Dismiss Notice
Join Physics Forums Today!
The friendliest, high quality science and math community on the planet! Everyone who loves science is here!

Squashing Shot Noise - help please

  1. Dec 23, 2012 #1
    Citizens of Physicsforums.

    I would greatly appreciate your advice.

    I've been wrestling with a pesky noisy phototransistor for quite a long time, and am at wit's end.

    The sensor is being used to transuce an audio signal in the 1-5kHz range. Unfortunately the shot noise is introducing a broadband "hiss" into the audio that cannot be filtered, or at least not completely because it overlaps the audio signal.

    Can anyone recommend a simple circuit to effectively "Dolby" filter such noise?

    Thanks in advance for sharing your ideas!

    Most obliged,

  2. jcsd
  3. Dec 23, 2012 #2
    Could you average the signals from several phototransistors? If the noise is random it should average to zero from multiple detectors.

    Could you modulate the source and attempt a lock-in method to remove the noise?

    Welcome to the Physics Forums.
  4. Dec 23, 2012 #3
    Before you worry all that, first, tell a little more about your design.
    1) Is it you are sensing a light signal that modulated by an audio signal of 1 to 5 KHz.
    2) Do you transmit in a pure dark environment or in environment with lights? If there is light from the environment, make sure the noise is not from the environment. One sure way is to totally block the photo transistor and see whether the hissing goes down.
    3) If you are modulating the light, how hard are you modulating the signal.....meaning what is the variation of the carrier? Make sure you modulate hard enough so you don't have to amplifier the photo transistor too much. Less gain means less noise.
    Do you have any filter after the photo transistor pre amplifier? You are detecting from 1 to 5 KHz. Usually we put low pass filter at 5KHz to roll off any higher frequencies. That will make a difference.

    If #2 fail, you need to show the circuit here to see whether you have too much noise gain. After all these, then you can worry about dolby and all.
  5. Dec 24, 2012 #4
    Thanks very much for your replies!

    The signal I am transducing is surprisingly enough, a harmonica reed. Actually a set of ten reeds, summed. The noise appears to intensify with each additional sensor, rather than canelling out :-( But I'm intrigued by the idea of signal averaging... which I would know how to accomplish if it were a digital signal but cannot imagine how to do without some sort of sample/hold.

    I'm relatively sure of the noise source as I experimented with a single isolated sensor with just a pull up resistor, illuminated by mini incandescent lamp. Completely sikent until the lamp is powered up.

    The idea of modulating over a carrier occurred to me. .. . reflecting how some strain gages are (used to be) driven. But I do not have the expertise to design such a circuit.

    I'll post the schematic tomorrow, but forewarn you that it is embarrassingly simple.

    Thanks again for taking the time to share your ideas!!!

  6. Dec 25, 2012 #5
    How did you connect them in this case? To average the analog voltage I would buffer and Low Pass Filter each sensor and then average them with resistors or an Analog Summer.

    For the modulation method, obtain a clock signal at an appropriate frequency, perhaps 150k-500kHz and use it to control a fast shutter that blocks your light source. Use this same clock to swap between inverting and not inverting your detector signal, and Low Pass Filter the output with a frequency roll off appropriate to your harmonica's range, 40kHz maybe? You could easily use this chip to do the job, as shown in the example labelled Lock-In Amplifier, replacing the first inverting amplifier with your buffered photodiode signal on pins 1, 16.
  7. Dec 25, 2012 #6
    Thanks again for your suggestions. Attached is the schematic, which I forewarned was rather simple. You'll see that there are already two stages of low pass filtering, but they dont seem to do the trick.

    This circuit is not my design, per se. I actually paid someone to design it for me. But looking at it now more closely I find a number of features that dont look right.

    1) The dropping resistors R4 and R5 seem to be too small. According to the sensor datasheet (attached), the collector current is nominally supposed to be 0.4mA, but the current value of 220 ohms suggests something like 100x that value.

    2) the datasheets for the op amps state that the minimum bias voltage is 10V, but it looks like the Vcc in the current circuit is 8.15V

    3) The feedback cap's are also not right. At first, I tried replacing them by computing the corner frequency. This improved the SNR, but barely. Then I tried empirically through trial and error, which was more effective. But still the problem is that the noise band overlaps the signal.

    4) The LED driver current was set too high (appx 20mA) and as a result the transistors were saturating.

