<html>
  <head>
    <meta http-equiv="Content-Type" content="text/html; charset=UTF-8">
  </head>
  <body>
    <p>An option we are looking at for a work project is the STM32MP1XX
      family of processors. <br>
      <br>
      It has a Arm A7 and Arm M4 in the same chip. The A7 runs Linux,
      and the M4 is for the real-time stuff. They have a bus between the
      two processors so they can look at the other's peripherals and
      memory. <br>
      <br>
      They have a development board that is under $69 you could base a
      Robot around. <br>
<a class="moz-txt-link-freetext" href="https://www.mouser.com/ProductDetail/STMicroelectronics/STM32MP157A-DK1/?qs=9r4v7xj2LnkgRwGGD4%252BIjQ%3D%3D">https://www.mouser.com/ProductDetail/STMicroelectronics/STM32MP157A-DK1/?qs=9r4v7xj2LnkgRwGGD4%252BIjQ%3D%3D</a><br>
    </p>
    <p>It is relatively new, and not for the faint of heart. I would
      recommend getting to understand the M4, the ST CubeIDE development
      tools, and how ST does things with the HAL before you jump into
      this beast. To program the M4, the IDE connects to Linux with
      Ethernet copies it over, and the A7 writes to the M4 memory. <br>
      <br>
      <br>
      When I was doing a resonant power supply we had a PID loop that
      was executed every clock, between 25 and 100 kHz.  Our goal was to
      adjust the input voltage (on a slower loop) so the frequency was
      50 kHz (resonance most efficient) and the clock by clock loop
      would deal with abrupt current changes by changing the frequency
      higher to shed power and make the frequency lower to provide more
      power. We had to compensate for the sample time in the PID
      equation.  It is possible to do. But Python would not do it. <br>
      <br>
      We were dealing with a power supply for a laptop PC. Where to save
      power they would shut parts of the processor and peripherals
      completely off when not needed, lowering the current, and turning
      things on quickly when needed causing the current to increase.
      These events would happen randomly sometimes multiple times a
      second. <br>
      <br>
      Kip<br>
    </p>
    <div class="moz-cite-prefix">On 2/18/21 3:58 PM, David Anderson via
      DPRGlist wrote:<br>
    </div>
    <blockquote type="cite"
      cite="mid:7a13d84e-177d-d38c-8b79-cfe709b45677@smu.edu">
      <meta http-equiv="Content-Type" content="text/html; charset=UTF-8">
      Howdy,<br>
      <br>
      Not to be overly pedantic...<br>
      <br>
      But another solution to these timing problems is to use a micro
      controller that is not running Linux to do your real time stuff. 
      The STM32 Nucleo board I'm using with an RTOS has a context switch
      time which is reliably less than 10 us (which is actually the
      least I can measure).   <br>
      <br>
      I understand the arguments for using Linux.  I've been using it
      both at work and home since the early 1990s.  But not on robots. 
      If the argument is that it makes robotly things easier, I'm not
      sure anyone's experience bears that out.<br>
      <br>
      Right tool for the job, as the old guys use to say.<br>
      <br>
      cheers!<br>
      dpa<br>
      <br>
      <br>
      <div class="moz-cite-prefix">On 2/18/21 11:03 AM, John Swindle via
        DPRGlist wrote:<br>
      </div>
      <blockquote type="cite"
        cite="mid:1830267517.2943411.1613667783411@mail.yahoo.com">
        <meta http-equiv="Content-Type" content="text/html;
          charset=UTF-8">
        <p style="background-color: #f4eaa5;color: #000000
          ;margin:5px;padding: 2px;text-align: left !important;
          align-content: center; display: block; border: 1px solid
          #000000; font-size: large; font-family: sans-serif;"><strong><em
              style="font-size: 11px;"> [EXTERNAL SENDER]</em></strong></p>
        <div style="color:black;font: 10pt Arial, Helvetica,
          sans-serif;">Going back to a previous topic regarding jitter
          in timing intervals.
          <div><br>
          </div>
          <div>Why can't parameters used in calculations be scaled by
            the actual sample interval? I understand 50ms is chosen
            because it gives optimum control without undue overhead.
