1

Opportunistic Integrity Monitoring forEnhanced UAV Safety

Mahdi Maaref, Joe Khalife, and Zaher M. Kassas

Abstract

An opportunistic advanced receiver autonomous integrity monitoring (OARAIM) framework forunmanned aerial vehicle (UAV) navigation is developed. This framework aims at fusing ambientterrestrial signals of opportunity (SOPs) with global navigation satellite system (GNSS) signals toprovide tight protection level (PL) bounds. A receiver is assumed to make pseudorange measurementson multiple GNSS satellites and terrestrial SOPs. The PLs ofthe proposed framework are analyzed,and the reduction in the PLs of a GNSS-SOP solution over that of a GNSS-only solution is studied.It is demonstrated that adding SOP measurements, which inherently come from low elevation angles,is more effective in reducing the PLs than adding measurements from new satellites. Experimentalresults are presented demonstrating the performance of theproposed system. The results show thatOARAIM reduces the vertical and horizontal PLs by 46.7% and 50.9%, respectively, compared tothe ARAIM method.

I. INTRODUCTION AND MOTIVATION

Autonomous unmanned aerial vehicles (UAVs) are predicted to revolutionize a wide range of sectors,such as surveying, farming, filming, construction, transportation, emergency response, infrastructureinspection, and package delivery. As these vehicles approach full-autonomy, the accuracy and integrityof their navigation system become ever more stringent [1], [2]. While the notion of accuracy is self-explanatory, the notion of integrity is less obvious, but itis of utmost importance in the safety criticalapplication of aviation. Integrity is a criterion to evaluate the reliability and to measure the level oftrust in the information produced by a navigation system. A high-integrity navigation system mustbe able to detect and reject faulty measurements and providean integrity measure of the confidencein the system performance at any time. Integrity monitoringcan be provided through the GNSSnavigation messages to indicate satellite anomalies, suchas the clock errors. However, this typeof integrity information is not useful for real-time applications, as it takes a few hours to identifyand broadcast the satellite service failure [3]. Thus, alternative frameworks for integrity monitoringhave been developed, which can be categorized into: internal and external [4]. External methods (e.g.,ground-based augmentation system (GBAS) and satellite-based augmentation system (SBAS)) leveragea network of ground monitoring stations to monitor the transmitted signals, while internal methods(e.g., receiver autonomous integrity monitoring (RAIM)) typically use the redundant information withinthe transmitted navigation signals. RAIM inherently possesses desirable characteristics due to its designflexibility and adaptability [5]. RAIM is a technique primarily based on checking the consistency ofredundant measurements. RAIM assesses the availability performance by calculating the protectionlevel (PL) on-the-fly, which is the radius of a circular area centered around the position solution andis guaranteed to contain the true position to within the specifications of RAIM, i.e., with a probability

This work was supported in part by the National Science Foundation (NSF) under Grant 1751205 and in part by theOffice of Naval Research (ONR) under Grants N00014-16-1-2809 and N00014-19-1-2511. The authors would like to thankKimia Shamaei for her help in data collection.

M. Maaref and Z. Kassas are with the Department of Mechanicaland Aerospace Engineering, The University of California,Irvine. Joe Khalife is with the Department of Electrical Engineering and Computer Science, The University of California,Irvine. (email:mmaaref@uci.edu, khalifej@uci.edu, andzkassas@ieee.org).

2

less than or equal to an acceptable integrity risk [6]. By comparing the PL with a pre-defined alertlimit (AL), the availability of the navigation system couldbe determined; specifically, if the PL isless than the AL, the navigation solution is deemed reliablefor the pre-defined integrity risk, andunreliable otherwise.

RAIM was initially proposed for GPS-based navigation. Recently, advanced RAIM (ARAIM)algorithms have been developed for multi-constellation navigation systems, which use measurementsfrom different global navigation satellite systems (GNSS)[7], e.g., Galileo-GPS, GLONASS-GPS,etc. Nevertheless, relying on GNSS signals alone poses an alarming vulnerability for UAV navigationdue to line-of-sight (LOS) blockage by high-rise structures, unintentional interference, intentionaljamming, and spoofing. Besides, due to the geometric configuration of GNSS satellites, the verticalerror of the GNSS navigation solution is too large for UAV navigation in urban environments [8]. Toaccount for GNSS limitations, alternative sensors have been integrated into UAV’s navigation system,and the integrity of these sensors has been the subject of recent studies. Recently, many differentRAIM schemes incorporating other sensing modalities have been proposed, such as (i) multi-GNSSconstellation RAIM (e.g., Galileo-GPS [9] and GLONASS-GPS[7]), (ii) GNSS-sensor RAIM (e.g.,GPS, inertial measurement units (IMUs), wheel speed encoders, and cameras [10]; GPS and lidar[11]; GNSS and IMU [12]; and GNSS, lidar, and IMU [13]), (iii)GNSS-SOP RAIM (e.g., [14]), and(iv) SOP RAIM [15], [16]. In addition to sensors, ambient radio signals in the environment, whichare not intended for navigation, have been recently considered as a supplement or an alternativeto GNSS signals [17]. These signals, termed signals of opportunity (SOPs), can be terrestrial (e.g.,cellular signals [8], digital television signals [18], AM/FM signals [19]) or space-based (e.g, low Earthorbit (LEO) satellites [20], [21]). SOPs possess desirablecharacteristics for navigation purposes: (i)ubiquity, (ii) high received power, (iii) large transmission bandwidth, (iv) wide range of transmissionfrequencies, and (v) geometric diversity. Recent researchhas demonstrated centimeter-level-accurateUAV navigation with cellular SOPs [22] and meter-level-accurate UAV navigation with LEO satellitesignals [20]. Moreover, it has been demonstrated that fusing GNSS and cellular SOPs results insignificant reduction in the UAV’s position uncertainty [8].

As the number of systems that rely on SOPs for navigation grows, the need for modeling mea-surement errors and monitoring the integrity of SOP-based navigation systems increases. Over thepast few years, research has been conducted to model different error sources that deteriorate SOPmeasurements. Table I summarizes some of the common error sources. However, the integrity ofSOP-based navigation systems has been barely studied in theexisting literature. This article presentsa new paradigm, termed opportunistic ARAIM (OARAIM), whichreduces the PLs of UAVs by fusingGNSS and terrestrial SOP pseudorange measurements. It is shown that by incorporating SOPs, thePLs can be made smaller than the ones from any combination of current GNSS constellations, asshown in Figure 1. This reduction is essential in order to meet stringent integrity standard neededfor safe UAV operations, especially in (i) GNSS-challengedenvironments and (ii) environments withpoor satellite-to-user geometry.

