USA Fencing SEMI Statement Regarding Saber Machine Timing

Discussion in 'Fencing Discussion' started by mtwieg2, Mar 22, 2018.

  1. mtwieg2

    mtwieg2 DE Bracket

    Joined:
    Nov 24, 2015
    Messages:
    332
    Likes Received:
    64
    Surprised there apparently hasn't been a thread about this yet.

    I had heard from a friend that a specific event in the NE area precipitated this announcement, but I can recall at least one previous occasion where competitors became convinced that the scoring machines in use at a competition had incorrect timing (regardless of stickers). That specific incident resulted in one competitor being black carded.

    The basic issue is that there is no easy way to demonstrate that a scoring machine has the correct firmware/timings. Some machines are programmed to give certain audible/visual indicators which are specific to firmware versions, which is a good solution (assuming the officials know how to interpret those indicators). For other machines, the machine provides no help, leading fencers to make up their minds based on "feel." The rules do not prescribe any resolution to such matters.

    I'm curious about other's feelings on how widespread of an issue this is, and potential ways to address it.
     
    Last edited: Mar 22, 2018
  2. BrodeurNJD30

    BrodeurNJD30 Made the Cut

    Joined:
    Dec 10, 2005
    Messages:
    270
    Likes Received:
    22
  3. jkormann

    jkormann Podium

    Joined:
    Nov 8, 2002
    Messages:
    2,911
    Likes Received:
    182
    The screaming in the video was the best.
     
    Privateer likes this.
  4. mtwieg2

    mtwieg2 DE Bracket

    Joined:
    Nov 24, 2015
    Messages:
    332
    Likes Received:
    64
    I had heard about this video but couldn't find it, thanks.

    Does this apply to all SG machines, or just the SG12-ST?
     
  5. AStoddard

    AStoddard Rookie

    Joined:
    Dec 2, 2015
    Messages:
    20
    Likes Received:
    8
    Inspiration for a simple lockout timing test comes from venuswasaflytrap on reddit.

    Maximum standard length of a sabre is 1.05m from pommel to tip.

    Formula for time taken to free fall under gravity is (2d/g)^0.5 (d is distance of drop, g is 9.8 m/s2) https://en.wikipedia.org/wiki/Equations_for_a_falling_body

    Timing change for lock out was from 120ms +/- 10ms to 170 ms +/- 10 ms.

    Take 2 body cords, and 2 sabres and connect them directly to the box laying them flat.

    The idea is to drop the A line clips onto the guards of the opposite weapon releasing them simultaneously with a difference in drop height such that ~ 150ms elapses between them striking the guards. Both lights will come on under new timings but the one that fell from higher up will be locked out under the old timings.

    Very conveniently if I have my math correct the difference between a drop of one sabre length and a half sabre length is 135 ms. If you drop from a sabre height and 3/5 of a sabre height you have 104ms and from 2/5 of a sabre height compared to a full sabre height 170ms.

    Use a third sabre to make two measurements. First measure a sabre length of bodycord A line (the one going to the clip) and fold it in half to find a half-sabre length a A line. Hold one bodywire clip a sabre height above the opposite guard and with your other hand let the other clip dangle a half sabre length below it above the opposite guard. Drop them both simultaneously and then continue to test with both slightly more and slightly less difference in drop height. If you can consistently get a double light with the lower clip falling from less than half a sabre height you are on the old timings. If you can consistently get a double light with the lower clip falling greater than half a sabre length above the guard you are on new timings.
     
  6. mtwieg2

    mtwieg2 DE Bracket

    Joined:
    Nov 24, 2015
    Messages:
    332
    Likes Received:
    64
    Yeah there are various ways you could use gravity/distance to make a rough timing test, though I doubt human hands would be able to get better than +/-10ms accuracy. If we assume that all machines have either 120ms or 170ms lockout (with very little deviation from those two), then such a test is reasonable. But if we allow the possibility of there being wider variation in the lockout timing (like 157ms or 135ms) then you need a much more accurate test. I actually built such a tester recently, which produces delays at 110, 130, 160, and 180 ms.

    But in reality, it's very unlikely that the timings on any modern scoring machine have actually "drifted" substantially, assuming they use crystal oscillators as a clock source (god help us if any machines out there used RC oscillators...). Like maybe 0.1% over the device's lifespan. There's no reason a scoring machine's accuracy should be off by more than a millisecond.
     
  7. jdude97

    jdude97 Podium

    Joined:
    Apr 10, 2013
    Messages:
    1,802
    Likes Received:
    195
    But what about tolerance in the manufacturing?
     
  8. mtwieg2

    mtwieg2 DE Bracket

    Joined:
    Nov 24, 2015
    Messages:
    332
    Likes Received:
    64
    For modern machines (meaning they run on microcontrollers, not analog timing circuits) there's really only two factors that determine the timings: the clock source and the code.

    I'll give a little background on what I mean by "the clock," in case anyone is interested (otherwise skip to the bottom of this post).
    Modern computers (including the tiny microcontrollers in scoring machines, our phones, and the superclusters at Facebook) are what are referred to as synchronous state machines. This means that every operation happens at regular time intervals. This interval is determined by a signal called the clock. For example a modern computer may have a processor with a clock frequency of 4GHz, which means it can change states every 250 picoseconds. The processors in scoring machines are way simpler and slower, and will probably use a clock in the neighborhood of 10MHz, meaning that the processor can change states every 100 nanoseconds.

