Rockwell Jupiter OEM Receiver - Multipath and Noise

Recently ,TTS, the Netherlands representative for Rockwell OEM GPS receivers, loaned me two of their Jupiter receivers for evaluation. In this page the results of my experiments are reported:

  1. In a so-called zero baseline setup I tried to measure the multipath- and noise errors on the raw C/A code and carrier (see my multipath- and noise pages for an explanation of the method).
  2. Two short baseline tests, one in a multipath benign environment and one in a heavy multipath environment, allowed me to gain knowledge of the Jupiters performance under favourable conditions and under realistic conditions.
  3. Finally, the tracking performance results in the urban canyon and in the woods are reported.


Experiment 1.

In the first experiment, I used Tri-M's Mighty Mouse antenna on a small groundplane (25 cm dia) on top of my roof, with an unobstructed view over the horizon. The antenna was connected to a tri-way splitter with a DC coupling to my reference receiver, and the the antenna inputs of the two Jupiters AC coupled via a 100pF capacitor. (the three receivers in a 'zero-baseline' setup). The Jupiters were set-up to generate raw measurement data in Rockwells binary format. The data was downloaded to the serial ports of my PC. About two hours of data were recorded and processed.


1.1. C/A Code Multipath

The error due to C/A code multipath can be determined by subtracting the integrated carrier observations from the C/A code observations. Unfortunately, it appeared to be impossible to determine this error : the code observation as delivered by the receiver followed the carrier observation to within a few cm's. Most probably this is a result of heavy code smoothing by the receiver itself  (for an explanation see my smoothing page). Although code smoothing is an effective means to reduce multipath errors on the code observation, it invalidates the method for determination of the error.


1.2. C/A Code Noise

Due to the same reason as stated above, the C/A code noise is masked by the smoothing process. However, another way of getting an estimate of the error is by calculating the position of the antenna in the single difference mode, using one Jupiter as the reference and the other one as the 'moving' receiver. Since both receivers share the same antenna, all common errors (including multipath) cancel, the position difference should be zero, and remaining deviations are the result of the smoothed receiver (and antenna) noise. The two figures below give a plot of latitude versus longitude, and height versus time.

The 'spikes' in the above plot are probably caused by one of the receivers acquiring a new satellite, some time is required before the smoothing process becomes effective. It is interesting to see from both figures how small the remaining error is once the smoothing process becomes effective.


1.3. C/A Carrier Noise

The C/A carrier noise averages to the low value of 2.5 mm (1 sigma), and appears to be rather independent of the satellites elevation.

In principle, this low amount of carrier noise allows carrier phase double difference processing (a technique to compute position differences or 'baselines' to the cm accuracy level ). Forming double differences of the carrier phase observations (see my theory page for an explanation) in a zero baseline setup should result in integer values, polluted with (twice) the C/A carrier noise. With a noise level of 0.1 cycle or 2 cm (a criterion which is easily met by the Jupiter) it is possible to separate the integer values from the noise. Once the integer values of at least three satellite pairs are known, the baseline can be calculated to the above mentioned accuracy level.

Unfortunately the Jupiter double differenced observations do not converge to integer values but to any real value, and moreover, the value changes slowly in time. The reason behind this behaviour is unknown to me.


Experiment 2a.

The first part of the second experiment was aimed at establishing the receiver performance in the single difference mode, and in a multipath benign environment.

I used my reference receiver with the geodetic antenna (with a choke ring ground plane) on top of the roof of my house as a high quality reference, logging raw data at a rate of once per second.

I placed the Mighty Mouse on the roof top of my car and connected another of my reference receivers, the Jupiter, and a Garmin 25LP OEM receiver to it via the three way splitter described above. I parked my car close to Amsterdam Schiphol Airport in an area with an unobstructed view, and about 3.7 km away from 'home'.

I logged raw data from all three receivers for one hour at a rate of once per second. The reference receivers allowed me to establish the 'true' baseline between 'home' and 'car' with an accuracy of 1 to 2 cm (double difference carrier phase solution).