Preliminary studies to assess the PL reduction due to using SOPs have been considered in [15],[30], [31]. These studies investigated a classical RAIM-based approach, where the maximum of onlyone measurement outlier at each time-step was considered. However, in the complicated wirelessenvironments (e.g., deep urban canyons, near the buildings, SOP blind spots, etc.) where the sig-nals are heavily affected by multipath and LOS-blockage, the assumption of experiencing only onemeasurement outlier may not be valid anymore. Moreover, at avery high altitude (e.g., UAVs flyingup to an altitude of 250 m above ground level (AGL)) signal interference will be experienced [32]and the received signals from SOPs may not be reliable. To thoroughly tackle these problems, thisarticle extends previous work through five contributions. First, in contrast to previous work, thisarticle aims to detect more than one outlier induced into measurements due to LOS signal blockage

3

TABLE ISOPERROR SOURCES AND ERROR MITIGATION ALGORITHMS.

Error source Reference Proposed algorithms

Clock errors [18]Developed a six-state Kalman filter to calibrate both quadratic and sinusoidal clockerror terms so as to prevent long-term pseudorange divergence and error growth in

the position solutions.

[23]Developed a model for imperfect clocks of the user device andtransmitter and

established a differential positioning methodology basedon the clock error models.

[24]Developed adaptive filters to estimate the unknown SOP clockerror statessimultaneously with the process noise covariance to characterize the SOPs’

oscillation stability.

Multipath [25]Developed a navigation framework based on a multi-state constraint Kalman filter

(MSCKF) to reduce the effect of time-correlated pseudorange measurement noise inmultipath environments.

[26]Studied the impact of the channel on positioning capabilities of LTE with respect tothe signal bandwidth. Obtained timing error distributionsby means of the histogram

of maximum likelihood estimates by using typical channel models.

Signalblockage and

NLOSreception

[15]Developed a framework to detect and exclude the outliers that induce large biases inthe cellular pseudorange measurements. These outliers arecaused by LOS blockage

and short multipath delays.

[27]Developed a path planning method, which employs so-called signal reliability maps

to provide information about regions where large errors dueto poor LOS ormultipath is expected.

Synchronizationand sectormismatch

[28]Studied the clock bias in different sectors of a cellular base transceiver station

(BTS) cell. Identified a dynamical model relating the clock biases in different BTSsectors and validated the model experimentally.

[29]Developed a self-synchronized localization algorithm that relies exclusively on

time-of-arrival (TOA) and time-difference-of-arrival (TDOA) measurementsperformed by a sensor network.

GNSS PLs

Fig. 1. UAV PLs with ARAIM (GNSS only) and OARAIM (GNSS-SOP).

or multipath. To this end, this article establishes a GNSS-SOP OARAIM framework and calculatesthe corresponding vertical PL (VPL) and horizontal PL (HPL). Second, the characteristics of cellularSOP measurements for integrity monitoring purposes are studied. In contrast to [15], where onlythe accuracy of the SOP measurements for ground vehicle was investigated, this article evaluatesthe recorded data for both ground vehicles and UAVs to obtaintwo integrity monitoring criteria,

4

namely, URA and failure rate. Third, a fault-tree and the associated fault probabilities for a combinedGNSS-SOP system is developed. Then, the corresponding integrity support message (ISM) parametersfor SOPs are discussed (e.g., user range error, URA, maximumnominal bias, etc.). Fourth, MonteCarlo simulations are performed to study the reduction of the GNSS-SOP PLs as a function of theexisting satellite-to-user geometry and SOP transmitter-to-user geometry. Fifth, experimental resultswith cellular SOPs are presented evaluating the efficacy of the proposed OARAIM framework on aUAV. The results show that the proposed GNSS-SOP framework can reduce the VPL and HPL by46.7% and 50.9%, respectively.

II. T ERRESTRIAL SOPS ERROR CHARACTERIZATION

In order to incorporate SOP measurements into an ARAIM-typeframework and determine theintegrity of the combined GNSS-SOP system, one must statistically characterize SOP pseudorangemeasurements, namely determine the user range accuracy (URA) and failure rates. While recentresearch have studied fault and error sources in SOP-based navigation [15], this article is interested inthe statistical properties of these resulting errors. URA is a statistical indicator of the ranging accuracyobtainable with SOPs, where accuracy refers to the degree ofconformance of the measurements withthe true ranges; while fault probability refers to the percentage of time that the measurements areunusable by the navigator. In contrast to GNSS signals, SOPsdo not have publicly available recordsfor their ranging characteristics. Therefore, the definition of a “fault” in these measurements is stillopen to interpretation. A study analyzing the statistics ofcellular SOP pseudoranges for ground vehiclenavigation was conducted in [15]. This article extends [15]to characterize cellular SOP pseudorangesfor UAV navigation. In contrast to [15] where only the accuracy of the SOP signals was studied, thisarticle considers both the URA and fault probability of the SOP measurements. This achieves thecorresponding ISM parameters for SOPs, such as probabilityof a single SOP fault and probabilityof SOP constellation fault, which will be discussed in the next Section. The methodology adopted in[15] and in this article can be applied to other types of terrestrial SOPs.

In this article, the data collected by the Autonomous Systems Perception, Intelligence, and Naviga-tion (ASPIN) Laboratory over several years of experimentalcampaigns was used to characterize thestatistics of cellular SOP pseudorange measurements. Cellular SOP pseudoranges were recorded for anaggregate of tens of hours from UAVs and ground vehicles (GVs). These pseudoranges were obtainedusing the Multichannel Adaptive TRansceiver Information eXtractor (MATRIX) [8] software-definedreceiver (SDR) in (i) different environments; (ii) at different carrier frequencies; and (iii) for differentsignal types, including long-term evolution (LTE) and code-division multiple access (CDMA) signals.Note that the data collected by GVs was used to characterize cellular SOP measurements, mimickingUAVs flying at low altitudes (e.g., during takeoff and landing phases and performing missions suchas goods delivery). Table II summarizes the characteristics of the recorded cellular SOPs.

