Oooh, this topic is too juicy for me to pass up I'm a software engineer and my team makes GPS receivers for aircraft, so I hope I can shed some light on the topic.

Firstly, the Navi found on the Acura isn't strictly GPS-driven. It's actually a GPS-assisted inertial navigation system.

The system primarily determines your position by sensing forward motion through a wheel rotation counter, and which way you are turning using a yaw rate gyro. Given an initial position and heading, the system can integrate the outputs of the two sensors to determine your position.

As you can imagine, the errors in an integrating system will grow larger and larger as time goes on. The system compensates for this in two ways: First, it tends to snap your position onto the nearest street on the map, and second, it uses the GPS to re-fix your position if your current position estimate completely disagrees with the GPS estimate.

The system performs the above corrections only once in a while (in the order of once every 10-30 seconds, in my estimate). Your position may appear to be off the road between corrections.

When you're driving down an area with several close parallel roads, the system may also snap your position to the wrong road. This is due in part to the large uncertainty in the GPS position solution, which may be in the order of 100 ft. or more at times. The estimate gets worse when the satellite geometry or signal is poor, eg. when you're driving in an area with lots of tall trees or buildings.

So, there you have it--the reason you appear "off road" depends on a whole host of factors. There are many more variables in the mix than you realize--the system will not snap you to a road if your estimated heading is greater than about 45 degrees away from where the road is pointing, for instance; or that the system will snap you to a one-way road only if you're heading the same direction.

It's quite a remarkable system, and one which has a host of compromises that allow the system to work in a variety of environments. You'd be surprised how well our Navi performs over other systems.

rominl: Yes, it matters a lot to a GPS receiver when you're in an "urban canyon" (or any kind of canyon for that matter). The main problems you get are masking of lower-elevation satellites, multipath (signal reflections), and signal attenuation due to foliage.

However, since our Navi is primarily an inertial sensor, it can survive for a long time without GPS signals, especially when it's using map aiding.

Metallic tint does affect the signal reception. As edgalang has mentioned, the GPS antenna is located under the parcel shelf. However, given the relatively small importance of the GPS sensor's accuracy, I think you'll be able to pick up "enough" signals during your drives outside downtown that you won't notice a significant degradation of performance.

If you're really anal-retentive about the GPS reception, I think you can buy a trunk-mounted (magnetic base) antenna from Alpine, which will plug right into our Navi unit. An antenna on your trunk will work much better than one enclosed in your car.

Jpritch: it's not gyro precession, it's integration error--noise in the sensor signals basically accumulate over time and overwhelm your true measurement.

GoHawks63: You're talking about Selective Availability (S/A). S/A is a randomizing algorithm which degrades the precision of your GPS solution. As of mid-2000, S/A has been officially turned off with no plans to turn it back on (thank you Bill Clinton!), so it doesn't factor into our solution any more.

And remember, our Navi is primarily inertial, so S/A wouldn't have made that much of a difference most of the time.

Here's a hint: The most accurate positioning you're going to get out of your Navi happens when you're traveling down a straight road (and your Navi is correctly tracking you), then make a 90-degree turn onto another road. The map snapping will recognize that you've rounded a corner, and will snap your position to the junction based on the map--that's as good a position as you're going to get.

Even with S/A turned off, the accuracy of standard (C/A) GPS signals cannot touch that of the military (P/Y) signal. The military signal has 10 times the chipping rate of the standard signal, giving it a much narrower correlation peak.

I've actually had my Navi unit replaced because of a defective gyro. It appears that the Navi has an adaptive filter for the gyros that automatically recalibrates for bias drift. However, the gyro on my unit appears to be bimodal--it would have one of two distinct biases, and the system will periodically be confused and constantly show a right turn.

I took a webcam with me and I video captured my drives for days on end until I finally had the event on a file. I showed the video to the service staff and they agreed to replace the entire unit.

The lever arm effect of antenna placement is insignificant in this system -- remember, we're not doing full 3-D INS here. As I said, the Navi is primarily speed and turn rate sensing, and will only use the GPS position estimate once in long while to check the validity of its estimate. A lever arm error of 0.5m is hardly noticeable with a tens-of-meters CEP.

Originally posted by AndySara

If the maps on the NAVI DVD were made before 2/5/2000, then that is the reason why it throws the NAVI off when you are driving down a city block that runs parallel to another that might be 75 feet away. That also explains why the NAVI is so indecisive at T or Y junctions. With the advent of new maps, and new measurement taken of maps that are several years old, our NAVI system should become more accurate.