Performance of Real-time Precise Point Positioning in New Zealand

Transcription

1 Performance of Real-time Precie Point Poitioning in New Zealand Thi i a Peer Reviewed Paper Ken HARIMA, Suelynn CHOY and Chri RIZOS, Autralia Kogure SATOSHI, Japan Key word: GPS, GNSS, PPP, Real-Time SUMMARY Mot of the GNSS baed high accuracy poitioning ervice available today are baed on differential technique like RTK. Thee kind of ervice are only available inide a network of Continuouly Operating Reference Station (CORS) or a few kilometer from uch tation. An emerging alternative to thee infratructure heavy method i the Precie Point Poitioning (PPP) technique. PPP a a technique aim at delivering centimeter level accuracy poitioning without the need of directly connecting to a local CORS. Although PPP till need to monitor GNSS atellite and compute correction for their ignal, thee correction are calculated computed from pare global network. Recent reearch and development ha invetigated into atellite broadcat of correction for PPP. The combination of pare CORS and atellite delivery make PPP baed poitioning ideal for ituation where local infratructure i unavailable or unreliable. The preent document tudie the potential performance of PPP in New Zealand. PPP i now a quite mature reearch topic and a number of real-time correction for PPP are publicly available for teting. Three of thee tream are ued to calculate real-time kinematic PPP olution in Auckland, Hating, Wellington, Yaldhurt (Chritchurch), Dunedin and Chatham Iland. The elected realtime tream have been teted or are likely to be available a, atellite delivery ervice. MADOCA meage produced by the Japanee Aeropace Exploration Agency have been in tet uing the Quai-Zenith Satellite Sytem ince CLK81 meage treamed from the IGS real-time erver are computed by atellite operator GMV uing their MagicGNSS oftware. Finally the CLK9B meage calculated by the French Space Agency (CNES) wa broadcat by the QZSS in tet performed in 2014 and Reult of real-time poitioning tet how that accuracie of 6 cm in horizontal poitioning and 15 cm in vertical poitioning can be achieved by kinematic PPP. The convergence time to 10 cm of horizontal accuracy i about 30 minute.

2 Performance of Real-time Precie Point Poitioning in New Zealand Ken HARIMA, Suelynn CHOY and Chri RIZOS, Autralia Kogure SATOSHI, Japan 1. INTRODUCTION Global Navigation Satellite Sytem (GNSS) Precie Point Poitioning (PPP) technique ha been an active reearch topic for nearly 20 year. Thi mode of poitioning conform to the original intention of Global Poitioning Sytem (GPS) or GNSS uage, that i, poition reolution with a ingle receiver independent of location or available local infratructure. Mot of the currently etablihed high accuracy poitioning ervice are baed on differential poitioning technique like Real-Time Kinematic (RTK). Thee ervice account for ytematic biae by comparing the obervation on the uer receiver with thoe obtained in a nearby reference tation. Service like RTK are only available inide a network of Continuouly Operating Reference Station (CORS) or a few kilometer from uch tation. PPP on the other hand, ue a robut modelling of GNSS error and precie information on atellite orbit, atellite clock and ignal biae to make accurate etimate of the uer poition (Kouba, 2009; Zumberge et al., 1997). Unlike differential approache PPP only make ue of information that i valid globally or over wide area, and thu only require a pare network of monitoring tation. Pot-proceed poitioning ervice baed on PPP are now mature and are available from multiple online ource, offering accuracie of 2-3 cm (Grinter & Janen, 2012). More recent tudie are focued on real-time delivery of PPP ervice and ambiguity reolution for PPP (Ge et al., 2007, Laurichee et. al 2009). A public Real-Time Service (RTS) to upport real-time high preciion GNSS application allow the teting of real-time PPP (IGS, 2015). The caveat of real-time PPP are the availability of low latency precie atellite orbit and clock information, and communication channel ued to dieminate thi information. From a practical viewpoint atellite-baed broadcating i an ideal communication link for PPP enabling the trategic advantage of PPP, namely it wide coverage area. A combination of PPP baed poition and atellite baed delivery will be completely independent of local infratructure, making it ideal for cae where local infratructure i either unavailable or unreliable. Thi include ocean, remote area but alo can be conceived in the aftermath of natural diater. In thi tudy, an analyi of the quality of the atellite orbit and clock correction available from real time tream i preented along with an evaluation of the performance of PPP algorithm uing thee real-time tream. Three real-time tream were elected for the tudy, all of them publicly available and all of them teted with atellite tranmiion. The MDC1 tream i generated by the Japan Aeropace Exploration Agency (JAXA) uing the Multi-GNSS Advanced Demontration tool for Orbit and Clock Analyi (MADOCA) oftware (JAXA, 2014). The CLK81 tream i generated by GMV uing the MagicGNSS oftware and broadcated uing IGS real-time ervice. The CLK9B

