Mitigating Jamming and Meaconing Attacks Using Direct GPS Positioning

Transcription

1 1 Mtgatng Jammng and Meaconng Attacks Usng Drect GPS Postonng Yutng Ng, Student Member, IEEE and Grace Xngxn Gao, Senor Member, IEEE BIOGRAPHIES Yutng Ng s a graduate student n the Aerospace Engneerng Department at the Unversty of Illnos at Urbana- Champagn. She receved her B.S. degree n electrcal engneerng, graduatng wth unversty honors, from the same unversty. Her research nterests are n advanced sgnal trackng algorthms, navgaton, control, RADAR and UAVs. Grace Xngxn Gao receved her B.S. degree n mechancal engneerng and her M.S. degree n electrcal engneerng from Tsnghua Unversty, Bejng, Chna n 2001 and She receved her PhD degree n electrcal engneerng from Stanford Unversty n From 2008 to 2012, she was a research assocate at Stanford Unversty. Snce 2012, she has been an assstant professor n the Aerospace Engneerng Department at Unversty of Illnos at Urbana-Champagn. Her research nterests are systems, sgnals, control, and robotcs. She s a senor member of IEEE and a member of ION. Abstract Drect GPS Postonng (DP) s a robust method that estmates the GPS navgaton soluton drectly from the receved GPS sgnal. In contrast, tradtonal methods, such as scalar trackng and vector trackng, compute ntermedate channel range and range resdual measurements ndependently, then use these as nputs nto the navgaton flter to estmate the navgaton soluton. However, a brute force mplementaton of DP s not computatonally effcent. Our research contrbuton s to evaluate the robustness of DP usng our more computatonally effcent DP mplementaton. Our novel and effectve DP recever archtecture ntalzes the navgaton soluton guesses as two groups: guesses varyng n poston and clock bas, guesses varyng n velocty and clock drft. We then perform vectorzed calculatons to get the expected delay and doppler between the recever and each satellte n vew. Followng that, we perform batch calculatons usng Fast Fourer Transforms (FFTs) to obtan the vector correlaton and vector spectrum. The navgaton soluton resdual derved from the vector correlaton and vector spectrum s then used as nput nto the navgaton flter. We mplemented our recever archtecture usng our research platform - PyGNSS. We then evaluated the robustness of our DP recever archtecture by subjectng t to smulated jammng and meaconng attacks. We demonstrate through our experments the robustness of our DP recever archtecture. Yutng Ng and Grace Xngxn Gao are wth the Department of Aerospace Engneerng at the Unversty of Illnos at Urbana-Champagn, Urbana, IL 61801, USA. I. INTRODUCTION Drect GPS Postonng (DP) drectly estmates and tracks the navgaton soluton from the receved raw GPS sgnal [1], [2]. Ths s unlke tradtonal scalar trackng whch estmates channel range and range rate measurements ndependently, then solves for the navgaton soluton va trlateraton [3] [7]. Smlarly, tradtonal vector trackng also produces the channel range and range rate resduals ndependently [8] [11]. Channel measurements estmated usng tradtonal methods are degraded under sgnal multpath and obstructon [12], [13], leadng to errors n the navgaton soluton. On the other hand, the shape of the vector correlaton n DP s robust to sgnal multpath and obstructon, leadng to a more robust navgaton soluton [14], [15]. In our pror work, we have mplemented a dfferent DP recever archtecture from the DP recever archtecture presented n ths paper. That archtecture uses coherent vector correlator measurements wth the Unscented Kalman Flter (UKF). Scalar trackng provded the carrer phase measurements to enable the coherent vector correlator. Lve experments were conducted and the results were analyzed for both statc and dynamc scenaros. In addton, we have provded a theoretcal proof that drect acquston s more robust than scalar acquston under meaconng scenaros. We also showed that drect postonng s more robust to jammng than scalar postonng usng smulated nose [1]. Work by other researchers can be grouped nto two categores: drect acquston and trackng. Drect acquston, also known as collectve detecton can be further separated nto coherent and noncoherent drect acquston [2], [14] [18]. Coherent drect acquston s too costly [2]. Noncoherent drect acquston has lower gan than coherent drect acquston. However, the noncoherent method nvolved less computaton and was shown to have superor detecton to scalar acquston [14], [15]. MATLAB mplementatons of drect acquston was provded n [16], [17]. Smulatons were used to compare the performance of drect acquston aganst scalar acquston by [19], [20]. A mathematcal treatment was provded by [21] [23]. Wth regards to trackng, a trackng loop for the vector correlaton was devsed by [24], based on pror work [25]. Also basng ther work on a trackng loop was [26], [27]. Many of the above mplementatons were computatonally ntensve, dd not estmate and track the full navgaton parameters and dd not provde a dedcated evaluaton of DP s robustness. Therefore, usng our computatonally effcent mplementaton of DP, we provde a dedcated evaluaton of DP s robustness through smulated jammng and meaconng attacks.

