current in the
    capacitor, if you chose the right value inductor it cancels out this curent
    in the capacitance exactly and hence your impedance whomps up. you dont need
    a resistor anymore , although you might want to tame its Q slightly, say


    for 30mhz ive got a inductor wound on a plastic toroid (made from 2mm dia
    strimmer wire (plastic) bent round a warm soldering iron tip) it has about
    30 turns i think (tuned by trial and error by taking off turn untill its
    past point of resonance then puting one back on)

    i tried a tunable inductor but even with big heavy screens everywhere it
    still picked up no end of interference.
    yes thats about it. u can use an inductor for the mosfet load too, or a
    curent source, or a higher voltage and a plain old resistor. if you tune the
    input and the output too much you are prone to oscilation, and of course
    phase variance, its just one big trade off noise/amplification/filtering
    versus phase variance. lots of playing around unless your expert at
    mathmatical modelng.
    of course with 10m your going to have more than one cycle of delay with
    50mhz, and so you measurment will overflow, good luck with 10mm resolution.
    returned signal strenght at 10m is going to be oh so low, you wld need i
    think an APD and not to forget a good lense, and a white target.

    i gues you would need to be sure your signals are the precise shape you
    calculate for. this isnt easy with hi frequency non sinewaves. maybe trial
    and error and some calibration might be required.

    i have two 30mhz signals (32khz apart) going into an xor gate and the signal
    that comes out looks awfull after the first low pass filter.
    you need to screen things well and seperate power supplys ( of course) and
    also carefull ground planes. if you seperate them to far u then have long
    signal wires wich can also be problomatic. also your beam has to travel down
    the center of your line of sight of the PD if you use a lense.
    wel i cheat a bit at the moment im using a hp 5328 timer counter, wich has
    sub picosecond average time interval measuring, however i use frequency
    shifting to lower the frequency down to 32 khz, but keeping the same phase
    angle so i can multiply the delay by 1000 or even much more, so i only need
    to measure down to 10ns wich u can easily get modules to measure that, or
    even just measure the average dc level of the pulse from the edge detector.

    i might even put an on board micro although doubt il have enough enthusiasm
    left to do that for a while, not sure if it is easy enought to get this to
    the stage where it could be a viable comercial product, do any such exist ?
    ive heard of some devices wich use multiple receivers at diferent focus
    points/angles to detect distance.

    the variable bandwidth with amplitude could be an issue if you use this
    device before the synchrounous detector.

    im using a 5mw red laser, a ca3209 APD fed into a bf998 dual gate mosfet
    wich is fed into a an emiter folower and a PIN agc circuit wich goes into a
    sa605 fm mixer ic although im thinking this is cuasing some problems at the
    moment. i might go back to the string of ecl diferential amplifiers i used

    the delay cuased by varying the gain with the PIN is fortunatly in the
    opesoite direction to the rest of the circuit, ie stronger signal means
    longer delay, but more atenuation actualy means less delay, unfortunatly its
    hard to get the two to cancel perfectly, but i live in hope.

    i might up the main frequency but first i need to find another two crystals
    that are 32khz apart wich is dificult. i might change the 32khz to 455khz
    and use proper IF filters rather than made from discrete wich i think is
    cuasing a problem, or even have 2 IF frequencies or even lower the frequency
    still further say 100hz then just do low pass filtering, but then wld need
    virtualy dc coupled stages.

    I had hoped to use a 32khz crystal as a IF filter but this wasnt very good
    for phase variance, although jitter was non existant. .. sigh..

    unfortunatly i cldnt get 455khz filters from RS or farnell, does anyone know
    where i can get them easily in the UK?

    i think you have a very chalenging project. good luck and keep me posted
    please :)

    sory if my english is hard to understand at times too, i have no real excuse
    for it tho.

