Connect with us

Filtering ideas

Discussion in 'Electronic Basics' started by [email protected], Mar 14, 2006.

Scroll to continue with content
  1. Guest


    I'm new at signal processing and was wondering if there's any new
    innovative methods in detect very weak signals over noise. Signal
    example less noise ->

    The signal lasts about 1/2 ms and a fraction of a uV. Noise could
    possible be hundreds times greater than signal. One advantage is the
    circuit controls exactly when the signal fires. So at least the
    circuitry knows exactly when to detect the signal. I would like to
    limit the sampling time to 0.5 second.

    The input coil is 3.5" dia., 3.5" tall, 1000 turns of 24G, 23 ohms. I
    calculate 60mH and guessing at about 8000 pF.

    Perhaps a common 1st stage would be a low noise op-amp followed by some
    filtering-- perhaps an Elliptic filter is good at filtering out high
    end ->
    Perhaps a PLL would make a good 3rd stage. I'm wondering if there are
    any other methods. Not sure if DSP could work since the STN is broad
    unless we're talking about a 24 bit ADC-- sounds like a lot of bits to

  2. w2aew

    w2aew Guest

    Depending on "what" you need to measure on your signal, it might be a
    good application for a lock-in amplifier. This technique is often used
    to measure very small signals buried in noise.
  3. A more modern term is "synchronous demodulation".

    Many thanks,

    Don Lancaster voice phone: (928)428-4073
    Synergetics 3860 West First Street Box 809 Thatcher, AZ 85552
    rss: email:

    Please visit my GURU's LAIR web site at
  4. Guest

    Anyone have any thoughts on this Elliptic filter ->

    It's the highest order Elliptic active filter I could find. It's not
    bad, but would need at least 4 stages.

  5. Joel Kolstad

    Joel Kolstad Guest

    You can make as high of an order filter as you like out of one single op-amp;
    the problem is that the sensitivity of the filter's response to component
    variations and op-amp imperfections increases super-linearly with respect to
    the order of the filter you're trying to build. This is why you tend to see
    most "high performance" active filters built out of bi-quad casecades (with
    typically two op-amps per section) and topologies with only one op-amp tend to
    either need trimming or else somewhat more relaxed specs. (And I've never
    seen someone try to pull off, e.g., a 4th order filter with one op-amp in
    anything other than SPICE, but who knows... somebody probably has!)

    As such, I'd suggest you take your circuit there and run it through the a
    Monte-Carlo simulation in SPICE and see if you like the results!

  6. Opamp elliptics are a bit of a waste of time unless you have a particular
    frequency to notch out.
    Certainly useless if you want that 1MHz break frequency.
    I'd guess a 30kHz filter might be appropriate.
    Would suggest just stacking some simple 2nd or 3rd order Butterworth
  7. Fred Bartoli

    Fred Bartoli Guest

    If you have a good DSO then you have the lock-in of choice.
    Lots of averaging with maybe a small preamplifier front end if the signal is
    too low.
    I have no pb looking at sub-uV signals directly feeding them to my scope (no
    preamp because I'm a lazybones). A small RC to BW limit the noise and 4K
    averaging nicely do the rest.
  8. Guest

    Thanks for input. I see that you're 100% correct. Have you tried a
    program called "Filter Solutions 2006?" It's a dream program. I'm very
    disappointed in the Elliptic filters. The best bandpass filter I found
    using a Pos SAB active implementation with lower & higher corner freq.
    = 1KHz & 10KHz is a Chebyshev I. It requires only 4 op-amps for a 7th
    order bandpass. Even after entering LM741 amp parameters I still get
    -208db at 60Hz and -89db at 25KHz. Nearly 1db at 1KHz and 10KHz.
    It's unbelievable! I'm curious what the disadvantages to a Pos SAB
    active implementation since it uses less than half the parts-- 4
    op-amps compared to 10. Here's the Spice Net List:

    V1 1 0 AC 1
    *V1 1 0 PULSE 0 1 0 5.033E-08 0
    R111 1 2 4E+04
    C112 2 7 1.069E-08
    R113 2 0 2500
    R109 2 3 1E+04
    C101 3 4 1.504E-08
    C102 4 5 1.828E-09
    R103 5 0 5.737E+05
    R104 4 7 2500
    R106 6 0 1E+04
    R107 6 7 1795
    C108 4 0 2.183E-09
    X101 5 6 7 OPAMP PARAMS: A=3E+11 B=4E+06
    R211 7 8 2E+04
    C212 8 13 6.06E-09
    R213 8 0 5000
    R209 8 9 1E+04
    C201 9 10 1.001E-08
    C202 10 11 7.2E-09
    R203 11 0 1.051E+05
    R204 10 13 5000
    R206 12 0 1E+04
    R207 12 13 2821
    C208 10 0 1.469E-09
    X201 11 12 13 OPAMP PARAMS: A=3E+11 B=4E+06
    R311 13 14 2E+04
    C312 14 19 6.288E-09
    R313 14 0 5000
    R309 14 15 1E+04
    C301 15 16 7.512E-09
    C302 16 17 1.434E-08
    R303 17 0 7.002E+04
    R304 16 19 5000
    R306 18 0 1E+04
    R307 18 19 2269
    C308 16 0 9.47E-10
    X301 17 18 19 OPAMP PARAMS: A=3E+11 B=4E+06
    R401 19 20 1E+04
    C402 20 21 1.451E-09
    R403 21 0 1.557E+04
    R404 20 23 3.513E+04
    R406 22 0 1E+04
    R407 22 23 7.787E+04
    C408 20 0 1.451E-09
    X401 21 22 23 OPAMP PARAMS: A=3E+11 B=4E+06
    ..AC DEC 200 0.1 2.5E+04
    ..PLOT AC VDB(23) -200 0
    ..PLOT AC VP(23) -200 200
    ..PLOT AC VG(23) 0 0.005
    ..TRAN 0.05 10 0
    ..PLOT TRAN V(23) -0.5 0.3

    *OpAmp Complex Model 1=+in 2=-in 10=Vo A=G*Bw Product(Hz)
    ..SUBCKT OPAMP 1 2 10
    +PARAMS: A=1.E+12 B=1.E+06
    Rin 1 2 2.e+06
    Cin 1 2 5.e-12
    E1 3 0 1 2 1.0
    Rp 3 4 1.0
    Cp 4 0 {1/(2*PI*B)}
    E2 5 0 4 0 {A/B}
    Rout 5 10 500.
  9. Jim Thompson

    Jim Thompson Guest


    Designing elliptic filters CAN be a bit of a headache if you're
    winging it.

    If you really want to do it right may I suggest...

    "Synthesis of Filters"
    Herrero and Willoner
    Prentice Hall EE Series © 1966
    Library of Congress 66-27547

    Once upon a time I had the whole General Parameter Filters chapter
    programmed into an ancient tape-card-reading hp calculator (or was it
    TI... it was early '70's).

    Probably ought to redo it in Larkin-suggested PowerBasic.

    ...Jim Thompson
  10. Joel Kolstad

    Joel Kolstad Guest

    It sure is! Unless you're cooling the with, e.g., liquid helium, thermal
    noise alone is well above -208dB in any reasonable bandwidth! :)
    Most likely sensitivity to component value variations -- run a Monte Carlo
    simulation on each and see.
  11. Yes, the prog is excellent. I recognised the circuit drawing. Don't believe
    those incredible dB figures. In normal practice you're lucky to get anything
    down to -80.
    To spot those filter differences, initially pick a SAB type circuit design,
    select a component on it, then on the drop down box go for a multiple run
    Monte-Carlo analysis using random values for all the R's and C's. You'll see
    some -really- ugly spreads on that original pristine response. Then do the
    same using leapfrog and GIC implementations.
  12. Mac

    Mac Guest

    Do you just want to detect the signal? Why bother if you already know when
    to expect it? Or is this an experiment to see whether detection is

    Anyway, I have three suggestions:
    1) Filter out-of-band noise.
    2) Average, if possible
    3) Use correlation to detect

    If you just want to observe the signal, you can probably just use an
    oscilloscope with averaging, since you have a trigger.

    If you really do want to detect the signal, well, I'm not a signal
    processing expert either, but correlation is probably the best way to
    detect it. I guess you would digitize, then correlate.

    Also, any properties of the signal which are known will help you detect
    it. For example, is it repeated regularly? If so, and if the period is
    known, your detection can be improved dramatically.

    You might want to try some other newsgroups. comp.arch.dsp?

  13. Bob Masta

    Bob Masta Guest

    Assuming that the stimulus and reponse are repetitive,
    the best way to handle this is with synchronous averaging.
    The idea is that you sample a series of response time points that
    are triggered by the stimulus. If there were no noise,
    then each set of points would look identical. The noise
    causes each actual response to be wildly different because
    of the added trash, but if you average enough of these
    together the true response appears. You get a 3 dB
    improvement for each doubling of the number of responses

    Note that this kind of averaging is done on a point-by-point
    basis: Point 1 of response 1 is averaged only with point 1
    of response 2, 3, 4, etc, independent of other time points.

    The beauty of synchronous averaging is that it does not
    affect the waveform in any way... there is no rounding, no
    phase shift, no step response, no ringing. But it may take
    a lot of averages to get the desired smoothness.

    This is the way researchers record evoked neural
    responses from scalp electrodes. The desired response
    is totally buried in huge amounts of noise, but because it
    is repeatable, it can be extracted after 100s of passes
    are averaged.

    I have some on-line tutorials about this at
    The synchronous averaging titles are about halfway
    down the page. Here is a direct link to the first
    At the bottom of each article are links to next
    and prior, etc.

    Hope this helps!