2 2 The rest of the paper s organzed as follows. Secton II of the paper descrbes our computatonally effcent DP recever archtecture used n ths paper, wth focus on obtanng the vector correlator and vector spectrum measurements, followed by subsequent trackng usng a navgaton flter. Secton III of the paper dscusses our mplementaton and experments based on feld data. We subjected our DP recever to smulated jammng and meaconng through post-processng usng our research platform - PyGNSS. We then compare the robustness of DP aganst scalar trackng, n Secton IV. Fnally, Secton V summarzes the paper. II. DIRECT GPS POSITIONING DP estmates and tracks the underlyng sgnal and navgaton state parameter set of recever poston, clock bas, velocty and clock drft drectly from the receved raw GPS sgnal. In other words, we are searchng for the underlyng recever parameter set X that most lkely generated the observed receved GPS sgnal Y. X : state vector of GPS recever (1) = x, y, z, c t, ẋ, ẏ, ż,c t T X : state vector of th satellte x, y, z, c t : 3D poston coordnates, clock bas (m) ẋ, ẏ, ż,c t : 3D velocty coordnates, clock drft (ms 1 ) c : speed of lght, (ms 1 ) Y : model of receved GPS sgnal (2) D (t) G (t) = X D (t)g (f codet + code)e j2 (f carr t+ carr ) : databt sequence of the th satellte : L1 C/A code sequence of the th satellte f code : code frequency of the th satellte (3) = f C/A + f C/A fdcarr f L1 code : code phase of the th satellte (4) = f C/A ( X x,y,z,eci Xx,y,z,ECI c +(X c t Xc t)) fcarr : carrer frequency of the th satellte (5) f IF = f IF + fdcarr : ntermedate frequency (IF), (Hz) f dcarr : carrer doppler frequency of the th satellte (6) = f L1 ( los x,y,z (Xẋ,ẏ,ż,ECI c Xẋ,ẏ,ż,ECI)+(X c t X )) c t los x,y,z : lne of sght vector n ECI coordnates (7) = X x,y,z,eci X x,y,z,eci X x,y,z,eci Xx,y,z,ECI carr : carrer phase of the th satellte f C/A : frequency of C/A code, (MHz) f L1 : frequency of L1 carrer, (MHz) The receved GPS sgnal Y s a superposton of sgnals Y receved from satelltes n vew (2). Representng each satellte as a channel, the receved GPS sgnal of each channel s governed by the channel delay and doppler parameters ( code,f carr). As shown, the channel parameters are a functon of the unknown recever state X and the known satellte states X through expected delay and doppler calculatons (3-7). Thus, the receved GPS sgnal Y can also be expressed as a functon of the unknown recever state X. The objectve s then to search for the best estmate of the recever state X that most lkely led to the receved sgnal Y. A. Effcent Search for Best Guess Navgaton Soluton To speed up the search process, we frst splt the underlyng parameter set nto two subsets: X xyz,c t : 3D poston and clock bas parameters (8) = x, y, z, c t T Xẋẏż,c t : 3D velocty and clock drft parameters (9) = ẋ, ẏ, ż,c t T The parameters wthn each subset are nterdependent. The parameters across subsets are frst order related n tme and thus can be ndependently estmated at an nstant n tme. We search for the 3D poston and clock bas resdual gven the current 3D velocty and clock drft. We also search for the 3D velocty and clock drft resdual gven the current 3D poston and clock bas. Ths was respectvely acheved through carrer and code wpeoff of the receved GPS sgnal on a per channel bass. In addton, databt wpeoff, nvolvng the removal of 180 degrees carrer phase flps due to the navgaton data bts, was performed usng our research platform - PyGNSS. Y carr wpeoff = Y code only G (f codet + code) (10) Y code wpeoff = Y carr only e j2 (f carr t+ carr ) (11) After performng carrer and code wpeoff on