Assessment of GPS Signal Quality in Urban Environments Using Deeply Integrated

GPS/IMU

Andrey Soloviev, Dean Bruckner, Frank van Graas, Ohio University

Lukas Marti, Robert Bosch Corporation, RTC

BIOGRAPHY

Andrey Soloviev is a senior research engineer at the Ohio University Avionics Engineering Center. He holds B.S. and M.S. in Applied Mathematics and Physics from the Moscow University of Physics and Technology and a Ph.D. in Electrical Engineering from Ohio University. His research interests include various aspects of inertial navigation, GPS/INS integration, GPS software receiver development, GPS carrier phase positioning, digital signal processing, Laser Radar (LADAR) localization technologies, and joint time-frequency data analysis. He received the ION Early Achievement Award in 2006. Dean Bruckner is a Graduate Research Associate studying avionics and GPS systems at the Avionics Engineering Center at Ohio University. He received a B.S. in Electrical Engineering from the U.S. Coast Guard Academy and an M.S.E.E. from the Naval Postgraduate School. Upon retiring from active duty in the U.S. Coast Guard in 2005, he enrolled in Ohio University’s Ph.D. program. Frank van Graas is a Fritz J. and Dolores H. Russ Professor of Electrical Engineering and Principal Investigator with the Avionics Engineering Center at Ohio University, Athens, OH. Dr. Van Graas is a Past President of The Institute of Navigation (98-99). In 1996, he received the Johannes Kepler Award from the Satellite Division of the ION. Dr. Van Graas’ research interests center on all facets of GPS, including aircraft precision approach and landing, attitude determination, and system integration. Lukas Marti is a senior research engineer with the Robert Bosch Research and Technology Center in Palo Alto, California. He holds a B.S. in Electrical Engineering from the University of Applied Sciences, Burgdorf, Switzerland and an M.S. and Ph.D. in Electrical Engineering from Ohio University, Athens, Ohio. His research has included the Local Area Augmentation System (LAAS), with emphasis on the analysis of GPS

signal anomalies, and the use of integrated GPS systems for vehicle navigation in urban environments.

ABSTRACT

This research evaluates the quality of GPS signals and their usability for localization in urban environments using GPS data collected in urban canyons. GPS signals collected on a Software Defined Radio (SDR) platform in urban canyons in downtown Athens, Ohio and Columbus, Ohio are processed using a deeply integrated GPS/INS scheme. This deep integration architecture allows for coherent signal integration over time intervals as long as 1 second. The deep integration mode that provides continuous carrier phase tracking is used herein. Performance of the deep integration scheme in urban canyons is compared to performance characteristics of commercially available low-sensitivity GPS receivers. Results characterize the received signals in urban environments in terms of signal strength, tracking continuity and multipath influence on signal tracking performance. Results attained show that signals from 5 to 6 Space Vehicles (SVs) are available for processing, even in dense urban canyons. Deep GPS/INS integration enables continuous carrier phase tracking, thus allowing for accuracies on the cm/s level in velocity in urban environments. In contrast, velocity performance of the commercial low-sensitivity GPS receivers considered yielded errors on the level of 1 m/s. Additionally, the results demonstrate that continuous carrier phase tracking is possible, even for those cases where buildings block the satellite Line Of Sight (LOS). Further, consistent carrier phase tracking is performed for at least 2 SVs for a test scenario where all LOS vectors are blocked by buildings, and up to 6 SVs for all other urban canyon scenarios. Tracking remains consistent for weak signals with Carrier-to-Noise Ratios (CNRs) down to 12 dB-Hz.

815

INTRODUCTION

Outdoor localization services based on Global Navigation Satellite Systems (GNSSs) have tremendously matured over the past decade and are widely available in a variety of applications. Current GNSS user performance, however, is fragmented by environmental boundaries. GNSS performance is generally sufficient for most localization applications in rural and suburban environments. In contrast, urban environments with a high density of tall buildings, generally referred to as urban canyons, still pose a very challenging environment for most GNSS receivers. When not integrated with other sensors, they display sub-optimal performance. In previous efforts at Ohio University, a real time, Software-Defined Radio (SDR) based, batch processing GPS receiver architecture was developed and tested. [1-3] This architecture was extended to include a low-cost Inertial Measurement Unit (IMU) to perform GPS/IMU deep integration, in order to enable the system to process signals with Carrier-to-Noise Ratios (CNRs) down to 15 dB-Hz. [4, 5] The deep integration mode developed provides consistent carrier phase tracking without requiring the knowledge of navigation data bits. Consistent carrier phase tracking at a 15 dB-Hz level was previously demonstrated in real flight environments. [6] This paper extends these efforts in two respects. First, the test scenarios are expanded to include land vehicle navigation in urban canyons. A two-stage test plan is implemented, with a preliminary test in downtown Athens, Ohio and a full test in downtown Columbus, Ohio. The test data collected is used to evaluate the availability and quality of urban GPS signals. The signal quality evaluation portion focuses on carrier phase

measurements needed to provide continually usable position, velocity and timing information. A higher-performance, tactical grade IMU is then incorporated to enable the use of integration intervals up to one second in duration, thereby improving post-correlation signal-to-noise ratio and, in turn, signal availability. In this way, the GPS signals available in urban canyons can be more fully described, and the expected performance of a range of lower cost IMUs can also be assessed by adding simulated errors to the tactical grade IMU signals. The critically important empirical relationship of IMU quality to GPS signal availability can then be described.

DATA ACQUISITION ARCHITECTURE AND PROCESSING

The data acquisition system architecture is illustrated in Figures 1 and 2. The system was installed in a Ford Econoline 350 cargo van with roof racks, as shown in Figure 1. Although only one GPS channel was needed for deep integration processing, two channels were available and were used to provide redundancy. The simplified system block diagram in Figure 2 shows the general data acquisition structure. Two GPS patch antennas, one in the L1 frequency band with an internal amplifier and the other an L1/L2 antenna with an external low-noise JCA amplifier, provided the GPS signals in the preliminary test. In the final test, the L1/L2 patch antenna and JCA amplifier were replaced with a NovAtel pinwheel L1/L2 active antenna. The two NovAtel OEM-4 receivers (model PWRPAK-4E-L1L2W) used in the preliminary test were replaced in the final test by a single SiRF StarIII receiver. With these exceptions, the system is the same throughout the tests.

NovAtel OEM-4 GPS receivers (later, one SiRF StarIII) Controller for laser sensor Software radio RF components Software radio digital components Inertial Measurement Unit and circuitry

GPS antennas Laser sensor

For batch processing

For sequential processing

For potential augmentation to GPS

NovAtel L1/L2 pinwheel antenna substituted here

Figure 1: Photographs of Van-Mounted Data Acquisition System

816

Laptop PC

SW radio RF front end

OEM-4

IMU

IMU, SW radio digital circuits

OEM-4

StarIII

GPS

FPGA PC-104

PC

Laptop PC Laser

DQI system

Figure 2: Simplified Block Diagram The SDR employs a downconvert-and-digitize front-end. [7] The GPS/IMU deep integration is performed in postprocessing. Details of the downconversion and postprocessing with the American GNC brand IMU are described variously in [2-6] and are not repeated here. The IMU system also shown in Figure 2 consists of a tactical grade Systron Donner Digital Quartz IMU (DQI) sensor, a PC-104 computer controller, Field Programmable Gate Array (FGPA) circuitry and a GPS receiver to time stamp the IMU data. [8] Finally, a Sick brand Laser Measurement Sensor (LMS) 200 provided continuously-panned distance data in a 180-degree arc in the horizontal plane with 0.5 degree resolution. [9] Distance was recorded with cm-level resolution with a maximum range of 80 m. This laser sensor is rated Class 1. Laser data was manually synchronized to IMU data approximately to the nearest quarter second, and was recorded on a laptop computer.