Next, the error component of the SOP measurements (an error equivalent to the GPS user rangeerror (URE)) was characterized using the empirical data andthe method discussed in [15]. Figure 3illustrates the different environments in California, USA, in which the measurements were collectedand the empirical probability density functions (pdfs) calculated from the collected measurements.Overlaid on the empirical pdfs are their corresponding Gaussian approximations. With respect tointegrity, an overbounding function is conservative if it predicts the occurrence of large navigationerrors to be at least as frequent as their actual occurrence [33]. In [34], it was shown that the assumptionfor a zero-mean Gaussian error distribution can be replacedby a requirement that the error distributionis symmetric, unimodal, and whose cumulative distributionfunction (cdf) is overbounding the errorsless than the mean and underbounding the errors greater thanthe mean, i.e.,

Go > Ga, Ga ≤ 0.5

Go ≤ Ga, Ga > 0.5,

5

TABLE IICHARACTERISTICS OF RECORDED CELLULARSOPS.

Environment PlatformNumber

oftowers

Signaltype Freq. [MHz]

Bandwidth[MHz]

Length[s]

Date[DD/MM/YYYY]

Open sky Stationary 2 LTE 2125 and 1955 20 - 13/5/2019

Semi-urbanand urban

GV 2 LTE 2145 and 1955 20 1500 15/11/2016

GV 3 LTE 2145 and 1955 20 2000 20/1/2017

GV 4 LTE 2145 and 739 20 and 10 1600 22/6/2018

GV 5 LTE 1955 and 739 20 and 10 580 27/6/2016

GV 3 LTE 739 10 825 5/11/2017

GV 5 LTE 1955 and 739 20 and 10 1800 22/8/2018

Deep Urban GV 2 LTE 2145 and 1955 20 345 12/10/2018

GV 2 LTE 2145 and 1955 20 500 24/8/2018

Open sky andSemi-urban

UAV 2 CDMA 882.75 1.23 3300 13/5/2017

UAV 7 CDMA 882.75 1.23 2900 22/2/2017

UAV 8 CDMA 882.75 1.23 2200 1/10/2017

UAV 8 CDMA 882.75 1.23 2600 1/10/2017

UAV 2 CDMA 882.75 1.23 3500 15/11/2017

Urban UAV 11 LTE739, 1955,

2125, and 214510 605 16/6/2019

whereGo andGa are the overbound cdf and the empirical cdf, respectively. To ensure the integrity riskaround the tails, the signle-cdf overbounding method [34] was adopted in this article. The overboundcdf parameters, namely cdf measurement error means and cdf standard deviations are summarized inTable III for the different environments and flight altitudes. Here, open sky refers to an environmentwhere the receiver has clear LOS to the transmitter, while low (high) altitude refers to altitudes less(greater) than 5 meters above the ground. As can be seen from Table III, cellular SOP pseudorangeerror’s standard deviations range from around 10 centimeters to nearly 5 meters. Calculating the SOPmeasurements’ fault probability is a more challenging problem, since no official SOP measurementintegrity standard has been established so far. According to the Air Force GPS Standard PositioningService Performance Standard [35], a fault is identified by ameasurement error greater than 4.42 timesthe broadcast URA standard deviation. However, URA information pertaining to SOP measurementsis generally non-existent. To address this issue, an empirical method similar to the one discussed in[7], [36] is employed in this article. This method was initially developed to evaluate the GLONASSfault rate without having access to the broadcast URA. As shown earlier in this section, the SOPmeasurement error is generally less than 4 m. Therefore, it is possible to consider an induced biaslarger than4.42 × 4 = 17.68 m as a fault. Using the cdf-overbound shown in Fig. 2, the probabilityof a cellular SOP transmitter being in a faulty state can be tentatively assumed to be10−5.

Remark 1 . As discussed in the next section, the OARAIM framework doesnot require anSOP-only based navigation solution. Therefore, the availability of SOPs boil down to having at leastone SOP measurement, which is a question of SOP coverage. Subsequently, it is important to notethat the fault probability of SOP measurements calculated in the framework herein is more preciselythe conditional fault probability: conditioned on the factthat there is SOP coverage. Practically, theprobability of having SOP coverage is very close to one sincethe areas of interest are entirely covered

6

Fig. 2. The approximated fault bias and its corresponding probability.

SanBernardino

Aliso Viejo

Riverside

Fig. 3. Characterization of cellular SOP pseudoranges: SOPs environments and corresponding pdfs.

by wireless services. Moreover, this paper assumes the receiver is in the presence of SOPs. Hence,the overall SOP measurement fault probability is also assumed to be10−5.

Remark 2 . Note that the proposed framework (i.e., OARAM for the UAV flight) can be extendedto incorporate measurements taken in other environments ofinterest, such as in deep urban canyonsand at high altitudes. However, at the time this article was written, it was prohibited by the U.S.Federal Aviation Administration (FAA) to fly UAVs in such environments [37]. It is worth pointingout a recent study of UAV connectivity to the cellular network demonstrated that the received cellularsignal power on UAV altitudes between 30 m and 120 m are stronger than the received power onground-based receivers, despite the downwards-tilted cellular antennas [38].

7

TABLE IIIML PARAMETERS OFGAUSSIAN FITS FOR CELLULARSOPPSEUDORANGES IN DIFFERENT ENVIRONMENTS.

Altitude Environment Overbound measurementerror mean [m]

Overboundmeasurement errorstandard deviation

[m]

- Open sky 0 m 0.22 m

Low altitudeUrban andsemi-urban

0 m 3.02 m

Deep urban 0 m 5.97 m

High altitude Semi-urban 0 m 0.97 m

Urban 0 m 2.64 m

SOPs (Ground)

UAV (Aerial)

GNSS signals (Space)

GNSSreceiver

SOPreceiver

Navigation filter

OARAIM

Navigationsolution

PLs

navigation systemUAV

Navigation

solution

and

pseudoranges

Fault

detection

and

exclusion

Fig. 4. Opportunistic navigation framework with OARAIM.