    Colin =^.^=
  12. colin wrote...
    Careful, using a sufficiently-wideband opamp can insure the summing-
    junction impedance will be low compared to the total shunt capacitance.
    Resistors have 0.05pF to 0.1pF of self capacitance, this should be your
    total feedback capacitance. With 3k resistor you'd have a -3dB rolloff
    at 530MHz. You want high R for low noise, so we'll try 100k, yielding
    a 16MHz rolloff. Then we can apply the standard R-C-R trick (this is
    more than 30 years old) to get a flat frequency response to 75MHz, or
    whatever you decide your bandwidth should be.

    The noise density from 100k will be sgrt(4kT/R) = 0.4 pA/rt-Hz, which
    is 3.5nA rms for a 50MHz -3dB bandwidth (applying the pi/2 brickwall
    correction, see AoE p 453). But the unavoidable "e_n - omega - Cin"
    noise will be higher, see any of my many posts about this subject.

    For example, if the opamp's e_n = 6 nV (for an opa655) and Cin = 5 pF,
    the e_n-C noise density will be 1.5 nA/rt-Hz at 50MHz, exceeding the
    resistor's Johnson noise. Moreover, since e_n-C noise density rises
    with f, you'll want to consider a sharp band-limiting filter.

    It may be necessary to use a composite amplifier configuration inside
    the loop, to insure that the gain meets f_T > 2pi Rf Cin fc^2 = 8 GHz.
    This may seem a bit daunting, but it's actually rather simple. R2 is
    adjustable, to compensate for the unknown value of Cf (you'll need an
    accurate input-current step to adjust R2, but that's another story).

    | Rf R2 adjustable
    | ,---/\/\---+---/\/\--/\/\----,
    | | '--||--' | C2 R3 | nA-sensitivity wideband
    | | Cf '--||--/\/\-- gnd | transresistance amplifier
    | | |
    | | __ ,-||--/\/\--+ correction network details
    | input O--+---|+ \ | __ | R2 C2 = Rf Cf
    | | >-+-/\/\-+-|- \ | R3 C2 sets bandwidth
    | ,-|-_/ | | >-----+---
    | | | gnd --|+_/
    | gnd --/\/\--+-/\/\--' composite amplifier

    With proper component selection you may use de-compensated opamps. For
    example the opa657 has an open-loop gain of 30 at 50MHz. This amp has
    slightly lower noise than the opa655, but higher capacitance.

    You may want to consider a BJT opamp, but these have high bias currents,
    and therefore have high input shot noise currents. For example, the 1GHz
    bandwidth ad8009 has an input current-noise density of 41 pA/rt-Hz, which
    greatly exceeds the 0.4 pA noise we saw for our 100k feedback resistor.
    No, sorry, faulty reasoning, go back and read / study some more.

    - Win

    (email: use hill_at_rowland-dot-org for now)
  13. Steve Parus

    Steve Parus Guest

  14. John Larkin

    John Larkin Guest

    At 650, you could probably use one of those Agilent integrated pin-amp
    ICs. They are intended for fiber-optic receivers, but you can just
    focus light on the chip anyhow. They are very cheap.

    Otherwise, a cheap silicon pin diode should work; see Tyco, OSI,
    PD-LD,, people like that.

  15. Mike

    Mike Guest

    No... As Win said, the total capacitance is low. The resistance is low too,
    but at 50MHz, the capacitance will tend to dominate in most cases -
    especially if you place the amplifier near the detector.

    -- Mike --
  16. Yannick

    Yannick Guest

    the current in the inductor is 180 out of phase with the current in the
    yes ofcourse, Z = (l/C)/(1/jwc + jwl) , Z to infinite so 1/jwc + jwl =
    0 => w = 1/root(LC)

    the problem is the phase will change very rapid with an increase in
    frequentie , so i don't think i can do this because i am doing
    multiple measurements with different frequenties, soo i have to know
    exactly what the phase difference will be at every frequentie but this
    wont be easy when you do 100 meausrements orso.
    yes thats why i use multiple measurements with different frequenties

    good luck with 10mm resolution.