    5) BTW, the frequency range from lowest note to highest note is: 274-2793Hz

    I experimented with an external active filter, just to see if a sharp cutoff might do the trick. But even with a 32dB/octave filter, the noise leaks in.

    Which is why I was fantasizing about a Dolby type filter ... it brings back memories of magnetic tape... it was impossible to filter out the hiss without throwing out the baby with the bathwater.

    Now that's all off my chest, I'd like to digest what you write about the chopping and PLL approach. As I confessed earlier, its really beyond my expertise. But I would be pleased to pay someone to design a circuit if it would solve this abiding stubborn problem.

    Thanks so much! I feel so much better now that I have someone to talk to!


    ps: I would also appreciate your opinion about R4/R5 - will reducing the collector current help?
  8. Dec 25, 2012 #7


    User Avatar
    Science Advisor
    Gold Member
    2017 Award

    It isn't clear how you are modulating your audio onto the light. What are the details of the present transducer? There may be a link or attachment that I have missed? I can't find anything.
  9. Dec 25, 2012 #8
    I apologize if the attachments didn't get uploaded. I am trying again to attach them here.

    I'm actually not modulating the light signal, per se. We're using SMT proximity sensors, with LED and phototransistor in the same package. The IR LED's are driven by a current source. And the reflection off the vibrating target (reed) is what creates the audio-frequency signal.

    If necessary, I'd be happy with a generic sinusoidal signal of the proper frequency and amplitude - as long as it was free of noise (hiss).

    thanks again


    Attached Files:

  10. Dec 26, 2012 #9


    User Avatar
    Science Advisor
    Gold Member
    2017 Award

    I was also interested in the physical setup. There will be a vast range in the modulation depth of the light coming off the vibrating object, depending on how it's actually illuminated and where you measure the light coming off. Basically, the Optics of the situation.
    Also, looking for low level signals at low frequencies is not always the best. Synchronous detection, using a 'shuttering' frequency takes the signal to be detected up to a frequency well away from shot noise, hum etc. You don't need a phase locking system because you can use the same source for modulation and detection. The only difference with the equipment is the oscillator and the mixer. The filter then becomes a Band pass instead of a Low pass. I think you'll find this is pretty standard and used whenever possible / necessary.
    However, I don't think you need be looking for a particularly low level signal in the first place if you make the most of the optical system in the first place. There is virtually no limit to how brightly you can illuminate your subject and how much you can vary the amplitude. This implies that you can get any signal to noise ratio that you need.
  11. Dec 26, 2012 #10


    User Avatar

    Staff: Mentor

    I expect the LED current was set to maximum in an effort to provide useable illumination of the reed, itself unlikely to be very reflective? Resistors R4 and R5 being the series 220Ω resistor between the 8.3V and 4.15V rails? It is difficult to see the typical IC on your datasheet, quite a few details have been obliterated by your highlighter pen. Fig 3 in the datasheet shows the transistor is almost a constant current device, so use of a small R4 means the transistor is unlikely to saturate unless a large expanse of reflective white paper or a mirror is encountered.
    Most LM833 datasheet graphs do start at ±5V, though one shows a dotted line extending down to just below ±4V. The design does seem unnecessarily borderline, the designer could surely have made the 8.3V line 10.0V, though he must have been satisfied it was working okay at 8.3V,
    I'd take some convincing to accept that is true.
    http://img803.imageshack.us/img803/4666/holly1756.gif [Broken]
    Last edited by a moderator: May 6, 2017
  12. Dec 26, 2012 #11
    Thanks again for your thoughtful suggestions. The most encouraging and exciting think you wrote concerned the idea of chopping or heterodyning (or is it homodyning?) the signal. I understand this in principle, from my experimentation with Radio as a kid. But I never designed such a circuit. I would imagine that an IC must be available, like the AD630 you suggested earlier. However, I would also hope that there is a less expensive way of doing this with discrete components. (The AD630 is about $20, and I think I'd need two: one for each board - which is prohibitive.)

    Regarding the illumination setup, I've attached a photo and diagram to illustrate the arrangement. The reeds are relatively reflective, made of polished brass/bronze.

    I apologize for the yellow hilighting in the datasheet. I didnt realize that it made the text illegible. Attached is a clean version. The typical value of Ic is at the bottom of page 3. It refers to the "on state" inasmuch as the sensor was most probably intended for on/off binary detection of a target rather than the linear application I am forcing out of it.