3 tream i generated by the French Space Agency (CNES) a part of their PPP-Wizard project. The elected real-time tream have been teted or are likely to be available a atellite delivered ervice. MADOCA meage produced by the Japanee Aeropace Exploration Agency have been broadcated by the Quai-Zenith Satellite Sytem (QZSS) ince CLK81 meage treamed from the IGS real-time erver are produced by atellite operator GMV. Finally CLK9B produced by the French Space Agency (CNES) wa broadcat by the QZSS in tet performed in 2014 and 2015 (Harima et. al., 2015) For thi tudy, orbit and clock correction taken from the real-time tream are combined with obervation from CORS in New Zealand to generate PPP olution. Dual-frequency obervable correponding to ignal from GPS and GLONASS atellite are ued to calculate the PPP olution. The olution are calculated uing the RTKLIB oftware (Takau and Yauda, 2009), an open ource oftware for high accuracy poitioning. Thi paper i tructured in five ection. A brief decription of the PPP algorithm i given in Section 2. An evaluation of the accuracy of real-time tream available for PPP i preented in Section 3. The performance of real-time PPP uing the real time tream i tudied in Section 4. A ummary of the finding i given in Section PRECISE POINT POSITIONING In differential approache (like RTK) error on the GNSS obervation are mitigated by differencing with meaurement taken in a nearby tation and ued directly to correct for error in meaurement at the uer receiver. In PPP by contrat, error are modeled a a combination of error with different origin. Thi allow the eparation of GNSS error between thoe with global validity, like atellite clock and orbit error, and local error uch a atmopheric effect and receiver biae. Global error can be etimated uing a pare network and ent to the uer for correction. Local error are either etimated along with the receiver poition or compenated by mean of linear combination. Numerou verion of thi core concept have been developed over the pat year. Thi ection decribe the PPP algorithm ued to calculate the poitioning olution preented in ection 4. Thee algorithm, implemented in the RTKLIB oftware, are baed on (Zumberge et al, 1997) for tandard PPP and (Ge et al, 2007) for ambiguity reolved PPP. 2.1 Standard PPP algorithm Standard PPP algorithm ue ionophere-free (IF) linear combination of meaurement from two different carrier frequencie to eliminate the effect of the ionopheric delay, which i the larget error with local variability: P IF L IF = ρ + dt dt + c 1 (B P1 B P1 = ρ + dt dt + c 1 (λ 1 N L1 ) + c 2 (B P2 B P2 ) (1) + B L1 B L1 ) + c 2 (λ 2 N L2 + B L2 B L2 ) (2)

4 where P IF i the ionophere-free combination of P 1 and P 2 code meaurement, L IF i the ionophere-free combination of the L 1 and L 2 carrier meaurement corrected for antenna phae offet and phae windup effect. c 1 and c 2 are the ionophere free linear combination coefficient, λ 1 and λ 2 the wavelength of L 1 and L 2 carrier. ρ i a range etimate, which include the atellitereceiver range and tropopheric delay, dt and dt are the receiver and atellite clock offet. B P1 and B P1 are the receiver and atellite bia for the P 1 code meaurement, B P2 and B P2 are the biae for P 2. B L1 and B L1 are the biae for L 1 phae meaurement, B L2 and B L2 are the biae for L 2. N L1 and N L2 are the integer ambiguitie. For tandard PPP, the required global information are the precie atellite orbit, which are ued to aociate ρ with the receiver poition, and the ionophere-free clock dt IF = dt + c 1 B P1 + c 2 B P2. (3) Ionophere-free linear combination of code and carrier meaurement corrected for the ionopherefree atellite clock, antenna phae offet and phae windup effect, are ued to etimate the receiver poition, the zenith tropopheric delay, the receiver ionophere-free clock, and a float ambiguity from the modified meaurement model B FLT = c 1 (λ 1 N L1 + B L1 B L1 B P1 + B P1 ) + B L2 B L2 B P2 + B P2 ) (4) +c 2 (λ 2 N L2 P IF L IF + dt IF + dt IF = ρ + dt IF (5) = ρ + dt IF + B FLT (6) An extended Kalman Filter i ued to etimate the parameter from the meaurement. The receiver clock i modelled a a white noie proce with the reult of a tandard, code baed olution a the mean. The zenith tropopheric delay i modelled a a random walk proce. The float biae are modelled a a contant value but monitored for cycle lip uing geometry-free carrier phae meaurement. Finally the receiver poition i modelled a a white noie proce with the reult of tandard code baed poitioning a the mean. 2.2 Ambiguity reolution in PPP While conolidating the integer ambiguitie and hardware biae into one floating ambiguity allow the reolution of poition without the tranmiion of atellite ignal biae. Mot of the recently propoed PPP algorithm ugget the tranmiion of thee individual biae to attempt integer ambiguity reolution. Ambiguity reolution take advantage of the fact that the phae ambiguitie N i are integer valued. By olving and fixing the value of N i to integer value, error aociated with thee parameter can be eliminated. Thi in theory will allow the PPP olution to approach RTK in accuracy. Several trategie have been propoed for ambiguity reolution in PPP. The method ued in the