the receved GPS sgnal, estmaton of the X xyz,c t and Xẋẏż,c t parameter subsets then occur ndependently and n parallel, drastcally reducng computatonal complexty and tme. Estmaton of the X xyz,c t parameter subset was acheved through performng one correlaton per channel. The code phase of the sgnal replca, whch the receved sgnal was correlated aganst, was frst shfted by an amount determned by the current 3D poston and clock bas parameters. See equaton (4) for the relatonshp between the code phase and the 3D poston and clock bas parameters. Ths shft allows us to drectly estmate the 3D poston and clock bas resdual X xyz,c t. Estmaton of the Xẋẏż,c t parameter subset was acheved through performng one fourer transform per channel. The carrer frequency of the receved sgnal was frst shfted

3 3 Fg. 1. Plot of normalzed correlaton ampltude aganst resdual code phase shft (left); magntude spectrum aganst resdual carrer frequency shft (rght). by an amount determned by the current 3D velocty and clock drft parameters. See equaton (6) for the relatonshp between the carrer frequency and the 3D velocty and clock drft parameters. Ths shft allows us to drectly estmate the velocty and clock drft resdual Xẋẏż,c t. For each channel, the normalzed correlaton magntude and frequency spectrum s shown n Fg.1. The trangular correlaton peak and snc spectrum peak would be shfted proportonal to X xyz,c t and Xẋẏż,c t resduals. The wdth of the trangular correlaton s related to the code chp duraton whle the wdth of the snc manlobe s related to the coherent processng nterval. The mathematcal expressons descrbng the correlaton and spectrum peak s gven as follows: R Y code only 8 < (1 + ( )), 1 < ( ) apple 0 (1 ( )), 0 < ( ) apple 1 (12) : 0, otherwse F Y carr only T cohsnc( (f f)t coh ) (13) = sn( (f f)t coh) (f f) : resdual code phase shft due to X xyz,c t f : resdual carrer frequency shft due to Xẋẏż,c t T coh : coherent ntegraton perod, 0.020s To reterate, the correlatons help search through the frst parameter set, (x, y, z, c t), whle the frequency spectrums help search through the second parameter set, (ẋ, ẏ, ż,c ṫ). As such, carrer wpeoff s performed before the correlatons and code wpeoff s performed before the Fourer transform. Our DP mplementaton n ths paper utlzed a coherent ntegraton perod of 20ms. The second step s generatng the canddate recever state error vectors, wth two sets, fxed velocty and fxed poston. The thrd step s assgnng correlaton ampltudes and spectrum magntudes from the correlatons and spectrums to the respectve canddate recever state error vectors. The noncoherent sum of the correlaton ampltudes and spectrum magntudes across each channel forms the vector correlaton and vector spectrum. The fourth step estmates the error nput vector e as the weghted mean of the canddate recever state error vectors. The weghts are determned from the vector correlaton and vector spectrum. In ths effectve and effcent manner, the nput to the jont trackng and navgaton flter s generated. Fg. 2. DP process flow, obtanng the error nput vector e and error nose covarance matrx W. The number of 3D poston and clock bas canddates can dffer from the number of 3D velocty and clock drft canddates. In the fgures above, they are labeled as M and N wth ndexes j and k respectvely. B. Estmaton and Flterng The jont trackng and navgaton flter used n our effectve and computatonally effcent mplementaton of DP s a Kalman Flter (KF) [28], [29]. Snce we are obtanng drect measurements of the state parameters, the geometry matrx s an dentty matrx and the equatons can be drastcally smplfed.