    Your recommendations concerning the intensity of illumination is, unfortunately, inconsistent with my observations. I found that the intensity of the noise is monotonically related to the intensity of the light (both when I use the built-in IR LED, or an independent grain-of-wheat incandescent.) The signal, on the other hand is unimodal. No kidding. The amplitude of the signal increases with LED current up to a point, but then drop off precipitously. If I move the target farther away from the sensor, then the point at which this drop occurs moves to a higher level of current. But still the same behavior is exhibited.

    Illumination issues notwithstanding, I plan to implement your suggestion of lowpass filtering each of the sensors, rather than the entire bank all together. But I am most interested in implementing your frequency shifting/signal chopping/modulation/demodulation idea. Could I possibly hire you as a consultant to sketch the circuit? (I am able to lay out the board.) Or can you recommend someone who might be available and willing to do this? While I myself am eager to learn, I feel like I'd need to go back to school, and pay attention!

    Attached Files:

  13. Dec 27, 2012 #12


    User Avatar
    Science Advisor
    Gold Member
    2017 Award

    I now see the physical layout and it's obviously fairly cramped in there! Looking at the way you have mounted the sensors. I have some relevant points to make, the way the angle of the reed affects the illumination of the photo transistor could be increased significantly. The light flux hitting the sensor will hardly be affected by the distance from the reed in its motion - it will be due to the changing angle of the reed deflecting light away from the sensor to a varying degree. I think, if you had the sensor units rotated 90 degrees from their present orientation. Also, if the sensor were positioned at the end of the reed, the variation of angle would be at least double what you could expect if it's placed as in your diagram. This would give a much bigger variation in the amount of light actually reflected to it. I think. A big deflection would direct the light to one side, completely away from the line joining source and sensor. Did you experiment with this or was it just an arbitrary choice?
    You say that the noise you get is apparently proportional to the light level. This must imply, I think, that the light level hitting the sensor is not changed a lot as the angle of the reed changes. The angle of the reed won't ever be great - perhaps a couple of degrees at that part of the reed and the trigonometry of the situation (and any curvature of the reed would be producing a reflection image with a position that varies only slightly. This could even imply that a narrow slot over both detector and source would produce would produce deeper modulation as the image would be deflected right away from the sensor. Needless to say, there would have to be a compromise for the slot size - too small and too little light would get through for the detector to register.
    From your picture, it looks as if the mods I am suggesting could all be incorporated - albeit being a pain to do a rebuild.
  14. Dec 27, 2012 #13
    I cannot thank you enough for the time and effort you've been putting in to this problem.

    I'm very eager to implement your idea of chopping/modulating/demodulating the signal. Just wish I knew how. The AD630 is definitely an elegant one-stop-shopping chip, but a bit impractical for this application.

    Regarding your suggestions on the light path etc., I do suspect that you are right about the light deflection due to the angular deflection of the target. On the other hand... the target is actually rather close to the sensor - within 2mm. And according to the Datasheet, the sensor is capable of sensing a flat object parallel to its face within that distance. (See attached Fig 1 from page 4.)

    Of all the many assumptions and educated guesses that went into this design, the one meticulous thing I did was conduct a parametric experiment in which a single sensor was attached to a 3-axis stage. This permitted me to evaluate the sensitivity to x, y, and z positioning. I also did some angular sensitivity studies. I eventually configured the sensor layout so as to achieve the greatest amplitude signal but without buzzing against the reed (hence the reason they are not located at the tippy-tip.)

    The illumination study that I did was less meticulous. I connected a single sensor in the most simple manner: with just a pull up resistor, and measured the voltage at the collector. Without exciting the LED, I illuminated the transistor with an old fashioned incandescent lamp, powered by a regulated supply. The result was white noise, audible as "hiss" when patched through an audio amplifier. Although I cannot say with certainty that it is shot noise, it is my best guess based on what I've read on various sources of noise, such as Johnson, Nyquist, dark, 1/f, etc...

    That said, I do plan to try your experiment of masking the sensor with a slit. But the solution that brings me the greatest hope is your modulation/demodulation solution.

    Speaking of which... I was wondering... what if... I were to transmit the signal using a wireless transmitter? I have one handy in the shop, although I think its based on FM. I've got nothing to lose.

    Thanks again for your help.


