Technical Literature. SmartRTK: A Novel Method Of Processing Standardised RTCM Network RTK Information For High Precision Positioning

Transcription

1 SmartRTK: A Novel Method Of Processng Standardsed RTCM Network RTK Informaton For Hgh Precson Postonng Aprl 008 Frank Takac, Werner Lenhart Techncal Lterature Takac, F. and Lenhart, W., (008), SmartRTK: A Novel Method Of Processng Standardsed RTCM Network RTK Informaton For Hgh Precson Postonng, Proceedngs of ENC GNSS 008, Aprl -5, Toulouse, France

2

3 SmartRTK: A Novel Method Of Processng Standardsed RTCM Network RTK Informaton For Hgh Precson Postonng Frank Takac, Werner Lenhart Leca Geosystems AG, Heerbrugg, Swtzerland BIOGRAPHY Frank Takac receved an MSc from RMIT Unversty, Melbourne, Australa n 001. He joned Leca Geosystems n 00 and has been nvolved n developng algorthms for a range of hgh-precson GNSS feld products and nfrastructure solutons. He s currently leadng the GNSS postonng algorthms group. Werner Lenhart receved an MSc from Graz Unversty of Technology, Austra n 001 and a PhD n 006. After workng as research assocate for the Austran Academy of Scence he was nvolved n several montorng projects ncludng GPS as Unversty assstant at Graz Unversty of Technology. He joned Leca Geosystems n 006 where he s workng n the GPS Product Management group. 1. ABSTRACT Network RTK s a maturng technology that overcomes range lmtatons of conventonal RTK by modellng dstance dependent atmospherc effects. However, correcton nformaton mght not be avalable for all satelltes observed n a network. Tradtonal RTK algorthms do not process raw observatons wthout correctons n the poston soluton. Nevertheless, these observatons stll contan valuable nformaton for postonng. In ths paper we present a SmartRTK soluton that consders all of the avalable observaton nformaton n the poston soluton. The results demonstrate that uncorrected observatons can be effectvely ncluded n order to mprove the precson of the poston soluton and to yeld more fxed rover postons. In addton, the SmartRTK approach employs an atmospherc decorrelator that uses optmal combnatons of the L1 and L observatons and atmospherc stochastc modellng to mtgate the effects of resdual model errors. The result s more homogeneous postonng accuracy throughout the network.. INTRODUCTION Real-tme knematc (RTK) postonng s an effectve tool for applcatons that reure hgh precson (centmetre level) coordnate accuracy. A conventonal RTK postonng system typcally comprses of a sngle reference staton whch transmts formatted nformaton such as code and carrer phase observatons to one or more moble rover unts n the feld. The reference staton data s combned wth local measurements collected at the rover usng propretary dfferental processng technues to yeld precse relatve coordnate estmates. The accuracy of conventonal RTK decreases as the dstance to the reference ncreases due to the spatal decorrelaton of dspersve and non-dspersve errors nduced by the onosphere and troposphere respectvely. Dependng on the prevalng atmospherc condtons, the operatng range of conventonal RTK postonng s usually lmted to a few tens of klometres. Network RTK s a maturng technology that has the potental to overcome several lmtatons of conventonal RTK. The typcal network RTK model comprses of three or more permanent reference statons connected to a central processng faclty that estmates the dstance dependent errors across the network. Correctons for these errors are combned wth raw reference staton observatons and dstrbuted to users n the feld as depcted n Fgure 1. Permanent Reference Staton Central Processng faclty Real-tme Network Informaton Rover B-drectonal communcaton lnk 0 50 km Fgure 1. The generalzed network RTK model. The central processng faclty collects real-tme data from three or more reference statons, estmates dstance dependent errors for the network and dstrbutes the precse correcton nformaton to n rovers n the feld. The network nformaton helps to mtgate the dstance dependent errors observed at the rover resultng n more homogeneous poston accuracy wthn the regon bounded by the reference statons. The concept of network RTK has been proven commercally. There are Presented at ENC GNSS 008, Aprl -5, Toulouse, France

4 currently several competng solutons ncludng the Master-Auxlary Concept (MAC), Vrtual Reference Staton (VRS), ndvdualsed Master-Auxlary correctons (MAX), Pseudo-reference Staton (PRS) and Flächenkorrekturparameter (FKP). Despte the proven benefts of network RTK, the technology stll has lmtatons. In order to meet the accuracy demands of RTK applcatons, the network nformaton s derved usng unambguous (fxed ambguty) carrer phase observatons and assocated models for the dspersve and non-dspersve effects. In addton to the measurement errors nherent n the carrer phase observatons, mperfect modellng of the dstance dependent errors can degrade the accuracy of the correctons. As these resdual errors grow, the effectveness of network RTK dmnshes. The performance of RTK s dependent, n part, on the number of avalable satelltes (Takac and Walford, 006). However, the network software may not be able to provde correctons for all satelltes n vew. A typcal case s low elevaton satelltes for whch the network software has not resolved the correspondng ambgutes. Nevertheless, the raw reference observatons stll contan valuable nformaton that can be useful for RTK postonng. However, network RTK s tradtonally consdered as an all-or-nothng soluton (Alves, 004). That s, raw and corrected observatons should not be mxed n the poston soluton. In an optmal soluton, the rover software should consder all of the avalable observaton nformaton and account for any resdual observaton errors remanng after modellng. In ths contrbuton, the novel concept of combnng raw and corrected observatons s examned usng the MAC network RTK approach. The effects of resdual observaton errors are mtgated usng combnatons of dual freuency measurements and stochastc modellng. The practcal beneft of ths new approach for RTK postonng s tested usng real-tme data. The results demonstrate ncreased avalablty of poston, better precson and more homogeneous accuracy throughout the network. Fnally, the applcaton of ths approach to other network RTK solutons s also examned. 3. NETWORK RTK TECHNOLOGY In practce, network RTK s realsed n several ways; for example, MAC, VRS and FKP. Each approach has assocated advantages and dsadvantages but they all share a common goal, whch s to provde accurate correcton nformaton for hgh-precson postonng. A detaled crtue of the varous network RTK solutons s beyond the scope of ths paper. Instead, a bref descrpton of each method s provded as a bass for dscusson. VRS, PRS and MAX are all varatons of the same theme and can be broadly categorsed as non-physcal reference statons. In each case, the network software computes dspersve and non-dspersve correctons optmsed for the poston of the user. The ndvdualsed correctons are appled to raw reference staton observatons to form vrtual observatons, whch are then broadcast to the user. For VRS and PRS, the physcal reference staton s also dsplaced so that the vrtual observatons refer to a non-physcal reference staton located n the vcnty of the approxmate rover locaton. Typcally, the baselne length s several metres for VRS and several klometres for PRS. In contrast to non-physcal network solutons, MAC and FKP broadcast raw reference staton observatons and network nformaton separately. In the MAC approach, the network nformaton s represented as sngledfference dspersve and non-dspersve correctons for all auxlary reference statons relatve to a master (Euler et al, 001). The FKP soluton represents the network nformaton usng the coeffcents of a plane surface centred at the locaton of a physcal reference staton (Wübbena and Bagge, 006). In both cases, the rover software decdes how the network nformaton s appled n the poston soluton. Generally, estmates of the dspersve and non-dspersve effects n the network are derved from unambguous carrer phase observatons. Typcally, correctons for the dstant dependent effects at the rover are computed usng lnear approxmaton models, although hgher order surfaces can also be employed. The effectveness of network RTK depends, n part, on the accuracy of the computed correctons. Fgure shows the relatonshp between the modelled nformaton and the true dstance dependent errors for a fcttous network of two reference statons. For smplcty, the dscusson s lmted to the lnear case. Error model error (ε) e b true error e a dfferental error (δ a ) rov 1 Ref a rov rov 3 Ref b Dstance Fgure Relatonshp between the modelled dstance dependent effects and the true errors. The true error s shown n red and the lnear approxmaton n green. The symbol δ s the dfferental error assocated wth the conventonal baselne soluton and ε s the model error (Adapted from Wübbena et al., 005). In Fgure, the network software receves precse carrer phase observatons from reference statons a and b and estmates the true errors e a and e b. The true error shown n red could be dspersve, non-dspersve or a combnaton of both. The dstance dependent errors are Presented at ENC GNSS 008, Aprl -5, Toulouse, France

5 modelled usng a lnear approxmaton shown n green. The lnear approxmaton s not perfect and dfferences to the true error ε represent the model error. The dfferental error δ s the dfference between the true errors observed at the reference staton and some other pont n the network. Network nformaton s expected to mprove the conventonal (sngle base) RTK soluton f the model errors are less than the correspondng dfferental errors at the rover locaton. For a rover operatng n the vcnty of a reference staton, the beneft of network nformaton s mnmal snce the dstance dependent errors effectvely cancel n conventonal dfferental processng schemes. As the user moves away from the reference towards rov, the magntude of the dfferental error ncreases. In ths example, the lnear model s a reasonable approxmaton of the errors at ths pont; therefore, network nformaton should s expected to mprove the poston soluton. However, the mpact network correctons at rov 3 wll be less sgnfcant because the model and dfferental errors are of smlar magntude. The same stuaton s also evdent at rov 1 whch s located outsde the network boundary. Estmates of the dspersve and non-dspersve effects for a gven satellte can only be derved once the assocated nteger ambgutes have been resolved. Generally, t s more problematc to resolve ambgutes for satelltes at low elevatons. Therefore, network correctons may only be avalable for a subset of the tracked satelltes. Tradtonally, raw and corrected observatons are not mxed n the same soluton. Ths s not optmal because the raw reference observatons stll contan valuable nformaton that s useful for postonng. For example, a user located at rov could stll make use of the raw reference observatons n the absence of network correctons. In order to combne all of the avalable nformaton n the poston soluton correctly, the rover software needs to have a detaled understandng of the processes appled to the data receved from the network. Standardsaton s a means of ensurng that the network nformaton s generated n a consstent and understandable manner. The MAC approach s realsed n v3.1 of the RTCM SC-104 standard for dfferental servces (RTCM, 007). The procedure for generatng MAC data s clearly defned n the standard and the format s used to verfy the concepts presented n ths paper. 4. THE MASTER -AUXILIARY CONCEPT A bref overvew of MAC s presented n ths secton. For a detaled revew, the reader s referred to Euler et al (001) and RTCM v3.1 (007). In essence, MAC data comprses the raw observatons of all reference statons n a network mnus nusance parameters such as clock errors and nteger ambgutes. In the context of MAC, a network comprses one master staton m and k auxlary reference statons as depcted n Fgure 3. Aux 1 Master m Rov r Aux Aux k Fgure 3. The defnton of a network n the context of MAC. Staton m s the master and statons 1 k, (k ) represent auxlary reference statons. φ m, Let be the raw carrer phase observaton between staton m and satellte n unts of metres such that: m, ρm + cdt m cdt I m, + Tm + λnm, ε m, φ = + (1) where: ρ m s the geometrc range c s the speed of lght. dt m, dt are the recever and satellte clock errors respectvely. I m, s the freuency dependent onospherc delay. T m s the tropospherc delay λ n m, s the carrer wavelength for freuency f s the true nteger cycle ambguty ε m, s the random measurement error. Introducng an auxlary staton k, the between staton sngle dfference observable can be wrtten as: φ mk, = φ = ρ k, mk φ + where ρmk = ρk ρm terms n (). m, cdt mk I mk, + Tmk + λnmk, + ε mk, () and so forth for the remanng The correcton dfference observable s generated by subtractng computed uanttes for the geometrc range, recever clock error and nteger cycle ambguty from the sngle-dfference observable gven n () such that: mk, s mk φmk, + cτ mk λamk, δφ = + (3) Presented at ENC GNSS 008, Aprl -5, Toulouse, France

6 where: δφ mk, s the correcton dfference observable for master staton m, auxlary staton k and satellte s mk s the computed geometrc range τ mk s the computed recever clock error mk a, s the computed nteger ambguty n unts of cycles. Fnally, the raw correcton dfferences gven by (3) are factored as dspersve and non-dspersve observables. The dspersve observable, denoted by the subscrpt γ, s gven by: f f δφmk, γ = δφmk,1 δφ mk, (4) f f f f 1 whle the non-dspersve observable, denoted by the subscrpt χ, s gven by: 1 the rover software has three optons. Frst, only process observatons for the satelltes that have correctons; second, gnore all of the network nformaton and only process raw observatons; thrdly, mx raw and corrected observatons n the soluton. The next secton wll examne the practcal applcatons of ths flexblty. 5. COMBINING NETWORK INFORMATION AND RAW OBSERVATIONS In standard network RTK postonng solutons, observatons wthout correspondng network nformaton are not processed. As dscussed n 3, ths s an arbtrary approach that excludes valuable nformaton from the soluton. When treated properly, raw and corrected observatons can be combned n a SmartRTK soluton to mprove postonng accuracy. To demonstrate the effect of combnng all the avalable nformaton, statc data was collected from the network depcted n Fgure 4. Aux 10 Rover Master 1 f1 f δφmk, χ = δφmk,1 δφ mk, (5) f f f f 1 Aux 4 Aux 3 In (4) and (5), f s the freuency of the L1 or L carrer denoted by the subscrpts 1 and respectvely. The dspersve and non-dspersve network errors are precsely determned usng fxed sngle-dfference ambguty values. However, t s well known that absolute nteger ambgutes can only resolved correctly on the double-dfference level. Therefore, the expanded sngle-dfference ambguty term n (3) s gven by: p a mk, nmk, Δamk, where: mk = (6) n, s the true ambguty for satellte. Aux km Fgure 4. The rover n network A s located approxmately 15km from the master, whch s also the closest reference staton. The data was frst processed usng a conventonal sngle base RTK soluton followed by a standard network RTK soluton, whch only consders corrected observatons. The data was processed a thrd tme usng the SmartRTK soluton that combnes all of the avalable nformaton. In all tests, the data was processed n a smulated RTK mode. Fgure 5 shows the number of avalable satelltes wth and wthout network correctons. p a mk, Δ s the dfference between the true ambguty and the estmated ambguty for a reference satellte p. The ambguty bas, also referred to as the ambguty level, s common to all estmated ambgutes for the baselne mk. Therefore, the bas wll be estmated as a modfed clock term n sngle-dfference processng schemes or cancel n the double-dfference soluton. The network software transmts dspersve and nondspersve correcton dfferences together wth the raw observatons of the master staton. The rover software s then free to decde how the network nformaton s appled n the poston soluton. If correcton dfferences are only avalable for a subset of the observed satelltes, Fgure 5. The number of avalable satelltes wth and wthout network correctons (network A). The number of avalable satelltes peaks at 8 durng the frst uarter of the sesson and never drops below 6. In comparson, there are only 5 avalable satelltes wth network correctons. The horzontal poston errors for each soluton are compared n Fgure 6. The standard network RTK and SmartRTK solutons have both been plotted aganst the sngle base results. Presented at ENC GNSS 008, Aprl -5, Toulouse, France

7 Fgure 6. Horzontal poston errors for the sngle base, standard network RTK and SmartRTK solutons (network A). The performance of each soluton s comparable. In ths example, the sngle base soluton s more precse than standard network RTK, especally n the frst half of the data set. Ths s a conseuence of weak satellte geometry rather than poor ualty network nformaton. In fact, the vertcal dluton of precson (vdop) s approxmately three tmes hgher than the sngle base soluton at the begnnng of the test. The avalablty of more satelltes n the sngle base soluton mproves the geometry. These extra satelltes are also consdered n the SmartRTK soluton. The heght postonng performance of SmartRTK s compared to the sngle base results n Fgure 8. All of the solutons show smlar horzontal postonng performance. Indeed, the precson of the horzontal component n all cases s sub-centmetre as shown n Table 1. Table 1. Horzontal poston statstcs (1σ) (network A). Soluton σ Hz Sngle Base Standard Net RTK SmartRTK In ths example, applyng network correctons does not yeld a sgnfcant mprovement n terms of horzontal poston accuracy. Ths result suggests that the dspersve and non-dspersve errors at the reference and rover are hghly correlated. In such cases, conventonal RTK s stll an effectve postonng tool. It s well known that the precson of the heght component derved from GNSS postonng s typcally 1- tmes less than horzontal precson. Ths s due an nherent weakness n satellte geometry caused by a lack of observed satelltes below the local horzon. Fgure 7 compares the precson of the heght component for the sngle base and standard network RTK solutons. Fgure 8. Heght poston errors for the sngle base and SmartRTK solutons (network A). The combnaton of all avalable nformaton yelds the most precse soluton. The combnaton of all avalable satelltes and network nformaton yelds the most precse soluton. Ths nference s supported by the heght poston statstcs for all solutons presented n Table. Table. Heght poston statstcs (1σ) (network A). Soluton σ Ht Sngle Base Standard Net RTK SmartRTK Statstcally, the standard network RTK soluton s the least precse, as depcted n Fgure 7, The SmartRTK soluton s the most precse; however, the mprovement s only margnal due to the hgh ualty of the raw data. One dsadvantage of the standard network RTK soluton arses when the number of avalable satelltes wth correctons s less than the crtcal threshold needed for postonng. In Fgure 9, the number of satelltes that have correctons falls below 5 after epoch However, there are at least 6 satelltes observed at the master and rover statons durng the whole perod. Fgure 7. Heght poston errors for the sngle base and standard network RTK solutons (network A). The standard network RTK soluton s less precse as a conseuence of less satelltes and weaker geometry. Presented at ENC GNSS 008, Aprl -5, Toulouse, France