4 4 The KF measurement update at epoch k: e : error nput vector (14) = e( X x,y,z,c t, Xẋ,ẏ,ż,c t) =[ x, y, z, c t, ẋ, ẏ, ż, c t] T W : error nose covarance nput matrx (15) = W (W Xx,y,z,c t,w Xẋ,ẏ,ż,c t) K : Kalman gan matrx (16) = ˆ k H T (H ˆ k H T + W ) 1 X : state error vector (17) = Ke ˆ k : predcted state error covarance matrx H : geometry matrx (18) = I 8, 8x8 dentty matrx X k : corrected state vector (19) = ˆX k + X k : corrected state error covarance matrx (20) =(I KH)ˆ k The error nose covarance nput matrces are estmated usng 20 past error nput vectors. The KF update nterval T s set to be the same as the coherent ntegraton nterval T coh. Havng T = T coh removes the need for measurement nterpolaton. Usng our flexble research platform - PyGNSS, we are not lmted to a T coh of 20 mllseconds, as shown n our work on jont GPS and vson drect postonng [30]. A T coh of 20 mllseconds was selected for ths paper as t allows us to convenently compare our results aganst lterature. A T coh of 20 mllseconds was also used n our work on Drect Tme Estmaton [31]. We performed navgaton bt wpe-off such that the KF update nterval and coherent ntegraton perod s synchronous across channels and can straddle navgaton bt boundares. The KF tme update equatons at epoch k +1: ˆX k+1 : predcted state vector (21) = FX k ˆ k+1 : predcted state error covarance matrx (22) = F k F T + Q k F : state propagaton matrx (23) = F ( T ) T T T 0 = T T : update nterval, (s) (24) Q k : state process nose covarance matrx (25) apple 0 0 = F F T 0 Q v,k Q v,k : velocty and clock drft component of Q k (26) 2 3 f(v k ) = 6 0 f(v k ) f(v k ) (c ) 2 : allan devaton of the frontend oscllator, (s) f(v k ) : saturaton functon for velocty v k (27) = /(mn(max(v 2 k, 5 2 ), 25 2 )) The ntal predcted state error covarance matrx ˆ 0 s ntalzed usng 20 past state vectors and should be smlar n order of magntude to the ntal state process nose covarance matrx Q 0. In addton, snce we are usng a constant acceleraton model, for dynamc scenaros such as that on a road vehcle, vehcle acceleratons are addressed through a tmevaryng state process nose covarance matrx, Q k [32], [33], set usng a saturaton functon f(v k ) on a 20 sample runnng average ampltude of the recever s velocty v = ẋ, ẏ, ż. The coeffcents of the saturaton functon were determned from a least squares ft to acceleraton tme data of a generc vehcle by [33]. The allan devaton of the frontend oscllator can be obtaned through the manufactorer s datasheet [34]. III. IMPLEMENTATION AND EXPERIMENT SETUP Ths paper evaluates the robustness of DP under stressful condtons. We do so usng smulated jammng and meaconng nput sgnals and our research platform - PyGNSS. We generated the nput sgnals n the followng manner: 1) A clean sgnal s frst collected under an deal, opensky envronment, jammng and meaconng s absent. 2) The clean sgnal s then processed to smulate two sgnfcant forms of attacks, jammng and meaconng. The methods used to smulate these attacks are dscussed below. In addton, n order to provde a deeper analyss, the severty of the attacks s adjusted by changng the ampltude of the attackng sgnal. 3) The attacked sgnal s then fed nto both DP and scalar trackng recever archtectures. A comparson of the trackng results s then provded. We smulated the followng two attacks on the clean sgnal: Jammng: A random complex voltage V = A cos + jasn s added to each sample, where A N (0, A) and U(0, 2 ). In other words, a whte nose sgnal wth Gaussan dstrbuted ampltude and Unform dstrbuted phase s superposed onto the orgnal clean sgnal to smulate the jammng effect. In addton, by ncreasng the standard devaton of the ampltude A, we are effectvely smulatng a stronger jammng source. Meaconng: Another sgnal collected at a separate locaton and/or a separate tme, known as the meaconng sgnal s superposed on the clean sgnal. The meaconng sgnal attempts to deceve the recever nto an ncorrect navgaton soluton by coverng the orgnal receved