I processed the Jupiter data in the single difference pseudorange mode with smoothed data from my reference receiver (remember: the Jupiter outputs only smoothed pseudoranges). Next, I determined the average and standard deviation of the position components and compared them to the double difference carrier phase solution.

The table below gives the statistics for the smoothed data based on 3041 observations (about 50 minutes worth of data). Bias is the difference between the averaged position and the truth, the standard deviation is a measure of the amount of scatter around the average position. Finally, the 95% error equals bias + 2 times the standard deviation.

error

Latitude
[m]

Longitude
[m]

Altitude
[m]

bias 0.87 1.20 1.46
st.dev 0.89 0.64 1.84
95% 2.65 2.47 5.15

It can be concluded, that the error in latitude equals the error in longitude, and the error in altitude is twice the horizontal error.

The plot below shows the behaviour of the Jupiter in the horizontal plane. The 'random walk' character of the plot is another indication of the heavy smoothing by the receiver (no, I wasn't drinking!).


Experiment 2b

The second part of the second experiment was a repetition of the first part, but in a not-so-friendly multipath environment.

The setup was identical to experiment 2a, with the only difference that I parked my car 15 m in front of a metal warehouse with a height of apr. 15 m. The warehouse acted as a nearly 'perfect' mirror for the satellite signals, maximising the detrimental effects of multipath reflections.

Again I logged raw data from all three receivers for one hour at a rate of once per second. The reference receiver again allowed me to establish the 'true' baseline (about 3 km) between 'home' and 'car' with an accuracy of 1 to 2 cm (double difference carrier phase solution).
Multipath turned out to be moderate during the first 37 minutes, turning into heavy during the last 20 minutes. Therefore I'll split the results into these two time slices.

First the moderate multipath case. The statistics for 2209 observations were:
error

Latitude
[m]

Longitude
[m]

Altitude
[m]

bias -1.22 0.33 -1.39
st.dev 0.46 0.21 1.76
95% 2.04 0.75 4.91

It's rather surprising, that this plot certainly is not worse than the low multipath case above. This certainly proves the ability of the Jupiter to mitigate the effects of low to moderate multipath ! The scatter plot below also proves this.

Unfortunately, the Jupiter is not very good at handling heavy multipath. It looks like as long as the trackers inside the receiver don't loose lock, the smoother works OK, but once a tracker looses (temporarily) lock, the smoother probably will be reset, resulting in a momentary large error. The statistics for 1179 observations are:
error

Latitude
[m]

Longitude
[m]

Altitude
[m]

bias -2.4 3.8 7.6
st.dev 6.6 6.0 9.1
95% 15.6 15.8 25.8

Check also the dramatic behaviour of the horizontal scatter plot below:


Experiment 3.

The last experiment was aimed at learning more about the tracking performance of the Jupiter in the urban canyon and under foliage. The set up was identical to experiment 2, and I drove my car through some narrow roads with apartment buildings, and through the Hoofddorp Woods close by. The horizontal plots below show the path that I followed. The upper plot gives the foliage track, the lower plot the track through the urban canyon.

Some remarks:

I looked at a number of statistics in order to try to quantify the tracking performance of the Jupiter.
First I looked at the number of observations with less than 4 satellites (no 3D position calculation possible). The track in the urban canyon and in the woods took 856 seconds to complete, hence under favourable conditions the same number of calculated positions should be available. In reality however, 17 observations had 3 or less, resulting in 839 positions.
Next, I looked at the PDOP of the remaining position solutions. The PDOP or Position Dilution Of Precision is a number which indicates the geometric strength of the available satellite constellation. Usually, one considers a PDOP of less that 6 to be acceptable. With this criterion, 27 more positions were discarded. This resulted in 812 observations, or 5 % 'lost'. Of the remaining 812 observations, only 665 had more than 4 satellites and an average pseudorange error of less than 5 m, or in other words: in 78 % of the time it was possible to calculate a position with a 5 m error confidence.


Compare the Jupiter results with the Garmin 25LP results

Go back to the index.