    hehe thanks:)
    my signal is a sinus, the triangle waves i talked about is not the
    signal but the shape of the samples (between 0 and 180 degree) in
    function of the frequentie (of the signal : sinus).

    i see you mix the HF signal with a HF signal + 32Khz,then you get an
    32Khz signal with the same phase difference... looks interesting.
    micro?? what do you mean?

    i wish you luck
    i am thinking about using a DDS frequentie synthesiser , Analog
    devices have 2 chips , one too 60Mhz and one to 400Mhz. with very high
    thanks, i will
    nono your english is perfect!
  17. Steve Parus wrote... They say very low noise, 2.8dB, but that's with a 50-ohm
    source impedance, right? If so, that's actually a HIGH
    noise, about 16 pA/rt-Hz, 120nA rms over a 50MHz bandwidth,
    compared to a 0.4 to 1.5pA/rt-Hz, 10nA rms noise level for
    the amplifier I described yesterday.

    - Win

    (email: use hill_at_rowland-dot-org for now)
  18. Yannick

    Yannick Guest

    yes that's a point but only if u use 2 frequenties,then you will do
    like u said , first a low frequentie to detect in wich period our
    signal will be and then a higher frequentie to know where we are in
    this particulary period(soo we got high resolution) but the problem is
    that we only can do phase measurement within 0 and 180 degree, soo you
    cant really see in wich half of the period we are measuring, thats why
    u need at least three measurements...

    i maybe will use 100 different frequenties, for example i start with
    50Mhz, then 51MHZ, then 52 and so on. then consider the distanced you
    measure a constant , and plot the phase difference between the sended
    signal and the received signal , you will see this will be a triangle
    wave (waving between 0 and 180 degrees) in function of the
    sendfrequentie. if you compare this wave (the wave with the received
    phase measurements) with the ideal wave, you can calculate wich is the
    right distance.(for every sample do : positive Root[(measured phase -
    calculated (ideal) phase)^2] then make the sum of it(yes indeed its
    like a fourier transformation but with triangle waves in stead of
    sinuses) the more samples you get the better your accuracy will be and
    noise will have less influence.
    yes indeed the higher the freqeuntie the better our resolution will
    be. but its hard to detect such high frequenties.
    indeed ofcourse that's also my plan. I am gonne program an atmel
    AT90S8535 microcontroller to send the signals to a DDS frequentie
    synthesiser wich gives the right frequenties in a minimum of time. the
    only problem with the analog device DDS synth is that the main clock
    is 120 Mhz soo when u want 60Mhz out you only get 2 samples , then
    after DA conversion thiss will make a square wave, and then u have to
    use antiimage filters to make a sinus again. but i am not sure how
    good this will be, what kind of DDS frequentie synthesiser will u be

    it's good someonse is also trying this challenge, gives me hope :),


  19. Yannick

    Yannick Guest

    indeed that's another way of calculating the distance out of the
    triangle wave but how would you measure its frequenty ? then you
    should again do DA conversion and filtering and then put this signal
    into a frequenty counter, or do i completly miss the point here?

    yes thats right.

    my main area of aplication would be
    ok, what kind of lense system are you using to detect the light? did
    you bought a module or made your own one?
    hmm that's clever.
    yes but i think i am going to use an elleptical filter with 47Mhz cut
    off frequenty like Analog devices is using on their evaluation board
    for the DDS.

    yes ofcourse ,and the atmel AT90s8535 does have a 10 bit AD convertor
    soo this is also usefull.

    I also had a different idea last week about doing this project :
    nothing more with phase calculation of sinusoidal signals but just
    with pulses:

    you send a very short puls , simultaniously you start charging a
    capacitor with a current source (soo it charges lineair). When the
    reflected pulse arrives on the detector you sample the voltage across
    the capacitor and this is in direct proportion with the distance the
    photons of the laser travelled. u should use a very fast SR flipflop
    (to start charging and uncharging the capacitor) and a low noise
    current source , and a high bit AD convertor, this should do the
    trick. What do you about this?

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