5 5 Fg. 3. DP remans robust whle scalar trackng s degraded by jammng. The dfferent jammng scenaros are color-coded and superposed on the same plot. Wthout loss of generalty, shown are the estmated code and carrer frequences for one of the satelltes n vew. sgnal. In our smulaton, the meaconng sgnal s collected from a recever locaton 1.25 mles away from the orgnal sgnal. IV. R ESULTS AND A NALYSIS Durng a regular trackng scenaro where there s nether meaconng nor jammng, both DP and scalar trackng recever archtectures easly track both code and carrer frequences, fc and f, successfully. We then evaluate how jammng and meaconng attacks mpact the performance of DP as well as scalar trackng. 1) Jammng: As the nose floor s rased by ncreasng the jammng ampltude through the parameter A, DP remans robust whereas the trackng performance of scalar trackng s degraded. Ths s shown n Fg.3 where the results from dfferent jammng scenaros are color-coded and superposed. 2) Meaconng: A meaconng sgnal from 1.25 mles away s unable to sgnfcantly deceve the DP recever that s already trackng onto the authentc sgnal. In order to deceve the DP recever, the meaconng sgnal has to frst fall wthn the man lobe then have smlar process parameters. As such, t s expected for the DP recever to reman robust to the above mentoned meaconng sgnal. The meaconng sgnal then acts lke a jammng sgnal. Results are shown n Fg.4. V. C ONCLUSION Drect Postonng (DP) has the potental to outperform tradtonal methods, such as scalar trackng and vector trackng, n terms of robustness. Ths s because t utlzes nformaton from the entre raw receved sgnal to generate drect measurements of the navgaton soluton. However, t s not wdely mplemented as prevous mplementatons of DP are not computatonally effcent. To shake up the statusquo, we proposed and mplemented a novel, effectve and computatonally effcent DP recever archtecture and went on to provde a dedcated evaluaton of ts robustness. We subjected our DP recever archtecture to smulated jammng and meaconng attacks. We dd so by collectng GPS Fg. 4. DP remans robust whle scalar trackng s degraded by meaconng. The dfferent meaconng strengths of no meaconng, 5:1 meaconng and 10:1 meaconng are color-coded and superposed on the same plot. Wthout loss of generalty, shown are the estmated code and carrer frequences for one of the satelltes n vew. sgnals n bengn envronments then post-processng them nto attacked sgnals usng our research platform - PyGNSS. We nput the attacked sgnals nto our DP recever archtecture, benchmarkng the trackng results aganst scalar trackng. We demonstrated wth our DP recever archtecture, DP s robustness, as compared to scalar trackng, wth respect to jammng and meaconng attacks. ACKNOWLEDGMENT Specal thanks to Ganshun Lm for hs contnuous support, Arthur Hs-Png Chu for runnng the jammng and meaconng smulatons, Sr Ramya Bhamdpat for kndly proofreadng the paper. R EFERENCES [1] A. Ramesh Kumar, Drect poston trackng n GPS usng vector correlator, Master s thess, Unversty of Illnos at Urbana-Champagn, [2] R. DEspost, GPS PRN code sgnal processng and recever desgn for smultaneous all-n-vew coherent sgnal acquston and navgaton soluton determnaton, n Proceedngs of the 2007 Natonal Techncal Meetng of The Insttute of Navgaton, 2001, pp [3] E. Kaplan, C. Hegarty, P. Ward, and J. Betz, Understandng GPS: Prncples and applcatons 2nd ed. Boston: Artech House, 2006, ch. Chapter 5: Satellte Sgnal Acquston, Trackng, and Data Demodulaton, pp [4] P. Msra and P. Enge, Global Postonng System: Sgnals, Measurements and Performance Second Edton. Lncoln, MA: Ganga-Jamuna Press, [5] P. Ward, The natural measurements of a GPS recever, n Proceedngs of the 51st Annual Meetng of the Insttute of Navgaton, 1995, pp [6] B. W. Parknson, Progress n Astronautcs and Aeronautcs: Global Postonng System: Theory and Applcatons. Aaa, 1996, vol. 2. [7] M. S. Braasch and A. Van Derendonck, GPS recever archtectures and measurements, Proceedngs of the IEEE, vol. 87, no. 1, pp , [8] Y. Ng and G. X. Gao, Advanced mult-recever vector trackng for postonng a land vehcle, n Proceedngs of the 26th Internatonal Techncal Meetng of the Satellte Dvson of the Insttute of Navgaton (ION GNSS+ 2014), Tampa, Florda, [9] S. Bhattacharyya, Performance and ntegrty analyss of the vector trackng archtecture of GNSS recevers, Ph.D. dssertaton, Unversty of Mnnesota, 2012.