            When the actual interval is, say, 47ms, why not scale the
            time-related parameters to 47/50 of what they nominally are,
            just for that interval? If the next interval is 74ms, scale
            the parameters to 74/50. Is this impractical? Is the
            uncertainty of measuring the time interval too large? This
            is, if Python says the time interval is 47ms, is the error,
            say, +/- 10ms?</div>
          <div><br>
          </div>
          <div>Related to this: If the sample-to-sample timing error is
            large or jittery, but if you have confidence that the
            long-term timing is 50ms per interval, then why can't the
            current calculations be smoothed, assuming that each
            calculation is jittery due to timing inaccuracy? Does the
            robot really make snap decisions based on a single interval?</div>
          <div><br>
          </div>
          <div>Most CPU datasheets require the clock to have 50 to
            100ppm accuracy. That sounds good, but for audio processing,
            it means streams coming from different sources will beat
            with one another. I have to resample the streams to match
            the actual sample rate of one of the streams. I use a
            metronome signal that each gadget processes. That doesn't
            mean any of the timings are correct (including the timing of
            the metronome). It just means they are all forced to look
            like they are the same. I say this because resampling is
            like what I'm suggesting: Scale the calculations to the
            actual interval.</div>
          <div><br>
          </div>
          <div>Am I missing something here?</div>
          <div><br>
          </div>
          <div>John Swindle</div>
          <div><br>
          </div>
          <div>
            <div
              style="font-family:arial,helvetica;font-size:10pt;color:black"><font
                size="2">-----Original Message-----<br>
                From: Murray Altheim via DPRGlist <a
                  class="moz-txt-link-rfc2396E"
                  href="mailto:dprglist@lists.dprg.org"
                  moz-do-not-send="true"><dprglist@lists.dprg.org></a><br>
                To: Chris N <a class="moz-txt-link-rfc2396E"
                  href="mailto:netterchris@gmail.com"
                  moz-do-not-send="true"><netterchris@gmail.com></a>;
                <a class="moz-txt-link-abbreviated"
                  href="mailto:dprglist@lists.dprg.org"
                  moz-do-not-send="true">dprglist@lists.dprg.org</a> <a
                  class="moz-txt-link-rfc2396E"
                  href="mailto:dprglist@lists.dprg.org"
                  moz-do-not-send="true"><dprglist@lists.dprg.org></a><br>
                Sent: Thu, Feb 11, 2021 4:06 am<br>
                Subject: Re: [Dprglist] PID-tuned Clock in Python?<br>
                <br>
                <div dir="ltr">On 11/02/21 3:13 pm, Chris N wrote:> I
                  don’t think you have a “clock accuracy” issue.  I’m
                  pretty sure<br clear="none">
                  > the hardware clocks, as in crystal + PLL etc., in
                  things like the<br clear="none">
                  > Pi, are plenty accurate for our needs.<br
                    clear="none">
                  <br clear="none">
                  Hi Chris,<br clear="none">
                  <br clear="none">
                  First, thanks very much, I appreciate your thoughtful
                  reply.<br clear="none">
                  <br clear="none">
                  Yes, I understand that it's not the accuracy of the
                  Raspberry Pi's<br clear="none">
                  system clock that is in question, but the accuracy of
                  time loops<br clear="none">
                  when implemented in a Python application using a
                  single or multiple<br clear="none">
                  threads, executing within a time-sharing operating
                  system like Linux.<br clear="none">
                  <br clear="none">
                  This is clearly where a microcontroller (like an
                  Arduino, STM32,<br clear="none">
                  ESP32, Pico RP2040, etc.) has its advantages. As I
                  noted, my<br clear="none">
                  MacBook was actually worse than my Raspberry Pi, which
                  in this<br clear="none">
                  light is perhaps not so surprising, i.e., the
                  MacBook's processor<br clear="none">
                  is a lot busier.<br clear="none">
                  <br clear="none">
                  My 3.5GHz multi-core i7 workstation is clearly giving
                  at least a<br clear="none">
                  whole core to the Python application and not
                  struggling, so its<br clear="none">
                  performance is very close to expected norms. But
                  that's not the<br clear="none">
                  case on the Pi, where I'll actually be using this
                  timing loop.<br clear="none">
                  <br clear="none">
                  [...]<br clear="none">
                  > Now regarding the PID idea:<br clear="none">
                  > <br clear="none">
                  > Say we have a loop that we want to run at a fixed