5 preent tudy i baed on (Ge et al, 2007): - Meaurement are corrected for ignal biae - The tandard PPP olution i calculated uing corrected meaurement. - The Melbourne-Wübenna combination P MW = λ 2 L λ2 λ1 1 λ 1 L λ2 λ1 2 λ 2 P λ2+λ1 1 λ 1 P λ2+λ1 2 (7) i ued to eparate integer ambiguitie from other element in B FLT. - The integer ambiguitie are reolved and ued to calculate a corrected value for B FLT - The corrected value for B FLT i ued a a meaurement to correct the PPP olution. The ingle-differenced Melbourne-Wübenna combination of L 1 and L 2 ignal, corrected for atellite biae, can be expreed a:. P MW r P MW = λ 1λ2 (N λ2 λ1 L1 N L2 N r L1 + N r L2 ), (8) here the upercript r repreent the reference atellite. Thi Melbourne-Wübenna combination can be ued to olve the integer ambiguity combination N L1 N L2 N r L1 + N r L2. Once the ambiguity combination i olved, the ingle-differenced N L1 ambiguity can be iolated from B FLT, corrected for atellite biae. B FLT r B FLT c 2 λ 2 N L1 N L2 N r L1 + N r L2 = λ 1λ2 (N λ2+λ1 L1 N r L1 ) (9) here N X i ued to expre the reolved verion of the ambiguity N X. Once the ambiguitie are reolved, a corrected ingle differenced value of B FLT B FIX r B FIX = λ 1λ2 N λ2+λ1 L1 N r L1 + λ 2 N L1 N L2 N r L1 + N r L2 (10) i calculated and ued a a meaurement to complement the equation (5) and (6) to calculate a corrected poition. The ambiguity difference N L1 N L2 are reolved to integer value uing the integer rounding method. The N L1 ambiguity i reolved uing the modified LAMBDA method (Chang et al, 2009). 2.3 Correction meage for PPP A explained in previou ection, tandard PPP olution give information on precie atellite orbit and ionophere-free atellite clock. Ambiguity reolved PPP additionally require hardware bia etimate for the code and carrier of the GNSS ignal. While numerou real-time tream that tranmit orbit and clock for GPS exit, tream containing multi-gnss are till few. GLONASS i till experimental for IGS-RT tream. Alo for real-time proceing, the information for PPP need to be available with a low latency, a thee value change over time. The available real-time tream follow a protocol etablihed by the Radio Technical Commiion for Maritime Service (RTCM, 2013). Thee meage contain correction to orbit and clock information broadcat by the GNSS atellite. Thi allow aving in

6 the tranmiion bandwidth while till allowing the uer to compute precie orbit and clock. Table 1 how the update rate of the different type of correction tranmitted by the different real-time tream. Orbit Clock Other MDC1 30 ec 1 ec URA: 1 ec CLK81 5 ec 5 ec CLK9B 5 ec 5 ec GPS Biae: 5 ec Table 1. Content of Real-time meage for PPP The MDC1 tream update clock correction every econd, and orbit every 30 econd, in addition it tranmit an etimate of the uer range accuracy (URA) each econd. The other tream update orbit and clock every 5 econd. At the time of writing, the only publicly available tream known to the author that tranmit ignal biae for real-time PPP are tream from CNES PPP-wizard project. 3. REAL TIME STREAMS FOR PPP The quality of the precie orbit and clock correction included in the augmentation meage ha a direct impact on the achievable accuracy and convergence time of the PPP olution. In thi ection we dicu the quality of the orbit and clock correction included in the real-time tream. Orbit and clock calculated uing data from the real-time tream were compared with pot-proceed verion of the ame parameter. The pot proce olution ued a a reference the IGS rapid product for GPS and IGS final product for GLONASS, thee product have an etimated accuracy of 3 cm for orbit and 75p RMS and 25p STD (equivalent to 2.25cm and 0.75cm repectively) for clock (IGS, 2014). Table 2 ummarie the etimated quality of real-time correction a compared to IGS product. Value in the table are daily tandard deviation of range difference for different poition on earth. The range difference, Signal in Space Range Error (SISRE), can be calculated a follow: SISRE = X RT at Xta ref X at Xta dt RT + dt ref (11) SISRE repreent the effect of orbit and clock error will have in the meaurement equation (5) and (6).