    Bob Masta

    D A Q A R T A
    Data AcQuisition And Real-Time Analysis
    Home of DaqGen, the FREEWARE signal generator
  14. Ken Smith

    Ken Smith Guest

    Since everyone else is talking about the signal, I'll ask a question about
    the "noise". Is it really random or is it mostly harmonics of the local
    mains frequency?

    If it is harmonics of the mains, you can help things a lot by timing the
    experiment such that it slides over one cycle of the mains in every N
    samples and then averaging N samples together. This tends to reject the
    low harmonics of the mains.

    For the mains frequency its self, you could also add a notch filter to
    remove it before the ADC.
  15. Guest


    I'm looking at an op-amp graph of total noise vs source resistance. The
    noise increases with an increase of source resistance. Does this also
    apply to reactance?

    The way I understand it is resistance is thermal noise, but reactance
    from inductors and caps don't caused any noise. Real L's and C's have
    some R, but in that sense they have noise.


    Some details:

    I have an input coil antenna that's 22 ohms R and about 10 mH. Nothing
    fancy, just a round loop coil with a lot of turns. Typical frequencies:
    from 1KHz to 1MHz. At say 1MHz the reactance would be just over 6 M
    ohms. The chart, "Total Noise vs Matched Source Resistance" at,C1,C1154,C1009,C1021,P1234,D3480

    According to the chart a 6 M ohm source resistance would have
    outrageous noise. Is it safe to say the 6 M ohm reactance (for 1MHz
    signals) will cause no noise? Rather the noise will come from the 22
    ohms R? According to the chart, 22 R at 1KHz is ~ 1 nV/SqrtHz. Not sure
    what it would be at 1 MHz, but it seems lower than 1KHz.

    The charts is for "Matched Source Resistance." The coil is one
    continuous resister, so if I place the 2 coil leads directly to the
    op-amp then is that considered matched? I'm using a typical
    differential op-amp has two input R's. So if the coil's total R is 22
    ohms then would that equate to two 11 ohm input R's? Very confusing,

    Many thanks,

  16. Noise is due to thermal noise *and* shot curent noise. The shot noise
    current of an amplifier is droped accross the source impedance (and
    Rin). If the source is an inducter, the voltage noise caused by the amp
    input current shot noise will increase with frequency despite the fact
    that the inductor itself does not generate any noise.

    Kevin Aylward B.Sc.
    SuperSpice, a very affordable Mixed-Mode
    Windows Simulator with Schematic Capture,
    Waveform Display, FFT's and Filter Design.

    "There are none more ignorant and useless,than they that seek answers
    on their knees, with their eyes closed"
  17. Ban

    Ban Guest

    W R O N G
    You are off by several orders of magnitude. it's 63k. But even this is
    impossible, because the capacity between the windings will not allow you to
    reach much more than 600 ohms. That would be a Q of 30. This coil would be
    useless above a few kHz. 1MHz will be attenuated by 40dB.
    But apart from that did you ever see a 10mH coil?

    The chart, "Total Noise vs Matched Source
    You do not understand that table either. Better to read up in Win's book on
    noise. too long to explain here.
    You just have too little knowledge to understand. Read AoE again.
  18. LT1028 is a good and expensive bipolar OP-amp.
    But at high Z it is a bad choise.
    Better to use a JFET- or MOSFET-device at high Z.

    Nature is normally very hard to fool!
    I suppose we are talking about a LC-circuit near resonance.
    That means the LC-circuit will act as a transformer for R,
    and it will show a noise source of the same magnitude as its
    equivalent parallell resistance, which can be 6Mohm.
    If it is far from resonance, it will still be a bad match to a bipolar
    And if it is an antenna, there may also be noise in the rf-field that it
    picks up.
  19. Guest

    Kevin Aylward wrote:

    Let me see if I understand this. A pure inductor cause shot noise, but
    no thermal noise? How can I calculate a differential op-amp's output
    noise if the source is mostly inductive? ->

    Note that in my case R1 & R2 are also reactive-- 22 ohm resistance and
    6 M ohm inductance.

    The capacitance is extremely small. So there's no resonance.

    I presume that the noise would be a lot greater if it was 6 M ohm
    resisters rather than inductors.

    Thanks for any help,
  20. --- snip ---

    A pure inductor does not cause shot noise, but the OP-amp does.
    The source impedance would shunt this shot noise, so when Z-source is
    low the shot noise is not dominant.
    When Z-source is high, it is!

    If X_inductive = 6 Mohm at 1 MHz then L would be ~ one Henry.
    ( Or do you mean milli-ohm? )
    The same value, 6 Mohm, for X_capacitive would imply C ~ 0.03 pF!
    That is small indeed, but I do not understand how it could be.
    My guess is that it would be about the same at the output of the OP-amp!
    The input capacitance of the amplifier can, if properly matched, balance
    the reactance. But then we have a resonant circuit.
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