                  rate. Say 20Hz / 50.0ms.<br clear="none">
                  > <br clear="none">
                  > Using PID to improve the timing accuracy of such
                  loops is certainly<br clear="none">
                  > an interesting idea, but I believe PID makes
                  things worse in this case.<br clear="none">
                  <br clear="none">
                  I'm not clear that this is the case yet. My
                  observation was the the PID<br clear="none">
                  loop was improving things somewhat significantly, but
                  more on that below.<br clear="none">
                  <br clear="none">
                  > See output from your clock_test.py below.   I ran
                  this on my Pi 3.<br clear="none">
                  > I deliberately used a tough background load to
                  amplify the effect <br clear="none">
                  > (stress –vm 4 –vm-bytes 128M) , but even with
                  normal loads the <br clear="none">
                  > negative effect of the PID can be observed, just
                  the errors would<br clear="none">
                  > be much smaller.<br clear="none">
                  <br clear="none">
                  Agreed, understood.<br clear="none">
                  <br clear="none">
                  > In line 6 you can see the 27ms error. <br
                    clear="none">
                  > [...]<br clear="none">
                  > The reason I think PID is a bad idea here is
                  because the nature of<br clear="none">
                  > the disturbance is simply too random and its very
                  intermittent. <br clear="none">
                  > The best you can do really is to use basic loop
                  timing logic to <br clear="none">
                  > ensure that the next iteration starts at the
                  right time, despite<br clear="none">
                  > the fact that this iteration took an unusual
                  amount of time or <br clear="none">
                  > sleep() took an unusual amount of time. With PID
                  you end up  > over-compensating and you are
                  effectively hurting the timing of<br clear="none">
                  > subsequent iterations.<br clear="none">
                  <br clear="none">
                  I certainly agree based on both logic and my
                  observations that the<br clear="none">
                  disturbances are intermittent, and from the
                  perspective of the<br clear="none">
                  Python program (i.e., it not knowing what else is
                  running), would<br clear="none">
                  appear entirely randomly.<br clear="none">
                  <br clear="none">
                  But I'm not clear the PID loop is over-compensating in
                  those cases.<br clear="none">
                  When a disturbance is within one 20Hz clock cycle,
                  there is no real<br clear="none">
                  fix for that in software. I entirely agree with you in
                  those cases.<br clear="none">
                  But when the CPU is really busy with some longer-term
                  process and<br clear="none">
                  that activity extends say over several seconds, the
                  PID (I believe)<br clear="none">
                  would tend to compensate better than with no PID.<br
                    clear="none">
                  <br clear="none">
                  > There are straight forward ways to deal with the
                  fact that  <br clear="none">
                  > time.sleep(x) doesn’t sleep for exactly x amount
                  of time, and the<br clear="none">
                  > fact that the amount of work which needs to be
                  done every iteration<br clear="none">
                  > is not 100% constant.  A python version of such a
                  fixed-rate loop > is here: <a shape="rect"
                    href="https://github.com/nettercm/timing%C2%A0%C2%A0"
                    target="_blank" moz-do-not-send="true">https://github.com/nettercm/timing  
                  </a>I typically use<br clear="none">
                  > similar loop timing logic in other languages and
                  sometimes even <br clear="none">
                  > on a microcontroller.<br clear="none">
                  > <br clear="none">
                  > In pseudo python it looks as follows. [...]<br
                    clear="none">
                  If I'm reading your code correctly, this is kinda what
                  I was doing<br clear="none">
                  before I adopted the PID into the mix. I had a fixed
                  trim value on<br clear="none">
                  the Rate, and when the 50ms loop was consistently a
                  bit slow or a<br clear="none">
                  bit fast I'd just modify the trim value. It was trying
                  to auto-correct<br clear="none">
                  that trim value that led me to using a PID loop. Now,
                  a fixed trim<br clear="none">
                  would obviously not take into account those nasty
                  intermittent surges<br clear="none">
                  or lags, but is certainly simpler to implement and a