7 Tokyo Bern Sydney Auckland Chath. I. Dunedin GPS SISRE for 5.03 cm 6.04 cm 4.90 cm 4.81 cm 4.77 cm 4.77 cm MDC1 GPS SISRE for cm cm cm cm cm cm CLK81 GPS SISRE 2.76 cm 9.41cm 2.93 cm 3.05 cm 2.95 cm 3.02 cm for CLK9B GLO SISRE for 8.81 cm cm cm cm cm cm MDC1 GLO SISRE for 8.85 cm cm cm 9.32 cm cm cm CLK81 GLO SISRE for CLK9B 8.43 cm cm 9.04 cm 7.90 cm 9.11 cm 9.73 cm Table 2. Signal in Space Range Error in Tokyo, Bern, Sydney and New Zealand. Alo a it will be hown in following ubection, a ignificant part of the difference between the real-time and reference product were large biae in the clock correction. Thee biae were between 10 and 30 cm for GPS and between 50cm and 1m for GLONASS, and contant throughout the day. Thee biae, while potentially affecting the convergence time of olution are not expected to have ignificant impact on poitioning olution in the teady tate. On one hand, a clock bia that i contant over the day (or one atellite pa) will be aborbed in equation (6) by the float ambiguity B FLT. On the other hand the error level on the code meaurement in equation 5 are in the order of meter. For thi reaon, a daily Standard deviation (STD) i ued in place of RMS a a meaure of the quality of real-time orbit and clock correction. With a few exception motly attributable to error in product for the G26 atellite, SISRE value are below 5 cm for GPS and below 11 cm for GLONASS. Alo with the exception of MDC1 product for GLONASS, the SISRE have imilar value around the world. The product tatitic on thi ection are baed on real-time product tranmitted on the 30 th of September MADOCA from JAXA (MDC1) Table 3 how the RMS and daily tandard deviation error for the MADOCA generated product. The orbit are expreed in atellite centric RAC coordinate. In thi coordinate ytem the Radial direction i approximately the direction to the center of the earth, the Along-track direction i the direction of the atellite velocity, and the cro track direction i perpendicular to the other two. The

8 error component with the mot influence in the SISRE, and conequently to poitioning accuracy, are the radial component and the clock. The orbit generated by MADOCA product have relatively large difference with repect to the IGS pot proce orbit. Thi i to be expected to ome extent a the MADOCA orbit are generated independently from the IGS. Thi i in contrat to the CLK9B tream which ue IGS ultra-rapid orbit and clock a a bae for their orbit. Radial Along Cro Clock RMS GPS 4.83 cm cm cm cm RMS GLO 5.09 cm cm 8.42 cm cm STD GPS 3.08 cm 6.49 cm cm 4.81 cm STD GLO 3.28 cm 7.76 cm 5.06 cm 9.47 cm Table 3. Error of MADOCA atellite orbit and clock etimate, 30 th of September The GPS clock were found to be a cloe if not cloer than other olution to the pot proce IGS clock. For GPS, the reultant SISRE i imilar to the one obtained uing the GMV (once the effect of atellite G26 are dicarded. Figure 1 how the RMS value of orbit error, radial error are hown in blue, along-track error are hown in green, cro-track error are hown in brown. There are atellite with large alongtrack and cro-track error, i.e. G13, G14, G15 and G22. The SISRE value correponding to thoe atellite are not ignificant for tet point in Sydney or New Zealand. However they do lead to relatively large SISRE error in Tokyo and Bern. In particular, the SISRE error for atellite G13 on Bern wa about 24.8 cm. However thee error are not expected to lead to ignificant performance degradation in PPP olution. A a lat note, the relatively large SISRE for GLONASS in Sydney and New Zealand are product of large clock error in atellite R04, which ha the equivalent of 45.5cm (1.5n) of clock error. If thi atellite i dicarded the SISRE have a maximum of cm. Value that are imilar to thoe of other real-time tream.

