Multiple GPS inputs for better accuracy/speed?

**2k1Toaster** · 08-02-2009, 02:15 PM

Originally Posted by trader007

doesnt the military have gps pucks that are accurate to within 3ft? do they make a usb version?

That was based off of a ground based timing signal that was shut off a while ago as it was no longer needed.

And gps can be accurate within a metre. This +/- 10m is a long time ago or on really crappy units. How accurately your gps calculates where you are is EXTREMELY dependent on the clock in it. It has to compare the time difference of receiving radio waves from different spots and we know those travel at the speed of light (literally). So the slightest error in clock timing, and your whole calculation is shot to hell, and you could be off by 10m. The more money you spend, the better the timing circuitry, the more accurate the results. Now there is always space wobble aswell, and that is what the military signal overcame for the most part as it basically just put the signal on the ground instead of from space since satellites move and there positions arent updated instantly. But it is pretty predictable, and a good receiver can "guess" how far off it is based on past events.

It's a wonder it works at all really...

I am willing to bet that if you wrote down every NMEA string and hand parsed the locations, they would be in the +/- 3m range wherever you are with whatever receiver you have now (like the BU-353). Some better, some worse, temperature effects timing circuits too. But programs like iGuidance and really any factory nav system guesses where you will be and should be as well as where you are. So if you are traveling down the interstate at 80mph and the ramp is right next to it and you werent guided to exit, but do o anyways, it is going to think that is error and pretend like you are still driving on the interstate (or vice-versa when supposed to take an exit and you dont). M$ Streets and Trips 2006 did not do this and I could see myself change lanes on the map. Of course this also means when construction routed you to the other side of the road or something, it would say you were off route and not recalculate but that is a different story.

**malcom2073** · 08-02-2009, 03:44 PM

Originally Posted by trader007

doesnt the military have gps pucks that are accurate to within 3ft? do they make a usb version?

There is something called subscription or SGPS. It gets you sub 10cm accuracy. It's commercial, but costs around $10k a year per unit.

**trader007** · 08-02-2009, 03:54 PM

Originally Posted by malcom2073

There is something called subscription or SGPS. It gets you sub 10cm accuracy. It's commercial, but costs around $10k a year per unit.

now THATS what im talkin about lol

i know its amazing gps works at all... even with cheap $20 adapters. i have a royaltek adapter right now, it seems the same exact specs as the bt353, but $10 cheaper. it shows my truck in the 1-lane driveway, i couldnt even tell if it was just a foot off.

10cm accuracy though... i wonder if you could setup one of those robot vacuum cleaners to go a specific route in your house based of space navigation...

**malcom2073** · 08-02-2009, 03:58 PM

Yeah, but what kind of GPS signal would you get inside? You're much better off doing IR wall detection and having it pre-map your house and use that for locating :P Not that I've given this any thought... :P

**Sicarius** · 08-02-2009, 11:32 PM

Originally Posted by 2k1Toaster

Only thing I can see you are overlooking is the programatically choosing which NMEA data to send to the program instead of the one closest to the time the app reads the serial port.

So if you have 8 readings all somewhat in the same spot or in the same line, and then 2 which are very far off (maybe a big tree or a bridge or something obstructed the time signals), then you would choose to send one of the 8 good ones. Regularly it would just be luck of the draw and you could get a bad one.

The whole problem with this is that it takes time. If you want to evaluate a 10Hz stream, for instance, to find the sample with the least error each second you would have to buffer the whole second's worth of data. If you decide that the first position was the best, it's now 1 second out of date. If you were driving along at 100km/h (=~27.8m/s) then when you finally decide to report that first position, you've moved 27.8 meters since you got it and that error is likely much worse than the measurement error to begin with. You're better off taking what you're given, when it's given to you.

**2k1Toaster** · 08-02-2009, 11:56 PM

Originally Posted by Sicarius

The whole problem with this is that it takes time. If you want to evaluate a 10Hz stream, for instance, to find the sample with the least error each second you would have to buffer the whole second's worth of data. If you decide that the first position was the best, it's now 1 second out of date. If you were driving along at 100km/h (=~27.8m/s) then when you finally decide to report that first position, you've moved 27.8 meters since you got it and that error is likely much worse than the measurement error to begin with. You're better off taking what you're given, when it's given to you.

The way I described, yes. It was an example. In reality, you would probably do a weighted average that cooresponds to none of the original points and send that

**Sicarius** · 08-03-2009, 12:26 AM

Originally Posted by 2k1Toaster

The way I described, yes. It was an example. In reality, you would probably do a weighted average that cooresponds to none of the original points and send that

Which only really makes sense if you're standing still or traveling in a perfectly straight line at constant velocity and you'll still be 0.5seconds behind using a straight average. Introduce any variations (turning, changing speed, changing elevation) and you have to make so many assumptions about position that the GPS is likely more accurate anyway.

When you have 10Hz reporting, you would be better off just suppressing those readings which have "too much" error rather that trying to correct for it.