                  lot less<br clear="none">
                  mysterious in use than a PID loop.<br clear="none">
                  <br clear="none">
                  So if the PID turns out to be causing actual problems
                  as you suggest,<br clear="none">
                  I can set the flag to disable it, and maybe
                  re-implement the fixed<br clear="none">
                  trim. I've been setting the PID's kp constant from an
                  RGB LED<br clear="none">
                  potentiometer like:<br clear="none">
                  <br clear="none">
                    <a shape="rect"
                    href="https://shop.pimoroni.com/products/rgb-potentiometer-breakout"
                    target="_blank" moz-do-not-send="true">https://shop.pimoroni.com/products/rgb-potentiometer-breakout</a><br
                    clear="none">
                  <br clear="none">
                  so if I end up disabling the PID I could manually set
                  the trim from<br clear="none">
                  the potentiometer, then just fix it in my YAML
                  configuration. This<br clear="none">
                  would be a "best guess" on a fixed, constant trim
                  value.<br clear="none">
                  <br clear="none">
                  I think we both agree that there's nothing anyone can
                  do to get a<br clear="none">
                  truly consistent loop using Python on Linux, but if
                  I'm willing to<br clear="none">
                  accept that those intermittent <50ms disturbances
                  are also unfixable,<br clear="none">
                  then the PID does still at least (IMO) manage the
                  longer lags and<br clear="none">
                  surges of the clock. This *seems* to be what I'm
                  seeing in the<br clear="none">
                  console outputs.<br clear="none">
                  <br clear="none">
                  The whole subject of PID is all a bit dream-catcher,
                  Voynich Manuscript,<br clear="none">
                  ouija board, patchouli incense kind of thing...
                  verra-mysterious.
                  <div class="yqt4755843592" id="yqtfd77735"><br
                      clear="none">
                    <br clear="none">
                    Cheers,<br clear="none">
                    <br clear="none">
                    Murray<br clear="none">
                    <br clear="none">
                    <br clear="none">
...........................................................................<br
                      clear="none">
                    Murray Altheim <murray18 at altheim dot com> 
                                        = =  ===<br clear="none">
                    <a shape="rect"
                      href="http://www.altheim.com/murray/"
                      target="_blank" moz-do-not-send="true">http://www.altheim.com/murray/
                    </a>                                    ===  ===<br
                      clear="none">
                                                                       
                                    = =  ===<br clear="none">
                        In the evening<br clear="none">
                        The rice leaves in the garden<br clear="none">
                        Rustle in the autumn wind<br clear="none">
                        That blows through my reed hut.<br clear="none">
                                -- Minamoto no Tsunenobu<br clear="none">
                    <br clear="none">
                    _______________________________________________<br
                      clear="none">
                    DPRGlist mailing list<br clear="none">
                    <a shape="rect"
                      ymailto="mailto:DPRGlist@lists.dprg.org"
                      href="mailto:DPRGlist@lists.dprg.org"
                      moz-do-not-send="true">DPRGlist@lists.dprg.org</a><br
                      clear="none">
                    <a shape="rect"
                      href="http://lists.dprg.org/listinfo.cgi/dprglist-dprg.org"
                      target="_blank" moz-do-not-send="true">http://lists.dprg.org/listinfo.cgi/dprglist-dprg.org</a><br
                      clear="none">
                  </div>
                </div>
              </font></div>
          </div>
        </div>
        <br>
        <fieldset class="mimeAttachmentHeader"></fieldset>
        <pre class="moz-quote-pre" wrap="">_______________________________________________
DPRGlist mailing list
<a class="moz-txt-link-abbreviated" href="mailto:DPRGlist@lists.dprg.org" moz-do-not-send="true">DPRGlist@lists.dprg.org</a>
<a class="moz-txt-link-freetext" href="http://lists.dprg.org/listinfo.cgi/dprglist-dprg.org" moz-do-not-send="true">http://lists.dprg.org/listinfo.cgi/dprglist-dprg.org</a>
</pre>
      </blockquote>
      <br>
      <br>
      <fieldset class="mimeAttachmentHeader"></fieldset>
      <pre class="moz-quote-pre" wrap="">_______________________________________________
DPRGlist mailing list
<a class="moz-txt-link-abbreviated" href="mailto:DPRGlist@lists.dprg.org">DPRGlist@lists.dprg.org</a>
<a class="moz-txt-link-freetext" href="http://lists.dprg.org/listinfo.cgi/dprglist-dprg.org">http://lists.dprg.org/listinfo.cgi/dprglist-dprg.org</a>
</pre>
    </blockquote>
  </body>
</html>