9 Figure 1. MADOCA orbit error for GPS (top) and GLONASS (bottom) atellite. Mot influential error are the error in the Radial direction (blue). Along-track (green) and Cro-track (brown) error have far le impact on perceived range error 3.2 Magic GNSS from GMV (CLK81) Table 4 how the RMS and daily tandard deviation error for the MagicGNSS generated product. Figure 2 how the RMS error of orbit calculated by the MagicGNSS oftware, whereby radial error are hown in blue, along-track error are hown in green, cro-track error are hown in brown. A Figure 2 clearly how, atellite G26 ha very large error in it calculated orbit and, although not hown in Figure 2, the clock correction. The orbit, clock error for thi atellite were above 70 cm for all component, and SISRE of up to 1.10 m. Thee error ignificantly affected the PPP performance preented in ection 4. Radial Along Cro Clock RMS GPS 9.08 cm cm cm cm RMS GLO 3.79 cm 9.83 cm 8.73 cm cm STD GPS 5.28 cm cm cm cm STD GLO 2.24 cm 5.11 cm 4.60 cm 7.41 cm Table 4. Error of MagicGNSS atellite orbit and clock etimate, 30 th of September If the effect of G26 i dicarded, daily STD for GPS orbit become 1.67cm for radial, 4.09 cm for along-track and 3.25cm for cro-track error with clock error equivalent to 7.77cm. In thi cae the orbit are ignificantly better than MADOCA and on par with PPP-Wizard. The clock error are till lightly larger than thoe of other ytem (driven by a 53cm clock error for G32) however the

10 over all, SISRE for Sydney and New Zealand are all below 4.52cm, which i better than MADOCA and comparable with PPP-Wizard. Figure 2. MagicGNSS orbit error for GPS (top) and GLONASS (bottom) atellite. Error in Radial, Along-track and Cro-track direction are hown in blue, green and brown. Large error can be een for atellite G26. Unfortunately problem with MagicGNSS ephemeri for atellite G26 were unknown at the time the PPP olution in ection 4 were calculated, o the error of atellite G26 affected the PPP olution. Thi i alo the reaon table 2 and 4 reflect the preence of G26. Had thi atellite error been detected and dicarded, performance imilar to thoe of other ytem could be expected. 3.3 PPP-Wizard from CNES (CLK9B) Table 5 how the RMS and daily tandard deviation error for the PPP-Wizard generated product. Figure 3 how the RMS error of orbit. Real-time orbit generated by the PPP-Wizard were the one cloet to the pot-proceed IGS product for both GPS and GLONASS. Thi in turn wa reflected in SISRE value ignificantly maller than the other olution, below 3.05 cm for GPS. Radial Along Cro Clock RMS GPS 3.11 cm 4.90 cm 4.90 cm cm RMS GLO 2.86 cm 8.32 cm 4.49 cm cm STD GPS 2.24 cm 3.95 cm 3.57 cm 5.19 cm STD GLO 1.76 cm 4.23 cm 2.40 cm 7.15 cm Table 5. Error of PPP-Wizard atellite orbit and clock etimate, 30 th of September The one exception for thi wa the 9.41cm SISRE on Bern. Here too, large error on the G26 atellite orbit and clock etimate influenced the reulting tatitic. If atellite G26 i dicarded, the SISRE value are all below 3.1 cm.

11 Figure 3. PPP-Wizard orbit error for GPS (top) and GLONASS (bottom) atellite. Error in Radial, Along-track and Cro-track direction are hown in blue, green and brown. It i alo to note that the PPP-Wizard tream alo tranmit phae biae deigned to allow PPP with ambiguity reolution and thu lead to preciion on the 2-3 cm level. Although the author have confirmed the benefit of ambiguity reolution in GPS only olution (Choy et. al., 2015), ambiguity fixing rate for the tet were below 30% in the tet on ection 4. It i believed thi i becaue tochatic parameter, pecifically the etimated noie of GLONASS obervable and ephemeri, were underetimated for the multi-gnss PPP olution. 4. PERFORMANCE OF REAL TIME PPP IN NEW ZEALAND Thi ection preent a tudy of the performance of PPP uing the real-time tream explained in ection 3. RTCM correction meage were received and decoded in real time. The orbit and clock correction information contained within were combined with the orbit and clock broadcat by GNSS atellite to obtain precie orbit and clock. The precie orbit and clock were ued in conjunction with obervable taken at variou CORS tation to obtain real-time PPP olution. Figure 4 how the tation ued in the tet and their location. PPP-Wizard orbit are baed on the IGS-ultra rapid product, which include the tation AUCK, WGTN, DUND, CHTI and a tation near the YALD tation in Chritchurch. MADOCA and MagicGNSS bae their product on a network that only include a tation near Dunedin in New Zealand. Thi mean that the HAST tation, which ha no nearby tation for any network hould be able to obtain the ame performance a hown here even in cae where local infratructure are abent. In the ame way, MagicGNSS and MADOCA baed olution on AUCK, WGTN YALD or CHTI

12 hould be repreentative of what potentially can be achieve by atellite baed PPP in cae where local infratructure i unavailable. Figure 4. CORS in New Zealand ued to calculate PPP Correction Real-time olution were calculated imultaneouly for the 6 ite, from the 28 th of September to the 3 rd of October (except during ervice outage time). The olution were retarted after either 6 or 12 hour. A total of 340 olution were calculated in thi way. Figure 5 how a ample of real-time PPP olution calculated on the 1 t of October from 18:00 to midnight UTC.