    Attached Files:

  15. Dec 29, 2012 #14


    User Avatar
    Science Advisor
    Gold Member
    2017 Award

    I haven't had much time for this as the family are around but I suggest you look up topics around Synchronous Demodulation low signal measurement. At these frequencies, four quadrant multipliers will work easily and the shuttering frequency need only be a few tens of Hz to avoid any odd spectral foldback for your high notes. (There's some basic constraints when you want to modulate and demodulate).

    I notice that these sensors are really meant for line following in robotics where the contrast is fairly high. I just wonder about how that distance sensitivity curve works and what exact optics is used for the light path - particularly with a semi-specular reflection that you will get off a metal reed definitely isn't a mirror. Using a slit may help - it could well be worth trying. The slit should be across the axis of the reed, I reckon and I think the sensor unit needs to be the other way round too.
  16. Dec 29, 2012 #15
    I will investigate your suggestions.
    Please enjoy your family... this project has languished for years, it can wait a few more days. In the meantime, I'm learning about various noise reduction ICs... hoping to stumble on an inexpensive and simple solution: ala noisy signal in - clean signal out.
    Thanks again.
    Happy New Year!
  17. Dec 29, 2012 #16
    I think I might have found an IC that does what you are recommending.

    AD7400A[/PLAIN] [Broken] sigma-delta modulator

    My hunt also led me down the path of dynamic noise reduction (DNR) and Dolby filtering, specifically a lineage of ICs originating with AD's SSM2000 circa 1996, which begat LM1894, which in turn is being replaced by TLV320AIC3256.

    The TLV is only $1.96 on Digikey. It has all kinds of bells and whistles, including stereo audio amps. But I'm intimidated by its 40-pin package.

    The LM1894 is on its way out, but still supported by TI. Its only $1.95, and features a much less imposing 14-pin SMT package.

    Whenever you are back online, I'd appreciate your opinion.
    Last edited by a moderator: May 6, 2017
  18. Dec 29, 2012 #17
    I'd suggest trying a photodiode into a transconductance amplifier as an experiment. I've had nothing but pain from my experiences with phototransistors. Aside from having a large output signal, everything else about them is terrible.
    For the transconductance amp, I'd use a low noise jfet or mosfet type op amp, such as the LMV751 with a +/-2.5 volt supply.
  19. Dec 30, 2012 #18


    User Avatar
    Science Advisor
    Gold Member
    2017 Award

    Dolby was originally invented (ifaik) as a way of reducing the audibility of tape noise for mini cassette recorders). It takes an audio signal and processes it so that channel noise which is subsequently added is less perceptible by boosting some frequencies on a variable basis. In your case, the noise is already on the audio signal so Dolby won't help, I'm afraid. the clever thing about the shuttering process is that it modulates the wanted signal to a part of the spectrum where there is less 'channel' noise. If the spectrum of your channel noise is, indeed, in the low frequency range then this will work. If it isn't, then there's nothing much you can do.
    Of course, using the optimum design of amplification can also help (having the appropriate input impedance for your amplifier.)
    I can see why you went for those sensors because they fit so nicely where they're needed but you may need to consider an alternative sensor. Did you consider some sort of capacitative pick up, for instance?
  20. Dec 30, 2012 #19

    jim hardy

    User Avatar
    Science Advisor
    Gold Member

    i may be wayyyy off track here

    but had a thought...

    Magnetic recording tape employs "Bias", a hf AC signal added to the signal to overcome tape's hysteresis and hiss. The technique was one of the War Prizes brought back from Germany at end of WW2.


    I'd tinker with the excitation current to the optocoupler to see if it helps. Perhaps add some ~40khz AC ? That you could do with existing circuit board....

    Again - it's just a crazy idea at this point.
    But i meant well.

    old jim
  21. Dec 30, 2012 #20


    User Avatar
    Science Advisor
    Gold Member
    2017 Award

    Afaik, the bias just reduces hysteresis distortion which, in itself gives a gritty, programme related noise like the quantising 'noise' in ADCs, which is a distortion rather than random noise, really. Using a high bias frequency is a bit like extreme oversampling, which spreads the quantising noise over a wide bandwidth, mostly outside the audible range. The Recording Audio pre and de emphasis curves do a lot to help the basic SNR too because of the spectrum of random tape noise.
    Mind you, Tape = Ghastly reproduction. It's a wonder we got away with it for so long.

    I thought some more about a capacitative pick up and realised that the dreaded spit could ruin it with the small clearances needed for getting any sensible values of C. But I guess that would also be a problem with any optics relying on clean reflections.
Share this great discussion with others via Reddit, Google+, Twitter, or Facebook