PRELIMINARY RESULTS

A preliminary test was performed on 14 April 2006 in downtown Athens, Ohio. The goals of the preliminary test were threefold:

• to compare performance of a sequential tracking architecture provided by an off-the-shelf conventional GPS receiver, in this case the NovAtel OEM-4 [10], and batch-processing with inertially aided long signal integration to enhance processing capability for signals with

Carrier-to-Noise Ratios (CNR) below the conventional threshold of 32 dB-Hz;

• to perform an end-to-end test of the data acquisition system; and

• to initially characterize the GPS signal in urban environments of benign to moderate difficulty.

In this paper, benign urban environments are considered those where signals from a minimum of three Space Vehicles (SVs) exist with CNRs consistently above 32 dB-Hz on all streets. Urban environments of moderate difficulty are those where signals from a minimum of three SVs exist with CNRs consistently above 32 dB-Hz on major streets, but with fewer SVs in small streets and alleys. Difficult urban environments are those where signals from a minimum of three SVs do not exist with CNRs consistently above 32 dB-Hz even on major streets. Note that this classification requires the signals from the same SVs to be consistently above the tracking threshold, not merely the instantaneous total of SVs with ID numbers that differ from one measurement to the next. Accordingly, Athens, Ohio contains areas considered benign to moderate according to this classification method. In using such classification methods, one should consider that a city typically contains areas that fall into more than one category. For this reason, categorizing cities by the extent and relative proportion of each category is more helpful than using only one category. The alley shown in Figure 3 below, located between two multi-story brick buildings, contained the most restricted view of the sky in the entire preliminary test. The other portions of the test were a less-restricted alley between two-story brick buildings and various city streets that, along with the two alleys, comprised a rectangular loop with a total distance of approximately one half mile.

Figure 3: Alley in Preliminary Test (Athens, OH)

817

The general post-processing methodology used in the preliminary test was to note where the sequential receiver was unable to track, and then “zoom in” on those spots in the recorded data using the deep integration processing. Figure 4 shows the number of visible satellites reported by the NovAtel OEM-4 receiver for the test period of approximately 40 minutes. The test period began with a 15-minute stationary period in a university parking lot, transitioned to a five-minute drive downtown, and concluded with two loops around the test course. The time intervals enclosed in the second and fourth vertical bars in Figure 4 below represent the time intervals spent in the more restricted alley. In these intervals, one

or two SVs are generally visible. The first and third vertical bars represent the time intervals in the less restricted alley. In these intervals, zero to four SVs are generally visible. The categories superimposed on the graph by the horizontal bars illustrate the level of navigation solution generally possible with 1, 2, 3 or 4 or more visible SVs. These are a complete 3D navigation solution, a 2D navigation solution, a 2D navigation solution using a calibrated receiver clock, and no solution. Figure 5 exemplifies the performance of the deep integration GPS/IMU system by means of three dimensional (3D) GPS signal images recovered for corresponding data batches.

Complete 3D nav solution

2D solution

2D solution with calibrated receiver clock

No solution

More restricted alley

Less restricted alley

Figure 4: Performance of Sequential Tracking Receiver

Relative signal energy

Code shift,chips

Doppler adjustment,

Hz

Relative signal energy

Doppler adjustment,

Hz Code shift,

chips Figure 5: Example Performance of Batch Processing Receiver: 3D images of a 0.2 s GPS signal batch for PRN 10 in the more restrictive alley (left) and PRN 5 in the less restrictive alley (right)

818

In the alleys, an integration period of 20 ms was required to track two SVs, and an integration period of 200 ms was required to track two to three additional SVs, as illustrated in Table 1. Thus, a representative sequential tracking architecture is shown to have limited capabilities for urban navigation, even in benign to moderate urban environments represented by downtown Athens, Ohio. Batch processing, open-loop tracking techniques with deep GPS/IMU integration significantly improve signal availability in post-processing in benign to moderate urban environments. Specifically, they bring the number of potentially usable signals from zero to four or five, although these are almost entirely indirect path signals. This technique shows promise for use in challenging urban environments, which are considered next. Table 1: Performance of Deep Integration Receiver in Preliminary Test—SVs with Consistent Carrier Phase Tracking

More restricted alley PRN CNR (dB-Hz) Integration interval

required (ms) 2 32 20 4 32 20 5 -- -- 6 19 200

10 24 200 30 23 200

Less restricted alley PRN

CNR (dB-Hz) Integration interval required (ms)

2 43 1 4 32 20 5 22 200 6 -- --

10 32 20 30 28 200

ADVANCED CONVENTIONAL RECEIVER TEST RESULTS

The full test of the system was performed on 19 May 2006 in downtown Columbus, Ohio. Columbus is one of the 20 most populous cities in the United States, and contains several areas of urban canyons that represent moderate to difficult challenges for GPS navigation. The photographs in Figure 6 illustrate buildings and structures found in the downtown area. The lower photo illustrates two consecutive pedestrian overpasses on S. Wall Street, each with a width of approximately 20 m along the direction of the test vehicle’s track.

Figure 6: Downtown Columbus, OH (upper photo Broad & 4th; lower photo S. Wall)

819

In these urban canyons, the first portion of the test was an examination of the performance of an advanced Commercial-Of-The-Shelf (COTS) GPS receiver, the SiRF StarIII. The term advanced used here indicates that the receiver’s design has been optimized for performance in urban environments, and is designed to be integrated with companion services such as additional communications links to provide non-ranging GPS satellite data and dead-reckoning sensors such as a gyroscope and wheel sensors. [11] The SiRF StarIII evaluation kit receiver was connected via a signal splitter to the same NovAtel L1/L2 pinwheel antenna used by the deep integration receiver. Like the NovAtel OEM-4 receivers described in the previous section, the StarIII was used in standalone mode only, and was not integrated with any other sensors. The results of the StarIII’s performance in the Columbus tests are shown below in Figures 7 through 9. Figure 7 shows the number of visible satellites reported by the SiRF receiver for the Columbus test. Accordingly, at least 5 visible SVs are reported. Enough satellites are thus available to compute an unrestricted navigation solution for the entire test trajectory.

The SiRF receiver version used in the test provides the complete navigation solution (position and velocity) while raw code and carrier measurement outputs are not supplied. Measurement quality is thus evaluated indirectly from position and velocity trajectories. Figure 8 represents position solution output of the SiRF Star III receiver. The data presented show position discontinuities on the order of 100 m that indirectly indicate the presence of discontinuities in pseudorange measurements at the same level. To evaluate velocity accuracy, SiRF velocity data were analyzed at two stationary parts of the test trajectory as illustrated by Figure 9. Particularly, SiRF receiver velocity outputs were compared against the reference velocity, which is zero. Velocity data presented in Figure 9 show meter-level velocity errors. It can be therefore inferred that the velocity solution does not completely utilize carrier tracking outputs that allow for velocity estimation accurate to the cm/s level if carrier phase measurements are used. [12]

Figure 7: Number of Visible SVs Reported by SiRF Star III for the Columbus Test

820

Examples of trajectory “jumps”

Figure 8: SiRF Star III Position Solution for the Columbus Test

Stop 1

Stop 2

Figure 9: Examples of SiRF Star III Velocity Solution for Stationary Parts of the Columbus Test

821

DEEP INTEGRATION STATIONARY TEST RESULTS

The stationary tests of the deep integration receiver presented in this section were performed as follows:

• Five urban canyon locations in downtown Columbus were selected, ranging from moderate to difficult, as defined earlier. The easiest moderate scenario was an average downtown Columbus street; the most difficult was underneath a wide pedestrian overpass, with no sky visible at all except for an approximate 10 degree square patch of sky on the horizon to the rear of the vehicle.

• The vehicle was stopped for several minutes

during data collection.

• A 1 s integration interval was applied in postprocessing to maximize signal detection.

• The quality of GPS signal tracking was assessed

in postprocessing in three respects:

o To evaluate availability, the total number of visible SVs that include both direct and indirect path SV signals detected and measured was recorded.

o To evaluate tracking consistency, the

carrier phase measurements of each SV were examined for discontinuities.

o To evaluate tracking accuracy,

integrated velocity was derived from the carrier phase measurements as described in [12] and compared to the true vehicle velocity, which was zero.

Figure 10 shows the urban surroundings for Scenario 1, which is approximately at the midpoint of relative difficulty for the five stationary scenarios. The test vehicle stopped for a couple of minutes to obtain the 50 second test interval that appears in subsequent figures. In addition to the forward, or Southeast (SE), view shown in Figure 10, the view to the driver’s left (NE) was blocked by a tall building. The view to the passenger’s right (SW) was relatively open, while the view to the rear (NW) was largely obstructed by tall buildings. Figures 11 and 12 illustrate the azimuth and elevation of each SV measured, relative to both the north-aligned locally level navigation frame and, when compared to Figure 10, the Columbus street grid.

Figure 10: Urban Surroundings, Scenario 1

Ver

tical

SV 4 SV 2 SV 10

SV 5

SV 30

West South

Figure 11: SV Line of Sight Unit Vectors, Scenario 1

Azimuth, deg

Ele

vatio

n, d

eg

Figure 12: SV Elevation and Azimuth, Scenario 1

822

Figure 13 shows 3D GPS signal images for SVs 4 and 5. The deep integration receiver computes a 3D image for each signal batch as described in [3] and [4]. The signal is identified when a peak is found, as shown in this figure.

Relative signal energy

Code shift, chips

Doppler shift, Hz

Relative signal energy

Doppler shift, Hz Code shift,

chips Figure 13: Relative Energy Peaks for SVs 4 and 5, Scenario 1 (CNRs are 24 dB-Hz and 16 dB-Hz) The deep integration receiver also measures accumulated Doppler and CNR, as illustrated in Figures 14 and 15, respectively. The term δ(Accumulated Doppler) is used herein to denote accumulated Doppler measurements that are compensated for the satellite motion component, as well as receiver clock, ionospheric and tropospheric first-order drift components. The satellite motion component is computed from ephemeris data as described in [12]. Polynomial approximation is applied to compensate for the first-order drift terms. Figures 14 and 15 demonstrate a consistent carrier phase tracking. In other words, successive accumulated Doppler measurements do not exhibit discontinuities. As the SVs move in relation to the stationary test van and downtown buildings, gradual changes are seen in both the carrier phase measurement and the CNR. For 1 s integration intervals used in these tests, a measured CNR of approximately 12 dB-Hz marked the threshold below which carrier phase tracking became inconsistent.

Time, s

CN

R, d

B-H

z

Time, s

consistent carrier phase tracking

δ(A

ccum

ulat

ed D

oppl

er),

m

Empirically derived carrier phase tracking threshold ~

12 dB-Hz

Figure 14: Accumulated Doppler and CNR for SV 4, Scenario 1

Time, s

CN

R, d

B-H

z

Time, s

δ(A

ccum

ulat

ed D

oppl

er),

m

consistent carrier phase tracking

Figure 15: Accumulated Doppler and CNR for SV 5, Scenario 1

823

Another two interesting effects are seen in Figure 16, namely multpath fading and resulting tracking discontinuities. As the SV moves in relation to the stationary test van and buildings, some of the multipath signals add destructively to the direct path or to each other. This effect reduces CNR to a level below the empirical tracking threshold. Figure 16 shows two carrier phase discontinuities, one for each of two episodes of multipath fading that occur approximately 14 seconds apart. When the receiver tracking function recovers, a discontinuity of 9 to 10 cm appears in the accumulated Doppler. Code phase tracking performance is similarly affected by multipath fading, as shown in Figures 17 and 18. At these instants, the phase adjustment exceeds the correlator spacing, and the code phase adjustment applied by the deep integration receiver exhibits considerable fluctuations. Similarly, the code minus carrier calculation displays similar fluctuation. The open-loop, batch processing architecture allows the code-phase tracking to resume within one signal accumulation interval, which is 1 s, after the fading disappears.

Time, s δ(A

ccum

ulat

ed D

oppl

er),

m

CN

R, d

B-H

z

Time, s

Half-cycle slips

Multipath fading

Figure 16: Half Cycle Slips due to Multipath Fading Observed in Accumulated Doppler and CNR Plots of SV 10, Scenario 1

Time, s

Cod

e Ph

ase

Adj

ustm

ent

(Ear

ly m

inus

Lat

e), m

Code tracking breaks (Phase adjustment exceeds correlator spacing)

Tracking recovery

Figure 17: Code Tracking Breaks due to Multipath Fading Observed in Code Phase Adjustment Plot of SV 10, Scenario 1. Note that code tracking is recovered using carrier phase measurements that are applied to coast through code tracking breakages.

CN

R, d

B-H

z

Time, s

Time, s

Cod

e M

inus

Car

rier

, m

Tracking recovery

Code tracking breaks

Figure 18: Code Tracking Breaks due to Multipath Fading Observed in Code Minus Carrier and CNR Plots for SV 10, Scenario 1

824

Figure 19 shows performance of the receiver integrated velocity computed from carrier phase measurements for Scenario 1. The integrated velocity herein is defined as a receiver velocity integrated over the carrier phase update interval. Details of integrated velocity computations can be found in [12]. The integrated velocity measurements, derived from the carrier phase measurements of up to five SVs (two of which are shown above), show consistent cm-level performance with standard deviations of 0.65 cm, 0.81 cm and 1.39 cm respectively in the north, east and up directions.

Table 2 summarizes the results of the stationary tests for the five scenarios, arranged from hardest to easiest. From this data, it is apparent that direct or indirect path signals from 5 to 6 SVs are available for processing in dense urban canyons, including scenarios where no direct path signals exist. Centimeter-level accuracies in carrier phase-based integrated velocity are obtained in all scenarios, even in the case where no direct path signals exist. Millimeter-level accuracies in integrated velocity are obtained in some scenarios. Finally, consistent carrier phase tracking is demonstrated for at least 2 SVs in the most difficult scenario (no direct path signals) and for up to 5 SVs in less challenging scenarios.

Time, s

Vertical

North

East std dev = 0.65 cm

std dev = 0.81 cm

std dev = 1.39 cm

Err

or, m

Figure 19: Integrated Velocity Errors, Scenario 1

Table 2: Results of Stationary Tests: Carrier Phase (CP) Parameters

Scenario number 2 3 1 4 5

Relative difficulty (hardest to easiest) 5 4 3 2 1

Accuracy level of CP-based integrated velocity: cm mm cm cm mm

• Integrated velocity error sigma in East direction (cm) 1.38 0.55 0.65 0.88 0.24

• Integrated velocity error sigma in North direction (cm) 0.84 0.48 0.81 0.28 0.44

• Integrated velocity error sigma in Vertical direction (cm) 2.59 0.90 1.39 1.14 0.49

Num of SV channels with consistent CP tracking 2 3 4 5 5

825

DEEP INTEGRATION DYNAMIC TEST RESULTS

Examination of the instantaneous frequency of multipath signals in a moving test vehicle reveals that significant differences from the instantaneous frequency of the direct path signal can exist. These differences exist for those cases where LOS vectors from the SV and from the reflecting object to the receiver differ significantly from each other, and the projection of the receiver velocity on the LOS vector from the reflecting object is non-zero. The vector difference between the two components of the receiver velocity rcvrV along these two directions is

termed V∆ and is illustrated in Figure 20 below. When postprocessed in the deep integration receiver by computing 3D signal images, as in Figure 21, the direct path signal energy peak is readily distinguishable from the multipath signal peak(s).

Receiver

rcvrV

Direct signal

Multipath signal

Reflecting object (e.g., wall)

multipathV

directV

∆V

|∆V| ~= | Direct path frequency – Multipath frequency |

Satellite

Figure 20: Frequency Difference between Direct Path and Multipath Signals

Relative signal energy

Multipath signal

Direct-path signal

Doppler shift, Hz Code shift,

chips Figure 21: Direct Path and Multipath Signal Peaks

These techniques are being applied in ongoing research to data taken in the test vehicle as it moved between the stationary scenarios presented above. Results of these dynamic tests will be presented in future papers.

SIMULATED INTEGRATION WITH LOW-GRADE IMU

For designs that move beyond the prototyping stage, a critically important design aspect is the relationship between IMU cost and overall system performance, here exemplified by consistent carrier phase tracking. To establish an empirical relationship between these two criteria, a simulation was performed as follows:

• Three IMU sensor performance models were created to span the performance space between the DQI (approximately $15,000 when purchased in 2005) and IMUs expected to be available in the next few years in the $1,000 range. These were designated Low Grade IMU 1, 2 and 3, as shown in Table 3 below.

• Inertial sensor data obtained from the DQI was

corrupted with accelerometer biases and gyroscope drifts. Inertial sensor errors were simulated as first-order Gauss-Markov processes with a time constant of 100 s.

• The maximum deep GPS/IMU integration period

for each sensor performance model was determined by identifying when the 3-sigma INS error, mapped into the position error space, exceeded one quarter wavelength of the GPS L1 carrier frequency, as shown in Table 3 and Figure 22 below.

• The reduced integration times and CNRs for

each sensor performance model were applied to the five stationary cases presented above, and overall system performance was determined in postprocessing.

Table 3: Simulated IMU Sensor Performance Models

Model characteristics

Low Grade IMU 1

Low Grade IMU 2

Low Grade IMU 3

Estimated cost $4,000 $2,000 <$1,000

Maximum integration interval

(s) 1 0.72 0.46

826

Integration time, s

Posi

tion

erro

r (3

sigm

a), m

0.46 s

threshold for carrier phase tracking aiding λL1/4

0.72 s

Figure 22: Simulated Error Growth for 3 IMU Sensor Performance Models Table 4 below shows the number of SVs that would have been visible in the five stationary tests with IMUs of these varying simulated performance characteristics. The 12 dB-Hz signal processing threshold used previously is increased for each IMU model by the amounts shown in Table 3 above. Table 4 also reveals that the number of SVs visible is fairly insensitive to reductions in IMU quality. Scenario 5, the most difficult of all the stationary scenarios, is the only one to show any reduction in SVs visible as a result of simulated IMU performance reductions. Table 4: Number of SVs Visible for IMU Sensor Performance Models

IMU Grade Number of SVs Visible

Scenario number 2 3 1 4 5

Relative difficulty 5 4 3 2 1

DQI 5 6 5 5 6

Low Grade IMU 1 5 6 5 5 6

Low Grade IMU 2 5 6 5 5 6

Low Grade IMU 3 3 6 5 5 6

Table 5 and Figure 23 below show the number of SV channels displaying consistent carrier phase tracking for each of the IMU sensor models. IMU cost is much more correlated to the number of channels with consistent carrier-phase tracking (and hence to overall system performance) than it is to the number of SVs visible. The

relationship is seen in these discrete points on a smoothly sloped surface. IMU performance levels that are better than those associated with a nominal cost of $4,000 per unit yield no apparent improvement in overall system performance. IMU performance levels worse than this level yield an almost linear reduction in overall system performance down to zero consistent carrier phase tracking channels in the most difficult scenario. Table 5: Number of SV Channels with Consistent Carrier Phase Tracking

IMU Grade # of SV Channels with

Consistent Carrier Phase Tracking

Scenario number 2 3 1 4 5

Relative difficulty 5 4 3 2 1

DQI 2 3 4 5 5

Low Grade IMU 1 2 3 4 5 5

Low Grade IMU 2 1 2 3 4 5

Low Grade IMU 3 0 2 3 4 5

Figure 23: Number of SV Channels with Consistent Carrier Phase Tracking

CONCLUSIONS

GPS signals in urban canyons are characterized as they appear to a conventional GPS receiver; an advanced receiver optimized for urban areas; and a batch processing, deeply integrated GPS/IMU receiver with open-loop tracking architecture. Significant performance improvements are noted for each successive architecture.

827

Signals from 5 to 6 SVs are available for processing by the deeply integrated GPS/IMU receiver even in very dense urban canyons. The quality of these signals for tracking purposes is assessed in three respects. First, carrier phase-based integrated velocity is shown to be accurate at least to cm level in all scenarios, and to mm level in some scenarios. Second, consistent carrier phase tracking is demonstrated for at least 2 SVs in the most challenging scenario, where no direct path signals exists, and up to 5 SVs in less difficult scenarios. Third, signal tracking can break down when CNR is below 12 dB-Hz (corresponding to a 1 s integration interval), or due to multipath fading. The difference in frequency between direct path and multipath GPS signals is shown to provide a clear way to distinguish between these signals. Frequency is thus a potentially useful factor for identifying and tracking GPS signals in dynamic scenarios. Finally, the relationship of cost versus performance for IMU quality in a deeply integrated GPS/IMU architecture is shown to be a fairly smooth slope, within bounds. The limiting performance factor appears to be the number of channels with consistent carrier phase tracking. The range of interest in IMU unit estimated cost ranges from less than $1,000 upwards to an expected $4,000. In the most difficult scenario where no direct path SV signals are received, the improvement in overall system performance increases nearly linearly with increased cost.

ACKNOWLEDGEMENTS

The authors wish to thank the Robert Bosch Corporation, RTC, for funding this research. The authors would also like to thank the Federal Aviation Administration for funding prior research into frequency multipath effects.

REFERENCES

[1] Marti, L. The Effects of the Radio Frequency Front-End Onto Signal Estimation. Proceedings of the Institute of Navigation GPS-2003, 9-12 September 2003, Portland, OR, USA.

[2] Gunawardena, S., van Graas, F. and Soloviev, A. Real Time Block Processing Engine for Software GNSS Receivers. Proceedings of the ION NTM, 26-28 January 2004, pp. 371-377, San Diego, CA, USA.

[3] van Graas, F., Soloviev, A., Uijt de Haag, M., Gunawardena, S. and Braasch, M. Comparison of Two Approaches for GNSS Receiver Algorithms: Batch Processing and Sequential Processing Considerations. Proceedings of the ION GNSS-2005, 13-16 September 2005, Long Beach, CA, USA.

[4] Soloviev, A., van Graas, F. and Gunawardena, S. Implementation of Deeply Integrated GPS/Low-Cost IMU for Reacquisition and Tracking of Low CNR GPS Signals. Proceedings of the ION NTM, 26-28 January 2004, San Diego, CA, USA.

[5] Gunawardena, S., Soloviev, A. and van Graas, F. Real Time Implementation of Deeply Integrated Software GPS Receiver and Low Cost IMU for Processing Low-CNR GPS Signals. Proceedings of the ION 60th Annual Meeting, 7-9 June 2004, Dayton, OH, USA.

[6] Soloviev, A., Gunawardena, S. and van Graas, F. Deeply Integrated GPS/Low-Cost IMU for Low CNR Signal Processing: Flight Test Results and Real Time Implementation. Proceedings of the ION GNSS-2005, 13-16 September 2005, Long Beach, CA, USA.

[7] Tsui, J. B. Fundamentals of Global Positioning System Receivers: A Software Approach. New York: John Wiley & Sons, Inc., 2000.

[8] Data sheet for DQI-digital quartz IMU (inertial measurement unit) (www.systron.com/PDFS/datasheets/DQI.pdf)

[9] Technical Description of the LMS 200, LMS 220, LMS 211, LMS 221, LMS 291 Laser Measurement Systems. Revised June 2003 (www.mysick.com/saqqara/pdf.aspx?id=im0012759.)

[10] OEM4 Family User Manual Volume 1: Installation and Operation. Revised December 22, 2005 (www.novatel.com/Documents/Manuals/om-20000046.pdf)

[11] GPS Chip Solution: SiRF StarIII GSC3e/LP (www.sirf.com/products/gps_chip.html)

[12] van Graas, F. and Soloviev, A. Precise Velocity Estimation Using a Stand-alone GPS Receiver. Proceedings of the ION NTM, January 22-24, 2003, Anaheim, CA, USA.

828