New Statistical based Tolerancing per GPS Lockout PIN

Discussion in 'Official V1Driver Discussion Section' started by mswlogo, Jun 27, 2017.

  1. mswlogo

    mswlogo Premium Member Advanced User Premium Member

    Messages:
    3,166
    Likes Received:
    2,793
    Country:
    United States
    Joined:
    Nov 1, 2014
    Location:
    MA
    I've wanted to do this for a long time but wanted a very stable base to work with first.

    Little background:

    Escort, as I understand, uses fixed "Segments" and tags a lockout to be in a specific Segment. I never liked this because it could land right on the boundary of two segments. In the long run that doesn't hurt (much) but you'd probably end up with two lockouts as its frequency wiggles.

    In V1Driver they are not fixed segments. Each Lockout is centered in the Tolerance you set.

    V1Driver, uses a slightly finer tolerance than Escort does (by Default), but you can't directly compare it; partly because of the Fixed Segments you might lockout 2 segments unnecessarily even on a stable False, which won't happen in V1Driver. It has Dynamic Segments (sort of).

    When reading comments on YaV1 and my own experience with it I found YaV1 (default) tolerance a bit tight. I think this made some users impatient and they would start using Manual Lockouts (as I did and I hate manual lockouts). So by default V1Driver users a default tolerance wider than the default Tolerance that YaV1 uses. Of Course both apps can be changed. But most users don't change it.

    I chose a default value that was a compromise of being reasonably "safe" and lock things out at a reasonable rate. I figured anything finer than Escort would be reasonably safe.

    Any way, now onto the new stuff:

    What if you didn't have to choose a rigid tolerance for all locations at all or worry about it?

    Not all falses are alike, some are very stable, some gradually drift, some wiggle a lot.

    Even as things stand with V1Driver or YaV1, you can set the tolerance very tight and just wait it out and eventually it will lock out everything. But you could get falses for quite some time, which defeats the purpose of GPS Lockouts. Because you may tune K band out (in some stuburn locations) by the time it's fully learned.

    Welcome to “Statistical Tolerancing” per lockout !!

    What V1Driver will now allow you to do is collect statistical errors on each GPS Lockout PIN. It will use standard statistics (that you can adjust) to determine if a new candidate hit is a statistical “Outlier” for that specific lockout rather than use a fixed tolerance for all lockouts. If a particular GPS Lockout wiggles a lot, it’s tolerance will be calculated high. If a GPS Lockout it stable the tolerance will be calculated tight.

    Once enough statistics are gathered you can view the Tolerance that will be used by clicking the info on a GPS Lockout Pin.

    You can set how statistically tight you want things in "Standard Deviations". The Default is 3 Standard Deviations.
    You can also choose how many samples are needed to start applying the Standard Deviation Logic. The Default is 14 Samples (that means a minimum of 14 days (using other defaults)).

    The gist of this new scheme is, it learns fairly quickly (safer than Escort and basically no change in current behavior), but now, over time, as it collects statistics, it will tighten things up a LOT and minimize the lockout range. It SHOULD remain just as quiet but run with MUCH narrower ranges where it can.

    You can turn this all off by setting the number of Samples to Collect to 0 !!

    Another bonus to this method is if a GPS Lockout gradually drifts, the model will gradually move with it. It does that now by creating a new PIN when it goes out of tolerance and the old PIN will get demoted. But now it will gradually move with it. It’s only when suddenly something is statistically an outlier (e.g. a LEO is within the former fixed tolerance) and it will now cease muting it (and start a new GPS Lockout if it remains consistent).

    In practice, I'll be honest, I don't know how well this will work (and why it's in Beta only and may be a long Beta Cycle). I think it should work well. The hardest part here is verifying it's behaving correctly. I may add more hooks to help with that.



    JeszJeYhVHQIo7Y-uZiBqGfaDeauo6jq.jpeg JeszJeYhVHQIo7Y-uZiBqGfaDeauo6jq.jpeg JeszJeYhVHQIo7Y-uZiBqGfaDeauo6jq.jpeg
     
    Last edited: May 11, 2018
    grayman, Vortex, Normsky and 12 others like this.
  2. Deacon

    Deacon TXCTG Advanced User Premium Member

    Messages:
    6,462
    Likes Received:
    6,443
    Country:
    United States
    Joined:
    Nov 13, 2016
    Location:
    Hill Country, TX
    That's a fantastic and sound idea. Very well done.
     
    belusch, thanks and mswlogo like this.
  3. InsipidMonkey

    InsipidMonkey Premium Monkey Advanced User Acceptus Premium Member

    Messages:
    3,944
    Likes Received:
    6,174
    Country:
    Iceland
    Joined:
    Mar 22, 2017
    Location:
    New England
    Sounds like a great idea! I'm really impressed with the constant development and improvement you bring to V1Driver. It almost makes me wish I had an iPhone.

