4.4.4 Testing GPS Surveying Systems:

INVESTIGATIVE TESTS

The defining characteristics of these tests is the desire to understand a certain phenomenon that impacts on GPS performance through a carefully designed experiment. Invariably such testing is carried out by academic institutions. Because there are so many factors impacting on baseline accuracy, it is necessary to constrain or standardise as many of the factors as possible. These could be done in the following way:

• Do all the tests on a single baseline, but varying the other factors such as time of day, length of observation session, etc.
  
• Test several different GPS instruments simultaneously and, if possible, over the same baseline.
  
• Use the same GPS datasets, but vary the processing options, and even the software, in order to gauge the impact of different modelling and processing strategies.
  
• In a controlled manner induce a disturbing influence in order to determine realistic operational accuracies across all plausible conditions.

Tests include those that study:

• the cycle of receiver-satellite geometry effects on baseline results,
• the impact of using different data pre-processing strategies,
• the quality of the raw tracking data, under different site conditions,
• the use of different styles of GPS antenna,
• the impact of using different data processing software, or
• the impact of using different observation models.

One institution undertaking considerable GPS receiver testing is the UNAVCO organisation (University NAVstar Consortium), in particular into the effects of multipath. Their reports can be accessed via Internet (section 3.4.2). Many other multipath investigations have also been reported in the literature.

Another interesting experiment is reported by HUSTI et al (1994). The test involved collecting phase data with a variety of GPS surveying receivers, on a number of baselines of varying length. The data was converted into the RINEX (Receiver INdependent EXchange) format (GURTNER, 1994). The quality of the data collected by each receiver was judged to be of similar high quality because when the data was processed using receiver-independent software (the "Bernese software") the results that were obtained were very similar. However, when each of the instrument datasets was processed using the proprietary commercial software, differences in the baseline results were noticed. Several tests such as these have been carried out, and the basic conclusion is that the mixing of GPS receivers and software is still a risky proposition.

The Netherlands Geodetic Commission has also compared the performance of different codeless GPS receivers under A/S conditions. SLUITER & HAAGMANS (1995) report the results of investigations into the susceptibility of the receivers to intentional and unintentional jamming. (Unfortunately, they do not name the best performing receiver!)

Another type of test seeks to investigate the variability of the baseline results. GPS positioning accuracy is influenced by such factors as (section 2.4.1):

• The observation type -- and hence the measurement noise.
• The signal disturbances present -- and responsible for systematic biases.
• The algorithms used -- appropriate for the level of biases and the accuracy sought, but tempered by operational issues.
• Operational issues such as satellite geometry, length of observation session, static or moving antenna, baseline length separating a pair of antennas, etc.

Assuming a certain hardware configuration, and acknowledging the presence of an acceptable level of residual biases (those that remain after the application of processing algorithms based on double-differencing simultaneously observed tracking data -- section 6.3.2), it is the operational issues that influence accuracy the most. The operational issues listed above will impact the quality of the positioning results in several ways, for example:

(a) the magnitude of the residual biases is most affected by baseline length (the longer the baseline, the larger the residual biases, and hence there is a commensurate reduction in accuracy), and

(b) the length of the observation session affects the sensitivity of the solution to residual biases, geometry, etc. (all other things being equal, solutions based on short observation sessions are generally less reliable than those from longer sessions).

Hence, there are two options:

• Test a single baseline by collecting 24 hours of GPS data, and partitioning the dataset into different session lengths (15, 30, 45, 60 ... minutes) before processing the data sessions.
• Test a variety of baseline lengths by collecting 24 hours of GPS data, and partitioning all the datasets into the same observation session length before processing the data sessions.

In both cases there are many results that then have to be systematically analysed in order to ascertain the overall performance characteristics of GPS (and especially how variable the baseline accuracy is) as a function of parameters that can be influenced such as: baseline length and observation session length; and as a function of parameters over which there is little control such as: satellite geometry, multipath and signal disturbances. An example is illustrated in the Figures below, where a 24 hour dataset collected on an approximately 4km baseline has been partitioned into consecutive 5 minute (Figure a) and 30 minute (Figure b) observation sessions, and the same data processing algorithm applied to the phase data (double-differenced ambiguity-free solutions).

Double-difference phase solution (ambiguity-free) for 4km baseline:
(a) 5 minute observation sessions (Figure left), (b) 30 minute observation sessions (Figure right).

Back To Chapter 4 Contents / Next Chapter / Previous Topic

© Chris Rizos, SNAP-UNSW, 1999