SVN49 and Other GPS Anomalies

Elevation-Dependent Pseudorange Errors in Block IIRs and IIR-Ms

Figure 1: Radial orbit differences between GPS SVN49 broadcast ephemerides and estimated orbits (Click image to enlarge.)

This is an updated version of the web article, Saving SVN49, that includes discussion of the elevation-dependent pseudoranging anomaly on the most recently launched GPS satellite and similar anomalies observed on other GPS satellites.

An investigation into the elevation-dependent pseudoranging anomaly on the most recently launched GPS satellite reveals that the U.S. Air Force has implemented an intentional alteration in the broadcast ephemerides and clock data in order to correct the anomaly. The investigators conclude that the solution will probably work for normal navigation users but may pose a difficulty for some high-precision users.

Tim Springer and Florian Dilssner
European Space Operations Centre

» Download this article (PDF)

Share via: Slashdot Technorati (Inside GNSS)Twitter

The GPS spacecraft SVN49 (Space Vehicle Number 49), also known as Block IIR-20(M) and PRN01, carries the demonstration payload for the new civil GPS L5 signal. Its March 24, 2009, launch enabled the United States to meet the International Telecommunications Union (ITU) deadline for securing primary rights to use of the RF band by GPS.

However, a report on SVN49 presented by the GPS Wing’s chief engineer, Lt. Col. David Goldstein to the European Navigation Conference in Naples, Italy, on May 4 mixed good news with bad.

The demonstration L5 signal began transmitting successfully on April 10. However, other GPS signals being broadcast by the satellite — particularly those at the L1 frequency — demonstrated larger than expected pseudorange errors that appear to be elevation-dependent. That is, they change with the varying elevation angle of the satellite as it rises and sets in the sky.

These signal anomalies, characterized by the U.S. Air Force as “out of family” transmissions, has kept the latest GPS satellite from being declared healthy.

The cause of the SVN49 anomaly, which appears most strongly on signals transmitted on the L1 frequency, is now understood, as well as the reason why certain receivers do not “see” the anomaly. The effect is caused by signal reflections coming from a special auxiliary port (designated J2) that is used to feed the L5 signal to the satellite’s antenna array.

These reflections cause a secondary-path signal with a delay of approximately 30 nanoseconds, which has the appearance of a multipath error. Ground receivers with advanced multipath mitigation techniques are not as sensitive to the anomaly — or at least much less so.

In the course of our investigations into the signal anomaly on SVN49, we learned that GPS SVN55 and, most likely, other Block IIR and IIR-M satellites carry some electronic equipment connected to the same auxiliary port as used for the L5 signals on SVN49.

(Editor’s Note: Independent sources have confirmed to Inside GNSS that a classified payload has been connected to the J2 port on several satellites. However, unlike SVN49, these classified packages include a component designed to negate or dampen the secondary-path signal.)

Assuming that any electronic signal connected to this auxiliary port would likely cause a similar kind of signal reflection — and associated disturbance in pseudorange accuracies as observed on SVN49, we took a closer look at the residuals of all GNSS satellites used in ESOC’s routine processing of observations for the International GNSS Service (IGS).

Our analysis revealed that SVN49-like L1 pseudorange anomalies are indeed present on several GPS Block IIR and IIR-M satellites, but these anomalies are much smaller in size.

U.S. Air Force operators are making a substantial effort to mitigate the anomaly on SVN49. Their initial effort involved significantly changing the broadcast ephemerides of this satellite, that is, the information transmitted regarding the satellite orbit and clock.

These alterations further complicate the handling of this satellite. The surprising fact here was that the changes in the broadcast ephemerides were of the level of ~150 meter whereas the reported signal problems were at the few meter level only.

In this article we take a detailed look at the signal anomaly observed for SVN49 and other satellites and at the quality and “treatment” of broadcast ephemerides and time being undertaken to mitigate the effects of the signal anomaly.

Analysis
As one of the International GNSS Service (IGS) analysis centers, the European Space Operations Centre (ESOC) in Darmstadt, Germany, routinely generates GNSS orbit products at the 20-millimeter level.

However, after the change in the broadcast ephemerides for SVN49 on May 1 we were unable to include this satellite in our routine processing because it not longer passed through our data preprocessing steps. (We will explain this point a little later.) This exclusion of the satellite coupled with the reported signal anomaly triggered the analysis presented here.

To study the SVN49 signal anomaly we use IGS data (available at http://igs.org) from the time period of April 30 to May 30. We begin on April 30 because this is one day before the Air Force changed their handling of SVN49, based on an investigation into the signal anomaly and development of a method for correcting it conducted by the GPS Wing. This gave rise to a significant change in the broadcast ephemerides values for the spacecraft.

For the analysis presented here we used a processing scheme that is very similar to the processing we use for the generation of our highest quality products, i.e., the postprocessed observations we submit to the IGS as contribution to the so-called IGS final products. But we significantly increased the tolerance in our data preprocessing to ensure that the data from SVN49 would not be excluded from our calculations.

The results of our analysis will allow us to compare our orbit and clock estimates for SVN49 to the broadcast information. This should clearly show what is currently being done with the SVN49 broadcast ephemerides. Furthermore, the residuals of our processing should clearly reveal the signal anomalies.

Orbit and Clock Differences
First, let us take a look at the differences between our estimated orbits and clocks and the orbit and clocks as given by the GPS broadcast ephemerides. Figure 1 (above, right) shows the radial orbit difference for all the active GPS satellites for days 120 and 121 of 2009 (April 30 and May 1). Only SVN56 was excluded because it had undergone a repositioning maneuver on April 30 during which its transmissions were set unhealthy.

The figure nicely illustrates the dramatic change in the broadcast ephemerides of SVN49 starting on May 1, 2009. After May 1 this difference has remained very constant in time with the radial orbit difference remaining at 150 meters and the clock difference at 500 nanoseconds. Note that the broadcast orbit is 150 meters above the estimated orbit.

Until April 30 the broadcast ephemerides of SVN49 behaved very similarly as those of all other satellites. After April 30 this changed dramatically for no apparent reason. However, the change in orbit and clock is in itself consistent such that an “ordinary” user of the system would actually not notice this effect.

The 150-meter orbit difference is fully reflected in the satellite clock bias. Hence, the user range error is almost unaffected by this change. We notice this 150-meter offset in our software because in our preprocessing we perform an orbit fit through the broadcast ephemerides.

This leads to some problems because the orbit resulting from the fit has to “obey” the equations of motion of the satellite center of mass. The resulting orbit fit therefore has an RMS agreement with the broadcast ephemerides at the 80-meter level, whereas for normal GPS satellites the agreement after fit is at the 1-meter level.

Consequently our resulting a priori orbit is no longer consistent with the a priori clock offsets coming from the broadcast ephemerides. This discrepancy, at the 80-meter level, leads to the exclusion of the satellite in our preprocessing.

SVN49 Signal Anomaly
As the next step, let us look at the reported signal anomaly on SVN49. In the work we do at ESOC the most important observation type is the carrier phase observation. The pseudoranges are used for preprocessing, to determine the absolute value of the receiver and transmitter clock offsets, and for integer ambiguity resolution.

This relative unimportance of the pseudorange observations is also reflected in the fact that the weight of the carrier phase observations in our postprocessing algorithms is 1,000 times larger than the weight of the pseudorange observations.

This also means that an anomaly in the carrier phase observations would seriously affect us whereas an anomaly in the code observations would have hardly any effect. Furthermore, we should note that as pseudorange and carrier phase observables we use the ionosphere-free linear combination of the observations on the two (L1/L2) frequencies.

Elevation Dependence
. . . the mean and the RMS of the SVN49 residuals are significantly higher than those of the other GPS satellites. The mean is typically around zero but for SVN49 it reaches approximately -1 meter. For SVN49 the RMS of the residuals reaches a magnitude of two meters whereas for the other GPS satellites it is at the one-meter level. The non-zero mean of the residuals is unpleasant because this could have a serious effect on our integer ambiguity resolution in which the code observations are used . . .

Other Satellites
As mentioned at the beginning of this article, we extended our investigation to other Block IIR and IIR-M satellites that ESOC routinely processes and found that several of these also exhibit a similar, though much smaller anomaly . . .

Unresolved Issue
In the scope of these anomaly investigations one additional interesting aspect surfaced. We noticed that the residuals of the IGS station ZIM2 (Zimmerwald, Switzerland) did not show the SVN49 anomaly. After some investigation we found that ZIM2 represented a model of GPS/GLONASS receiver that was tracking unhealthy satellites. Some further “data mining” revealed two other such receivers with data for SVN49, ROSA (Rosana, Brasil) and GANP (Ganovce, Slovakia) . . .

Conclusions
We have clearly shown that a significant issue exists with the code observations of GPS SVN49 (PRN01) whereas the carrier phase observations seem to perform nominally. The issue manifests itself as an elevation-dependent bias with a peak-to-peak difference of approximately four meters in the ionosphere-free observations (PC).

We have furthermore demonstrated that the broadcast ephemerides of this satellite are altered by approximately 150 meters (and 500 nanoseconds for the clock) and have shown that this is done to reduce the effect of the observed anomaly. An artificial shift of the satellite antenna offset by 150 meters does correct a significant amount of the observed elevation-dependent code bias.

The carrier phase observations seem to be fine, as we can do integer ambiguity resolution without too much problems. However, using the code observations for aiding the ambiguity resolution, for example, using the well-known Melbourne-Wübbena (MW) linear combination, is not advisable.

The figure illustrates that the effect of the anomaly on the MW observations is at the one-meter level, which means that it is of similar size as the wavelength of the MW linear combination (0.86 m). Therefore, finding the correct integer number of cycles will be difficult.

The different size of the bias in the ionosphere-free observations and the Melbourne-Wübbena observations is because these observations are based on different linear combinations of the dual-frequency observations. The bias is present only, or at least mainly, in the observations of the L1 frequency and thus its “amplification” is different for the PC and the MW linear combinations.

We conclude that with the currently implemented “correction scheme,” which uses an artificial satellite antenna offset of -150 meters, most regular “navigation” users of GPS will be able to use the satellite without any loss of accuracy.

However, high-end users, especially those performing ambiguity resolution using code observations, will have to model the anomaly more accurately, e.g., using a third-order polynomial as we have done here. Of course, the model should be derived for the code observations on the L1 frequency to ensure that both the PC and the MW observations are properly corrected.

The GPS Wing is currently gathering information from manufacturers and user communities regarding the effect of the SVN49 anomaly on a wide range of GPS products and applications. The research will be used to reach a decision on whether or not to set the satellite healthy and, if set healthy, what to do to mitigate the signal anomaly prior to setting it healthy.

For the complete story, including figures, graphs, and images, please download the PDF of the article, above.

Acknowledgment
The authors gratefully acknowledge IGS for providing GPS data and ephemerides.

Manufacturers

The GPS/GLONASS receiver model that was able to track SVN49 without modification, as reflected in Figure 4, is the NetR5 from Trimble, Sunnyvale, California. The GPS receiver that also was able to track the anomalous signals was the Trimble NetRS.