    Any chance of an Android version?
     
    mswlogo likes this.
  4. CaptainGizmo

    CaptainGizmo Learning to Drive General User

    Messages:
    36
    Likes Received:
    24
    Country:
    United States
    Joined:
    Jun 3, 2017
    You are amazing, and I admire your technical knowledge and passion to continue development of V1D.


    Sent from my iPad using Tapatalk
     
    belusch and mswlogo like this.
  5. jdong

    jdong Premium Member Advanced User Premium Member

    Messages:
    6,755
    Likes Received:
    9,188
    Country:
    United States
    Joined:
    Jun 5, 2013
    As far as validating it, just throwing an idea out there: Might be worth having a bar visualization of the spectrum for each band and draw on top the areas that are locked out at the current location.

    One area of situational awareness that seems sorely lacking with today's detectors and UI is "how protected am I against the moment for ____"? If you had the underlying digital detector exposed you could answer that question really well in terms of the noise floor and what signals are being analyzed/suppressed. But even without that, communicating that you're in a zone where a large part of K band or close-to-24.150 is locked out might be valuable to cautious drivers, and also allow people to spot check the model to make sure lockout zones aren't blooming unnecessarily in both distance and bandwidth.
     
  6. Deacon

    Deacon TXCTG Advanced User Premium Member

    Messages:
    6,462
    Likes Received:
    6,443
    Country:
    United States
    Joined:
    Nov 13, 2016
    Location:
    Hill Country, TX
    Currently locked out signals are already displayed (grayed out but still listed). The hard part (or rather the part requiring net new functionality, I think) would be either paying attention well enough to know or otherwise finding a way to easily track and report on the status of those lockouts, their drift, and their radius.
     
  7. jdong

    jdong Premium Member Advanced User Premium Member

    Messages:
    6,755
    Likes Received:
    9,188
    Country:
    United States
    Joined:
    Jun 5, 2013
    Right, the latter part is where some sort of visualization of the lockout frequency range would be helpful
     
  8. mswlogo

    mswlogo Premium Member Advanced User Premium Member

    Messages:
    3,166
    Likes Received:
    2,793
    Country:
    United States
    Joined:
    Nov 1, 2014
    Location:
    MA
    I totally agree. That was an issue I always had with Escort. You had really very little idea what was going on. And it's even more difficult with the "Manual Locking" Radar Detectors (they should at least allow you to view a Map of some sort).

    One thing I really spent a LOT of time on V1Driver is telling the driver what's going on (for both Developer and User to build up confidence that it's behaving).
    Like Why was something Muted or not Muted. But this new feature is a little tricky because it's buried deep (which is good, it slipped in very cleanly so far)

    One idea I had was drawing circles around each PIN and the diameter of the Circle is proportional to the Tolerance on that PIN. That way you could get a "Birds Eye View" of the Tolerances in use.
    But it's not easy doing custom drawing on the Map. It would of course only be a "Diagnostic" view. Another way is to add some more picker wheels (Debug Only) to select Tolerances you want see.
    That way you could see, why hasn't that PIN acquired enough statistics to use the Standard Deviation Logic. Or why is that PIN so wide on tolerance, should it have split it, into two Narrow Ranges (this is where things get really tricky to assess)?

    Another nice statistic would be what is my "average" tolerance on all the Lockout PIN's. That should gradually go down over time (especially older ones, which is why doing it with Pickers might be best because you can already choose by date and pin type).
    The "Pickers" are Database Queries :)
     
    belusch and jdong like this.
  9. Deacon

    Deacon TXCTG Advanced User Premium Member

    Messages:
    6,462
    Likes Received:
    6,443
    Country:
    United States
    Joined:
    Nov 13, 2016
    Location:
    Hill Country, TX
    That really is super cool. I love the idea of essentially figuring out what kind of drift you see and how far out you see it, and making the lockouts as loose as they need to be but no looser. Or said another way, tighten them down as hard as possible and still be effective, to minimize the risk of locking out legit radar sources.

    VR should really make you an offer so good you can't refuse and leverage your talents.
     
  10. WJS

    WJS A low profile can be the best defense Intermediate User Premium Member

    Messages:
    516
    Likes Received:
    352
    Country:
    United States
    Joined:
    Sep 8, 2015
    Location:
    CT
    This is really exciting and I can't wait to test it. For the longest time I've had the app in the background because it just works flawlessly and it's one of the main reason I won't bail on the V1. This new idea just sounds like fun!
     
  11. mswlogo

    mswlogo Premium Member Advanced User Premium Member

    Messages:
    3,166
    Likes Received:
    2,793
    Country:
    United States
    Joined:
    Nov 1, 2014
    Location:
    MA
    I added a few screen shots to show the UI Changes to the OP.

    - - - Updated - - -

    I used to be on Android and changed to iOS as my primary platform for this Project. My goal was I wanted it to work well on my Primary Phone (not a dedicated device in the car).
    I really can only run one platform as my Primary Phone. Trying to juggle both platforms would be messy and I'm still learning the ins and outs as an iOS User and as a developer.
    I think it's critical I use it day in and day out to monitor and improve it. And there just isn't enough time in the day to properly do both platforms (and keep my wife).

    And before anyone asks, using a Single Source for both platforms is VERY difficult to get peak performance out of each respective platform (lots of compromises).

    The Swift language it's written in is just amazing, as well as the XCode IDE development environment and is actually a lot of fun to code in.
    Features wouldn't have come so fast and efficient if not for the Swift language and platform itself (big one being what's called "Core data" which is SQLite under the hood) .
    Thank goodness I didn't use Object-C (blah) and chose to jump in with both feet and use Core Data (which was a huge learning curve, but paid off big time).
     
    HeartDoctor, belusch, thanks and 2 others like this.
  12. InsipidMonkey

    InsipidMonkey Premium Monkey Advanced User Acceptus Premium Member

    Messages:
    3,944
    Likes Received:
    6,174
    Country:
    Iceland
    Joined:
    Mar 22, 2017
    Location:
    New England
    Fair enough, thanks for the reply!

    Maybe one of these days I'll swipe the fiancees iPhone and give it a shot.
     
  13. mswlogo

    mswlogo Premium Member Advanced User Premium Member

    Messages:
    3,166
    Likes Received:
    2,793
    Country:
    United States
    Joined:
    Nov 1, 2014
    Location:
    MA
    FYI, I added a Video of AutoTolerance feature in the opening post. Sorry it's a little long winded because of the time it takes to have it do it's thing properly.
     
    Normsky, belusch, Furuno2127 and 2 others like this.
  14. LexusISF

    LexusISF Premium Member Advanced User Premium Member

    Messages:
    1,635
    Likes Received:
    1,354
    Country:
    Canada
    Joined:
    Jun 11, 2017
    Location:
    Vancouver, B.C.
    Thanks mswlogo for what you have created todate and contined support and commitment to listen to suggestions/tweaks/ideas. We are lucky to have you woth us, and I agree with the comment, no question you have helps with sales of the V1. I can speak from experience without your app, I probably would have bought a Uniden R3.
     
  15. te37

    te37 Learning to Drive...secretly Advanced User Premium Member

    Messages:
    1,235
    Likes Received:
    877
    Country:
    United States
    Joined:
    Dec 11, 2013
    Location:
    Where am I not?
    I think you are currently keeping the V1 alive! Amazing work! Valentine Research needs to send you come cash haha
     
    mswlogo likes this.
  16. mazdaprofour99

    mazdaprofour99 Learning to Fly Beginner User

    Messages:
    75
    Likes Received:
    53
    Joined:
    May 3, 2017
    Thank you for what you do and the time you put into maintaining and updating this program. I would not have bought the v1 if this great program was not available.
     
    mswlogo likes this.
  17. mswlogo

    mswlogo Premium Member Advanced User Premium Member

    Messages:
    3,166
    Likes Received:
    2,793
    Country:
    United States
    Joined:
    Nov 1, 2014
    Location:
    MA
    This feature is now in Beta. Sorry for the delay. I had an issue with AutoMute/AutoSnooze that took a while to sort out.
     
    thanks likes this.
  18. Vortex

    Vortex Making Videos Administrator Advanced User Premium Member

    Messages:
    19,023
    Likes Received:
    34,226
    Country:
    United States
    Joined:
    Jul 19, 2012
    Location:
    Washington State
    Home Page:
    I'm gonna watch the videos in a bit when I get a little time, but this looks fantastic. I love that we'll now have some concrete data and numbers to show for the signals. It's hard to keep track of tons of signals over time mentally, and so having this all logged and calculated I like a lot more than just leaving it at "well, signals tend to drift over time due to temperature, age, and blah blah." Now we'll have something more definitive to show. I'm gonna have to start running my V1 again and start messing around with this. :)
     
    mswlogo likes this.
  19. mswlogo

    mswlogo Premium Member Advanced User Premium Member

    Messages:
    3,166
    Likes Received:
    2,793
    Country:
    United States
    Joined:
    Nov 1, 2014
    Location:
    MA
    @HeartDoctor asked what folks are seeing on their Statistical based tolerancing in another thread on K band out of box muting. I thought it better for folks to report here.

    I'd like to know as well.

    How many do you have?
    And what kind of numbers you seeing.

    I only have a few because I've had to restore my database a few times due to testing and switch phones.
    But a few are 10x narrower and a few are slightly larger than default.

    If anyone has a lot of them I wouldn't mind getting a copy of your database if you don't mind sharing.
     
  20. HeartDoctor

    HeartDoctor PSL + 5 never more;) Advanced User

    Messages:
    1,373
    Likes Received:
    2,121
    Country:
    United States
    Joined:
    May 30, 2016
    Location:
    SE Florida
    06CEE141-DA7B-47FE-A505-1F923DD1309F.jpeg 86C17840-E72C-401F-BDF6-C0991D38D664.jpeg

    These are my first 2 to reach statistical significance.