    A program has no way to perceive or measure the passage of time directly. But programs are extremely good at counting events. So if it detects two events, it can tell you exactly how many times it has changes state during that time. When the clock is a known, constant frequency, then you also know that each state change took one clock period, and therefore you can derive an actual measurement of time. For example, if my processor has a 10MHz clock and the processor counts ten thousand state changes between two events, then that means that those two events happened one millisecond apart (10,000/10,000,000).

    The precision of this time measurement is limited by the clock frequency, which is going to be negligible in our case (like <1 microsecond). Waaayy back in the old days, scoring machines used the 50/60Hz mains as a clock, meaning your precision could be as poor as 20ms. I don't know if these are sold anymore, but you certainly shouldn't see any in competition...

    The accuracy of the timing measurements is also limited by the accuracy of the clock source. If our program expects the clock to be 10MHz, but in reality it's 9MHz, then its measurement of time will be ten percent slower. Is such a scenario plausible? Well let's talk about clocks a bit more.

    There are a few simple ways to generate a clock. One of the common methods is what's called an RC oscillator. Often a processor will have an RC oscillator built in, making it the cheapest option. However, RC oscillators are far less accurate. Their initial tolerance can be atrocious, like 30% off. Even with factory calibration, they can drift by several percent with age and temperature.

    The other common method of generating a clock is a crystal oscillator, which is a component external to the processor itself. Crystal oscillators are:
    1. Accurate. A typical crystal will have a tolerance of 100ppm (meaning 0.01%). At the very worst its frequency may change by as much as 0.1% over its natural lifespan.
    2. Reliable. It's very difficult to screw up a crystal oscillator. And when they do fail, it's usually due to some severe physical damage to one of a couple components. And when such a failure happens, the result is that the clock just stops operating entirely, or operates at a hugely different frequency (like double the intended frequency). There's no mechanism I'm aware of that can cause a crystal oscillator to be off by just a few percent.
    3. Cheap. Nowadays, the cost of adding a crystal oscillator is basically pennies.

    The final factor that can cause errors in time measurements is, of course, bad code. I'm not even going to open that can of worms.

    The bottom line is:
    The only way timing errors on the order of ~10% could arise in scoring machines is:
    1. The product engineers were so unbelievably stupid or penny-pinching that they chose to use crappy RC oscillators instead of a crystal oscillator.
    2. Buggy firmware. Like, embarrassingly bad code.
    3. They accidentally manufactured it with the wrong crystal, whoopsie!

    Any machine design that is subject to any of these issues shouldn't even be allowed to be used for competition, imo.
     
    Last edited: Mar 24, 2018
    jdude97 likes this.
  9. Mac A. Bee

    Mac A. Bee is a Verified Fencing ExpertMac A. Bee Podium

    Joined:
    Feb 5, 2008
    Messages:
    1,567
    Likes Received:
    146
    Andrew, Thank you. Found a vendor-provided one at this weekend's RYC+
     
  10. Zebra

    Zebra DE Bracket

    Joined:
    Dec 25, 2013
    Messages:
    917
    Likes Received:
    239
    Which vendor? And did BC know about it and restrict that strip to foil/epee?
     
  11. Mac A. Bee

    Mac A. Bee is a Verified Fencing ExpertMac A. Bee Podium

    Joined:
    Feb 5, 2008
    Messages:
    1,567
    Likes Received:
    146
    We swapped that machine out for a Favero.
     
    Last edited: Mar 26, 2018
  12. Strytllr

    Strytllr DE Bracket

    Joined:
    Nov 6, 2003
    Messages:
    575
    Likes Received:
    70
    I tried this on my SG 11 machines which I have never sent in for the saber timing upgrade and this did not work for them.
     
  13. Mac A. Bee

    Mac A. Bee is a Verified Fencing ExpertMac A. Bee Podium

    Joined:
    Feb 5, 2008
    Messages:
    1,567
    Likes Received:
    146
    Nerd Drift: Warp speed?;)
     
  14. jkormann

    jkormann Podium

    Joined:
    Nov 8, 2002
    Messages:
    2,911
    Likes Received:
    182
    It's a Kyber Crystal of course.
     
  15. Mac A. Bee

    Mac A. Bee is a Verified Fencing ExpertMac A. Bee Podium

    Joined:
    Feb 5, 2008
    Messages:
    1,567
    Likes Received:
    146
    Kyber crystals are for force. Dilthium crystals power warp drives.:rolleyes:
     
  16. jkormann

    jkormann Podium

    Joined:
    Nov 8, 2002
    Messages:
    2,911
    Likes Received:
    182
    I'm an older fencer. I believe in the "force" more than I believe in my speed. :)
     
  17. oiuyt

    oiuyt Podium

    Joined:
    Apr 26, 2000
    Messages:
    10,268
    Likes Received:
    997
  18. tsbphd

    tsbphd Rookie

    Joined:
    Oct 24, 2017
    Messages:
    22
    Likes Received:
    4
    So, it looks like USAF is acknowledging the issue as presented by CyrusofChaos in the embedded video from March.
     
  19. mfp

    mfp Podium

    Joined:
    May 10, 2002
    Messages:
    1,962
    Likes Received:
    285
    "We recommend that faulty machines not be used for tournaments in which rolling points will be awarded. Use for other tournaments is at the discretion of the organizer."

    Seriously? The USA Fencing Board of Directors mandated the saber machine timing be used in "ALL sanctioned USA Fencing tournaments" beginning Aug 1st 2016 but US-SEMI only recommends non-compliant scoring machines not be used for some tournaments and for others states it's up to the discretion of the organizer?
     
    jdude97 likes this.
  20. Inquartata

    Inquartata Podium

    Joined:
    Jul 12, 2001
    Messages:
    36,583
    Likes Received:
    1,289
    "forfeiting of stickers"?
     

Share This Page