6 6 [10] J. J. Brewer, The dfferental vector phase-locked loop for global navgaton satellte system sgnal trackng, Ph.D. dssertaton, Ar Force Insttute of Technology, [11] J. J. Splker Jr, Vector delay lock loop processng of radolocaton transmtter sgnals, Mar , us Patent 5,398,034. [12] P. W. Ward, Performance comparsons between FLL, PLL and a novel FLL-asssted-PLL carrer trackng loop under RF nterference condtons, n Proceedngs of the 11th Internatonal Techncal Meetng of the Satellte Dvson of The Insttute of Navgaton (ION GPS 1998), 1998, pp [13] P. W. Ward and T. D. Fuchser, Stablty crtera for GNSS recever trackng loops, Navgaton, vol. 61, no. 4, pp , [14] P. Axelrad, B. K. Bradley, J. Donna, M. Mtchell, and S. Mohuddn, Collectve detecton and drect postonng usng multple GNSS satelltes, Navgaton, vol. 58, no. 4, pp , [15] P. Axelrad, J. Donna, and M. Mtchell, Enhancng GNSS acquston by combnng sgnals from multple channels and satelltes, n Proceedngs of the 22nd Internatonal Techncal Meetng of The Satellte Dvson of the Insttute of Navgaton (ION GNSS 2009), 2001, pp [16] J. W. Cheong, J. Wu, A. G. Dempster, and C. Rzos, Effcent mplementaton of collectve detecton, n IGNSS symposum, 2011, pp [17], Asssted-GPS based snap-shot GPS recever wth FFTaccelerated collectve detecton: Tme synchronsaton and search space analyss, n 25th Internatonal Techncal Meetng of the satellte Dvson of the Insttute of Navgaton (ION GNSS 2012), Nashvlle USA. Cteseer, 2012, pp [18] L. R. Paulo Esteves, Mohamed Sahmoud, Collectve detecton of multgnss sgnals, InsdeGNSS, pp , may [19] O. Baler, D. Raphael, and A. J. Wess, Maxmum-lkelhood drect poston estmaton n dense multpath, Vehcular Technology, IEEE Transactons on, vol. 62, no. 5, pp , [20] A. J. Wess, Drect poston determnaton of narrowband rado frequency transmtters, Sgnal Processng Letters, IEEE, vol. 11, no. 5, pp , [21] P. Closas, C. Fernández-Prades, J. Fernández-Rubo et al., Maxmum lkelhood estmaton of poston n GNSS, Sgnal Processng Letters, IEEE, vol. 14, no. 5, pp , [22] P. Closas, C. Fernandez-Prades, J. Fernández-Rubo et al., Cramér rao bound analyss of postonng approaches n GNSS recevers, Sgnal Processng, IEEE Transactons on, vol. 57, no. 10, pp , [23], A bayesan approach to multpath mtgaton n GNSS recevers, Selected Topcs n Sgnal Processng, IEEE Journal of, vol. 3, no. 4, pp , [24] J. Lu, X. Cu, M. Lu, and Z. Feng, Drect poston trackng loop based on lnearsed sgnal model for global navgaton satellte system recevers, IET Radar, Sonar & Navgaton, vol. 7, no. 7, pp , [25], Vector trackng loops n GNSS recevers for dynamc weak sgnals, Systems Engneerng and Electroncs, Journal of, vol. 24, no. 3, pp , [26] F. D. Nunes, J. Marcal, and F. M. Sousa, Low-complexty VDLL recever for mult-gnss constellatons, n Satellte Navgaton Technologes and European Workshop on GNSS Sgnals and Sgnal Processng (NAVITEC), th ESA Workshop on. IEEE, 2010, pp [27] T. Ln, J. T. Curran, C. ODrscoll, and G. Lachapelle, Implementaton of a navgaton doman GNSS sgnal trackng loop, n Proceedngs of ION GNSS, 2011, pp [28] R. E. Kalman, A new approach to lnear flterng and predcton problems, Journal of Fluds Engneerng, vol. 82, no. 1, pp , [29] R. E. Kalman et al., Contrbutons to the theory of optmal control, Bol. Soc. Mat. Mexcana, vol. 5, no. 2, pp , [30] Y. Ng and G. X. Gao, Jont GPS and vson drect poston estmaton, n Proceedngs of the 2016 IEEEION Poston Locaton and Navgaton Symposum (IEEEION PLANS 2016), Georga, Savannah, [31], Robust GPS-based drect tme estmaton for PMUs, n Proceedngs of the 2016 IEEEION Poston Locaton and Navgaton Symposum (IEEEION PLANS 2016), Georga, Savannah, [32] F. D. Busse, J. P. How, and J. Smpson, Demonstraton of adaptve extended Kalman flter for low-earth-orbt formaton estmaton usng CDGPS, Navgaton, vol. 50, no. 2, pp , [33] Mathworks. (2015, jun) State estmaton usng tme-varyng kalman flter. [Onlne]. Avalable: [34] Mcrosem. (2015, jun) Quantum chp scale atomc clock product nfo. [Onlne]. Avalable: