TEOCO, a provider of assurance, analytics and optimization solutions to communications service providers, has expanded the testing capabilities of its AIRCOM Device Test Lab in Columbia, Maryland, with additional test systems from Rohde & Schwarz. Combined with Rohde & Schwarz’s conformance and carrier test case support, the R&S TS-LBS Location Based Services (LBS) and the R&S CMW-PQA Performance Test Systems enable TEOCO to support new industry and carrier testing requirements for next-generation wireless technologies including carrier aggregation, IMS, VoLTE, RCS, E911 over IMS, LTE A-GNSS, LTE OTDOA and LTE eCID.
Release 10 and beyond, also referred to as LTE-Advanced, allows for a substantial uplift in the capacity and throughput of LTE, in addition to mobile device performance improvements. In LTE-Advanced, carrier aggregation (CA) is a key feature that allows the combination of multiple carriers to increase bandwidth and ultimately data rates in the network. To meet this need, the R&S CMW-PQA test system performs automated testing of a devices downlink, uplink and bidirectional data performance with or without carrier aggregation under simulated network conditions.
LBS is already a key technology enabling a myriad of new applications that people depend on every day. LTE and additional satellite constellations are being leveraged to improve availability and performance of location technologies indoors and outdoors. Higher customer expectations are driving the need for more advanced testing methodologies, and the R&S TS-LBS test system supports field-to-lab testing where real-world conditions are captured with high performance 16-bit resolution, and replayed in the lab for more accurate simulation.
“We are excited to expand our testing capability and capacity for LTE-Advanced and LBS-enabled devices,” said Hemant Minocha, Executive Vice President at TEOCO. “And given the increase in test complexity and costs, we are pleased to be working with a partner such as Rohde & Schwarz whose expertise in conformance and carrier acceptance testing and breadth of test cases helps future-proof our investment while delivering the quality and results our customers have come to expect.”
Averna, a developer of test solutions and services for communications and electronics device-makers worldwide, has acquired U.S.-based Cal-Bay Systems, a privately owned provider of test systems, test and measurement solutions, automated testing and vibration monitoring tools. This move is intended to strengthen Averna’s presence in the consumer electronics, medical device, and vibration monitoring markets, as well as providing the company with strategic positioning on the U.S. West Coast.
Averna has acquired 100-percent share of Cal-Bay Systems for an undisclosed amount and will take on management of its U.S. and European offices. The current owners, Buck Smith and Patrick Kelly, will continue to participate in the day-to-day operations and expansion plans. Cal-Bay Systems, headquartered in San Rafael, California, brings to Averna 20 years of test expertise and customer relationships in comprehensive, automated test and measurement processes for mission-critical applications.
“We are thrilled with the acquisition of Cal-Bay, as it is very synergistic with our portfolio of test engineering services and expertise, and constitutes another building block of our growth plan,” said Averna President and CEO André Gareau. “Cal-Bay’s customer base and strategic location on the U.S. West Coast are a great addition to our current operations. This is also a fantastic growth opportunity for Cal-Bay’s team and we welcome them into the Averna family.”
Gareau continued, “These are exciting times for our company and shareholders. This acquisition brings a wealth of business opportunities for our company, and could not have been accomplished without the talented team that we have assembled.”
Rohde & Schwarz will be demonstrating its SMBV100A simulator at ION GNSS+ 2014, and is offering a new Wireless Standards poster for visitors to its booth. Rohde & Schwarz will be exhibiting in Booth I during the show, which will be held in Tampa, Florida, September 10-11.
Based on a digital vector signal generator, the SMBV100A supports realtime and hybrid configurations up to 24 dynamic GPS, GLONAS, Galileo, BeiDou and QZSS satellites and can be synchronized for multi-channel RF solutions. Rohde & Schwarz will be demonstrating its latest SMBV100A solutions and technologies along with dedicated solutions to support navigation and GNSS testing including:
Support realistic user environments: obscuration, multipath, antenna characteristics and vehicle attitude
Realtime external trajectory feed for hardware in the loop (HIL) applications
High signal dynamics, simulation of spinning vehicles and precision code (P code) simulations
Support for ground-based augmentation systems (GBAS)
Avionics test solutions: VOR, ILS, DME, and TACAN
Field-to-lab capture and playback solutions
Interference hunting and direction finding solutions
New Wireless Standards Poster – Register Now!
Conference attendees who stop by Booth I can register for a new Wireless Standards Poster. The convenient wall chart provides an overview of all the major standards for digital cellular, public safety, TV white space, wireless connectivity and GNSS technologies.
Not able to attend? Click here to register to have a poster sent to you.
FRINGES. No, I’m not talking about the latest celebrity hairstyles nor the canopy of an American doorless, four-wheeled carriage from yesteryear (think Oklahoma!). I’m talking about interference fringes. But there is a connection to these other uses of the word fringe as we’ll see. You’ve all seen interference fringes at your local gas station, typically after it has just rained. They are the alternating bands of color we perceive when looking at a gasoline or oil slick in a puddle of water. They are caused by the white light from the Sun or artificial lighting reflected from the top surface of the slick and that from the bottom surface at the slick-water interface combining or interfering with each other at our eyeballs. The two sets of light waves arrive slightly out of phase with each other, and depending on the wavelengths of the reflected light and our angle of view, produce the colorful fringes. If the incident light was monochromatic, consisting of a single frequency or wavelength, then we would perceive just alternating bright and dark bands. The bright bands result from constructive interference when the phase difference is a near a multiple of 2π whereas the dark bands result from destructive interference when the difference is near an odd multiple of π.
Interference fringes had been seen long before the invention of the automobile. They are clearly seen on soap bubbles and the iridescent colors of peacock feathers, Morpho butterflies, and jewel beetles are also due to the interference phenomenon rather than pigmentation. Sir Isaac Newton did experiments on interference fringes (amongst other things) and tried to explain their existence — wrongly, it turned out. But he did coin the term fringes since they resembled the decorative fringe sometimes used on clothing, drapery, and, yes, surrey canopies.
It was the English polymath, Thomas Young, who, in 1801, first demonstrated interference as a consequence of the wave-nature of light with his famous double-slit experiment. You may have replicated his experiment in a high-school physics class. I did and I think I did it again as an undergraduate student taking a course in optics. Already by that point I was aiming for a career in physics or space science but I didn’t know that as a graduate student I would do research involving interference fringes. But not using light waves.
My research involved the application of very long baseline interferometry or VLBI to geodesy. VLBI had been developed by radio astronomers to better understand the structure of quasars and other esoteric celestial objects. At either ends of a baseline connecting large radio telescopes, perhaps stretching between continents, the quasar signals were recorded on magnetic tape and precisely registered using atomic clocks. When the tapes were played back and the signals aligned, one obtained interference fringes as peaks and troughs in an analog or digital waveform. Computer analysis of these fringes not only provided information on the structure of the observed radio source but also on the distance between the radio telescopes — eventually accurate enough to measure continental drift.
But what has all of this got to do with GPS? In this month’s column, we look at a technique that uses fringes generated by signals arriving at an antenna directly from GPS satellites and those reflected by snow surrounding the antenna to measure its depth and how it varies over time. GPS for measuring snow depth; who would have thought?
“Innovation” is a regular feature that discusses advances in GPS technology and its applications as well as the fundamentals of GPS positioning. The column is coordinated by Richard Langley of the Department of Geodesy and Geomatics Engineering, University of New Brunswick. He welcomes comments and topic ideas.
Snowpacks are a vital resource for human existence on our planet. They provide reservoirs of fresh water, storing solid precipitation and delaying runoff. One sixth of the world population depends on this resource. Both scientists and water-supply managers need to know how much fresh water is stored in snowpack and how fast it is being released as a result of melting.
Snow monitoring from space is currently under investigation by both NASA and ESA. Greatly complementary to such spaceborne sensors are automated ground-based methods; the latter not only serve as essential independent validation and calibration for the former, but are also valuable for climate studies and flood/drought monitoring on their own. It is desirable for such estimates to be provided at an intermediary scale, between point-like in situ samples and wider area pixels.
In the last decade, GPS multipath reflectometry (GPS-MR), also known as GPS interferometric reflectometry and GPS interference-pattern technique, has been proposed for monitoring snow. This method tracks direct GPS signals, those that travel directly to an antenna, that have interfered with a coherently reflected signal, turning the GPS unit into an interferometer (see FIGURE 1). Its main variant is based on signal-to-noise ratio (SNR) measurements, although GPS-MR is also possible with carrier-phase and pseudorange observables. Data are collected at existing GPS base stations that employ commercial-off-the-shelf receivers and antennas in a conventional, antenna-upright setup. Other researchers have used a custom antenna and/or a dedicated setup, with the antenna tipped for enhanced multipath reception.
FIGURE 1. Standard geodetic receiver installation. The antenna is protected by a hemispherical radome. The monument (tripod structure) is ~ 2 meters above the ground. GPS satellites rise and set in ascending and descending sky tracks, multiple times per day. The specular reflection point migrates radially away from the receiver for decreasing satellite elevation angle. The total reflector height is made up of an a priori value and an unknown bias driven by the thickness of the snow layer.
In this article, we summarize the SNR-based GPS-MR technique as applied to snow sensing using geodetic instruments. This forward/inverse approach for GPS-MR is new in that it capitalizes on known information about the antenna response and the physics of surface scattering to aid in retrieving the unknown snow conditions in the site surroundings. It is a statistically rigorous retrieval algorithm, agreeing to first order with the simpler original methodology, which is retained here for the inversion bootstrapping. The first part of the article describes the retrieval algorithm, while the second part provides validation at a representative site over an extended period of time.
Physical Forward Model
SNR observations are formulated as SNR = Ps/Pn. In the denominator, we have the noise power, Pn, here taken as a constant, based on nominal values for the noise power spectral density and the noise bandwidth. The numerator is composite signal power:
. (1)
Its incoherent component is the sum of the respective direct and reflected powers (although direct incoherent power is negligible). In contrast, the coherent composite signal power follows from the complex sum of direct and reflection average voltages (not to be confused with the electromagnetic propagating fields, which neglect the receiving antenna response and also the receiver tracking process):
(2)
It is expressed in terms of the coherent direct and reflected powers, as well as the interferometric phase,
, (3)
which amounts to the reflection excess phase with respect to the direct signal.
We decompose observations, SNR = tSNR + dSNR, into a trend
(4)
over which interference fringes are superimposed:
. (5)
From now on, we neglect the incoherent power, which only impacts tSNR, not dSNR, and drop the coherent power superscript, for brevity.
The direct or line-of-sight power is formulated as
(6)
where is the direction-dependent right-hand circularly polarized (RHCP) power component incident on an isotropic antenna; the left-handed circularly polarized (LHCP) component is negligible. The direct antenna gain, , is obtained evaluating the antenna pattern in the satellite direction and with RHCP polarization.
The reflection power,
, (7)
is defined starting with the same incident isotropic power, , as in the direct power. It ends with a coherent power attenuation factor,
(8)
where θ is the angle of incidence (with respect to the surface normal), k = 2π/λ, is the wave number, and λ = 24.4 centimeters is the carrier wavelength for the civilian GPS signal on the L2 frequency (L2C). This polarization-independent factor accounts only for small-scale residual height above and below a large-scale trend surface. The former/latter results from high-/low-pass filtering the actual surface heights using the first Fresnel zone as a convolution kernel, roughly speaking. Small-scale roughness is parameterized in terms of an effective surface standard deviation s (in meters); its scattering response is modeled based on the theories of random surfaces, except that the theoretical ensemble average is replaced by a sensing spatial average. Large-scale deterministic undulations could be modeled, but their impact on snow depth is canceled to first-order by removing bare-ground reflector heights.
At the core of , we have coupled surface/antenna reflection coefficients, , producing respectively RHCP and LHCP fields (under the assumption of a RHCP incident field). These terms include antenna response power gain and phase patterns, evaluated in the reflection direction, and separately for each polarization. The surface response is represented by complex-valued Fresnel coefficients for cross- and same-sense circular polarization, respectively. The medium is assumed to be homogeneous (that is, a semi-infinite half-space). Material models provide the complex permittivity, which drives the Fresnel coefficients.
The interferometric phase reads:
.(9)
The first term accounts for the surface and antenna properties of the reflection, as above. The last one is the direct phase contribution, which amounts to only the RHCP antenna phase-center variation evaluated in the satellite direction. The majority of the components present in the direct RHCP phase (such as receiver and satellite clock states, the bulk of atmospheric propagation delays, and so on) are also present in the reflection phase, so they cancel out in forming the difference.
At the core of the interferometric phase, we have the geometric component, φI = kτi, the product of the wave number and the interferometric propagation delay. Assuming a locally horizontal surface, the latter is simply:
(10)
in terms of the satellite elevation angle, e, and an a priori reflector height, HA. Snow depth will be measured in terms of changes in reflector height.
The physical forward model, based only on a priori information, can then be summarized as:
(11)
where interferometric power and phase are, respectively:
(12)
. (13)
In all of these terms the pseudorandom-noise-code modulation impressed on the carrier wave can be safely neglected, given the small interferometric delay and Doppler shift at grazing incidence, stationary surface/receiver conditions, and short antenna installations.
Parameterization of Unknowns
There are errors in the nominal values assumed for the physical parameters of the model (permittivity, surface roughness, reflector height, and so on). Ideally we would estimate separate corrections for each one, but unfortunately many are linearly dependent or nearly so. Because of this dependency, we have kept physical parameters fixed to their optimal a priori values, and have estimated a few biases. Each bias is an amalgamation of corrections for different physical effects. In a later stage, we rely on multiple independent bias estimates (such as for successive days) to try and separate the physical sources.
Each satellite track is inverted independently. A track is defined by partitioning the data by individual satellite and then into ascending and descending portions, splitting the period between the satellite’s rise and set at the near-zenith culmination. Each satellite track has a duration of ~1–2 hours. This configuration normally offers a sufficient range of elevation angles, unless the satellite reaches culmination too low in the sky (less than about 20°), in which case the track is discarded. In seeking a balance between under- and over-fitting, between an insufficient and an excessive number of parameters, we estimate the following vector of unknown parameters:
. (14)
FIGURE 2 shows the effect of the constant and linear biases on the SNR observations. Reflector height bias, HB, changes the number of oscillations; phase shift, φB, displaces the oscillations along the horizontal axis; reflection power, , affects the depth of fades; zeroth-order noise power, , shifts the observations up or down as a whole; and first-order noise power, , tilts the SNR curve. A good parameterization yields observation sensitivity curves as unique as possible for each parameter.
FIGURE 2. Effect of each parameter on SNR observations; curves are displaced vertically (6 dB) for clarity.
The forward model, now including the biases, can be summarized as follows:
(15)
where the modified interferometric power and phase are given by:
, (16)
. (17)
The total reflector height, H = HA – HB (a priori value minus unknown bias), is to be interpreted as an effective value that best fits measurements, which includes snow and other components.
Bootstrapping Parameter Priors. Biases and SNR observations are involved non-linearly through the forward model. Therefore, there is the need for a preliminary global optimization, without which the subsequent final local optimization will not necessarily converge to the optimal solution.
SNR observations would trace out a perfect sinusoid curve in the case of an antenna with isotropic gain and spherical phase pattern, surrounded by a smooth, horizontal, and infinite surface (free of small-scale roughness, large-scale undulations, and edges), made of perfectly electrically conducting material, and illuminated by constant incident power. Thus, in such an idealized case, SNR could be described exactly by constant reflector height, phase shift, amplitude, and mean values.
As the measurement conditions become more complicated, the SNR data start to deviate from a pure sinusoid. Yet a polynomial/spectral decomposition is often adequate for bootstrapping purposes.
Statistical Inverse Model Formulation
Based on the preliminary values for the unknown parameters vector and other known (or assumed) values, we run the forward model to obtain simulated observations. We form pre-fit residuals comparing the model values to SNR measurements collected at varying satellite elevation angles (separately for each track). Residuals serve to retrieve parameter corrections, such that the sum of squared post-fit residuals is minimized. This non-linear least squares problem is solved iteratively using both a functional model and a stochastic model. The functional modeling includes a Jacobian matrix of partial derivatives, which represents the sensitivity of observations to parameter changes where the partial derivatives are defined element-wise. Instead of deriving analytical expressions, we evaluate them numerically, via finite differencing. The stochastic model specifies the uncertainty and correlation expected in the residuals. Their a priori covariance matrix modifies the objective function being minimized.
Directional Dependence
It is important to know at which elevation angles the parameter estimates are best determined. Here, we focus on the phase parameters instead of reflection power or noise power parameters.
We can utilize the estimated reflector height and phase shift to evaluate the full phase bias function over varying elevation angles. Similarly, we can extract the corresponding 2-by-2 portion of the parameters’ a posteriori covariance matrix, containing the uncertainty for reflector height and for phase shift, as well as their correlation, which is then propagated to obtain the full phase uncertainty (see FIGURE 3).
FIGURE 3. Uncertainty of full phase function, propagated from the uncertainty of reflector height and of phase shift, as well as their correlation.
The uncertainty attains a clear minimum versus elevation angle. The least-uncertainty elevation angle pinpoints the observation direction where reflector height and phase shift are best determined (in combined form, not individually). The azimuth and epoch coinciding with the peak elevation angle act as track tags, later used for clustering similar tracks and analyzing their time series of retrievals.
If we normalize phase uncertainty by its value at the peak elevation angle, then plot such sensing weights (between 0 and 1) versus the radial or horizontal distance to the center of the first Fresnel zone at each elevation angle, we obtain FIGURE 4. It can be interpreted as the reflection footprint, indicating the importance of varying distances, with a longer far tail and a shorter near tail (respectively regions beyond and closer than the peak distance). The implications for in situ data collection are clear: one should sample more intensely near the peak distance (about 15 meters) and less so in the immediate vicinity of the GPS antenna, tapering it off gradually away from the antenna. As a caveat, these conclusions are not necessarily valid for antenna setups other than the one considered here.
FIGURE 4. Reflection footprint in terms of a sensing weight (between 0 and 1) defined as the normalized reciprocal of full phase uncertainty, plotted versus the radial or horizontal distance from the receiving antenna to the center of the first Fresnel zone at each elevation angle; valid for an upright 2-meter-tall antenna; the receiving antenna is at zero radial distance.
Results
We now examine the snow-depth retrievals from the GPS multipath retrieval algorithm and assess both the precision and accuracy of the method. Multiple metrics have been developed to assess the quality of the results. The accuracy of the method has been evaluated by comparing with in situ data over a multi-year period. Three field sites were chosen to highlight different limitations in the method, both in terms of terrain and forest cover: grassland, alpine, and forested. We will look at the forested site in some detail.
Satellite Coverage and Track Clustering. All GPS-MR retrievals reported here are based on the newer GPS L2C signal. Of the approximately 30 GPS satellites in service, 8-10 L2C satellites were available between 2009 and 2012 (8, 9, and 10 satellites at the end of 2009, 2010, and 2011, respectively). Satellite observations were partitioned into ascending and descending portions, yielding approximately twenty unique tracks per day at a site with good sky visibility. GPS orbits are highly repeatable in azimuth, with deviations at the few-degree range over a year, translating into ~50-100-centimeter azimuthal displacement of the reflecting area (corresponding to the first Fresnel zone at 10°-15° elevation angle for a 2-meter high antenna). This repeatability permits clustering daily retrievals by azimuth. It also allows the simplification that estimated snow-free reflector heights are fairly consistent from day to day, facilitating the isolation of the varying snow depth during the snow-covered period.
For a given track, its revisit time is also repeatable, amounting to practically one sidereal day. The deficit in time relative to a calendar day results in the track time of the day receding ~4 minutes and 6 seconds every day. This slow but steady accumulation eventually makes the time of day return to its starting value after about one year. As all GPS satellites drift approximately at the same rate, the time between successive tracks remains nearly repeatable. Its reciprocal, the sampling rate, has a median equal to approximately one track per hour, with a low value of one track within two hours and a high of one track within 15 minutes; both extremes occur every day, with low-rate idle periods interspersed with high-rate bursts. The time of the day reduced to a fixed day (such as January 1, 2000) could also be used to cluster tracks. Neighboring clusters, which are close in azimuth and/or in reduced time of the day, are expected to be more comparable, as they sample similar conditions and are subject to similar errors.
Observations. FIGURE 5 shows several representative examples of SNR observations. A typical good fit between measured and modeled values is shown in Figure 5(a), corresponding to the beginning of the snow season. Generally the model/measurement fit is good when the scattering medium is homogeneous; it deteriorates as the medium becomes more heterogeneous, particularly with mixtures of soil, snow, and vegetation. There are genuine physical effects as well as more mundane spurious instrumental issues that degrade the fit but do not necessarily cause a bias in snow-depth estimates. These include secondary reflections, interferometric power effects, direct power effects, and instrument-related issues.
FIGURE 5. Examples of observations: (a) good fit; (b) presence of secondary reflections; (c) vanishing interference fringes; (d) atypical interference fringes.
Secondary reflections originate from disjoint surface regions. Interference fringes become convoluted with multiple superimposed beats (see Figure 5(b)). As long as there is a unique dominating reflection, the inversion will have no difficulty fitting it, as the extra reflections will remain approximately zero-mean.
Random deviations of the actual surface with respect to its undulated approximation, called roughness or residual surface height, will affect the interferometric power. SNR measurements will exhibit a diminishing number of significant interference fringes, compared to the measurement noise level (see Figure 5(c)). This facilitates the model fit but the reflector height parameter may become ill-determined: its estimates will be more uncertain. Changes in snow density also affect the fringe amplitude.
Snow precipitation attenuates the satellite-to-ground radio link, which affects SNR measurements through the direct power term. First, this shifts the SNR measurements up or down (in decibels); second, it tilts the trend tSNR as attenuation is elevation-angle dependent; third, fringes in dSNR will change in amplitude because of the decrease in the coherent component of the direct power.
Partial obstructions can affect either or both direct and interferometric powers. In this case, SNR measurements, albeit corrupted, are still recorded. This situation is in contrast to complete blockages as caused by topography. The deposition of snow and the formation of a winter rime on the antenna are a particularly insidious type of obstruction, as their presence in the near-field of the antenna element can easily distort the gain pattern in a significant manner. In the far-field, trees are another important nuisance, so much so that their absence is held as a strong requirement for the proper functioning of multipath reflectometry.
Satellite-specific direct power offsets and also long-term power drifts are to be expected as spacecraft age and modernized designs are launched. In addition, noise power depends on the state of conservation of receiver cables and on their physical temperature. Less subtle incidents are sudden ~3-dB SNR steps, hypothesized to originate in the receiver switching between the L2C data and pilot subcodes, CM and CL.
Quality Control. Anomalous conditions may result in measurement spikes, jumps, and short-lived rapidly-varying fluctuations. For snow-depth-sensing purposes, it is necessary and sufficient to either neutralize such measurement outliers through a statistically robust fit or detect unreliable fits and discard the problematic ones that could not otherwise be salvaged.
The key to quality control (QC) is in grouping results into statistically homogeneous units, having measurements collected under comparable conditions. In our case, azimuth-clustered tracks are the natural starting unit. Secondarily, we must account for genuine temporal variations in the tendency of results, from beginning to peak to the end of the snow season. The detection of anomalous results further requires an estimate of the statistical dispersion to be expected. Considering that the sample is contaminated with outliers, robust estimators (running median instead of the running mean, and median absolute deviation over the standard deviation) are called for, if the first- and second-order statistical moments are to be representative. Given estimates of the non-stationary tendency and dispersion, a tolerance interval can then be constructed such that it bounds, say, a 99% proportion of the valid results with 95% confidence level. We also desire QC to be judicious, or else too many valid estimates will be lost. Notice that in the present intra-cluster QC, we compare an individual estimate to the expected performance of the track cluster to which it belongs; later, we complement QC with an inter-cluster comparison of each cluster’s own expected performance.
Based on our practical experience, no single statistic detects all the outliers. We use four particular statistics that we have found to be useful: 1) degrees of freedom, essentially the number of observations per track (modulo a constant number of parameters); 2) using the scaled root-mean-square error (RMSE) to test for goodness-of-fit, that is, how well measurements can be explained adjusting the unknown values for the parameters postulated in the model; 3) reflector height uncertainty; and 4) peak elevation angle, which behaves much like a random variable, as it is determined by a multitude of factors.
Combinations. We combine multiple clusters to average out random noise. Noise mitigation aims at not only coping with measurement errors but also compensating for model deficiencies, to the extent that they are not in common across different clusters. Before we combine different clusters, we have to address their long-term differences. The initial situation is that snow surface heights will be greater downhill and smaller uphill; we take this into account on a cluster-by-cluster basis by subtracting ground heights from their respective snow surface heights, resulting in snow thickness values, which is a completely physically unambiguous quantity. Snow thickness is more comparable than snow heights across varying-azimuth track clusters. Yet snow tends to fill in ground depressions, so thickness exhibits variability caused by the underlying ground surface, even when the overlying snow surface is relatively uniform. Further cluster homogeneity can be achieved by accounting for the temporally permanent though spatially non-uniform component of snow thickness.
The averaging of snow depths collected for different track clusters employs the inversion uncertainties to obtain a preliminary running weighted median, calculated for, say, daily postings, with overlapping windows or not. The preliminary post-fit residuals then go through their own averaging, necessarily employing a wider averaging window (say, monthly), which produces scaling factors for the original uncertainties. The running weighted median is then repeated, producing final averages. The variance factors reflect the fact that some clusters are better than others.
Thus, the final GPS estimates of snow depth follow from an averaging of all available tracks, whose individual snow depth values were previously estimated independently. A new average is produced twice daily utilizing the surrounding 1–2 days of data (depending on the data density), that is, 12-hour posting spacing and 24-hour moving window width. The averaging interval must be an integer number of days, so as to minimize the possibility of snow-depth artifacts caused by variations in the observation geometry, which repeats daily.
Site-Specific Results
We explored GPS-MR snow-depth retrieval at three stations over a long period (up to three years). Throughout, we assessed the performance of the GPS estimates against independent nearly co-located in situ measurements. We also compared the GPS estimates to the nearest SNOTEL station. SNOTEL (from snowpack telemetry) is an automated system for collecting snowpack and related data in the western U.S. operated by the U.S. Department of Agriculture. Although not co-located with GPS, SNOTEL data are important because they provide accurate information on the timing of snowfall events.
The three sites we used were 1) a site in the T.W. Daniel Experimental Forest within the Wasatch Cache National Forest in the Bear River Range of northeastern Utah, with an elevation of 2,600 meters; 2) one of the stations of the EarthScope Plate Boundary Observatory, a grassland site located near Island Park, Idaho; and 3) an alpine site in the Niwot Ridge Long-term Ecological Research Site near Boulder, Colorado. While we have fully documented the results from each site, due to space limitations we will only discuss the results from the forested site (known as RN86) in this article. This is a more challenging site than the other two, due to the presence of nearby trees. Furthermore, it was subject to denser in situ sampling of 20-150 measurements spatially replicated around the GPS antenna, and repeated approximately every other week for about one year.
We show results for the 2012 water-year, the period starting October 1 through September 30 of the following year. Where GPS site RN86 was installed, topographical slopes range from 2.5° to 6.5° (at the 2-meter spatial scale), with average of ~5° within a 50-meter radius around the GPS antenna. RN86 was specifically built to study the impact of trees on GPS snow depth retrievals (see FIGURE 6). Ground crews manually collected in situ measurements around the GPS antenna approximately every other week starting in November 2011. Measurements were made every 1–2 meters from the antenna up to a distance of 25-30 meters. In the second half of the year, the sampling protocol was changed to azimuths of 0° (N), 45° (NE), 135° (SE), 180° (S), 225° (SW), and 315° (NW). With these data it is possible to obtain in situ average estimates, with their own uncertainties (based on the number of measurements), which allows a more meaningful comparison.
FIGURE 6. Aerial view of the forested site (RN86) around the GPS antenna (marked with a circle).
There is reduced visibility at the current site, compared to other sites. Track clusters are concentrated due south, with only two clusters located within ±90° of north. Therefore, the GPS average snow depth is not necessarily representative of the azimuthally symmetric component of the snow depth. In the presence of an azimuthal asymmetry in the snow distribution around the antenna, the GPS average would be expected to be biased towards the environmental conditions prevalent in the southern quadrant. To rule out the possibility of an azimuthal artifact in the comparisons, we have utilized only the in situ data collected along the SE/S/SW quadrant.
The comparison shows generally excellent agreement between GPS and in situ data (see FIGURE 7). The first four and the last one in situ data points were collected with coarser spacing and/or smaller azimuthal coverage, which may be partially responsible for different performance in the first and second halves of the snow season. The correlation between GPS and in situ snow depth at RN86 amounts to 0.990, indicating a very strong linear relationship. Carrying out a regression between in situ and GPS values, the RMS of snow-depth residuals improves from 9.6 to 3.4 centimeters. The regression intercept and slope (with corresponding 95% uncertainties) amount to 15.4 ± 9.11 centimeters and 0.858 ± 0.09 meters per meter, respectively. According to these statistics, the null hypotheses of zero intercept and unity slope are rejected at the 95% confidence level. This implies that at this location GPS snow-depth estimates exhibit both additive and multiplicative biases. The latter is proportional to snow depth itself, meaning that, compared to an ideal one-to-one relationship, GPS is found to under-estimate in situ snow depth at this site by 14 ± 9%, although the uncertainty is somewhat large.
FIGURE 7. Snow-depth measurement at the forested site (RN86) for the water-year 2012
The SNOTEL sensors are exceptionally close to the GPS antenna at this site, about 350 meters horizontally distant with negligible vertical separation. Yet the former is located within trees, while the latter is located at the periphery of the forest and senses the reflections scattered from an open field. Therefore, only the timing of snowfall events agrees well, not the amount of snow. Although forest density is generally negatively correlated with snow depth, exceptions are not uncommon, especially in localized clearings exposed to intense solar radiation, where shading of the snow by the trees reduces ablation.
Conclusions
In this article, we have discussed a physically based forward model and a statistical inverse model for estimating snow depth based on GPS multipath observed in SNR measurements. We assessed model performance against independent in situ measurements and found they validated the GPS estimates to within the limitations of both GPS and in situ measurement errors after the characterization of systematic errors. The assessment yielded a correlation of 0.98 and an RMS error of 6–8 centimeters for observed snow depths of up to 2.5 meters at three sites, with the GPS underestimating in situ snow depth by ~5–15%. This latter finding highlights the necessity to assess effects currently neglected or requiring more precise modeling.
Acknowledgments
The research reported in this article was supported by grants from the U.S. National Science Foundation, NASA, and the University of Colorado. Nievinski has been supported by a Capes/Fulbright Graduate Student Fellowship and a NASA Earth System Science Research Fellowship. The article is based, in part, on two papers published in the IEEE Transactions on Geoscience and Remote Sensing: “Inverse Modeling of GPS Multipath for Snow Depth Estimation – Part I: Formulation and Simulations” and “Inverse Modeling of GPS Multipath for Snow Depth Estimation – Part II: Application and Validation.”
Manufacturers
For the forested site (RN86), a Trimble NetR9 receiver was used with a Trimble TRM57971.00 (Zephyr Geodetic II) antenna with no external radome.
FELIPE G. NIEVINSKI is a faculty member at the Federal University of Santa Catarina, Florianópolis, Brazil. He has also been a post-doctoral researcher at São Paulo State University, Presidente Prudente, Brazil. He earned a B.E. in geomatics from the Federal University of Rio Grande do Sul, Porto Alegre, Brazil, in 2005; an M.Sc.E. in geodesy from the University of New Brunswick, Fredericton, Canada, in 2009; and a Ph.D. in aerospace engineering sciences from the University of Colorado, Boulder, in 2013. His Ph.D. dissertation was awarded The Institute of Navigation Bradford W. Parkinson Award in 2013.
KRISTINE M. LARSON received a B.A. degree in engineering sciences from Harvard University and a Ph.D. degree in geophysics from the Scripps Institution of Oceanography, University of California at San Diego. She was a member of the technical staff at the Jet Propulsion Lab from 1988 to 1990. Since 1990, she has been a professor in the Department of Aerospace Engineering Sciences, University of Colorado, Boulder.
FURTHER READING
• Authors’ Journal Papers
“Inverse Modeling of GPS Multipath for Snow Depth Estimation—Part I: Formulation and Simulations” by F.G. Nievinski and K.M. Larson in IEEE Transactions on Geoscience and Remote Sensing, Vol. 52, No. 10, 2014, pp. 6555–6563, doi: 10.1109/TGRS.2013.2297681.
“Inverse Modeling of GPS Multipath for Snow Depth Estimation—Part II: Application and Validation” by F.G. Nievinski and K.M. Larson in IEEE Transactions on Geoscience and Remote Sensing, Vol. 52, No. 10, 2014, pp. 6564–6573, doi: 10.1109/TGRS.2013.2297688.
• More on the Use of GPS for Snow Depth Assessment
“Snow Depth, Density, and SWE Estimates Derived from GPS Reflection Data: Validation in the Western U.S.” by J.L. McCreight, E.E. Small, and K.M. Larson in Water Resources Research, published first on line, August 25, 2014, doi: 10.1002/2014WR015561.
“GPS Snow Sensing: Results from the EarthScope Plate Boundary Observatory” by K.M. Larson and F.G. Nievinski in GPS Solutions, Vol. 17, No. 1, 2013, pp. 41–52, doi: 10.1007/s10291-012-0259-7.
• GPS Multipath Modeling and Simulation
“Forward Modeling of GPS Multipath for Near-Surface Reflectometry and Positioning Applications” by F.G. Nievinski and K.M. Larson in GPS Solutions, Vol. 18, No. 2, 2014, pp. 309–322, doi: 10.1007/s10291-013-0331-y.
“An Open Source GPS Multipath Simulator in Matlab/Octave” by F.G. Nievinski and K.M. Larson in GPS Solutions, Vol. 18, No. 3, 2014, pp. 473–481, doi: 10.1007/s10291-014-0370-z.
This integrated circuit supports simultaneous reception and processing of the GPS L1/L5, Galileo E1/E5a, and GLONASS G1 signals with 40 tracking channels. The dual-band analog RF front-end is integrated on the same mixed-signal chip as the baseband hardware, including an embedded processor to close the tracking loops: overall, a compact, low-power, and low-cost solution.
By Fabio Garzia, Stefan Köhler, Santiago Urquijo, Philipp Neumaier,Jörn Driesen, Sybille Haas, Thomas Leineweber, Tao Zhang, Sascha Krause, Frank Henkel,Alexander Rügamer, Matthias Overbeck, and Günther Rohmer
Multi-constellation multi-band global navigation satellite system (GNSS) receivers can efficiently exploit the advantages derived from the modernization of existing GNSS constellations, such as GPS and GLONASS, as well as from the launch of new ones like Galileo and BeiDou. Utilizing multiple systems can significantly improve the availability of a navigation solution in urban canyons and heavily shadowed areas. Increased satellite availability also guarantees higher measurement redundancy and improved reliability. Moreover, the excellent inherent noise and multipath mitigation capabilities of the new and modernized wideband signals in the L5/E5a band, combined with the ionosphere error mitigation given by frequency diversity, significantly improves the accuracy in both measurement and position domains.
Still, most commercial fully-integrated single-chip mass market GNSS receivers use only a single-frequency band for their positioning, velocity, time (PVT) solution: either GPS L1 C/A or Galileo E1 and GLONASS G1. For example, the Teseo chips are single-chip solutions that support multiple constellations but only on one frequency band. This approach reduces design costs and enables the lowest consumption of power, but neglects the advantages of wideband signal processing – which offers increased robustness thanks to two simultaneous frequency band receptions and the capability of mitigating the ionosphere error.
Another approach for realizing multi-constellation multi-frequency solutions is to combine different chips for the analog front-end and the digital baseband. One fully integrated single-chip analog multi-band front-end for the simultaneous reception of GPS L1/L5, Galileo E1/E5, and GLONASS has been presented. However, this chip included only the front-end and requires an additional, separate digital-baseband solution.
The purpose of the NAPA project (NAvigation chip for Pedestrian navigation and higher precision Applications) is to close this gap by providing a fully integrated, compact, low-power, and low-cost solution in which the analog and digital parts of the GNSS receiver are integrated together on the same chip. The NAPA receiver offers all the advantages of multi-constellation reception with additional dual-frequency support.
The NAPA chip features a monolithic, single mixed-signal chip implementation of a multi-system, multi-band analog front-end and the related digital baseband core, including an embedded processor. The NAPA chip can be used as a stand-alone GNSS sensor, because no additional components are required to obtain a PVT solution. The ASIC was implemented in a low-power technology and adopts some ad-hoc low-power architectural features. In regard to costs, an ASIC solution is more convenient than FPGA, provided the non-recurring engineering costs (NRE) are amortized by the amount of chips manufactured and sold. The NAPA chip supports multi-system (GPS, Galileo, and GLONASS) and multi-band (GPS/Galileo L1/E1, L5/E5a, GLONASS G1) processing. Figure 1 shows the frequency band being selected for receiving and processing in the NAPA chip. With two fully deployed GNSS — GPS and GLONASS — NAPA chips can already be used in many commercial applications. Thanks to the spectral overlay of the GPS L1/L5 and Galileo E1/E5a signals, the chip is also ready for Galileo. The frequency selection features both the narrow-band legacy signals L1/G1, which can be used for fast acquisition. For highest tracking accuracy, the wideband GPS L5 and Galileo E5a BPSK(10) modulated signals can be utilized.
Figure 1. GNSS signals received and processed by the NAPA chip.
The higher accuracy is obtained primarily by the attenuation of the ionospheric error. The ionosphere is a dispersing media that can introduce a bias error between 1 and 20 m. Forming a linear combination of two independent frequency-band measurements, the ionospheric bias can be measured and almost completely removed. In addition, Precise Point Positioning and Wide/Narrow-laning combinations are possible, thanks to the second received frequency band. The first allows for the combination of precise satellite positions and clocks with multi-frequency measurements, providing cm/dm solutions. The second adopts fast ambiguity solutions for carrier-phase positioning and cycle-slip detection.
In this article, we present the NAPA chip in detail. We describe the architecture of the analog front-end and its digital counterpart and the innovative features of each. Then we provide details about chip implementation, manufacturing, and test setup. Finally, we present the first verification results and draw conclusions.
Architecture Overview
The NAPA chip architecture, depicted in Figure 2, is composed of two separate blocks integrated on the same silicon die: the analog core provides the functionality of a two-frequency radio-frequency (RF) front-end, whereas the digital part implements the main GNSS processing tasks, including the correlator channels and an embedded processor, and takes care of the RF front-end control. The interface between the two blocks is completely digital and provides synchronizers to ensure a valid clock domain crossing (CDC).
Figure 2. Overall NAPA architecture with emphasis on the digital core blocks.
Analog Front-End. The analog RF front-end supports the simultaneous reception of GPS L5 / Galileo E5a and GPS L1 / Galileo E1 / GLONASS G1 signals as well as modes where only one reception path is activated.
Both passive and active GNSS antennas are supported, thanks to integrated low noise amplifiers (LNA). There are two separate signal reception paths for the two frequency bands. The L1/E1/G1 path is characterized by a quasi-zero-IF conversion that mixes the middle frequency between L1/E1 and G1 to zero frequency. The L1/E1 reception bandwidth is up to 14 MHz so as to incorporate the MBOC modulations of Galileo E1 and future GPS L1C signals. A programmable automatic gain control (AGC) controls the complex analog baseband signals before they are digitized with a 4-bit dual-channel analog digital converter (ADC).
The second reception path receives an L5/E5a signal with up to 20 MHz bandwidth for the BPSK(10) modulated signals. This path uses a low-IF architecture. The signal is down-converted to an intermediate frequency (IF) of 15.345 MHz. The image frequency is suppressed by a polyphase filter. The real-valued analog signal is controlled by an AGC and converted to the digital domain using a single 4-bit ADC. A common phase locked loop (PLL) is used with specific L1/E1/G1 and L5/E5a dividers to generate the mixers’ local oscillator (LO) frequencies. The PLL loop filter is integrated on-chip to minimize external elements. Moreover, automatic filter and voltage-controlled oscillator (VCO) calibrations are included to mitigate process tolerances. The PLL can handle input clock frequencies between 10 and 80 MHz with a recommended clock frequency of 36.115 MHz.
An SPI core was implemented on the front-end part to facilitate control of the different front-end features. This means it is possible to tune the PLL, to switch off a complete front-end path if the second frequency band is not used and to activate different on-chip calibration procedures.
The frequency plan of the front-end is depicted in Figure 3. Due to the quasi zero-IF architecture, the complex L1/E1 baseband signal is located on an IF of -13.64 MHz and the GLONASS G1 frequency division multiple access (FDMA) signals on an IF of +12.94 MHz, with respect to the GLONASS G1 center frequency of 1602 MHz. The real-valued L5/E5a signals are provided by the second ADC and located on an IF of 15.345 MHz.
Figure 3. RF front-end frequency plan.
The ADC samples are generated with a frequency of 74.4871875 MHz for both the single channel L5, as well as for the dual-channel L1/E1/G1 ADCs. The ADC clock is also directly connected to the baseband digital core and is used as the main clock for the GNSS hardware modules. The embedded processor in the digital core receives a second clock, which is twice as fast as the GNSS hardware one.
Digital Baseband SoC. The baseband is characterized by a system-on-chip (SoC) architecture based on a SPARC-compatible 32-bit LEON2 microprocessor running at approximately 150 MHz. The GNSS functionality, including acquisition and tracking, are implemented using dedicated hardware modules.
The processor’s primary functions are to correctly configure the RF front-end and control the different parts of the receiver. In particular, it triggers acquisition, initializes, and starts the tracking channels with the signals detected during acquisition and takes care of closing the frequency/phase/delay locked loops (FLL/PLL/DLL) used for signal tracking. The tracking loops have strict real-time constraints; communication between the channels and the processor features a high-speed infrastructure.
Structurally, the processor is connected to a hierarchical on-chip Advanced Microcontroller Bus Architecture (AMBA) composed of a high-performance bus (AHB) and a peripheral bus (APB). The AHB provides a direct connection between the processor, the real-time GNSS modules, and the system memory, a monolithic 1 MByte block that hosts the main program at run-time. Different programs can be loaded if needed by using the external SD-card interface.
In addition to the processor, there are four additional AHB masters: the bootloader, the SD-card controller, the real-time GNSS modules, and the on-chip processor debugger. The bootloader is in charge of the bus control at system start-up. The SD-card controller has integrated direct-memory access (DMA) capabilities to move data between the SD card and the system memory. The real-time GNSS modules can write the tracking results directly to the system memory. Finally, the integrated processor debugger allows real-time debugging and is used mainly in the verification phase. The APB provides a connection to generic peripherals, and control and status interface of the GNSS modules without real-time constraints, as well as the control and status interface of the RF front-end. Since the GNSS modules operate in a separate clock domain that runs at half the frequency of the processor domain, some synchronization logic is necessary to ensure correct CDC.
The adoption of an SoC architecture provides higher flexibility than conventional static hardware solutions. In addition to typical GNSS applications, the user can also implement some signal monitoring and processing algorithms in software. The eCos-embedded operating system is provided to ease software development.
Generic Peripherals. The digital core is equipped with several peripherals that enable the communication with the outside world. The two separate universal asynchronous receiver/transmitter (UART) interfaces can run at 115.2 kbps. A dedicated serial peripheral interface (SPI) master is also provided with a maximum of 10-MHz clock frequency. For example, these interfaces can be used to provide NMEA data to some external display device or raw data (pseudoranges, code phases) in order to calculate a PVT solution. It is also possible to directly access the measurements generated from the correlator hardware and to control the tracking NCOs, which means users can choose their own algorithms for the loop closure. A possible application is the realization of vector-delay tracking using the NAPA ASIC and an external processor.
The SD-card interface facilitates the loading and storage of large amounts of data, for example, memory codes and almanacs. The possibility of making signal snapshots periodically and saving them to an SD card for later analysis has also been foreseen. This could be useful in special applications in which the receiver hardware is not accessible to the user all of the time.
In addition, 10 general-purpose I/O pins (GPIO) are provided. They can be controlled via software and can provide a very basic interface (for example, to connect to external LEDs or switches).
Acquisition Module. The acquisition module adopts a parallel code phase search in the Fourier domain by using a 16-k Samples Fast Fourier Transform (FFT) core. The adopted algorithm is known as parallel code-phase search.
The L1/E1/G1 signals coming from the front-end are first filtered and then sent to the acquisition module to allow a fast detection of the satellites in the L1/E1/G1 bands with their respective code delays and Doppler frequencies. The acquisition of GLONASS G1 FDMA signals is possible thanks to a software-configurable hardware mixer that can be set with the different G1 carrier frequencies. No direct hardware acquisition is supported for the L5/E5a band signals. The tracking of L5/E5a band signals is possible by performing a hand-over from L1/E1 band or a Tong search using the tracking channels.
The acquisition process is performed iteratively over all the possible satellites and over a set of Doppler values. These values are obtained by dividing the complete range of possible Doppler variations into bins. The smaller these bins are, the more accurate the acquisition result, but the more time is required to complete the entire process.
The acquisition has an additional layer of configurability because of the adoption of coherent and incoherent accumulations. These accumulations are supported in hardware but are completely software-controlled. This provides another possibility for achieving higher accuracy, but at the cost of a larger execution time due to an increase in the amount of accumulations.
To speed up acquisition, we introduced a dedicated logic based on a novel patented algorithm. With this algorithm, we are able to detect the Doppler of the L1/E1 satellites present in the signal with an accuracy of 2 Hz. By performing this Doppler search step before the actual acquisition, we are able to generate a list with Doppler values that can be used instead of the bins. This gives more accurate results thanks to the algorithm’s inherent accuracy (see Figure 4) and allows a reduction in the acquisition time since the amount of Doppler values are usually smaller than the bins. Another advantage of this algorithm is the possibility to detect the transition to an indoor context (such as where there is a lack of satellite signals) by simply looking at the Doppler list, without performing any acquisition.
Figure 4. Comparison between standard and Doppler-list based acquisition of an L1 signal.
A single iteration step for the acquisition of a GPS L1 signal requires no more than 1 ms for each accumulated epoch. To achieve a good compromise between accuracy and speed, we typically use four epochs of incoherent accumulation, which means approximately 4 ms execution time. For Galileo L1 with four incoherent accumulations, an iteration step takes approximately 16 ms. This time has to be multiplied by the number of satellites and bins to estimate the execution time of the complete process.
Integrated Acquisition Memories. The acquisition module is characterized by dedicated memory blocks used for the fast FFT processing. It also provides the possibility to use these on-chip memories to store a snapshot of the incoming signals. In particular, we can store up to 81,920 samples of raw data for the complex L1 and real L5 IF signals for further analysis or processing, even off-chip. This enables sophisticated spoofing detection methods, for example, as well as interferer detection and characterization methods. Spoofing detection can be implemented by monitoring the 2D-acquisition search space. Interferer detection and characterization can employ short-time Fourier transforms (STFT) on the snapshot.
Using the chip as a simple snapshot receiver without having to use the on-chip dedicated GNSS hardware is also a possibilty. For this purpose, the integrated peripherals like UART and SPI ports are provided as interfaces.
Tracking Module. The 40 versatile tracking channels can be mapped to any combination of GPS, Galileo, and GLONASS signals on the two reception bands. One possible combination would be to track 10 GPS and 10 Galileo satellites simultaneously on both L1/E1 and L5/E5a bands. Alternatively, the user can include GLONASS signals by using fewer GPS / Galileo combinations. The assignment of these tracking channels to the actual GNSS signals can be changed at run-time in order to adapt to different reception situations or to assist the selected signal processing methods.
Each channel is characterized by a five-tap correlator. For the BPSK modulated signals without side peaks, such as GPS L1/L5, Galileo E5a, and GLONASS G1, we use only three values (early, late, and prompt). For Galileo E1 BOC(1,1) signals, five values are foreseen (very early and very late in addition to the previous), so that false peak lock conditions can be detected and a bump-jumping algorithm can be applied. The switch between these modes can be done at run-time and determines the amount of correlation values to be exchanged between correlators and processor.
Low-Power Features. The GNSS modules operate in their own clock domain. This clock domain is divided in clock-gated regions. There is a common region for the bus interfaces, one region for the acquisition, and one for each tracking channel. This allows a fine-grain shut-down of the GNSS modules that are not currently in use. For example, the acquisition can be deactivated when there are enough signals in tracking or the unused tracking channels can be disabled. This allows a reduced power consumption for the idle modules. This activation/deactivation procedure is controlled through a set of registers connected to the APB and is performed via software.
External Front-End Interface. To allow for more flexibility, we provided an additional RF front-end interface. The interface is also depicted in Figure 3. This interface features one 2-bit complex and an additional 2-bit real input, as well as a clock input. The user can decide to directly connect the digital baseband core to an external RF front-end with compatible sampling rate parameters, and exclude the on-chip RF front-end. This makes it possible to use the NAPA chip for validating other RF front-end devices, or it can be adapted to special customer needs.
Boot-Up Sequence. The SoC includes a hard-coded bootloader that is in charge of the bus control at start-up. In this phase, the processor is switched off. The bootloader loads a 24-kByte program from the SD-card to the system memory and starts the processor. In this phase, the processor runs with the external oscillator clock. Having performed the RF front-end initialization, the processor can switch to the front-end PLL generated processor clock that runs at approximately 150 MHz. This switch is completely transparent to the processor. Then the actual main GNSS receiver program is loaded into the system memory and executed.
The NAPA Chip
The NAPA chip has been manufactured in a low-power 1.2 V 65 nm TSMC technology. The 4.5 mm x 5.0 mm chip die was mounted in a QFN68 package; first test samples are available. The core requires a 1.2 V power supply, the pads 1.8 V. Figure 5 shows a picture of the die and its interconnections. The two parts, the analog core and the digital baseband, are clearly distinguishable. The chip is currently in the verification phase.
Figure 5. NAPA chip.
Within the project, the development and testing of the NAPA design was carried out on basically two platforms. During the hardware development phase, the baseband core has been prototyped on a FPGA device and tested using a special file-player setup, as explained in the following section. Having taped out the chip and received the first samples from the foundry, a test board has been developed in order to verify NAPA chip functionality.
FPGA Test Setup. In the development phase, the NAPA baseband core has been implemented on a Xilinx Virtex6 FPGA device. A Xilinx ML605 development board has been used for the test setup. The main limitation of the testing in this phase was the lack of an analog RF front-end prototype. In order to make early testing of GNSS functionality possible, we adopted a file player developed by Fraunhofer IIS in a previous project. This file player uses a desktop PC to reproduce a digital signal data-stream stored in a binary file on the PC. The stream is sent through a dedicated interface to a commercial digital acquisition board. This board receives a clock synchronized with the baseband core’s clock in the FPGA and delivers the signals directly to the FPGA pins. The complete setup is depicted in Figure 6. The setup in use can be seen on the left part of the opening figure.
Figure 6. FPGA test setup.
Test Board. In the verification phase, which is currently ongoing, the first unpackaged test chip dies have been glued directly to the test PCB and bonded on board without any housing. After receiving the packaged chips, the QFN68 could be regularly soldered on the PCB. A block diagram of the board is depicted in Figure 7. The board hosts the typical switch buttons and LEDs for quick control and status detection as well as some specific interfaces. The clock can be provided through a dedicated SMA clock connector as well as a discrete oscillator. Two sub-miniature push-on (SMP) connectors are also provided for separate the L1 and L5 antenna inputs. The two UART ports, the debugger UART, and the SPI master port are connected using a FTDI chip. This chip allows the simultaneous connection of these ports to a desktop PC’s USB port. A parallel connector is provided to interface external front-end ADC signals and clock. The GPIOs are accessible through the same connector. A dedicated socket is added for a mini-SD card.
Figure 7. Block diagram of NAPA test board.
Preliminary Results
The chip on the test board was first tested using the same file player of the FPGA setup. This way, we could evaluate the correct functionality of the digital baseband core without the need to activate and configure the on-chip front-end. After the successful tests, we focused on the on-chip front-end configuration, and we used the antenna connectors to provide valid GNSS signals. We tested the chip using three different configurations: a GNSS signal simulator, a static roof antenna, and a small active patch antenna.
In the three configurations, we successfully acquired GPS L1 and Galileo E1 signals. We were also able to perform tracking on GPS L1 and L5I, as well as Galileo E1b and E5aI. Figure 8 shows the spectrum of a snapshot of L1 and L5 paths made using the on-chip dedicated snapshot hardware and sent through the UART port with a dedicated binary protocol for offline processing. For this special test, we used an arbitrary waveform generator to provide noiseless Galileo and GLONASS signals in the L1 and L5 frequency bands, supported by the NAPA chip. After performing a FFT of the two snapshots, we can clearly see these signals. In the L1 plot, the E1b signal is present in the negative frequency range with the two peaks typical of the BOC(1,1) modulation. The FDMA GLONASS G1 is in the positive frequency range with its trapezoidal characteristic. It is also possible to see a side lobe of the E1a BOCcos(15,2.5) in the proximity of the zero frequency. In the L5 plot, we can see the main peak of BPSK E5a signal on the right and its mirrored image on the left, due to the fact that L5 signal path is real.
Figure 8. Spectrum of L1 and L5 band showing a Galileo E1 and E5a signal.
Acknowledgment
This project has been funded by the Bundesministerium für Bildung und Forschung (BMBF) (German Federal Ministry of Education and Research), which is gratefully acknowledged.
Actions Necessary to Reduce Vulnerability and Ensure Availability
By Brad Parkinson
(From the 25th Anniversary GNSS History Special Supplement)
Introduction
Brad Parkinson
About 40 years ago, we had a vision for positioning, navigation, and timing (PNT). That vision was more than successful, and became known as GPS. In some respects we have been almost too successful: PNT is frequently taken for granted. PNT, in the form of GPS, has become a powerful worldwide enabler for productivity and for safety. Estimated yearly value runs to many tens of billions of dollars.
For several years, I have been concerned about comments that denigrate GPS because the signal strength is relatively weak. The speakers have gone on to say it can be completely replaced with inertial or other techniques. Recently, comments by government officials further energized me to look at the full picture.
What can we do to reduce the vulnerability and ensure that the expectations of the users are going to be met? I summarize my solution as the PTA program and will elaborate in this article. At a top level, the term PTA means: Protect, Toughen, and Augment GPS to assure PNT. Note I say PNT, not GPS. The central issue is assuring access of PNT to the user, not the source of the information. I strongly believe that PTA is both achievable and absolutely necessary. Protecting PNT is particularly important to Europeans as they are just about to launch their fledgling Galileo system.
Speeches and travel only reach a limited number. When GPS World invited me to write a piece for the magazine’s 25th anniversary issue, it seemed an ideal opportunity to expand knowledge of the PTA program. The following is an edited form of a talk I have given a number of times, most recently at the European Navigation Conference in Rotterdam in April 2014.
GNSS initiatives and the GNSS community are growing rapidly, and certainly we are very enthusiastic about the progress of Galileo. But some places in the U.S. community are saying, “Well, this GPS band is underutilized; devoting all that bandwidth to a single system is not prudent.”
I beg to differ with that view. If you look at the separate signals in the L1 band around the world, by the year 2023 they will grow to be well more than 400 individual signals. Those signals service over 2 billion users, from emergency service providers to precision agriculture to crustal monitoring and many, many more. I have an entirely separate talk on “GPS for Humanity,” but that is not our subject today.
Calling the GPS frequency band “underutilized” simply points out ignorance, even among our supporters. For example, we say PNT to emphasize that GNSS provides four dimensions. Certainly, timing is the forgotten fourth dimension of GPS, and even our politician friends rarely understand the importance of this aspect. Yet we know that highly accurate timing, supplied by GPS, is absolutely critical for power distribution, for telecommunications, and for the financial sector.
It is instructive to summarize the penetration of the PNT “Stealth Utility” into the fabric of our society.
Market Size. Overall, GPS has more than 2 billion users worldwide. This represents a very diverse user group; we providers are continually seeing new and innovative ways to use GPS.
Figure 1, for which I am indebted to Frank van Diggelen, gives an estimate of the number of receivers currently fielded. Notice the number of military receivers: less than half a million. The gray bar depicts the industrial uses such as survey and machine control, which come in at about 4.5 million; these tend to be extremely high enhancers of industrial productivity.
Figure 1. GNSS market size, 2012.
We have to change the chart scale to depict bigger market segments. For example, recreation, automotive, and computing are shown on the lower half of the chart. In fact, mobile phones will still not fit on the chart. Attesting to the size of the estimated mobile phone base: one company alone will produce more than 900 million GPS-equipped smartphones this year. The pie diagram shows the dominance of mobile devices, but much higher productivity gains come from high-precision devices whose impact is very disproportionate to numbers of receivers.
We asked some economists, just what is all this worth? They looked at a subset of all the industries and concluded that GPS has a positive net effect to the tune of at least $32 billion annually. They had an expanded study that suggested about $90 billion annually. So, for those who question the value of GPS, the answer is that the net yearly returns to our national investment are more than 1000 percent. (Note: National investment is about $3 billion annually.)
To ensure these enormous economic benefits of PNT, there are two fundamental needs, and we providers must assure that they are met. The first and most important need is availability.
Availability. When we say availability, it is defined in a certain way; it means that PNT is available at the application-specified accuracy. We usually measure that accuracy at the 90th percentile: only 10 percent of the time can that error be exceeded.
Integrity. The second user need is the required integrity. That means that when the user expects a specific accuracy, the system is not lying to him. Integrity assurance is very much a focus of both the International Civil Aviation Organization (ICAO) and, in the United States, the Federal Aviation Administration (FAA). In many cases they require that PNT errors not exceed specified bounds more than once in 10 billion measurements (1 x 10-7). This integrity level requires so many samples, it is virtually impossible to verify experimentally; we have not had that many airplane landings, but it can be calculated. The metric we use is how many minutes GPS is not available — unavailability — at the specified accuracy and integrity. That is more easily understood than availability that aproaches 99.9XXX percent. The usual goal is that unavailability be zero.
We have an independent assessment of how well we are doing: FAA’s Wide Area Augmentation System (WAAS). They put out a report card with a lot of numbers. GPS clearly deserves a grade of A+.
And it will get better. The U.S. government’s PNT Advisory Board, which I co-chair, recently advocated that the full navigation message be added at the new civil frequencies, the L2C and L5C signals. The Air Force has now complied, thanks to strong support from General Willie Shelton. This makes two more civil signals fully available. They currently expect 2.9 meter ranging accuracy, but by the end of the year the Air Force operators expect the same full accuracy as the rest of the signals, on the order of 0.5 meter of ranging error.
This is an outstanding picture.
So What’s the Problem? A statement made by a high-level U.S. government official in my presence exemplifies the problem: “GPS is much too vulnerable. We must replace it with new inertials and chip-scale atomic clocks.”
I found this statement appalling. Unfortunately, it was a meeting where you don’t normally speak up, and I didn’t. Nonetheless, to me, that was totally wrong.
GPS indeed has a very weak signal, and it depends on having clear line-of-sight to four satellites. But in my opinion, a much better statement is what I call the PTA solution. Our goal should be to:
Protect the system and the signal.
Toughen the receiver and the system.
Augment GPS as needed to ensure users’ PNT requirements are met.
The focus is ensuring positioning, navigation, and timing (PNT), not merely ensuring GPS.
Fundamental Prerequisites for PNT
The first prerequisite for GPS-based PNT is a receivable, clear, and truthful(truthful implies full integrity) ranging signal. There are five main challenges to this.
Too-powerful authorized signalsnearby. This aspect snuck up on our community. The FCC authorizers were about to license a powerful signal in the frequency band adjacent to GPS, drowning out any hope of receiving the GPS signal. This can be called the authorized jammer. All PNT providers must be very vigilant about this; we have seen ignorant elements of the government poised to do great harm with well-intended but destructive actions, without knowledge of the unintended consequences.
Natural Interference. This interference, the cause of delays and attenuation, is reasonably well understood, and the subject of much research, dating back to when we first defined GPS. Random events such as solar flares can potentially cause great harm.
Inadvertent Natural or Manmade Jamming. A nearby device that creates spurious, destructive emissions can be a serious problem for GPS receivers. This class tends to be manageable by well-designed receivers.
Collateral Interference. An example is a person who wants to evade tracking but is inadvertently jamming nearby GNSS receivers in addition to his own local receiver.
Deliberate Jamming or Spoofing. This is perhaps the major concern for developers and users. I will discuss this further later.
There is a second major prerequisite: satellite geometry. The user who cannot see enough of the sky is called “sky-impaired.” There are two possible underlying problems:
The satellite constellation has “brown-out” because of failures or inadequate numbers; or
The user is operating in a mountainous or urban area with high, local shading angles.
Overcoming sky-impairment requires a denser constellation, or use of multiple GNSS.
Protect, Toughen, Augment
What can we — as developers, operators, and manufacturers — do to overcome the PNT availability challenges for our users? My solution is PTA. The good news is that quite a few of the actions I recommend are underway — in fact, many of GPS World’s readers are active participants.
I am going to examine these three PTA principles, expand on them a bit, and hopefully explain a few things that help focus on a broad solution.
Protect the System and the Signal
This can be organized into seven actions: three PreActions and four ReActions. PreActions are before there is serious interference, and ReActions obviously come after interference is occurring.
First, the PreActions.
Protect the Spectrum. The chart in Figure 2 represents the frequency plan for the L1 band, and displays some of the sources of the 400 signals I referenced earlier. The blue star, GPS L1 C/ A, is the only fully operational and reliable signal in the world right now. The red star is the U.S. GPS military signal. You can see it has important power lobes close to the band edge. The black star is M-code, the new military signal of the United States.
Figure 2. Frequency plan for the L1 band.
The Galileo power curve, which is pale green, has very significant nodes close to the band edge. Of course, the Galileo PRS (the magenta star) is right on the band edge. The imperative for these wider bandwidths is that they produce sharper correlation edges and consequently produce greater measurement precision. This leads to greater accuracy, and greater usefulness and utility for many PNT users.
Reallocation of radio bands adjacent to GNSS poses a significant threat. The band edge of the proposed high-power communication signal (sometimes called broadband) appears as the black vertical line. It is obviously very close to the edges of many of the colored PNT signals. Tests conclusively demonstrated unacceptable levels of interference with L1 C/A.
Consider the proposed, high-powered terrestrial signal one quarter-mile from a GPS receiver. This produces a power ratio of 5 billion (broadband) to one (GPS). To visualize that power ratio, consider Niagara Falls, which produces about a billion watts. Compared to that, GPS power is a tablespoon of water dropped from five feet, once per second (about 0.2 watts). This is the power ratio that was almost authorized with 40,000 ground-based transmitters in the U.S. At a city block away, the effect is 10 times worse.
To quantify interference effects, some initial tests were run and measured broadband effects used for analysis. Cell-tower locations near Las Vegas, Nevada, approximated the broadband transmitter locations. The nearby airport, McCarran Field, has three RNAV (GPS) approaches. As expected, GPS users on the ground would be significantly jammed, but the effect on aircraft would be nine times worse than the impact on ground receivers. This is due to altitude (line of sight), geometry, and the sensitivity of aircraft receivers.
The 12 broadband transmitters around McCarran Field would jam all of the RNAV GPS approaches to all three runways. Signals of this type would effectively shut down or severely limit operations at the airport.
Signals in the GPS band will increase in the next decade as the newer GNSS become operational. The proposed, adjacent broadband is even more incompatible with these newer signals since they will be closer in frequency. Note that the whole approach was rejected, solely on the basis of L1/CA. It was not even tested against the other, more susceptible, modern signals. The worst would have been yet to come, had they been authorized to broadcast in the adjacent band.
Adjacent bands can continue to broadcast non-GNSS signals originating in space because the power levels will be comparable with the PNT spectrum. But we must be very vigilant to stop any high-power terrestrial signals from being allowed. They would become, effectively, authorized jammers. There should be no spectrum reallocation to ground transmitters until technology has been thoroughly demonstrated to solve any problems, (particularly for the high-precision users) and there is enough time to re-equip the users.
Europeans should have two other important frequency authorization concerns. First, there is a legal barrier within the United States to using Galileo signals. They have not been formally authorized. I think it is a bureaucratic glitch, but it is something we in the United States have to solve; we do want to use all GNSS signals. Stay tuned!
There is another concern. A group at the Electronic Communications Committee, European Commission, recommends allowing pseudolites in the L1 GNSS band. As an experienced user of pseudolites for aircraft landing and some other applications, I believe this is a very risky idea; pseudolites can be very useful, but frequencies should be found elsewhere to avoid unexpected interference.
Stiff Legal Penalties for Interference. The second PreAction is to enact stiff legal penalties for GPS jamming, both in terms of jail time and fines. The goal is to deter the ubiquitous $33 GPS jammer that one can buy on the Internet.
On the U.S. FCC website, the agency lists the penalties for having a GPS jammer. Forfeitures range up to $16,000, and they might even put you in jail. The Australians take a much stronger view: up to five years imprisonment or $850,000 in some cases. Some people are alarmed by these heavy penalties and call them brutal. However, they are not always imposed, and if jamming and spoofing is intentional, especially where the landing of airplanes is concerned and lives are at stake, I think a strong deterrent is warranted.
Stop Jammer Manufacturing, Sales. The third pre-action is to prevent proliferation by shutting down manufacturing and web sales of jammers. What is the status?
The FCC website states that manufacturers should comply with the law: stop marketing these devices in the United States and stop selling and shipping to addresses in the United States. The loophole is you apparently can manufacture these devices if you sell them outside the U.S. Now, I have a little difficulty with this. I have pointed this out to the DHS and others; hopefully, stronger action will be taken.
The FCC told me in an open meeting a few months ago that they were shutting down the websites where these devices are sold. But about three weeks ago, I went online and immediately found a website that sells nine different devices to jam GPS and cellphone devices. Indeed, there were jammers, all very affordable, for jamming just about everything. More recently, the FCC assessed a multi-million dollar penalty against such a jammer manufacturer. We will see if this actually happens. I hope they accelerate these efforts.
Now for the ReActions.
Detect Jamming. To stop jamming, the first step is to know when it is occurring. There are a variety of ways to do this. Some devices or concepts are already on the table: for example, a Chronos CTL3510 GPS Jammer Detector, an Exelis Signal Sentry Jammer Detector, and the J911 cell phone detection and reporting of jamming, an example from NavSys.
The idea behind the NavSys J911 is that all GPS-equipped smartphones have the capability to detect jamming. This does not pinpoint jammer location, but alerts authorities to the problem. Phone location can be reported to a central database for the next two actions.
Pinpoint Jammer Location. Techniques range from directional antennas to time-difference-of-arrival using Fast Fourier Transforms. The latter was demonstrated for the FAA at Stanford more than 10 years ago: location pinpointed within five meters. Cell towers could implement such techniques, since they have accurate time and could run correlations. There are already commercial GPS jamming locators: something called a JLOC (NaySys Jammer Locator). The British are using similar techniques for jammer detection on some of their freeways.
Eliminate Jammer. Having pinpointed the jammer, the next step is to physically eliminate it. What is the status? At Newark Airport there is an FAA, ground-based GPS augmentation system antenna right next to the turnpike. They are part of a blind landing system. In early 2010, there was an infamous jammer interfering with the FAA GPS receiver. It took three months to locate the offending truck driver and shut down the jammer. The good news is that, more recently, in the same general location, they located a similar moving jammer within 24 hours after the interference started. However, these are very special locations. Recent studies have suggested that interference sources are much more widespread. Note: Only certain enforcement personnel are authorized to seize the jammer and arrest its operator.
Prosecute. Having located the offender, the law should then be applied to prosecute. Leeway should be applied, commensurate with the circumstances. In this New Jersey case, the authorities say the perpetrator is liable for a forfeiture of $31,875.
Toughen Receivers
There are at least five well-known ways to toughen receivers, thereby increasing jam resistance:
Increased satellite signal spreading (such as L1C, L5) allowing greater processing gain;
Integration with inertial navigation components;
Digital beam-steering or null-steering antennas;
Increased satellite power such as L5 (a difficult and fairly expensive technique);
Local antenna shading, for example, the top of an airplane, which is shaded from the jammer.
These improvements cascade and are cumulative, but a remaining issue is to make such techniques more affordable.
To illustrate these anti-jamming techniques, consider the effective area of a 1-kW jammer located on the Capitol building in Washington, D.C. A basic high-quality GPS receiver, within a line-of-sight range of 20 miles, will stop providing PNT. Simply using the newest L1C spread-spectrum GPS signal reduces the jamming area by about two thirds, allowing operation to about 10 miles from the Capitol. Adding inertial aiding allows PNT to within three miles, and adding digital beam-forming antennas and using aircraft natural shading brings the effective radius to about 0.1 mile, about the size of the capital building.
The point is toughening the PNT receiver with the technologies mentioned is an extremely effective strategy. It would require over 60,000 jammers to cover the same area as the original non-toughened GNSS receiver.
Some techniques are very affordable today, while others, such as digital beam-forming antennas, remain too expensive for the ordinary user. In addition, there is a potential U.S. problem of export restrictions. Unfortunately, many of these existing restrictions have simply incentivized non-U.S. development of equivalent capabilities.
Augment
The last element of the PTA construct is to augment or substitute PNT sources. We are all aware of the coming revolution in multiple PNT sources from new GNSS. An all-GNSS receiver diversifies the frequencies and the signals, thereby reducing vulnerability to interference. It also improves availability for the sky-impaired user because of densification of satellites sources. Using satellites from multiple constellations can significantly improve availability, provided integrity requirements are met.
With these additional GNSS constellations, there are three major levels of cooperation:
Compatible: no mutal interference;
Interoperable: working to allow common time and geodesy system;
Interchangeable: using accurately calibrated biases and offset. Any four SVs will suffice.
The major issue again is probably integrity, because to ensure economic value, availability requires known integrity. As far as the U.S. FAA and ICAO are concerned, for precision aircraft operations the integrity value should be that the system be “out of spec” less than once in 1 billion times. To be productive they also would like zero minutes of unavailability. That may seem extreme, but commercial aviation and public safety demand it. Regarding integrity, some new GNSS are clearly making faster progress than others.
It is useful to further examine the densifying opportunity of additional GNSS. The chart in Figure 3shows how densification can impact the user. The number of satellites (SVs) available in the sky (assumed optimal distribution) is shown. The colors refer to whether 0, 1, or 2 SVs are out of commission for maintenance or repositioning (typical maximum is 1 for GPS). The measure of effectiveness is minutes of outage per day. Consider a shading angle of 60 degrees, representing a user near a rugged mountain slope area or a city. With the nominal 24 SV GPS constellation (the GPS specification is 24 despite the U.S. having 31 active SVs), the outages, due to geometry alone, are six to ten hours. Improvement with additional satellites is dramatic and quite non-linear. With 33 satellites (about a 37% increase in density) outages are zero minutes per day to 33 minutes if one satellite is out for maintenance (reduction by a factor of over 10!). Of course, SVs could be from different GNSS constellations if they are truly interchangeable and have the required integrity. The clear message is that about 33 SVs are needed to cover reasonably high elevation angles.
Figure 3. How densification of additional GNSS can affect the user.
Integrity Monitoring. Currently, the U.S. GPS control segment continuously monitors GPS satellites. If a fault is found, they set the satellite inoperative until the problem is resolved, which may take many minutes. This alarm time is not fast enough for precision aircraft landing and approach (the requirement is six seconds to alarm). For these rapid integrity alarms, the United States relies on the FAA’s WAAS, and Europe uses EGNOS to monitor the basic GPS L1 C/A signal. Soon, the EGNOS message will include Galileo integrity alerts. Unfortunately, the United States does not yet have a plan for reciprocal WAAS monitoring of Galileo signals. In fact, formal approval to even use these signals has not yet been granted by the U.S. FCC.
Self Integrity (RAIM). If an all-GNSS receiver has more than six satellites in view, the user can use the Receiver Autonomous Integrity Monitoring (RAIM) technique. This allows the user to cross-check each measurement against others to find erroneous satellites and guard against spoofing. Take the recent GLONASS situation. With a good RAIM PNT receiver, the user could quickly isolate the large errors from the combined set of GPS/GLONASS measurements. In fact, some deployed receivers did just that. If all GNSS are totally interchangeable, it will be enormously helpful to implement RAIM.
The recent, prolonged GLONASS outage saddened us all because it reduced the credibility of all GNSSs. We hope the Russians will be forthcoming in announcing what happened and the corrections that are being made; hopefully, it won’t happen again.
Fortunately, there is a third independent, real-time tracking network of 200+ sites, known as the Global Differential System (GDGPS). Although NASA administers GDGPS, local-country scientists maintain and operate individual sites in near real time. GPS is monitored down to centimeter precision.
A central issue for GDGPS is whether the integrity monitor capability itself has integrity. Because of redundancy and independence, a form of inverse RAIM, hereby named System Autonomous Integrity Monitoring (SAIM), can be used. Figure 4 depicts the number of independent looks or ranging measurements to a single satellite over various points on the Earth. You can see in the dark areas the value is 60, and even in the relatively unmonitored areas around South America, the redundancy is 20. At a typical spot, perhaps off Spain, it depicts 50-fold redundancy. By cross-checking the dozens of GDGPS measurements for each satellite, a strong integrity cross-check can be created. The GDGPS plan is to also monitor Galileo as it becomes operational. Thus, GDGPS has excellent prospects to provide real-time integrity assessments for all users and all operational constellations. We need plans to connect all users to these potential integrity alarms.
Figure 4. The number of independent looks or ranging measurements to a single satellite over various points on the Earth.
There are three classes of ground-based augmentations:
Pseudolites. Ground augmentations could also include pseudolites broadcasting GPS-like signals for additional ranging. While somewhat helpful, this technique cannot cover large areas and can act as a strong interference source if the signal is in any GNSS frequency band. For this reason, in my opinion, pseudolites should never be authorized in GNSS frequencies.
Distance-Measuring Equipment. Modernized DME, planned as a GPS supplement by the U.S. FAA, is very valuable for the airborne users. Most ground users derive no benefit from DME because they do not have line of sight to the widely scattered transmitters. Ohio University’s Frank van Gras is working for the FAA on a DME plan should GPS not be available. It involves moving from the so-called legacy DME to the enhanced DME to ensure continuous aviation operations.
eLoran. eLoran, covering expandable local regions, uses a powerful signal at an entirely different frequency. It is two-dimensional, but in calibrated areas differential (eDLoran) is perhaps as accurate as 10 meters for harbor areas and similar purposes.
I chaired a study of eLoran for the FAA in 2006. Initially skeptical, the study members finally concluded (unanimously) that eLoran:
meets the needs of all identified critical applications: 10–20 meter navigation accuracy for harbor entrance; 0.3 mile required navigation performance (RNP 0.3); stratum 1 frequency precision and 50-ns time accuracy.
is a modern system: new infrastructure, solid state transmitters, state-of-the-art time and frequency equipment, uninterruptible power supplies; new operating concepts, time of transmission, all-in-view signals, message channel with differential corrections, integrity; new digital user equipment, processes eLoran and GPS signals interchangeably, compact H-field antennas eliminate p-static.
is affordable: Less than $143M to fully complete eLoran, avoid costs of decommissioning existing Loran-C infrastructure; operations and maintenance currently $37M/year, reduced with eLoran-enabled automation.
And our group concluded it was the most prudent and cost-effective general augmentation or backup to GPS.
The National PNT Advisory Board also unanimously recommended that we deploy eLoran. The departments of Transportation and Homeland Security supported it; then, after a change of administrations, in a budget crunch, it was defunded, and the dismantling of existing Loran C stations began. Congress now may be taking action, and the recent GLONASS outages should give an impetus to that.
Who Will Implement PTA?
To my knowledge, many elements are currently being pursued, some by GPS World readers. But I can identify no entity that has the authority, the knowledge, the breadth, and the resources to create a single, well-focused program. This reminds me of a fable from Aesop regarding ants. When no leadership emerges, the ants have to band together to solve the problem. Yes, I am suggesting that we are the ants and we all must contribute to the solution, as well as seeking governmental agencies to step up to the responsibility.
In that regard I have a “to do” list. We must:
Protect PNT.
Vigorously defend the spectrum.
Work with lawmakers to increase legal penalties for PNT interference.
Work with manufacturers and law enforcement to improve timeliness and accuracy of interference identification (crowd-sourcing, every cell phone a detector).
Field jammer location equipment.
Toughen PNT.
Develop industry (ICAO/RTCA/RTCM) standards for deep inertial integration and directional antennas.
Develop vector receivers (all GNSS).
Continue to implement ARAIM and inertial for integrity (+WAAS/EGNOS).
Encourage users to move to rugged receivers.
Augment PNT.
Expand integrity notifications to include GDGPS.
Develop RTCA standards for seamless DME and GPS/GNSS.
Implement eLoran and develop RTCM standards for seamless use.
Develop an international process for integrity certification of all GNSS (GLONASS, Galileo, and BeiDou).
In conclusion, the rumors of the death of GPS, in my opinion, are greatly exaggerated. Let’s not throw out the baby with the bath water. Instead let’s accelerate and expand PTA to Protect our band, and Toughen our receivers, and Augment GPS to ensure that PNT is available for all users now and in the future.
In the words of American poet Robert Frost,
The woods are lovely, dark and deep,
But we have promises to keep,
And miles to go before we sleep,
And miles to go before we sleep.
Thank you.
BRAD PARKINSON has been the Edward C. Wells Endowed Chair (emeritus) at Stanford University, where he is a recalled professor of aeronautics and astronautics.
He co-founded the well-known Stanford GPS Laboratory and led the development of many innovative uses of GPS, including blind aircraft landing, precision farm tractors, and the prototype of the FAA’s WAAS. He also directed development and was a co-PI for the successful test of Einstein known as Gravity Probe-B sponsored by NASA. He worked in various executive or board capacities at Trimble Navigation, Intermetrics, Rockwell International, and The Aerospace Corporation.
As an Air Force colonel, from 1972 to 1978, he was the chief architect and first director of the NAVSTAR GPS development program, retiring from the service after orbiting the first GPS satellites and proving GPS capabilities. He is a fellow of five professional societies and recipient of dozens of awards, including:sharing the 2003 Draper Prize with Ivan A. Getting for leading the development of the Global Positioning System.
Professor Langley (fourth from left) discusses the UNB geodesy program with Canadian Science Minister Ed Holder (third from left.)
The Canadian Minister of State for science and technology, Ed Holder, visited the University of New Brunswick on July 28 to announce the awarding by the Natural Sciences and Engineering Council of $2.4 million to 28 UNB researchers.
He was joined by Keith Ashfield, member of Parliament for Fredericton, where UNB is based, and Craig Leonard, the New Brunswick Minister of Energy and Mines.
A highlight of the visit was a tour of the Department of Geodesy and Geomatics Engineering to see the work of Prof. Richard Langley and his students. Langley received $170,000 in Natural Sciences and Engineering Research Council (NSERC) funding in the competition. The funding will support the work of his group in improving augmented multi-constellation satellite-based precise positioning in a wide range of environments. Langley is GPS World’sInnovation editor, a post he has held since the magazine’s inception.
Canadian Science Minister Ed Holder looks at GPS World magazine, which has featured Innovation columns edited by Richard Langley for more than two decades.
Although GPS was the first widely available satellite navigation system, it has now been joined by the Russian GLONASS system, and will soon be accompanied by the European Galileo system, the Chinese BeiDou system, and the Japanese QZSS — all of which have test satellites now in orbit. There are interesting problems to be solved in gaining maximum benefits from this plethora of GNSS for precise positioning and navigation, and Langley and his team will address a number of them.
The team is also involved in the analysis of data from the GPS-based instrument on the Canadian CASSIOPE scientific satellite launched at the end of September 2013. The instrument, which precisely determines the position of the satellite and provides information on the state of the Earth’s ionosphere, was designed at UNB.
The NSERC Discovery Grants Program is an integral component of the government’s efforts to develop, attract and retain the world’s most talented researchers at Canadian universities. The program funds discovery research in a multitude of scientific and engineering disciplines, which builds a broad base of research capacity across the country.
Professor Langley gave the following presentation at the NSERC Discovery Grants Scholarships Rollout Announcement at UNB on July 28:
The 3DM-GX4-45 by LORD MicroStrain is a miniature, industrial-grade GPS-aided inertial navigation system that uses high-performance MEMS sensor technology. It combines a triaxial accelerometer, triaxial gyro, triaxial magnetometer, temperature sensors, pressure altimeter, and dual on-board processors running a sophisticated Extended Kalman Filter (EKF) to provide excellent position, velocity, and attitude estimates.
It offers a range of fully calibrated AHRS measurements, including acceleration, angular rate, magnetic field, deltaTheta and deltaVelocity vectors. GPS measurements include LLH position, ECEF position and velocity, NED velocity, UTC time, GPS time, and SVI. The receiver is a 50-channel u-blox 6, which receives GPS L1 C/A code, and the SBAS signals WAAS, EGNOS, and MSAS.
The 3DM-GX4-45 provides accurate navigation and orientation under dynamic conditions for applications such as GPS-aided navigation; unmanned vehicle navigation; camera stabilization; robotic control; and reconnaissance, surveillance, and target acquisition.
LTE brings a promise of improved location accuracy with new positioning technologies and their integration using hybrid techniques. Although established technologies such as A-GNSS (A-GPS and A-GLONASS) provides excellent performance in environments with a clear view of the sky, performance is often poor indoors, where detection of satellite signals is limited. In LTE, current standards support Observed Time Difference of Arrival (OTDOA), an advanced cellular positioning technology that can augment A-GNSS and provide a more accurate location fix for indoor scenarios.
With large-scale VoLTE rollouts imminent, leading operators are confronted with the need for extensive and complex testing of LTE positioning technologies to ensure VoLTE E911 works well from day one. Additionally, the FCC, whose current E911 regulations apply only to outdoor environments, has proposed stringent indoor requirements as a response to increased mobile usage for emergency calls and lack of accurate positioning information on calls that originate indoors.
“Roughly 70 percent of 911 calls are placed from wireless phones and a majority of these calls originate indoors, so there is a real urgency in providing better location accuracy for mobile users, wherever they are calling from,” said Nigel Wright, vice president at Spirent Communications. “Spirent is currently working with all the key industry players to evaluate OTDOA and its integration with other positioning technologies, and to enable operators to meet the location requirements for VoLTE E911 and the evolving FCC requirements.”
Spirent 8100 LTS has won widespread acceptance as the leading platform for location testing in the wireless industry, and with this latest capability is now able to support OTDOA Position Calculation Function (PCF). Minimum performance testing for OTDOA looks only at the raw measurements from the device, whereas use of OTDOA PCF enables full verification of a device’s position accuracy performance. Recognizing its importance, leading carriers have established their own OTDOA positioning performance requirements beyond bare minimum standards. Ensuring that devices fully meet these requirements as well as the evolving FCC regulations for E911 requires comprehensive testing.
Two satnav superpowers battled it out aboard a superyacht in the Mediterranean this summer, as a spoofing detector designed to differentiate between real and fake GPS signals came to grips with a spoofing device previously responsible for hijacking a sophisticated drone helicopter, deceiving it into landing when it was trying to hover, and for misdirecting the same luxury yacht in tests last summer.
Mark Psiaki, Cornell University professor of mechanical and aerospace engineering, and graduate student Brady O’Hanlon spent a week aboard the White Rose of Drachs, a luxury superyacht, testing their second-generation spoofing detector as the boat cruised from Monaco around the boot of Italy to Venice at the head of the Adriatic Sea. Also on board was a researcher from assistant professor Todd Humphreys’ Radionavigation Laboratory at the University of Texas at Austin. Humphreys tested his latest spoofer aboard the same yacht last year; this year, Psiaki and O’Hanlon embarked for a follow-up experiment to see if they could outsmart the spoofer.
The Cornell team’s spoofing detection system electronics quietly at work detecting evildoers on the bridge of the White Rose.
Both researchers have published earlier versions of their work in GPS World magazine, Psiaki in “GNSS Spoofing Detection,” the Innovation column in the June 2013 issue, and Humphreys in “Drone Hack” in the August 2012 issue.
The former story relates how Humphreys and Psiaki began their investigations as far back as 2008. “There was no intention to help bad actors deceive GNSS user equipment. Rather, our goal was to field a formidable ‘Red Team’ as part of a ‘Red Team/Blue Team’ (foe/friend) strategy for developing advanced ‘Blue Team’ spoofing defenses.”
In international waters this summer, the Cornell and Texas teams could conduct their research unhindered; on land, it’s very difficult to get permission to hack a GPS signal, even for research purposes, Psiaki said.
The Cornell two-antenna system installed on the roof of the White Rose bridge next to the superyacht’s GPS antenna.
Aboard the White Rose, Humphreys’ team initiated an attack of the boat’s GPS receiver, overlaying a disguised false signal on top of the real one, and attempting to send the boat off-course without generating any obvious warning signs. Stationed in a different area of the boat, Psiaki and O’Hanlon’s device set itself to detect the false signals through real-time analysis of their properties, and to provide protection against any attack by issuing a definitive warning whenever false signal characteristics were identified.
“We tested numerous spoofing scenarios,” recalled Psiaki. “We proved the efficacy of the new two-antenna version of one of our spoofing detection systems. It is the functional equivalent of our previous moving-antenna spoofing detection system. With two antennas we can simulate the effects of antenna motion without any need for moving parts. The only problems we encountered were with the initial spoofing drag-off, at which point the true and spoofed signals interfere with each other, and signal tracking can be tricky.
“We recorded wide-band data for all these cases. We think that we know how to enhance our defenses to hold on to the signals and recognizing spoofing during the initial drag-off. We also think that we know how to recover the true signals after an attack. The recorded wide-band data should enable us to develop and test these refinements in the lab, i.e., without the need to go back to sea — not that we would mind having to take another cruise on the White Rose of Drachs.”
In one test, the yacht’s GPS receiver was spoofed into believing that it was veering off its course, set northwards to Venice, and heading south to Libya at a very high speed. The Cornell detector was able to warn the White Rose’s bridge crew about the attack before the yacht was 20 meters off course.
The White Rose’s GPS-driven chart showing it off the coast of Libya (black line) when it was actually in the Adriatic, cruising from Montenegro to Venice (blue line). The spoofing detector knew all along that this was a false reading.“This photo shows the White Rose’ Litton GPS receiver with ridiculous speed and altitude readings — we were in a hurry to get from the Adriatic to Libya and therefore spoofed a straight line route that took us across, actually beneath, Italy and Sicily, at speeds exceeding 900 kts in order to get there in 50 minutes. “
“We want to progress to the point where not only can we tell it’s a false signal, but we can also say, ‘Here is the true signal; here is the true position,’” Psiaki added.
The owner of the White Rose of Drachs, an anonymous businessman, allows the boat to be used for scientific purposes during off seasons.
The Cornell and White Rose team: (from left) Brady O’Hanlon, Cornell ECE Ph.D. student, Andrew Schofield, master of the White Rose of Drachs, and Mark Psiaki, Cornell Prof. of Mechanical & Aerospace Engineering.
Psiaki will present a paper on the superyacht experiments at the Institute of Navigation’s GNSS+ conference in September in Tampa, Florida, and GPS World will publish an article based on this paper in the November issue.
This story draws on initial reporting by Anne Ju in the July 28 Cornell Chronicle, with additional material and photos supplied by Mark Psiaki.
Northrop Grumman and Boeing have responded to a U.S. Air Force call for contractors interested in building a follow-on set of GPS III satellites, according to a report in Space News.
Lockheed Martin is under contract to deliver the first eight GPS III satellites, but the award for up to 22 further IIIs remains open. Difficulties with the payload for the first batch of satellites mean that although the Lockheed has three space vehicles ready, it has no signal payload to put aboard them. Subcontractor Exelis is at work on that. Delivery delays have prompted the Air Force to look about for alternatives.
Lockheed Martin itself began investigating options for its supply line last year.
Air Force “Sources Sought” Call
The U.S. Air Force issued an official “Sources sought” notice in June on a production-ready GPS space vehicle, equipped with an alternate payload, for consideration alongside the Lockheed Martin-built GPS III vehicle. The first phase of the contract would include two firm-fixed price contracts worth $100–$200 million to demonstrate a competitor to GPS III.
Key requirements are that the satellite must offer a payload alternative to that built by Exelis; the satellite must be ready to launch by 2023; and the production line must turn out two to three new satellites per year.
The second phase features a competition between Lockheed Martin and one or more other companies for as many as 22 satellites. A final contract award would be made in 2017 or 2018.
Current GPS III contractor Lockheed Martin reportedly sent an engineering team to help Exelis expedite a resolution of payload holdups, while simultaneously investigating a switch to other suppliers, beginning with the ninth satellite in the GPS 3 series. Lockheed Martin says five companies responded to its solicitation last year.
Air Force Gives Free Hand. Gen. Ellen Pawlikowski, head of the Air Force Space and Missile Systems Center (SMC), told the national Space Symposium in Colorado in June, “Obviously we want a GPS III that does what its supposed to do, delivered on time, and it’s up to Lockheed to manage its subcontractors. My view is if Lockheed is not happy with their subcontractors nav payload, and they believe that they can get a lower risk approach to delivering a nav payload by seeking a secondary source for that, then that’s clearly a decision for them to make.
“They [Lockheed ] know we are disappointed at the delays that we have seen, the technical issues that their subcontractor has had, and probably they are considering whether an alternative source could provide them a better opportunity.“
Lockheed Martin spokesman Chip Eschenfelder issued a statement: “Exelis has made good progress on the first GPS III space vehicle, SV01 navigation payload. All GPS III SV01 navigation payload components have successfully completed unit acceptance and environmental testing, with the exception of one component, the mission data unit.
“To date, significant MDU hardware testing indicates signal cross talk issues are resolved. The SV01 navigation payload forecast delivery to Lockheed Martin is fall 2014.”
Boeing built the platform and major payload components for the GPS IIF satellites and is one of three companies that received contracts in January 2013 to study how to improve the accuracy, coverage, and efficiency of GPS using smaller satellites.
Northrop Grumman Aerospace of Redondo Beach, California, has already delivered deployable antenna sets to Lockheed Martin for the first six GPS III satellites. The division has delivered more than 1,000 antennas for previous generations of GPS spacecraft, Northrop Grumman said.
A Hansel and Gretel Approach to Cooperative Vehicle Positioning
By Scott Stephenson, Xiaolin Meng, Terry Moore, Anthony Baxendale, and Tim Edwards
MEET GEORGE JETSON.Those of us of a certain age will remember the animated TV sitcom The Jetsons, which featured George Jetson, “his boy Elroy, daughter Judy, and Jane, his wife.” It portrayed life in 2062, 100 years after the series debuted in 1962. George and his family used many futuristic gadgets including robot maids, talking alarm clocks, flat-screen TVs, and flying automated cars. Many of those devices are already available, well ahead of schedule. But flying cars are not quite with us yet. However, asphalt-hugging automated vehicles are already here, albeit still in limited numbers. Google created a buzz recently with tests of its self-driving car. Google’s cars were developed as an outcome of the Defense Advanced Research Projects Agency’s 2005 Grand Challenge in which teams created autonomous vehicles and raced them through a challenging road course.
Self-driving cars use a host of sensors to determine their position with respect to their surroundings and to navigate a chosen route legally and safely. Although wide-spread ownership of self-driving cars might still be a ways off, drivers of conventional vehicles will soon benefit from the research being conducted to provide them with positional awareness of other vehicles in their vicinity. This work may be characterized as part of the larger effort in developing intelligent transportation systems or ITS.
What is ITS? In the words of ITS Canada, it’s “the application of advanced and emerging technologies (computers, sensors, control, communications, and electronic devices) in transportation to save lives, time, money, energy and the environment.” This definition applies to all modes of transportation, including ground transportation such as private automobiles, commercial vehicles, and public transit, as well as rail, marine, and air modalities. The term ITS includes consideration not only of the vehicle, but also the infrastructure, and the driver or user, interacting together dynamically.
Just looking at ground transportation, there are many ITS developments underway, some of which are already implemented to some degree including systems for vehicle navigation, traffic-signal-control, automatic license-plate recognition, parking guidance, and road lighting to name but a few.
An important aspect of ITS is cooperative vehicle communication, which includes transmission of data vehicle–to–vehicle or vehicle–to–infrastructure (and vice versa — known by the abbreviation V2X. Data from vehicles can be acquired and transmitted to other vehicles or to a server for central fusion and processing. These data can include accurate real-time vehicle coordinates, which can be used to improve driver situational awareness and to monitor traffic flow for example. This use of V2X is known as cooperative vehicle positioning.
Several technologies are being developed for accurate cooperative vehicle positioning including lidar, radar, image-based cameras, ultra-wideband, and signals of opportunity. But GNSS also has a role to play. In this month’s column, team of British researchers turn to a children’s fairy tale for inspiration in their development of a cooperative vehicle positioning approach using carrier-phase observations — another innovative application of real-time kinematic or RTK GNSS technology.
“Innovation” is a regular feature that discusses advances in GPS technology and its applications as well as the fundamentals of GPS positioning. The column is coordinated by Richard Langley of the Department of Geodesy and Geomatics Engineering, University of New Brunswick. He welcomes comments and topic ideas.
There is little doubt in the benefit gained from cooperative modes of road transport, as agents working together generally perform better. In simple terms, this is the holistic idea that the whole is greater than the sum of its parts, commonly known as synergy. On top of this clear advantage, the complex systems theory of emergence suggests that novel strategies will develop from the as-yet-undefined patterns and structures. It is clear, however, that to facilitate this development certain technological advances need to be achieved. In this case, individual road agents need to accurately identify their location, and communicate easily and safely with other agents. This is a shift away from protective and passive systems toward preventative and active transport safety.
Cooperative driving, or vehicle-to-vehicle or vehicle-to-infrastructure driving (V2X), is proposed as the next major safety breakthrough in road transport. An example of the concept is shown in FIGURE 1. It involves agents in the road transport environment communicating on local and national levels in real time, to maximize the efficiency of movement, dramatically reduce the number of accidents and fatalities, and make transportation more environmentally friendly.
Figure 1. Vehicle-to-vehicle communications as envisioned by the United States Department of Transportation.
In the U.S., the National Highway Traffic and Safety Administration has commented that connected vehicle technology “can transform the nation’s surface transportation safety, mobility and environmental performance,” with industry experts predicting the widespread uptake of the technology within five to six years. This provides an opportunity for road vehicles to share GNSS information.
To an extent, this is possible with current technology. Communication is fairly pervasive and pretty robust, with the explosion in personal handheld mobile devices, using the GSM/GPRS, 3G, and 4G cellular communications networks. Positioning systems exist now that will provide a reasonably accurate and reliable location most of the time. However, the type of applications included in cooperative driving demand much higher performance from these positioning systems. For instance, as shown in the example in FIGURE 2, two vehicles approaching an intersection at relatively high speeds require accurate and reliable high output position information, and an ability to communicate with one another, in order to assess the likelihood of collision.
Figure 2. Vehicles approaching a road intersection would benefit from V2X communication.
These requirements are partly inter-linked, and can be mutually beneficial. For instance, communications methods can be used to share information to aid positioning, and some existing positioning systems can also be utilized to share information.
Many recent solutions in vehicle tracking research have shifted the GNSS receiver to a supplemental role in the positioning system, favoring an inertial device as the core of the integrated solution. The clear advantage is that an inertial device operates continuously, although other sensors are required to achieve the required navigation performance. The GNSS receiver is demoted because of its inherent limitations, namely the requirement of a clear view of the satellites and the availability of correctional information.
Most vehicle positioning research over the past two decades has focused attention on GNSS-centered systems, as evidenced by the abundant use of satnav devices used to assist in-car navigation. Despite its apparent monopoly over vehicle positioning in the commercial sector, the most
successful systems developed to guide autonomous vehicles either relegate GNSS to one of a suite of sensors, or almost disregard it altogether. This is often due to its apparent lack of positioning accuracy or availability. Popular terrestrial positioning sensors include lidar, radar, image-based cameras, ultra-wideband (UWB), and signals of opportunity. Clearly, the combination of different complementary sensors is important, but it would be a mistake to discount the more advanced GNSS positioning techniques that are available, especially with the expansion of the four global GNSS services.
Cooperative Positioning
The positioning of GNSS receivers relative to one another is a common application in transportation, such as during the aerial refueling of an airborne fighter jet by a tanker. In this case, it is important to know accurately the relative position of the two airplanes, but not necessarily their absolute position.
Relative positioning of road vehicles is more complex. By their nature, road vehicles are almost always close to other vehicles or road infrastructure, and there are many separate agents in each scenario. Vehicles can also travel large distances, and in terms of GNSS positioning, this may mean vastly different atmospheric conditions. Hence, relative positioning in road transport is useful if all GNSS receivers relate to the same datum, which in most cases is effectively absolute positioning.
Some previous work carried out by others concentrated on using GNSS code (pseudorange) and Doppler measurements for the relative positioning of vehicles, because it offers a simpler implementation method and is not susceptible to the cycle slips attributed to carrier-phase measurements. However, this means sacrificing the higher accuracy solution available from carrier-phase measurements. A major obstacle to GNSS positioning for V2X applications is the likely scenario of mixed receiver and antenna technology between vehicles. This has a major influence on the performance of relative positioning. By comparing various V2X relative positioning solutions, researchers found that an increase in positioning accuracy was typically accompanied by a decrease in availability and an increased demand for transmission bandwidth between the vehicles.
RTK GNSS Positioning. Real-time kinematic (RTK) GNSS positioning can be used to provide a solution at an accuracy of better than 5 centimeters (horizontal). This relies on the static reference receiver being located within 20 kilometers of the roving receiver, observing a good selection of common satellites with dual-frequency receivers.
When RTK positioning is used, the distance to the reference station has a bearing on the successfulness of the integer ambiguity resolution. A short baseline will benefit from a closer correlation of errors, due to the GNSS signals traveling through very similar parts of the atmosphere. Assuming each receiver is observing common satellites, this similarity will typically result in a higher success rate in the ratio test using the common Least Squares Ambiguity Decorrelation Adjustment, or LAMBDA, technique. This is particularly important following a GNSS outage.
GNSS positioning of road vehicles using RTK or network RTK (where a network of reference stations replaces a single RTK reference station) can provide highly accurate (< 5 centimeters), high integrity, real-time tracking information with little delay and at a high output rate. The proliferation of network RTK GNSS positioning systems has increased dramatically over the last decade. Network RTK GNSS positioning can minimize the spatial decorrelation of errors that is a characteristic of single-reference RTK positioning as distance increases between reference and rover receivers. This allows the wide mobility range demanded from automotive applications.
The transmission protocol of network RTK corrections is typically RTCM v3.0 or higher, and the composition of the correction information varies depending on the commercial service provider. The most common type of correction message format is that for a virtual reference station (VRS), although the most comprehensive and versatile method is the master-auxiliary concept (MAC). See references in Further Reading for details.
In V2X and other intelligent transportation systems (ITS) applications, the position must be accurate, reliable, available, and continuous. Previous research has shown that network RTK GNSS positioning can deliver a highly accurate and precise solution in an ideal observation environment. In one test, more than 99 percent of the observations lay within 2 centimeters of the truth solution, with a very small number of anomalous results of up to 20 centimeters.
The availability of a network RTK solution is determined by the availability of GNSS signals and the network RTK corrections. As network RTK positioning uses carrier-phase observations, GNSS outages and cycle slips significantly affect the performance of a receiver. However, the re-initialization of the fixed integer ambiguity resolution following a GNSS outage (such as caused by an overhead bridge) can be relatively fast. But from a cold start, the ambiguity resolution can take up to two minutes. This limits the widespread adoption of the technology for vehicle positioning.
NGI Road Vehicle and Electric Locomotive Testbeds. We have carried out research at the Nottingham Geospatial Institute (NGI) using state-of-the-art testing facilities. These bespoke in-house facilities allow repeated controlled experiments, and are a useful tool in the development of ITS and V2X technology.
To test the positioning performance thoroughly and under real-world conditions, we carried out experiments using the NGI’s road vehicle, which is equipped with a collection of on-board ground-truth systems.
Also, the roof of the Nottingham Geospatial Building (home of NGI) is the location of a remotely operated electric locomotive running on a 200-millimeter-gauge railway track. A photograph of the locomotive and plan of the track are shown in FIGURE 3. The locomotive can carry a selection of various positioning instruments, such as GNSS receivers, inertial navigation system (INS) devices, and tracking prisms, and can travel at a speed of over three meters per second. The position of the track is accurately known, and has previously been scanned at a resolution of 2 millimeters.
Figure 3. The NGB2 reference base station and electric locomotive track on the roof of the Nottingham Geospatial Building.
Three control solutions are used to assess the performance of the cooperative positioning techniques in real-world tests: An RTK GNSS control solution provided by a local static continuously operating reference station (CORS); a network RTK GNSS solution based on the MAC standard; and a dual-frequency GPS/INS system. Each vehicle also can be independently tracked using survey-grade total stations or a proprietary UWB positioning system.
Sharing Network RTK Corrections
If vehicles could communicate with one another on the road, this would help overcome the communication system limitation in network RTK positioning of road vehicles. For instance, if vehicle A has an external connection to a network RTK service provider (such as a mobile Internet connection) and a local connection to a second vehicle (B), then it could share its network RTK correction messages directly. Effectively, vehicle A would re-broadcast the correction information it has received from the corrections provider to the receiver on vehicle B. However, this would rely on the functional capability of the receiver of vehicle B, as network RTK real-time processing can be computationally intensive.
Not all network RTK correction messages can be shared in this way, and the range over which the correction messages are still valid needs to be determined. As vehicles communicating with V2X devices are likely to be relatively close (a few hundred meters at most), the feasibility of sharing network RTK information is good.
However, the network RTK VRS technique may offer more advantages. It is the most common form of network RTK used around the world, and requires significantly less bandwidth (approximately 10 kilobits per second at 10 Hz). The rover receiver is also less burdened by processing requirements. A VRS system operating on buses in Minnesota restricts the baseline to 2 miles, by updating the VRS location every 2 minutes.
Correction messages typically have a lifespan of 10 seconds. After this time, the receiver determines the messages to be too old and does not compute a fixed-integer position. It can, however, use the information to calculate a differential GNSS (DGNSS) position. Therefore, the relayed message must arrive at the receiver on vehicle B well within 10 seconds. Previous trials at NGI found that the typical message latency of the original correction message reaching vehicle A via a GSM/GPRS connection is 0.85 seconds. The additional V2X communication to transfer the message to vehicle B should not add a significant delay.
Capturing Network RTK Messages. To demonstrate the potential benefit of sharing network RTK messages between vehicles, network RTK messages were captured on board a vehicle and shared with a second vehicle. Vehicle A is the NGI van, and vehicle B is the NGI electric train. Most off-the-shelf network-RTK-enabled GNSS receivers are designed to communicate directly with the network RTK server using a connected communication device (GSM modem, UHF/VHF radio, cell phone, and so on), which typically provides a stable connection to minimize data loss.
To intercept the network RTK correction message, the GNSS receiver was set up to simply accept the correction message from a smartphone via Bluetooth. In this case, the connection to the network RTK service provider is established between the smartphone and the network RTK server. An application running on the smartphone (as shown in FIGURE 4) requests information from the network RTK server, logs the data, and passes the message directly to the Bluetooth-connected GNSS receiver on vehicle A. By intercepting the correction message, it can also be forwarded on to a second receiver, in this case on vehicle B.
Figure 4. Flowchart showing the capturing and sharing of network RTK correction messages (left), and the NTRIP client program running on an Android smartphone (right).
Sharing Messages with Second Receiver. FIGURE 5 shows the positioning solutions generated by a shared-network-RTK correction message. The original message was captured by the smartphone application operating on board vehicle A (the NGI van), and applied to GNSS observations made by a receiver on vehicle B (the NGI train). The baseline between the two vehicles was less than 100 meters, and the location of the VRS requested from the network RTK server was the NGI building (in geodetic coordinates to three decimal places). As Figure 5 clearly shows, the shared VRS corrections are equally valid for any receiver operating in the vicinity of the VRS. The thick red line is the fixed position of the train track, and the thin blue line represents the positions generated by the GNSS receiver using the shared network RTK corrections.
