Connect with us

Noise Diode

Discussion in 'Electronic Design' started by Marceli Firlej, Aug 14, 2004.

  1. Which Zener diode could be used for white noise generator?
  2. I read in that Marceli Firlej <[email protected]
    Any one rated substantially above 6 V. 12 V devices are often used. They
    are really 'avalanche' diodes; true Zener operation occurs at breakdown
    voltages below 6 V and is not noisy.

    You get better Gaussian probability distribution if you use two diodes
    from the same batch and couple them to the + and - inputs of an op-amp.
    The output is the difference between the two noises and is good quality
    band-limited Gaussian white noise.
  3. John Larkin

    John Larkin Guest

    The classic RF noise source is something like a 16-volt zener... I'm
    not sure why. A 10 or 12-volt zener, biased at, say, 1 mA, will give
    something like 350 nv per root Hertz noise with pretty good
    statistical properties. Lower currents may do funny things. Low-power
    (small junction) zeners give good stats and are wider band than bigger

    350 nv/rtHz would give you about 50 uV over the audio range, so it'll
    need some amplification!

    So, any old 1/4 watt 12 volt zener should be pretty good. You can buy
    selected "noise diodes" for big bucks if you need guaranteed wideband
    precision; maybe NoiseCom has some appnotes.

  4. Ken Smith

    Ken Smith Guest

    A 12V one in a SOT-23 package.
  5. Robert Baer

    Robert Baer Guest

    And drive them at "low" current (tens of microamps seem to be ideal).
    Zeners tend to get quiet as the operating current approaches the rated
    test current.
    The drop can be rather fast, and every zener (same voltage and batch)
    can have a different current VS noise dropoff characteristic.
  6. I read in that Robert Baer
    I used to think that, but some experiments that Win reported a while
    back indicated that a few milliamps would be a better choice. Anomalous
    effects occurred at very low currents, like 'burst avalanching', IIRC.
  7. John Woodgate wrote...
    Is there a reason to believe the difference should be any better
    than the sum?
  8. Ken Smith

    Ken Smith Guest

    If the low frequency noise modulates the amplitude of the high frequency
    noise, the difference would be much better. I suspect that the ampltude
    of the RF components do decrease when the low frequencies lower the
    "average" voltage.
  9. Guy Macon

    Guy Macon Guest

    If 60 Hz or your local AM radio station is leaking into the two
    circuits, then the difference is better than the sum, but if there
    is no such interference and the noise from the diodes is random,
    it shouldn't matter whether yu take the difference or the sum.

    Also, for those who are planning on using the noise from the
    diodes to generate a random stream of bits, I refer you to the
    "Recurring topics in comp.arch.embedded" web page at
    [ ], which says:

    What's a Von Neuman compensator?

    Take a biased stream of 1's and 0's two at a time.

    If the pair is 00, do nothing.
    If the pair is 01, output a 0.
    If the pair is 10, output a 1.
    If the pair is 11, do nothing.

    (some implementations switch the outputs so that 01 = 1 on odd
    cycles and 01 = 0 on even cycles. The hardware random number
    generator on Intel's latest motherboards uses this method.)

    a Von Neuman compensator will remove the biases caused by switching
    threshholds that are not exactly at 50%, and other biases common to
    the analog to digital stages in typical hardware random number
    generators, and will also remove the biases caused by asymmetrical
    duty cycles and other biases common to typical counter/timer based
    random number generators.
  10. John Larkin

    John Larkin Guest

    Zener noise can be asymmetric in shape. I think that, given the choice
    of summing ++ or +-, the +- might be more symmetric. That sort of
    creeps up on the Central Limit Theorem a little quicker.

  11. John Larkin

    John Larkin Guest

    I've played with regular 1N759-types a bit. At 1 mA, the noise is
    usually pretty Gaussian and spectrally flat, maybe 350 nv/rthz. As
    current goes down, one sometimes sees asymmetric noise, then either
    geiger-counter-like spikes or very ratty sawtooth oscillations. Some
    of the NoiseCom noise zeners do work at very low currents, 10s of uA,
    but they are special parts, I suspect maybe ultra-small-area diodes.

    Agreed that individual types can vary a lot.

    A neon lamp makes lots of nice noise. Old GR noise generators used a
    gas tube of some sort with an external magnetic field for some reason.
    I have schematics somewhere...

    You can just use native opamp noise, but the statistics and spectra
    can be awful.

  12. Jim Thompson

    Jim Thompson Guest

    On Sun, 15 Aug 2004 11:48:01 -0700, John Larkin

    I'd like to see schematics. The magnetic field might be to keep the
    discharge from "dancing" around, perhaps adding a "popcorn" noise
    element ?:)

    ...Jim Thompson
  13. John Larkin

    John Larkin Guest

    The manual's at work; I can post pics to abse Monday. I think there
    are some directions for adjusting the magnet position, too. I bought
    an old GR toob-type noise gen on ebay for some jitter making, but it
    hums a lot and I've got to find time to replace all the 'lytics.

  14. Mike

    Mike Guest

    Assuming the diodes are independent (which certainly seems like a good
    assumption), the density of the sum is the convolution of the densities.
    For the difference, the result is the same.

    I've been playing with various distributions here in MathCad this morning,
    and while the sum and difference may result in different offsets, once the
    offset is removed the distributions appear to be identical.

    -- Mike --
  15. John Larkin

    John Larkin Guest

    Are you saying that, given two basicly identical random signals having
    lopsided probability distributions, the sum won't be lopsided? The
    difference will certainly have a symmetric distribution.

    If the signals were low-duty-cycle random positive pulses, there
    certainly would be a difference in result in probability distribution
    between the summing and differencing schemes. Zener noise can be
    pos/neg asymmetric, getting very pulsey at lower currents and can
    still be visually a little spikey at 1 mA.

  16. Mike

    Mike Guest

    I was indeed saying that, but that was before I had my coffee and started
    thinking. You're absolutely right.

    -- Mike --
  17. John Larkin

    John Larkin Guest

    Well, I typed my query *after* I had my latte, so naturally I had
    unfair advantage. But I did phrase it as a question in case I turned
    out to be wrong.

  18. John Larkin wrote...
    It's highly unlikely that two "nearly identical" zener diodes will
    have basically identical random signals.
  19. John Larkin

    John Larkin Guest

    No two zeners will be identical, but their noise properties should be
    close if they're the same types.

    The issue was John W's observation that the difference between the
    noise of two zeners has better stats than the noise of just one, or
    the noise of the sum. Obviously, either the sum or the difference will
    be better than the noise of just one, but I think he's right in that
    the diff is better than the sum, since zener noise tends to be a bit
    asymmetric, and always in the same direction.

  20. Ken Smith

    Ken Smith Guest

    Consider this:

    If the low frequency components modulate the amplitude of the high
    frequency components, the PSD will still look flat but the noise isn't as
    random as one would like.
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