Firstly, Real Time Kinematic correction involves a stationary base station unit and a moving rover unit for differential correction. The drift in the base station's satellite timings is used to correct the rover's position. A Japanese project to implement RTK in an open source library was created two years ago. I remember reading discussions of RKTLib that trailed off noting its initial lack of support for various models. It looks like development has continued to add features and expand support in the intervening time, and I'm wondering if there's any chance it could see use around here, since they seem to have accrued several semi-affordable models.
Second, as of this year there now exist APM-size boards which will implement dual-frequency correction for centimeter accuracy. Dual frequency receivers use different amounts of interference at the different bands to model atmospheric distortion directly, vastly improving accuracy. I'm looking particularly at the Hemisphere miniEclipse, the Novatel OEMV-1DF, and the Geneq SXBlue III. The unknown variable here is price in batches of 1000, which is not made public. Are any of these boards viable candidates for the DIYDrones store?
Is there an actual problem that you are trying to solve, or is the quest for ultimate GPS accuracy its own goal?
To a a man with only a hammer for a tool, every job looks like a nail.
High-precision GPS is an enabling technology for future applications, in particular applications which involve autonomous charging, takeoff, and landing. 15 minutes of flight time is very short for a lot of remote sensing and industrial activities, particularly given the 400ft altitude limit in the US that we may / may not have to comply with. A reliable 10cm precision would enable relatively painless low-voltage charging stations, and vastly reduce the amount of labor that goes into collecting data over wide areas. Simple charging stations, in quantity, would enable repeatable coverage of a wide area by a swarm of multirotors.
Furthermore, either of these would reduce the gulf between US/Europe, where we have sparse, but reliable state-run differential base stations that are compatible with these units (WAAS, EGNOS), and the rest of the world's relatively weak capability to use waypoints. This has been a theoretical consideration in promoting our techniques abroad.
All of the dual freq boards are more than $1K in small quantities.
I've purchased and used (>$5K ea) dual frequency boards and flown them on an RC helicopter.
To get sub cm level precision on these you need a serious antenna.
The units I had were not very vibration tolerant, they kept losing the RTK solution on the vibrating helicopter.
The Omnistar high precision correction service actually seemd a bit more robust than the local reference server.
My guess is that it was not as precise in a static situation.
With this working the GPS precision in flight was easily better than +/-10cm. this used a 4lb choke ring antenna.
I tried a smaller lighter antenna and never got good lock. I probably needed more gain. Since I could carry the weight I did as I had other problems to solve at the time.
Before trying the dual frequency system I purchased an L1 20Hz RTK system. It worked fine stand alone.
adding telemetry transmitters, RC transmitters, and vibration each individually degraded the L1 RTK so it would sometimes take 20min to get an RTK lock and then loose it randomly in flight. When it worked it was awesome, when if failed the helicopter results were spectacular, jumps of 50+/-ft in altitude. I was using GPS as my altitude reference so a sudden 50ft change in altitude could be sporty if you were under the helicopter and or were not paying attention with the override transmitter in hand and were on top of engaging the manual override.
Amazingly valuable information here. Practical geodesy is not a particularly simple subject to learn academically, and you've been a great help.
In a multi-sensor smart device, a large amount of the short-term variability in position associated with losing lock can be adjusted for with accelerometers and Kalman filters, if the switching is relatively fast - or at the very least it can serve as a safety measure so that the unit's unpredictable behavior occurs slowly. Uncorrected GPS can do a decent job of detecting *motion* during a lock combination all by itself, it's position error caused by low-frequency ionospheric changes and satellite lock changes that becomes the problem. A 'position fusion' approach may prove fruitful, if it can be practically developed given the expensive hardware.
Another thing which might help is a dual frequency GPS/GLONASS reciever, doubling the 12 or so satellites that are in the sky at any given time in order to make lock a lot more reliable.
I'm quite intimidated by the antenna requirement - what is the failure mode associated with having too small of an antenna?
Are you at liberty to discuss the solution you ended using after all this experimentation?
This was all for a rocket project.
Our bit starts at the 8min point of the video:http://watch.discoverychannel.ca/daily-planet/november/daily-planet...
The wrong antennas were (custom )ordered for the L2 GPS. The GPS needed 35 or 40dB of gain, but I ordered L1,L2 Omnistar small antennas with 25db of gain. (What I had been using with the L1 GPS) I ordered the L2 boards nd the antenna at the same time and did not realize my error until it was too late to order more custom small triband antennas. So the small antennas for the L2 boards were alyawys too noisy.
I've also been working on a deeply integrated GPS/IMU for the next rocket project.
Its currently on Hold because I found an off the shelf GPS that will hold lock at 8 to 10g.
See comments on my blog unreasonablerocket.blogspot.com
The latest post and comments actually have pertinant information.
I forgot to add:
1) Highly accurate, high-frequency GPS allows photogrammetric tasks to be much more constrained.