13 Figure 5. Real-time PPP olution calculated for the 1 t of October PPP olution were calculated uing MADOCA(black), MagicGNSS (red) and PPP-Wizard (blue) correction. The PPP olution were calculated uing obervable from the YALD tation near Chritchurch. The red line repreent real-time PPP olution obtained uing MagicGNSS product, the black line repreent olution obtained uing MADOCA product, and the blue line repreent olution obtained uing PPP-Wizard product. The center of coordinate are ITRF coordinate obtained uing Land Information New Zealand (LINZ) poitioning ervice (LINZ, 2015). The cale in the figure are in meter, with the top graph howing Eat-Wet poition error, the middle graph howing North-South error and the bottom graph howing the Vertical error a ±0.5m cale (a oppoed to the ±0.2 m cale for the horizontal component). The RMS horizontal accuracy wa 4.6 cm for olution uing PPP-Wizard, 6.4 cm for olution uing MADOCA, and 9.7 cm for olution uing MagicGNSS. The RMS vertical accuracy wa 8.7 cm for olution uing PPP-Wizard, 10.1 cm for olution uing MADOCA, and 12.7 cm for olution uing MagicGNSS. PPP-Wizard and MADOCA baed PPP olution can be expected to converge to better than 10cm of accuracy within 30 minute. The RMS of horizontal error for MagicGNSS baed olution converged to an accuracy better than 10 cm in 127 minute. 4.1 Poitioning Accuracy Table 6 how the RMS value for horizontal and vertical error for real-time PPP olution obtained uing different correction tream in variou CORS. The firt hour of olution were dicarded in order to give the olution time to converge and tabilize. An average of 14 to 20 olution are ued to calculate the tatitic in table 6.

14 PPP uing MDC1 PPP uing CLK81 PPP uing CLK9B Auckland Chath. I. Dunedin Hating Wellington Yaldhurt 6.22cm 6.13cm 5.80cm 6.33cm 7.43cm 9.66cm 6.62cm 10.29cm 7.45cm 11.35cm 9.74cm 11.46cm 8.71cm 12.34cm 5.03cm 7.69cm 10.78cm 11.82cm 3.39cm 8.10cm 9.15cm 13.53cm 4.34cm 10.58cm 9.86cm 12.44cm 5.30cm 8.10cm 9.75cm 12.14cm 4.98cm 7.42cm 9.76cm 13.61cm 4.64cm 9.69cm Table 6. Horizontal and vertical error of real-time kinematic PPP olution uing real-time correction from MADOCA (MDC1), Magic GNSS (CLK81) and PPP-Wizard (CLK9B). Firt 1 hour of meaurement were removed in order to allow olution to converge and tabilize. The performance of real-time tream are clearly ditinguihable, with PPP-Wizard achieving horizontal accuracie better than 5.30 cm for all tation and vertical accuracie of cm or better. The accuracie were better than 7.43cm horizontal and cm vertical for MADOCA. Finally MagicGNSS give average error of up to 10.78cm horizontal and 13.61cm vertical. The difference in performance wa not ignificant for different CORS, corroborating the fact that ditance from a reference tation (ued to calculate the PPP correction) ha little influence on the accuracy of the olution. It i upected that the MagicGNSS baed PPP olution would have achieved a performance level imilar to that of MADOCA if atellite G26 wa excluded. PPP algorithm that ue currently available real-time PPP product will greatly benefit from algorithm to detect and dicard thee outlier. 4.2 PPP olution convergence Figure 6 illutrate the convergence of horizontal poition in real-time PPP olution uing the variou correction tream. The red dot repreent PPP olution calculated uing MagicGNSS correction, the blue dot repreent olution uing PPP-Wizard correction and the black dot repreent olution calculated uing MADOCA correction. Figure 7 on the other hand illutrate the convergence of vertical poition in real-time PPP olution uing the variou correction tream.