III. OARAIM F RAMEWORK

In this section, an OARAIM framework is developed to performintegrity monitoring for GNSS-SOP-based navigation. A well-designed integrity monitoring framework provides the UAV with thePLs, i.e., horizontal and vertical regions centered at the UAV’s true position, which are guaranteedto contain the UAV’s estimated position with a certain levelof confidence. In this article, a baselinemultiple hypothesis solution separation (MHSS) ARAIM, which was introduced in [36], is used tocalculate the PLs. In the sequel, ARAIM will refer to MHSS ARAIM, for simplicity. ARAIM is arobust framework for combining navigation signals from different navigation sources with differentsignal properties, e.g., different URA values and different probabilities of single or multiple simulta-neous faults. As such, ARAIM is well-suited for combining SOP signals with GNSS signals to formOARAIM. In addition to providing PLs, OARAIM performs faultdetection and exclusion to mitigatethe effect of SOP and/or GNSS system faults on the navigationsolution. Figure 4 summarizes theOARAIM GNSS-SOP framework for UAV navigation.

OARAIM first constructs a fault tree. By definition, a fault-tree refers to a set of assumptions aboutthe environment in which a RAIM algorithm is applied. The measurements are supposed to be inone out of a set of different branches of the fault tree, to each of which ana priori probability

8

Three faults Four or more faults

Three GPS faults

Mode 6

Two GPSs andone SOP faults

Two SOPs andone GPS faults

Mode 10

Two faults Three or more faults

One fault Two or more faults

One or more faults

Mode 0

No fault

Three SOP faults

Mode 7

Mode 8 Mode 9

Two GPS faults

Mode 3

One GPS andone SOP faults

Two SOP faults

Mode 4Mode 5

Single GPS fault

Mode 1

Single SOP fault

Mode 2

∼ 1

Mode 6

Mode 7

Mode 8

Mode 9

Mode 3

Mode 4

Mode 5

Mode 1

Mode 2

NGPS × 10−5× (1 − 10−5)NGPS−1

NSOP × 10−4× (1 − 10−4)NSOP−1

(

NGPS2

)

× (10−5)2 × (1 − 10−5)NGPS−2

(

NSOP2

)

× (10−4)2 × (1 − 10−4)NSOP−2

NGPS × (10−5) × (1 − 10−5)NGPS−1

(

NGPS3

)

× (10−5)3 × (1 − 10−5)NGPS−3

(

NSOP3

)

× (10−4)3 × (1 − 10−4)NSOP−3

(

NGPS2

)

× (10−5)2 × (1 − 10−5)NGPS−2

(

NSOP2

)

× (10−4)2 × (1 − 10−4)NSOP−2

∼ 0

×(1 − 10−4)NSOP

×(1 − 10−5)NGPS

×(1 − 10−4)NSOP

×(1 − 10−5)NGPS

×NSOP × (10−4) × (1 − 10−4)NSOP−1

×(1 − 10−4)NSOP

×(1 − 10−5)NGPS

×NSOP × (10−4) × (1 − 10−4)NSOP−1

×NGPS × (10−5) × (1 − 10−5)NGPS−1

Mode 10

Mode 0

probability

probability

probability

probability

probability

probability

probability

probability

probability

probability

probability

Fig. 5. GPS-SOP fault tree and the associated fault probabilities.

of occurrence is assigned. Therefore, the fault tree can be employed to identify different sources offaults. OARAIM performs multiple statistical tests to detect faults, and then it attempts to excludethe detected faults. The HPL and VPL are subsequently calculated. In the following, the GNSS-SOPfault tree is constructed and the OARAIM algorithm is brieflydescribed.

A. Fault tree and fault modes

OARAIM considers a list of faults that need to be monitored and determines the correspondingprior probabilities that must be assigned to each mode. For simplicity, a combined GPS-SOP fault treewill be discussed. Extension to other GNSS constellations is expected to be straightforward. In [39], amethod was presented to determine the faults that need to be monitored and the associated probabilitiesof faults. Using the same methodology, in this article, a maximum of three simultaneous faults areconsidered. Also, the probability of a constellation fault(i.e., a fault that affects all transmitters) forboth GPS and SOP transmitters are assumed to be sufficiently improbable. This assumption relies onthe historical record of these signals. GPS records show that there is no evidence of a constellationfault since the first GPS satellites were launched [15]. Moreover, no SOP “constellation” faults wereexperienced in any of the tests listed in Table II. The resulting GPS-SOP fault tree is depicted inFigure 5.

To calculate the mode probabilities, the probability of GPSsatellite and SOP transmitter failuresmust be known, namely{PGPS,i}

NGPS

i=1 and {PSOP,i}NSOP

i=1 , respectively; whereNGPS andNGPS arethe numbers of visible GPS satellites and SOP transmitters,respectively. In this article, all SOP

9

transmitter failure probabilities were set to{PSOP,i}NSOP

i=1 = 10−4, as calculated earlier, and all GPSsatellite failure probabilities were set to{PGPS,i}

NGPS

i=1 = 10−5, according to the historical recordsdetailed in [40]. Therefore, the GPS-SOP fault tree and the corresponding prior probability of thefault modes can be expressed as shown in Figure 5, where mode 0corresponds to the fault-free mode,which is the most likely event. Modes 1 to 9 correspond to the faulty operations, including one, two,and three simultaneous faults, while mode 10 is assumed to never occur.

B. OARAIM algorithm

OARAIM shares some common inputs and constant parameters used by ARAIM [39]. While somevalues are independent of the signal type (e.g., total integrity budget, probability of false alarm,etc.), other values are SOP-specific. The OARAIM inputs are tabulated in Table IV. In contrast totraditional RAIM frameworks, where pseudorange measurement errors are assumed to have zero-mean, ARAIM accounts for unknown but bounded pseudorange biases denoted by{bnom,GPS,i}

NGPS

i=1 .For GPS measurements, these biases bound nominal errors, mainly due to the code correlation peakdeformation [41]. The values of the biases are extracted from the ISM and can be limited to 0.75m [5]. A similar value can be conservatively used for biases in SOP measurements, denoted by{bnom,SOP,i}

NSOP

i=1 , as SOP signals are unaffected by atmospheric errors. A summary of the OARAIMalgorithm is given below. The steps shared between OARAIM and ARAIM are not presented here forbrevity, but they can be found in [39].