Figure 5. Sharing the network RTK message from vehicle A to vehicle B.
The VRS message type was chosen because it requires much less bandwidth, takes less processing capacity, and is prevalent among legacy receivers. Network RTK users typically require download speeds of 1.8 kilobits per second (VRS) and 5.6 kilobits per second (MAC). This is well within the typical speeds available from cellular wireless communications, which offer 80 kilobits per second downlink speeds from 2.5G systems to beyond 40 megabits per second for recent 4G systems.
The GNSS receiver on vehicle B is operating in an ideal location, with a clear view of the sky and a high number of visible satellites, which improves the probability of successful RTK ambiguity resolution.
Generating Pseudo-VRS Corrections
The potential benefit to GNSS positioning of using V2X communication between various road vehicles and infrastructure can be expanded by the implementation of pseudo-VRS positioning. This system resembles the children’s fairy tale Hansel and Gretel, where in order to help remember the route through a forest that guides them back to their home, Hansel drops markers along the path (in separate cases small white pebbles, and then breadcrumbs). By using the markers, the children can navigate their way through the forest, but without them they are left lost and disoriented.
The pseudo-VRS system uses a similar principle, where vehicle A marks its path by leaving behind small packets of information that can be used by other nearby vehicles. The small packets of information are VRS-like, and are broadcast using V2X communication devices and technology. Like the breadcrumbs in the fairy tale that are eaten by birds shortly after being dropped by Hansel, these VRS-like packets of information have a short lifespan.
VRS Requirements. It has been long established that a short baseline between reference and rover receivers leads to more accurate and successful relative GNSS positioning. A short baseline can effectively deal with satellite orbit and atmospheric errors, which become difficult to deal with as the baseline length grows, and is the reason why RTK GNSS positioning is typically limited to baselines shorter than 20 kilometers. A typical RTK baseline may be between 1 and 10 kilometers long, but it is still beneficial to reduce the baseline further, particularly if there is a large difference in elevation. This is enabled by the VRS network RTK technique. By using the observation data from several permanent reference stations that surround the rover location, a virtual reference station is created close to the location of the rover, including virtual observation measurements and position. This VRS information is transmitted to the rover, and the rover receiver treats the information like that of a real reference station. This technique can deliver better than 5-centimeter accuracy up to 35 kilometers.
The principle builds on the transfer of measurements made at the real reference stations to the VRS. The carrier-phase measurement at the real reference station ( ), shown in Equation 1, is made up of the geometric distance between the receiver and satellite (), the integer ambiguity ( ), and the receiver and satellite clock bias (). The key to the VRS technique is that the integer ambiguity and the receiver and satellite clock bias are not location dependent, so they can be transferred directly to the virtual reference station from the real reference station.
(1)
By differencing the carrier-phase equation of the real and virtual reference stations ( and , respectively), the ambiguity and clock errors are canceled. The result is shown in Equation 2.
(2)
By combining the carrier-phase measurement equations at the real and virtual reference stations, only two unknown terms remain. The first includes the position of the VRS ( ), which is, in principle, arbitrary and is typically the approximate location of the rover receiver. The second is the observable of the VRS ( ), which can now be obtained without actually measuring it. (In practice, the technique is a little more complex, as satellite orbit and atmospheric errors and biases need to be modeled for the VRS position). The VRS information can then be packaged using the RTCM standards and delivered to the rover receiver to enable network RTK VRS positioning.
Pseudo-VRS. Using the established VRS techniques and standards described above, we propose to use the GNSS observations and subsequent position information to simulate the existence of a VRS (see FIGURE 6). Imagine vehicle A carries a GNSS receiver together with the means to calculate its position accurately (for instance, it is also receiving differential corrections or has other positioning devices on board). So long as the receiver can successfully resolve the integer ambiguity, it can also produce each component required to describe a VRS. Clearly in this case, the receiver on vehicle A is a “real” reference station, but the existing VRS standards can be exploited to transfer this information to other local GNSS receivers. For instance, a receiver operating on vehicle B can use the information as a local real-time differential correction service.
Figure 6. The flow of data during the generation and sharing of pseudo-VRS data.
Because the VRS technique is well established (the most popular form of network RTK positioning), legacy receivers are able to take advantage of this pseudo-VRS information. RTCM standards are also well defined for the transfer of GNSS information in this form.
The pseudo-VRS information is valid for several seconds, so the delays introduced in transferring the information from one vehicle to a second can easily be accommodated. Like any communication device based on radio waves, V2X communication devices are likely to be subject to a level of delay and message loss that requires redundancy in the system. It is important that during one epoch the whole pseudo-VRS message is delivered, as there is little similarity between one epoch and the next. The original reference receiver is likely to be on a moving vehicle.
Effectively, the pseudo-VRS imitates the VRS in Equation 2 by providing the virtual reference station coordinates and carrier-phase observable. The information is also delivered to the second receiver in the same format RTCM message. A slight difference here is that only one-way communication is needed — the original coordinates of the VRS do not need to be supplied by the second receiver.
The pseudo-VRS processing is carried out using the RTKLIB open source software. RTKLIB has limited options to vary the position of the base station during RTK positioning, so the program is seeded with customized configuration files and run independently for each epoch. This creates an additional feature: The processing of each epoch has no effect on any other.
Vehicle-to-Vehicle Communication. As we just consider the exploitation of V2X devices in this article, the nature of the communication medium is not under test. For this reason, off-the-shelf wireless routers (2.4 GHz) were used to communicate between vehicles, using fixed local IP addresses. However, the performance of the routers under cooperative driving tests is limited by range, multipath, and signal obstruction.
Real-World Tests
To generate significant test results, some of the following tests use recorded and replayed data.
Test Setup. To test the performance of a pseudo-VRS positioning system, and the success of different configurations, real-world tests were carried out at the Nottingham Geospatial Institute. Two vehicles were used. Vehicle A was the NGI’s road vehicle, and vehicle B was the NGI’s electric locomotive. As the position of the locomotive test track is very accurately known, this can be used to measure the performance of the pseudo-VRS system.
Vehicle A was equipped with six GNSS receivers, a tactical-grade INS system, and a wheel odometer, and tracked using a total station and 360º prism. This provided multiple position solutions to ensure significant results.
Vehicle B was equipped with a GNSS receiver, and tracked using a proprietary UWB system for related V2X tests.
Also, on the roof of the NGB, and lying inside the track perimeter, is the NGB continuously operating reference
station. This hyper-local reference station allows local RTK solutions, and acts as a barometer of GNSS activity when tests are episodically carried out.
FIGURE 7 shows an aerial image of the test scenario. The Google background shows the NGB to the west, and surrounding roads to the south and west (still under construction during the image acquisition). The thin yellow line is a ground distance of 100 meters. The red dots signify the position of vehicle A (in the east), and the purple dots show the position of vehicle B (on the roof of the NGB building). The accuracy of the Google image is unknown, and is used here purely for illustrative purposes.
Figure 7. Aerial image of the test.
Test Results. These tests are designed to show the performance of a pseudo-VRS system using a V2X communication system. However, the results shown here were created using recorded raw data. The test results will help to design the correct RTCM message to share between vehicles in future tests.
To simulate the operation of a pseudo-VRS system, vehicle A must share its known absolute position and some raw RINEX information for each epoch with vehicle B. Vehicle B can then use this information, together with its own observed RINEX data, for the same epoch to calculate its known absolute position. In practice, there will be a slight delay in the delivery of the information from vehicle A (much like in a traditional RTK system), so that information from concurrent epochs are unlikely to be used.
The RTKLIB software cannot directly handle the variation of a base station’s coordinates (and output an absolute solution), so a small separate script was designed to utilize the processing capability of the software in a pseudo-VRS system.
FIGURE 8 shows the results of pseudo-VRS positioning. During dual-frequency tests, 99.67 percent of observations achieved fixed ambiguity (1197/1201). During single-frequency (broadcast ionosphere) RTK, 61.45 percent (738/1201) observations achieved fixed ambiguity. The ratio test threshold was 2.0. Around the area of 454930E 339708N, the number of common visible satellites dropped from eight to seven, and then again from seven to six three seconds later. This caused each of the three solutions to degrade slightly. The dual-frequency RTK solution briefly lost its fixed ambiguity solution (for two epochs, or 0.1 seconds), before regaining the fixed solution. The single-frequency RTK solution could not achieve a fixed ambiguity solution again until the number of common visible satellites returned to seven (five seconds after the initial satellite was lost). The DGNSS solution saw a similar degradation in its solution during this period.
Figure 8. Results from pseudo-VRS positioning.
The mean coordinate errors for the three solutions are 0.054, 0.707, and 0.323 meters (1 standard deviation, 3D), as shown in Table 1. This is compared to a solution calculated using the local CORS base station. The error in horizontal and vertical follows the typical ratio of 1:2. Test results were also completed using a lower pseudo-VRS update rate. At 1 Hz, the results prove even better. Although the latency of the correction is up to 1 second (positioning is calculated epoch by epoch), the results were better than updates at 20 Hz. The dual-frequency RTK solution achieved a fixed ambiguity at every epoch (100 percent), and when compared to the known track position appeared correctly fixed. The single-frequency RTK solution achieved a fixed ambiguity for 70.02 percent (897/1201) of the observations; a slight improvement over the 20-Hz results.
Table 1. Results from pseudo-VRS positioning.
Table 2 shows the performance of the pseudo-VRS system under different latency scenarios. This is important because a message transmitted by vehicle A may be delayed or newer messages may be disrupted. Once the latency of the correction message reaches 8 seconds, the performance of the positioning solution begins to drop. The number of fixed ambiguity solutions falls, and the resulting positioning accuracy also decreases. However, the solution can still deliver 20- to 30-centimeter accuracy with a message latency of up to 30 seconds.
Table 2. Effect of message latency on positioning quality.
Conclusions
This article has outlined the potential benefit of V2X technology to cooperative vehicle positioning. A vehicle that knows its absolute position accurately can assist a second vehicle to position itself using established GNSS techniques.
The pseudo-VRS base-station location must have reasonably accurate coordinates. Without this, the correct integer ambiguity cannot be resolved, and there is the risk of an incorrect resolution giving false success. This requires good reliability and integrity of the position of vehicle A, a characteristic that can be provided by network RTK positioning but likely needs further support from alternative positioning solutions.
Acknowledgments
The authors acknowledge Leica Geosystems for the provision of an academic license for the SmartNet network RTK service. We thank Yang Gao and Qiuzhao Zhang of the University of Nottingham for their assistance and detailed discussion during the experimental tests. The work was supported by the U.K.’s Engineering and Physical Sciences Research Council. This article is based on the paper “A Fairy Tale Approach to Cooperative Vehicle Positioning” presented at the 2014 International Technical Meeting of The Institute of Navigation held in San Diego, California, January 27–29, 2014.
Manufacturers
For our tests, vehicle A (NGI’s road vehicle) was equipped with six Leica Geosystems AGGS10 GNSS receivers with individual AS10 antennas, an Applanix Corp.POS RS with Honeywell International Inc.CIMU tactical grade INS system, and was tracked using a Leica Nova TS50 total station. Vehicle B (NGI’s electric locomotive) was equipped with a Leica GS10 GNSS receiver and AS10 antenna.
SCOTT STEPHENSON is a postgraduate student at the Nottingham Geospatial Institute (NGI) within the University of Nottingham, Nottingham, U.K.
XIAOLIN MENG is an associate professor, theme leader for positioning and navigation technologies, and an M.Sc. course director at NGI.
TERRY MOORE is the director of NGI at UoN, where he is the professor of satellite navigation and an associate dean within the Faculty of Engineering.
ANTHONY BAXENDALE is head of Advanced Technologies & Research at MIRA Ltd. (formerly the Motor Industry Research Association), an automotive consultancy company headquartered near Nuneaton in Warwickshire, U.K.
TIM EDWARDS is a principal engineer responsible for intelligent mobility research activities within the Future Transport Technologies Group at MIRA Ltd.
FURTHER READING
• Authors’ Conference Paper
“A Fairy Tale Approach to Cooperative Vehicle Positioning” by S. Stephenson, X. Meng, T. Moore, A. Baxendale, and T. Edwards in Proceedings of ION ITM 2014, the 2014 International Technical Meeting of The Institute of Navigation, San Diego, California, January 27–29, 2014, pp. 431–440.
• Intelligent Transportation Systems
Proceedings of IEEE ITSC 2013, the 16th International IEEE Conference on Intelligent Transportation Systems, “Intelligent Transportation Systems for All Modes,” The Hague, The Netherlands, October 6–9, 2013.
“Driverless Cars” by R. Murray in Computing and Control Engineering, Vol. 18, No. 3, June-July 2007, pp. 14–17.
• GNSS and Inertial Navigation Systems
“GPS and Inertial Systems for High Precision Positioning on Motorways” by J.E. Naranjo, F. Jiménez, F. Aparicio, and J. Zato in Journal of Navigation, Vol. 62, No. 2, April 2009, pp. 351–363, doi: 10.1017/S0373463308005249.
• Vehicle-to-Vehicle and Vehicle-to-Infrastructure Technologies
“Implementation of V2X with the Integration of Network RTK: Challenges and Solutions” inProceedings of ION GNSS 2012, the 25th International Technical Meeting of The Satellite Division of the Institute of Navigation, Nashville, Tennessee, September 17–21, 2012, pp. 1556–1567.
“Relative Positioning for Vehicle-to-Vehicle Communication-enabled Vehicle Safety Applications” by C. Basnayake, G. Lachapelle, and J. Bancroft in Proceedings of the 18th ITS World Congress, Orlando, October 16–20, 2011.
“Can GNSS Drive V2X” by P. Alves, T. Williams, C. Basnayake, and G. Lachapelle in GPS World, Vol. 21, No. 10, October 2010, pp. 35–43.
• Network RTK
“Network RTK for Intelligent Vehicles” by S. Stephenson, X. Meng, T. Moore, A. Baxendale, and T. Edwards in GPS World, Vol. 24, No. 2, February 2013, pp. 61–67.
“A Comparison of the VRS and MAC Principles for Network RTK” by V. Janssen in Proceedings of IGNSS2009, the 2009 Symposium of the International Global Navigation Satellite Systems Society, Gold Coast, Queensland, Australia, December 1–3, 2009.
“Introduction to Network RTK” by L. Wanninger, IAG Working Group 4.1: Network RTK (2003–2007). Online article. Last modified June 16, 2008.
RTCM Standard 10403.1 for Differential GNSS (Global Navigation Satellite Systems) Services – Version 3, developed by RTCM Special Committee No. 104, Radio Technical Commission for Maritime Services, Arlington, Virginia, October 27, 2006.
“Accuracy Performance of Virtual Reference Station (VRS) Networks” by G. Retscher in Journal of Global Positioning Systems, Vol. 1, No. 1, 2002, pp. 40–47.
“An Overview of Multi-Reference Station Methods for cm-Level Positioning” by G. Fotopoulos and M.E. Cannon in GPS Solutions, Vol. 4, No. 3, January 2001, pp. 1–10, doi: 10.1007/PL00012849.