15 Figure 6. Convergence of horizontal poitioning for PPP olution calculated uing MADOCA(black), MagicGNSS (red) and PPP-Wizard (blue) correction. Value on the figure are the RMS of olution calculated between 28 th of September and 3 rd of October The PPP olution were calculated uing obervable from the CORS hown in Figure 4. Figure 7. Convergence of vertical poitioning for PPP olution calculated uing MADOCA(black), MagicGNSS (red) and PPP-Wizard (blue) correction. Value on the figure are the RMS of olution calculated between 28 th of September and 3 rd of October The PPP olution were calculated uing obervable from the CORS hown in Figure 4. There are 360 dot of each color in figure 6. Each dot repreent the RMS of horizontal error correponding to a one minute period (firt minute for the firt dot, the econd minute for the econd dot). The RMS i calculated for correponding ample acro all olution obtained uing a particular real-time tream. Figure 7 preent the equivalent tatitic for vertical error. A it can be een in figure 6, the olution uing MADOCA and PPP-WIZARD both converge to accuracie better than 10 cm at around the 30 minute mark, pecifically 27 minute for MADOCA and 25 minute for PPP-Wizard. The MagicGNSS baed olution took 127 minute to converge below the 10 cm mark.

16 Vertical accuracie converge below 15 cm of accuracy in 20 minute for MADOCA and 17 minute for PPP-Wizard. MagicGNSS converged within 20 cm of the vertical accuracy in 40 minute. It i to note that although thee value have been calculated by averaging more than 100 olution (and the author believe thi to be repreentative) convergence time in PPP can vary a lot from olution to olution. 5. SUMMARY Mot of the GNSS baed high accuracy poitioning ervice available today are baed on differential technique like RTK. Thee kind of ervice are only available inide a network of Continuouly Operating Reference Station (CORS) or a few kilometer from uch tation. An emerging alternative to thee infratructure heavy method i the Precie Point Poitioning (PPP) technique. PPP a a technique that aim to deliver centimeter level accuracy poitioning without the need of directly connecting to a local CORS. PPP i well uited for cae where local infratructure i unavailable or unreliable. Thi paper evaluate the three publicly available real-time correction tream for PPP. MADOCA correction generated by JAXA, MagicGNSS correction generated by GMV and PPP-Wizard correction generated by CNES. The Signal in Space Range error were better than 5 cm for all three product given that ome outlier were eliminated. Thee outlier, particularly atellite G26 for the MagicGNSS were hown to have profound impact in PPP performance. Any practical ytem to be ued in PPP mut be equipped with monitoring ytem to detect and dicard thee outlier. Horizontal accuracy of real-time PPP wa found to be 4.6 cm for olution uing PPP-Wizard, 6.4 cm for olution uing MADOCA, and 9.7 cm for olution uing MagicGNSS. Vertical accuracie were 8.7 cm for olution uing PPP-Wizard, 10.1 cm for olution uing MADOCA, and 12.7 cm for olution uing MagicGNSS. The MagicGNSS olution are expected to have better performance a long a outlier like G26 in thi can be detected and dicarded. MADOCA and PPP- Wizard can be expected to converge within 10 cm of horizontal accuracy and 15 cm vertical accuracy within 30 minute. Although the accuracie and convergence time are not yet at the level of commercial RTK ervice, PPP baed olution are much more flexible and cot effective to deploy. With the ue of atellite delivery, which i under tudy or development by multiple agencie around the world, thee type of ervice could become available anywhere in the world, independent of the availability of local infratructure.

17 ACKNOWLEDGEMENTS Thi reearch i funded through the Autralian Cooperative Reearch Centre for Spatial Information (CRCSI Project 1.11) and i a collaborative project between the CRCSI and the Japan Aeropace Exploration Agency (JAXA). The CRCSI reearch conortium conit of RMIT Univerity, Univerity of New South Wale, Victoria Department of Environment and Primary Indutry, New South Wale Land and Property Information and Geocience Autralia. The author would alo like to thank JAXA for maintaining the MDC1 real-time tream, the CNES for maintaining the CLK9B real-time and IGS for allowing acce to the CLK81 tream. REFERENCES Chang, X. W., Yang X., Zhou T. (2009), MLAMBDA: a modified LAMBDA method for integer leat-quare etimation, Journal of Geodey, Vol. 79, No 9, 2005, p Ge M., Gendt G., Rothacher M., Shi C., Lui J., (2007), Reolution of GPS carrier-phae ambiguitie in Precie Point Poitioning (PPP) with daily obervation," Journal of Geodey, vol. 82, pp , GMV (2015), MagicGNSS Brochure, Acceed Augut Grinter, T., Janen V., (2012). Pot-Proceed Precie Point Poitioning: A Viable Alternative?, Proceeding of the 17th Aociation of Public Authority Surveyor Conference (APAS2012), Wollongong, New South Wale, Autralia, March Harima K., Choy S., Li Y., Choudhury M., Rizo C., and Kogure S. 2015, Pilot Study on the ue of Quai-Zenith Satellite Sytem a a GNSS Augmentation Sytem for High Preciion Poitioning in Autralia, Proceeding of the IGNSS Sympoium, July 2015, Gold Coat, Autralia, Paper 27 IGS, (2015). Real-time Service, Acceed September IGS, (2014). International GNSS Service, Acceed January JAXA, (2014), Interface Specification for QZSS (verion 1.6). November 2014 Kouba, J., (2009), A Guide to uing International GNSS Service (IGS) Product, Acceed October Laurichee D., Mercier F., Berthia J., Broca P., Cerri L., (2009), "Integer Ambiguity Reolution on Undifferenced GPS Phae Meaurement and It Application to PPP and Satellite Precie Orbit

18 Determination," NAVIGATION: Journal of The Intitute of Navigation, vol. 56, pp , LINZ, (2015). Land Information New Zealand PoitioNZ ervice, Acceed December RTCM (2013), "RTCM STANDARD for differential GNSS," in RTCM , ed. Arlington, Virginia: Radio Technical Commiion for Maritime Service, Takau T. and Yauda A. (2009): Development of the low-cot RTK-GPS receiver with an open ource program package RTKLIB, International Sympoium on GPS/GNSS, Jeju, Korea, November Zumberge, J. F., Heflin M. B., Jefferon D.C., Watkin M. M. Webb F. H., (1997). Precie Point Poitioning for The Efficient and Robut Analyi of GPS Data From Large Network, Journal of Geophyical Reearch, 102, B3, BIOGRAPHICAL NOTES Ken Harima obtained hi Bachelor degree in Elctronic Engineering from Univeridad Simon Bolivar, Caraca Venezuela, in He obtained PhD degree from the Univerity of Tokyo in 2012.He i currently a reearch fellow at the School of Mathematical and Geopatial Science in RMIT Univerity. He reearch interet are GNSS augmentation and SBAS ytem. Suelynn Choy completed her PhD in 2009 in the area of GPS Precie Point Poitioning (PPP) at RMIT Univerity, Autralia. Since then, he work a a full-time academic taff at the School of Mathematical and Geopatial Science in RMIT Univerity. She teache land urveying, geodey, and GNSS navigation to undergraduate and graduate tudent. Her current reearch interet are in the area of multi-gnss PPP and uing GNSS for atmopheric and ground remote ening. Suelynn i the co-chair of the IAG (International Aociation of Geodey) Working Group 4.5.2: PPP and Network RTK under Sub-Commiion 4.5: High Preciion GNSS Algorithm and Application. Chri Rizo i a graduate of the School of Surveying, The Univerity of New South Wale(UNSW), Sydney, Autralia; obtaining a Bachelor of Surveying in 1975, and a Doctor of Philoophy in 1980 in Satellite Geodey. He i currently a member of the School of Civil & Environmental Engineering, UNSW. Chri reearch experitie i high preciion application of GPS. He i a member of a Fellow of the Autralian Intitute of Navigation, a Fellow of the U.S. Intitute of Navigation, a Fellow of the International Aociation of Geodey (IAG), an honorary profeor of Wuhan Univerity (P.R. China), and i currently Preident of the IAG ( ). Satohi Kogure i Miion Manager for QZSS operation and technical demontration in Satellite Navigation Office, Japan Aeropace Exploration Agency (JAXA). He received an MS in aeronautical engineering from Nagoya Univerity in 1993 and an MS in aeropace engineering from Univerity of Colorado at Boulder in He ha been working for the development of Japanee atellite poitioning ytem, QZSS a a atellite ytem engineer ince He i a member of Japan Society for Aeronautical and Space Science, Intitute of Poitioning, Navigation and Timing of Japan a well a U.S. Intitute of Navigation.

19 CONTACTS Dr. Ken Harima. Royal Melbourne Intitute of Technology (RMIT) Univerity GPO Box 2476V, Melbourne, Victoria 3001, Autralia Tel: ken.harima@rmit.edu.au Dr. Suelynn Choy Royal Melbourne Intitute of Technology (RMIT) Univerity GPO Box 2476V, Melbourne, Victoria 3001, Autralia. Tel: Fax: uelynn.choy@rmit.edu.au Dr. Chri Rizo The Univerity of New South Wale UNSW Sydney NSW 2052 AUSTRALIA Tel: Fax: c.rizo@unw.edu.au Mr. Satohi Kogure Japan Aeropace Exploration Agency Tukuba Space Center, Sengen, Tukuba, Ibaraki, Japan Tel : Fax : kogure.atohi@jaxa.jp