Step 1: Evaluate each fault mode: This step derives the following parameters for each fault mode:• The variance of the fault-tolerant position, which is defined as the estimated position using all

measurements, except the measurements in the fault mode being evaluated.• The difference between the fault-tolerant position and theall-in-view position (i.e., the estimated

position using all measurements). and the variance of this difference.• The worst-case impact of{bnom,GPS,i}

NGPS

i=1 and{bnom,SOP,i}NSOP

i=1 on the position estimate• Solution separation tests and solution separation thresholds• A chi-squared test and its threshold. Note that the thresholds for solution separation tests as well

as a chi-squared test must be constructed based on the probability of false alarm.Step 2: Calculate PLs: If all of the solution separation tests pass,OARAIM calculates the VPL.

Using the methodology described in [39], the VPL is a solution to the below equation.

2Q

(

V PL− b3(0)

σ3(0)

)

+

Nfault,modes∑

k=1

Pfault(k)Q

(

V PL− T(k)3 b3

(k)

σ3(k)

)

=

PMHIV ER

[

1−Pnot,monitored

PMHIV ER + PMHIHOR

]

,

whereQ is the tail distribution function of the standard normal distribution, b3 is the worst caseimpact of the nominal bias in vertical axis,σ3 is the variance of the fault-tolerant position estimate,Nfault,modes is the number of fault modes withPfault associated probability,Pnot,monitored is theprobability of the faults that are not included in the fault modes,T3 is the solution separation testthreshold, andPMHIV ER andPMHIHOR are the integrity budget for the vertical and horizontalcomponents, respectively.

Step 3: Exclude the faults: If any of the solution separation tests fails, ARAIM attempts theexclusion using the methodology described in [5]

Step 4: Formulate the vertical positioning performance criteria:• Criteria 1: 95% accuracy parameter, which is the achievablepositioning accuracy in the vertical

domain 95% of the time. According to the Localizer Performance with Vertical guidance (LPV)-200 standard, the 95% accuracy must be limited to 4 m.

10

• Criteria 2:10−7 fault-free position error bound, which is the achievable positioning accuracy inthe vertical domain 99.99999% of the fault-free time. According to the LPV-200 standard, the10−7 fault-free position error bound must be limited to 10 m.

• Criteria3: Effective monitor threshold (EMT), which is a parameter that takes into account thefaults with a prior≥ 10−5. According to the LPV-200 standard, EMT must be limited to 15m.

Figure 6 summarizes the OARAIM algorithm.

TABLE IVINPUTS TO THEGPS-SOP OARAIMALGORITHM .

Input Description Obtained from

{zGPS,i}NGPS

i=1

GPS pseudorangemeasurements

GPS front-end and trackingloop

{zSOP,i}NSOP

i=1

SOP pseudorangemeasurements

SOP front-end and trackingloop

{σURA,GPS,i}NGPS

i=1

Standard deviation of GPSURA

ISM

{σURA,SOP,i}NSOP

i=1

Standard deviation of SOPURA

From the overbounding cdf

{σURE,GPS,i}NGPS

i=1

Standard deviation of the GPSuser range error

ISM

{σURE,SOP,i}NSOP

i=1

Standard deviation of the SOPuser range error

Table III

{bnom,GPS,i}NGPS

i=1

Maximum bias for a GPSmeasurement

ISM

{bnom,SOP,i}NSOP

i=1

Maximum bias for a SOPmeasurement

Similar to the GPS maximumbias

{PGPS,i}NGPS

i=1

Probability of a single GPSfault

Historical records. Currentlyused value is10−5

{PSOP,i}NSOP

i=1

Probability of a single SOPfault

Experimental campaign.Proposed value is10−4

PConst,GPS

Probability of GPSconstellation fault

Historical records. Currentlyused value is 0

PConst,SOP

Probability of SOPconstellation fault

Experimental campaign.Proposed value is10−4

IV. PERFORMANCEEVALUATION

This section evaluates the performance of the OARAIM framework numerically and experimentally.

A. Simulation Results

A numerical simulation study was conducted to evaluate the reduction in HPL and VPL by incor-porating an additional SOP measurement versus an additional GNSS measurement. Since all GNSSsatellites are typically above the UAV, the satellites’ elevation angles theoretically range between 0and 90 degrees. GNSS receivers typically restrict the lowest elevation angle to a pre-defined elevationmask (typically around 10 degrees), to discard GNSS signalsthat are heavily degraded due to theionosphere, troposphere, and multipath. In contrast, the proposed OARAIM framework sweeps twicethe GNSS-only elevation angle range, i.e., -90 to 90 degrees, due to the fact that UAVs can fly directlyabove terrestrial SOPs.

11

GNSSreceiver

SOPreceiver

fzGPS;igNGPS

i=1

(See Fig. 4)Fault tree

NGPS

NSOP

Fault modesPrior Compute tests

statistics

(See Table 3)OARAIM inputs

solution

fails?

Any

chi-squaredtest fails?

Exclusionpossible?

Yes

No

Yes

Integrity is not guaranteed

Alarm TriggersIntegrity is guaranteed

Standard Operation

Compute PLs

No

Yes

Yes

(See [39])requirements

fails?

Compute EMT(See [5])

No

No

YesNo

fzSOP;igNSOP

i=1

VPLHPL

EMT

PLs

requirementsfails?

EMT

Probabilities

separation test

Fig. 6. OARAIM algorithm.

To illustrate the expected reduction in the PLs as a functionof the elevation angle of a newlyadded transmitter, 10,000 Monte Carlo realizations were generated. In each realization, the azimuthand elevation angles ofNGPS GPS satellites were drawn from uniform distributions over the interval[−180, 180] degrees for the azimuth angles and the interval[−90, 90] degrees for the elevation angles.Then, an additional measurement from a new transmitter was introduced. The new transmitter’sazimuth angle was also drawn from a uniform distribution over the interval [−180, 180] degrees,while its elevation angle was swept between -90 to 90 degrees. Figure 7 shows the reduction in theVPL and HPL as a function of the new transmitter’s elevation angle, for differentNGPS values anddifferent GPS elevation angle mask.

The following may be concluded from Figure 7. First, while adding measurements from additionalsatellites decreases PLs, measurements from transmittersat low elevation angles (shaded green region)are on average more effective at minimizing PLs than transmitters at elevation angles between 0 and90 degrees. Therefore, adding terrestrial SOP measurements, which are coming from transmitters atpractically -90 to 0 degree elevation angles, are on averagemore effective at minimizing the PLs thanadding GNSS satellite measurements. Second, the reductionis almost constant at negative elevationangles. This conveys an important conclusion that regardless of the UAV’s flight height (whether abovethe SOPs or at the same height with respect to SOPs), OARAIM can provide the same PL reduction.Third, in practice, GNSS receivers impose an elevation maskof about 15 degrees, which increasesthe PLs due to limited satellite-to-user geometry. In such cases, adding SOPs compensates for thislimitation and tightens the PLs.