2) Using GPS altitude as a means of putting a reference to the air pressure sensor lets one fly low-altitude 3D patterns that are set beforehand using a geospatial reference DEM.
I'm presently working on a project using Novatel OEMStar units with quite nice antennas. the OEMStar are dual GPS/GLONAS
My job initially is to establish base accuracy by operating two static units separated by about 3km. They are operating independently. The results are not so good.
Although the unit average position is extremely good, individual units can wander up to 5 metres from mean position in a few seconds. I assume thats caused by changes in constellation.
The real problem is both units don't track the same. This may be because they have slightly different sky visibility.
My application eventually will involve tracking moving equipment that has to be placed within 2 metres at any time over a range of a couple of thousand kilometres. The positions of multiple equipment units are reported by satellite link and compared to reference station(s)
This wander seems to be an inherent problem, so I'd like to know if anyone has a good working solution?
I've been told that real-time kinetic adjustments help, but as far as I can see it's basically dead-reckoning.
Any advice appreciated.
NovAtel have a fair bit of info published in their application notes, though it can take a while to wade through and find what's relevant to you. Optimal solution will depend on a bunch of things. For starters:
- How quickly do you need to achieve <2m accuracy? Can the mobile unit sit for half an hour, or 5 minutes, or 5 seconds?
- Will / can there be other base stations in specific zones of interest, or even set down temporarily in an area while mobile moves around within a few km of the base?
- Will the mobile unit be stopping to record specific locations (like a survey) or constantly moving (vehicle) and how quickly?
- How open is the terrain?
Time lets you average and stabilise in a particular location. This then lets you establish a reference base from which to do DGPS (within ~ a few km) if you have telemetry between that base and the mobile unit.
Open terrain means good satellite visibility, and consistent visibility.
The higher precision RTK solutions need consistent visibility as they need to maintain not just a channel/time lock on each signal, but a phase lock.
From OEMStar, you'd gain some benefit having GLONASS as well as GPS. When using both, there's benefit in having more than 14 channels, so OEMV-1G or OEM6xx would be better. You can achieve this while staying under $1000 per unit. (My bases can see ~18 satellites total)
If you can plop down even a temporary base station, that would help. Give it a very good antenna and time to average its position, then transmit differentials you your mobile unit. Or, subscribe to a satellite-broadcast differential system. Or, if in Canada, use the free national DGPS system.
To get better accuracy, more quickly, without DGPS, look at the more expensive models that use both L1 and L2 bands. The bands are so different that ionospheric effects can be determined and compensated on a single unit. I'm not sure how many L2 satellites are deployed, but believe it is a current solution. Note you'd also need dual-band antennas.
What are you calling a "quite nice antenna"? At the moment I'm trialling a $100 Garmin marine GPS+GLONASS antenna on bases, though my NovAtel supplier is recommending a $800 antenna for those locations and has lent me one to try.
Wow! That was quick!
My field units are trains and rail service vehicles. We have to determine if there is a collision potential by identifying which track or siding a unit is on. Track centre separation is about 4m.
Our planned position update rate is perhaps 1 per second - it can be slower, but the point is the units are moving a lot of the time so we won't get much averaged position.
We can put in basic logic models that say a transition from up-line to down-line is impossibe at that location, ditto from single-line to siding. It would be some sort of state machine.
My preference is to not do this modelling. I've seen a lot of software that tries to figure out where a train is, and that usually fails miserably and always relies on an operator to correct the state.
NB - we don't have track sensors so can't use that as an input.
I'm still working on the basic control and alarm model. We have the field communications sorted so all positions are reported continuously via a diverse communicatons network ready to be processed by the safety management system - that perhaps uses a static base-station for reference.
Another option is to use local base-stations at critical points - sidings in particular. The network is almost all single track with sidings.
The question then is should differential GPS be used locally? If so how, as trains and support vehicles move to different locations.
I realise all this is sort of off-topic for DIYDrones, but any serious use of GPS for long-range aircraft will have to address basically the same issues. An obvious example is GPS assisted landing at diverse airfields.
I also forgot to mention my clients are very much pushing for driverless trains - BIG trains. This whole control and safety issue starts to get very real very quickly.
Nice draw back to aircraft. I'll point out here that among the DGPS protocols supported by NovAtel is RTCA, which was developed specifically for use in aircraft approach/landing!
As you say, you could use established telemetry to pass DGPS data. How is up to you. You could do some matching based on approx location. Or, if you have enough bandwidth, send all the differentials everywhere but have some local network node filter out the data applicable to its approximate location. Placing a base at each siding and multi-track junction sounds like an option. It adds to the telemetry requirements, and to the list of equipment to maintain. I think an L1+L2 receiver is probably your best option, for the sake of simplicity and maintenance.
You might want to consult someone more experienced in high-end GPS systems for a 2nd opinion.
Take a look at this. The test is showing cm accuracy.