Shop OBEX P1 Docs P2 Docs Learn Events
Woodward PWM filter — Parallax Forums

Woodward PWM filter

rbehmrbehm Posts: 102
in General Discussion
We are all using PWM to create analog outputs.
These need filters to smooth them and remove ripple. But the simple RC filters are either slow or no very effective.
Here is a clever solution from Winfield Hill (of AoE fame) based on an idea from Stephen Woodward. It mere adds another RC to gain much better ripple rejection without increasing the settling time.

https://dropbox.com/s/xvwjvlun1rg1vca/4x.25_PWM_ripple.pdf?dl=0

Comments

  • MicksterMickster Posts: 2,610
    Hey I used to know Steve Woodward. He's also an airgun hobbyist. We had plans to develop an electronic airgun, some years ago

    :cool:
  • JonnyMacJonnyMac Posts: 8,926
    edited 2019-10-17 14:24
    Mickster wrote: »
    Hey I used to know Steve Woodward. He's also an airgun hobbyist. We had plans to develop an electronic airgun, some years ago

    :cool:

    My friend, Rick Galinson, has built three or four of these Propeller-powered paintball miniguns. This video was shot at a Parallax event.

    https://youtube.com/watch?v=lOkw1lMGsfw
  • MicksterMickster Posts: 2,610
    Fun, fun, fun...I want/need one :lol:

    Is it pure coincidence that Galinson is similar to Gatlin gun? :cool:
  • MicksterMickster Posts: 2,610
    There's a lot of interest in less-than-lethal home defense...I'd strap one to the foot of the bed :cool:
  • Tracy AllenTracy Allen Posts: 6,656
    edited 2019-10-18 00:06
    Great little filter, insightful. I found Stephen Woodward's short and sweet publication in EDN here...
    Cancel-PWM-DAC-ripple-with-analog-subtraction
  • TubularTubular Posts: 4,622
    Yeah thats neat

    With P1's counters providing an optional complementary output, all thats needed are one or two extra passive components (assuming enough pins are available)

  • Beau SchwabeBeau Schwabe Posts: 6,547
    edited 2019-10-18 03:54
    That filter does some other interesting things if you just put a sine wave, triangle wave, or saw wave to it.
  • jmgjmg Posts: 15,145
    rbehm wrote: »
    We are all using PWM to create analog outputs.
    These need filters to smooth them and remove ripple. But the simple RC filters are either slow or no very effective.
    Here is a clever solution from Winfield Hill (of AoE fame) based on an idea from Stephen Woodward. It mere adds another RC to gain much better ripple rejection without increasing the settling time.

    Interesting that Spice shows the waveform of his dual polarity lead/lag 2R/2C filter, to be an exact clone of a 2nd order LPF, using R/C/R/C
    - The traces literally sit on top of each other.

    ie he can throw away the Analog Switch, and get exactly the same response result :)
    Tubular wrote: »
    With P1's counters providing an optional complementary output, all thats needed are one or two extra passive components (assuming enough pins are available)

    See above, all you need the the extra passives, no need to use another pin ...

  • Tracy AllenTracy Allen Posts: 6,656
    edited 2019-10-18 18:11
    Reinhardt,
    Incidentally, thanks for the heads' up about the forthcoming sequel, Art of Electronics -- The x-Chapters. Who would not want to taste the frosting on the best electronics book ever written?

    Where did you find the link to that specific snippet? A quick search found this on EEblog, Winfield Hill himself participating, with links to other draft snippets in Dropbox.

    Preorder expected sometime in 1Q20.

  • Phil Pilgrim (PhiPi)Phil Pilgrim (PhiPi) Posts: 23,514
    jmg wrote:
    See above, all you need the the extra passives, no need to use another pin ...
    I've looked "above" but don't find what you're referring to. Without an active inverter -- whether it be external or via a second pin -- where does the inversion signal come from?

    -Phil
  • Tracy AllenTracy Allen Posts: 6,656
    We should note that this PWM filter references conventional PWM, constant period, variable duty cycle, like this one,
    fast-pwm-up-to-1-6-mhz
    not the duty mode of the cog counters.

    However, it should work with the duty mode at 50% duty cycle and quite a bit around that, so long as the joined periods do not stray too far from 1/clkfreq. Well, it could smoothing duty mode PWM -> analog at clkfreq=5MHz, for lower power operation.
  • jmgjmg Posts: 15,145
    Phil Pilgrim (PhiPi) wrote: »
    jmg wrote:
    See above, all you need the the extra passives, no need to use another pin ...
    I've looked "above" but don't find what you're referring to. Without an active inverter -- whether it be external or via a second pin -- where does the inversion signal come from?
    There is no need for any inversion signal - it can be discarded entirely.
    'above' was my comment right above the 2nd quote.

    Spice says this `new filter` is actually identical to a standard 2nd order, simplest form of identical R,C,R,C
    As per attached : Compares 'woodward-pwm-filter' with a simpler 2nd order and says 0.0V difference.


    Spice_2ndOrder_Compare_Diff2R2C.PNG
    1848 x 874 - 61K
  • rbehmrbehm Posts: 102
    Tracy Allen wrote: »
    Reinhardt,
    Incidentally, thanks for the heads' up about the forthcoming sequel, Art of Electronics -- The x-Chapters. Who would not want to taste the frosting on the best electronics book ever written?

    Where did you find the link to that specific snippet? A quick search found this on EEblog, Winfield Hill himself participating, with links to other draft snippets in Dropbox.

    Preorder expected sometime in 1Q20.

    Hi Tracy,

    Win Hill mentioned this in a thread on usenet sci.electronics.design.
  • Tracy AllenTracy Allen Posts: 6,656
    jmg,
    Did you also run your Spice simulation for PWM step from x% to y%, that is, how does it compare with figure 4x.156 in Hill's analysis? Tradeoffs between ripple and response.
  • jmgjmg Posts: 15,145
    Tracy Allen wrote: »
    jmg,
    Did you also run your Spice simulation for PWM step from x% to y%, that is, how does it compare with figure 4x.156 in Hill's analysis? Tradeoffs between ripple and response.
    The Spice results I have are at 20% duty, and they track exactly, in both rise time and ripple. ie the Filters have precisely identical transfers.
    Seems there is no free lunch, and 2R and 2C give a second order response, no matter how wired.
Sign In or Register to comment.