In order to study the performance of the OARAIM algorithm under fault-free and faulty conditions, asimulation was performed withNSOP = 2. For this test, GPS signals were obtained from a stationaryreceiver at the Madrid Deep Space Communications Complex (MDSCC). The receiver’s positionwas fixed atrr ≡ (106)·[4.849210,−3.603297, 4.11492]T expressed in an Earth-Centered-Earth-Fixed(ECEF) coordinate frame. The elevation and azimuth angles of the GPS satellite constellation above thereceiver over a 24-hour period was computed using GPS ephemeris files collected at the MDSCC. TheGPS observations were extracted from the recorded ReceiverIndependent Exchange Format (RINEX)file. Then, the SOP signals were simulated using a high-fidelity SOP simulator that has been usedin the previous research [42]. The SOP and receiver’s clock qualities were modeled as typical oven-controlled crystal oscillator (OCXO) and typical temperature-compensated crystal oscillator (TCXO),respectively [24]. The simulation settings are given in Table V. In the first scenario, both SOPs were

12

Fig. 7. The reduction in VPL and HPL after adding an additional measurement at an elevation angleelnew, which wasswept between -90 to 90 degrees. The green regions show negative elevation angles. The PL reduction in these regions canbe obtained only using SOP measurements.

fault-free, while in the second scenario, a fault of a magnitude of 30 m was injected into the secondSOP measurement. To illustrate the accuracy and integrity performances simultaneously, a so-calledStanford diagram was plotted in Figure 8, where the positionerror (PE), PL, and AL are shownfor four scenarios: GPS-only (black dots), GPS-SOP in fault-free operation (blue dots), GPS-SOPwithout OARAIM fault exclusion (red dots), and GPS-SOP withOARAIM fault exclusion (bluedots). By definition, PLs are not calculated after fault detection. However, similar to [6], the PLsare shown in the Stanford plot for a longer duration for subsequent analyses. The following may beconcluded from this diagram. First, by comparing the blue and black dots, it can be seen that addingSOPs eliminates system unavailability. Second, injectingthe fault into an SOP measurement causeda misleading operation (red dots); however, the OARAIM algorithm rejected the faulty measurementto achieve nominal operation (green dots). Third, by comparing the red and green dots, it can be seenthat as expected, the exclusion results in an increases in the PL.

B. Experimental Results

In order to evaluate the performance of the proposed OARAIM framework in a real-world scenario,a DJI Matrice 600 UAV was equipped with a dual-channel National Instrument (NI) universal softwareradio peripheral (USRP)-2955 to sample real LTE SOPs at fourLTE carrier frequencies: 739, 1955,2125, and 2145 MHz. These frequencies are allocated for the U.S. cellular providers AT&T, T-Mobile,and Verizon. The ground-truth reference for the UAV’s trajectory was taken from a Septentrio AsteRx-iV integrated GNSS-IMU system, which is capable of producinga real-time kinematic (RTK) solution.Figure 9 shows the experimental setup.

The UAV flew for 4 minutes, while collecting LTE signals from 11 LTE transmitters in theenvironment. The stored LTE signals were then processed by the LTE module of the MATRIX SDRto produce LTE SOP pseudoranges, which were then fused with raw GPS pseudorange measurementsobtained from the Septentrio receiver to produce the navigation solution along with the correspondingOARAIM integrity measures.

13

TABLE VSIMULATION SETTINGS

Parameter Definition Value

{zGPSm}NGPS

m=1GPS pseudorange measurements

Extracted from the navigation RINEX file (20July 2006 to 21 July 2006)

NGPS Number of used GPS satellites 5 to 9

NSOP Number of used SOP transmitters 2

PHMIHOR Integrity budget for the horizontal component 2× 10−9

PHMIVERT Integrity budget for the vertical component 9.8× 10−8

PFAVERT Continuity budget allocated to the vertical component 3.9× 10−6

PFAHORContinuity budget allocated to the horizontal

component9× 10−8

rr UAV’s position (106)·[4.849210,−3.603297, 4.11492]T

T Sampling time 0.1 s

{σGPSm}NGPS

m=1GPS measurement noise standard deviation 3 m

{σSOPn}2

n=1SOP measurement noise standard deviation 2 m

{rSOPn}2

n=1Towers’ position (106)·[4.848656,−3.60789, 4.115529]T

(106)·[4.848123,−3.60769, 4.116136]T

Two sets of results were produced to evaluate the impact of SOP measurements on navigation andsafety: (i) a navigation solution and ARAIM integrity measures using GPS measurements only and (ii)a navigation solution and OARAIM integrity measures using GPS and cellular SOP measurements.The two-dimensional (2-D) and three-dimensional (3-D) position root-mean squared errors (RMSEs)and maximum position errors are tabulated in Table VI for both navigation solutions: GPS-only andGPS-SOP. Table VII summarizes the average VPL and HPL over the entire UAV flight for GPS-only RAIM, GPS-only ARAIM, and GPS-SOP OARAIM. Figure 9 alsoshows (i) UAV’s traversedtrajectory, (ii) GPS-only and GPS-SOP skyplots showing satellite-to-user and SOP transmitter-to-usergeometry, and (ii) GPS-only and GPS-SOP PLs.

TABLE VINAVIGATION SOLUTION PERFORMANCE COMPARISON.

Solution 2-D RMSE[m]

3-D RMSE[m]

2-D Max.error [m]

3-D Max.error [m]

GPS-only 2.34 4.37 9.11 18.08

GPS-SOP 1.13 3.63 8.28 15.86

Reduction 51.4% 16.98% 9.14% 12.25%

TABLE VIIPL REDUCTION PERFORMANCE COMPARISON.

Approach VPL [m] HPL [m]

GPS-only RAIM 23.5 17.1

GPS-only ARAIM 12 11.4

GPS-only OARAIM 6.4 5.6

OARAIM reduction fromARAIM

46.6% 50.8%

14

Position error (PE) [m]

Protectionlevel(P

L)[m

]

Nominal operation

GPS-SOP (no fault)

GPS-SOP without using OARAIM fault exclusion

GPS-SOP with using OARAIM fault exclusion

Alert

limit(A

L)[m

]

AL)

GPS-only

Fig. 8. The Stanford diagram demonstrating the accuracy andintegrity performances simultaneously. Although the PL isnot calculated after the fault, it is calculated for a longerduration for subsequent analysis. To generate the results shownin this plot, GPS signals over a period of 24 hours were combined with simulated SOP signals from 2 SOP transmitters,obtained from a high-fidelity simulator.

V. CONCLUSION

To improve the availability of the integrity monitoring system, the PLs must remain small. Thisarticle showed that by incorporating SOPs, the PLs can be made smaller than the ones from anycombination of current GNSS constellations. To this end, the article presented an OARAIM frameworkfor enhanced UAV safety. OARAIM enables safe UAV navigationwith GNSS and SOPs, producingtight PLs, while identifying and excluding faults, if present. The article characterized the URAand failure rates of SOP pseudorange measurements for integrity monitoring purposes by studyingdata collected over extensive experimental campaigns in different environments. It was concludedthat cellular SOPs have similar characteristics to GPS pseudorange measurements. A fault tree wasconstructed for GPS-SOP-based navigation and the OARAIM algorithm was presented. Simulationresults were presented demonstrating that SOPs are more effective than GNSS at decreasing theUAV’s PLs. Experimental results were presented showing that the OARAIM reduces the vertical andhorizontal PLs by 46.7% and 50.9%, respectively, compared to the ARAIM method. The PL reductionin OARAIM translates to higher availability of the integrity monitoring system, allowing the UAVnavigation system to meet more stringent integrity standards than ARAIM with GNSS only. As such,UAVs can navigate safer and for longer periods of time with OARAIM.

ACKNOWLEDGEMENT

The authors would like to thank Kimia Shamaei for her help in data collection.

15

Multi-frequencyGNSS antennas

AsteRx-imodule

VN-100 IMU

GNSS-IMU system

LTE receiver Cellular antennas

NI 2955 USRP

Power supply

Storage and software

MATRIX LTE SDR

MATLAB-based estimatorand

Laptop

Battery

Johan

0◦

90◦270◦

180◦

◦

120◦240

300

0◦

90◦270◦

◦

120◦240

300

180◦

GPS-only GPS-SOP

GPS-only RAIM

VPL = 23.5 m

GPS satellite

SOP transmitter

HPL = 17.1 m

PRN2

SOP1SOP2

SOP4 SOP5SOP6SOP7

SOP8

PRN25PRN29

PRN6

PRN5

PRN12

PRN19

SOP3 SOP9

SOP10

UAV's Traversed path

Fig. 9. Experimental environment, experimental setup, andexperimental results showing the traversed trajectory, the GPS-only and GPS-SOP skyplots, and the PLs. The results show thatOARAIM reduces the vertical and horizontal PLs by 46.66%and 50.87%, respectively, compared to the ARAIM method.

16

REFERENCES

[1] R. Loh, Y, Bia, and T. Roe, “UAVs in civil airspace: safetyrequirement,”IEEE Aerospace and Electronic SystemsMagazine, vol. 24, no. 1, pp. 5–17, January 2009.

[2] D. Erdos, A. Erdos, and S. Watkins, “An experimental UAV system for search and rescue challenge,”IEEE Aerospaceand Electronic Systems Magazine, vol. 28, no. 5, pp. 32–37, May 2013.

[3] E. Kaplan and C. Hegarty,Understanding GPS/ GNSS: Principles and applications, 3rd ed. Artech House, 2017.[4] D. Imparato, A. El-Mowafy, and C. Rizos, “Integrity monitoring: From airborne to land applications,” inMultifunctional

Operation and Application of GPS. IntechOpen, 2018, pp. 23–43.[5] V. Kropp, “Advanced receiver autonomous integrity monitoring for aircraft guidance using GNSS,” Ph.D. dissertation,

University of Munich, Germany, 2018.[6] S. Bhattacharyya and D. Gebre-Egziabher, “Kalman filter-based RAIM for GNSS receivers,”IEEE Transactions on

Aerospace and Electronic Systems, vol. 51, no. 3, pp. 2444–2459, July 2015.[7] T. Walter, J. Blanch, M. J. Choi, T. Reid, and P. Enge, “Incorporating GLONASS into aviation RAIM receivers,” in

Proceedings of International Technical Meeting of the Institute of Navigation, January 2013, pp. 239–249.[8] Z. Kassas, J. Khalife, K. Shamaei, and J. Morales, “I hear, therefore I know where I am: Compensating for GNSS

limitations with cellular signals,”IEEE Signal Processing Magazine, pp. 111–124, September 2017.[9] A. Ene, J. Blanch, and T. Walter, “Galileo-GPS RAIM for vertical guidance,” inProceedings of National Technical

Meeting of The Institute of Navigation, January 2006, pp. 18–20.[10] R. Toledo-Moreo, D. Betaille, and F. Peyret, “Lane-level integrity provision for navigation and map matching with

GNSS, dead reckoning, and enhanced maps,”IEEE Transactions on Intelligent Transportation Systems, vol. 11, no. 1,pp. 100–112, March 2010.

[11] A. Kanhere and G. Gao, “Integrity for GPS-LiDAR fusion utilizing a RAIM framework,” in Proceedings of ION GNSSConference, September 2018, pp. 3145–3155.

[12] T. Needham and M. Braasch, “Gravity model error considerations for high-integrity GNSS-aided INS operations,” inProceedings of IEEE/ION Position, Location and Navigation Symposium, 2018, pp. 822–832.

[13] A. Hassani, N. Morris, M. Spenko, and M. Joerger, “Experimental integrity evaluation of tightly-integrated IMU/LiDARincluding return-light intensity data,” inProceedings of ION GNSS Conference, September 2019, pp. 2637–2658.

[14] M. Maaref, J. Khalife, and Z. Kassas, “Enhanced safety of autonomous driving by incorporating terrestrial signalsof opportunity,” in Proceedings of IEEE International Conference on Acoustics, Speech and Signal Processing, May2020, accepted.

[15] M. Maaref and Z. Kassas, “Measurement characterization and autonomous outlier detection and exclusion for groundvehicle navigation with cellular signals,”IEEE Transactions on Intelligent Vehicles, 2019, accepted.

[16] M. Maaref and Z. Kassas, “Autonomous integrity monitoring for vehicular navigation with cellular signals of opportunityand an IMU,” IEEE Transactions on Intelligent Transportation Systems, 2020, submitted.

[17] Z. Kassas, P. Closas, and J. Gross, “Navigation systemsfor autonomous and semi-autonomous vehicles: Current trendsand future challenges,”IEEE Aerospace and Electronic Systems Magazine, vol. 34, no. 5, pp. 82–84, May 2019.

[18] C. Yang, T. Nguyen, and E. Blasch, “Mobile positioning via fusion of mixed signals of opportunity,”IEEE Aerospaceand Electronic Systems Magazine, vol. 29, no. 4, pp. 34–46, April 2014.

[19] S. Fang, J. Chen, H. Huang, and T. Lin, “Is FM a RF-based positioning solution in a metropolitan-scale environment?A probabilistic approach with radio measurements analysis,” IEEE Transactions on Broadcasting, vol. 55, no. 3, pp.577–588, September 2009.

[20] Z. Kassas, J. Morales, and J. Khalife, “New-age satellite-based navigation – STAN: simultaneous tracking and navigationwith LEO satellite signals,”Inside GNSS Magazine, vol. 14, no. 4, pp. 56–65, 2019.

[21] J. Khalife, M. Neinavaie, and Z. Kassas, “Navigation with differential carrier phase measurements from megacon-stellation LEO satellites,” inProceedings of IEEE/ION Position, Location, and Navigation Symposium, April 2020,accepted.

[22] J. Khalife and Z. Kassas, “Precise UAV navigation with cellular carrier phase measurements,” inProceedings ofIEEE/ION Position, Location, and Navigation Symposium, April 2018, pp. 978–989.

[23] M. Rabinowitz and J. Spilker, Jr., “A new positioning system using television synchronization signals,”IEEETransactions on Broadcasting, vol. 51, no. 1, pp. 51–61, March 2005.

[24] Z. Kassas, V. Ghadiok, and T. Humphreys, “Adaptive estimation of signals of opportunity,” inProceedings of IONGNSS Conference, September 2014, pp. 1679–1689.

[25] K. Shamaei, J. Morales, and Z. Kassas, “A framework for navigation with LTE time-correlated pseudorange errors inmultipath environments,” inProceedings of IEEE Vehicular Technology Conference, April 2019, pp. 1–6.

[26] J. del Peral-Rosado, J. Lopez-Salcedo, G. Seco-Granados, F. Zanier, and M. Crisci, “Evaluation of the LTE positioningcapabilities under typical multipath channels,” inProceedings of Advanced Satellite Multimedia Systems Conferenceand Signal Processing for Space Communications Workshop, September 2012, pp. 139–146.

[27] S. Ragothaman, M. Maaref, and Z. Kassas, “Autonomous ground vehicle path planning in urban environments usingGNSS and cellular signals reliability maps – part I: Models and algorithms,”IEEE Transactions on Aerospace andElectronic Systems, 2018, submitted.

17

[28] J. Khalife, K. Shamaei, and Z. Kassas, “Navigation withcellular CDMA signals – part I: Signal modeling and software-defined receiver design,”IEEE Transactions on Signal Processing, vol. 66, no. 8, pp. 2191–2203, April 2018.

[29] O. Jean and A. Weiss, “Passive localization and synchronization using arbitrary signals,”IEEE Transactions on SignalProcessing, vol. 62, no. 8, pp. 2143–2150, April 2014.

[30] M. Maaref and Z. Kassas, “UAV integrity monitoring measure improvement using terrestrial signals of opportunity,”in Proceedings of ION GNSS Conference, September 2019, pp. 3045–3056.

[31] M. Maaref and Z. Kassas, “Aerial vehicle integrity monitoring improvement by fusing gps and terrestrial signals ofopportunity,” NAVIGATION, Journal of the Institute of Navigation, 2019, submitted.

[32] A. Colpaert, E. Vinogradov, and S. Pollin, “Aerial coverage analysis of cellular systems at LTE and mmwave frequenciesusing 3D city models,”Sensors, vol. 18, no. 12, December 2018.

[33] J. Rife, S. Pullen, P. Enge, and B. Pervan, “Paired overbounding for nonideal LAAS and WAAS error distributions,”IEEE Transactions on Aerospace and Electronic Systems, vol. 42, no. 4, pp. 2371–9877, October 2006.

[34] B. DeCleene, “Defining pseudorange integrity-overbounding,” in Proceedings of International Technical Meeting ofthe Satellite Division of The Institute of Navigation, September 2000, pp. 1916–1924.

[35] GPS Directorate, “Global positioning system standardpositioning service performance standard (GPS SPS PS),”http://www.gps.gov/technical/ps/, September 2008.

[36] L. Heng, “Safe satellite navigation with multiple constellations: Global monitoring of GPS and GLONASS signal-in-space anomalies,” Ph.D. dissertation, Stanford University, USA, 2012.

[37] Federal Communications Commission, “Overview of small UAS notice of proposed rulemaking,”https://www.faa.gov/uas/nprm/, February 2015, accessed November 25, 2015.

[38] Qualcomm Technologies, Inc., “LTE unmanned aircraft systems,” Tech. Rep. 1.0.1, May 2017. [Online]. Available:https://www.qualcomm.com/documents/lte-unmanned-aircraft-systems-trial-report/

[39] J. Blanch, T. Walter, P. Enge, Y. Lee, B. Pervan, M. Rippl, and A. Spletter, “Advanced RAIM user algorithm description:Integrity support message processing, fault detection, exclusion, and protection level calculation,” inProceedings ofION GNSS Conference, September 2012, pp. 2828–2849.

[40] T. Walter and J. Blanch, “Characterization of GNSS clock and ephemeris errors to support ARAIM,” inProceedingsof ION GNSS Conference, April 2015, pp. 920–931.

[41] D. Imparato, “GNSS-based receiver autonomous integrity monitoring for aircraft navigation,” Ph.D. dissertation, DelftUniversity of Technology, Netherlands, 2016.

[42] M. Maaref and Z. Kassas, “Ground vehicle navigation in GNSS-challenged environments using signals of opportunityand a closed-loop map-matching approach,”IEEE Transactions on Intelligent Transportation Systems, pp. 1–16, June2019.