8 In ths network, the closest reference staton s Aux 10 whle the master staton s located approxmately 43km from the rover. The poston errors for the sngle base and standard network RTK solutons are shown n Fgure 11. Fgure 9. Poston errors for the standard network and SmartRTK solutons (network A). SmartRTK mantans a fxed soluton even when the number of avalable satelltes wth correctons falls below 5. The conseuence of gnorng uncorrected satelltes s the loss of the fxed poston. In contrast, SmartRTK mantans the soluton wthout the need for rentalsaton. The results presented n ths secton serve as proof-of-concept for the SmartRTK approach of combnng raw and corrected observatons n the poston soluton. However, the data was collected durng a uet perod of atmospherc actvty where the dstance dependent errors effectvely cancel n sngle base RTK processng. In practce, ths s not always the case and dstance dependent errors for uncorrected satelltes can become sgnfcant. Furthermore, corrected observatons may also be affected by resdual model errors. It s mperatve to treat resdual errors properly for the best overall performance. 6. ATMOSPHERIC DECORRELATION Raw and corrected dfferental observatons can be based by resdual errors. In the case of raw measurements, the resdual errors grow as the reference-rover baselne length ncreases. For corrected observatons, mperfect modellng of the dstance dependent effects s the cause of resdual bases (see secton 3). To llustrate the effect of resdual errors, data was collected from the network depcted n Fgure 10. Aux 10 Master Rover 0 0 km Aux 3 Fgure 11 Poston errors for the sngle base and standard network RTK solutons (network B). Despte the avalablty of network nformaton, the standard network RTK soluton s stll affected by sgnfcant resdual dstance dependent errors. In the frst test, the precson of the horzontal poston for the sngle base soluton was sub-centmetre. In ths experment, poston errors as large as 0.09 m are evdent ndcatng that sgnfcant resdual dstance errors are present n the data. The poston statstcs for the sngle base and standard network RTK solutons are presented n Table 3. Table 3. Poston error statstcs (1σ) (network B). Soluton σ Hz σ Ht Sngle Base Standard Net RTK The mpact of applyng network correctons cannot be properly assessed n ths case. Although the precson of the standard network RTK soluton s better than the sngle base, the results can t be compared drectly because the master staton s not the closest reference staton. The magntude of the poston errors for the standard network RTK soluton are sgnfcantly larger compared to results recorded n the frst experment (Table 1 and Table ), despte the relatvely close proxmty of the nearest reference staton (1km). It s evdent that the standard network RTK soluton s stll affected by resdual bases. An analyss of the observaton resduals reveals that resdual onosphere s the domnant error source. An example plot for PRN 14 s shown n Fgure 1. Aux 7 Aux 0 Fgure 10 In network B the rover s located approxmately 43km from the Master. The closest reference staton s Aux 10, approxmately 1km from the rover. Presented at ENC GNSS 008, Aprl -5, Toulouse, France

9 soluton s compared to the standard network RTK soluton n Fgure 14. Fgure 1 Resdual model errors can stll reman n the data after network correctons have been appled. Ths chart s an example of the resdual onospherc errors for PRN 14. It s well known that the frst order onospherc effect can be removed by a formng a lnear combnaton of the L1 and L observables. Unfortunately, the nose of the onofree observable (L3) s approxmately 3 tmes greater than L1. In conventonal RTK, the decson to swtch to an L3 soluton s usually a functon of the baselne length. However, network RTK tres to model the dstance dependent errors so baselne length s a less meanngful metrc for predctng resdual errors. In fact, the typcal baselne length n a VRS soluton s n the order of only a few metres. A more robust approach of assessng the resdual dstance dependent errors s needed for network RTK. If the network conssts of four or more statons, then the precson of the predcted correctons can be used to assess the ualty of the network nformaton (Chen et al., 003). The precson of the computed correctons wll be hgh f the approxmaton model used for the dstance dependent effects matches the spatal shape of the actual errors and vce versa. An example of the precson of the onospherc correctons computed by the rover software for PRN 14 s gven n Fgure 13. Fgure 14 Poston errors for the SmartRTK and standard network RTK solutons (network B). The combnaton of all avalable nformaton and the atmospherc decorrelator yelds the most precse results. The precson of the SmartRTK soluton s vsbly more precse than the standard network RTK soluton. The large poston errors around epoch 9950 have been successfully mtgated, whch s reflected n the poston statstcs gven n Table 4. SmartRTK typcally makes use of one or two extra satelltes n the poston soluton as shown n Fgure 15. Fgure 15 The number of satelltes contrbutng to the standard network and SmartRTK poston solutons. SmartRTK typcally uses one extra satellte n ths example. Table 4. Poston error statstcs (1σ) (network B). Fgure 13 The precson of the nterpolated onospherc correctons for PRN 14. Note that the onospherc resduals are absolute. The precson of the computed correctons can be used to assess the ualty of the network nformaton. In Fgure 13, the precson of the computed correctons correlates well wth the actual onospherc resduals. The ualty nformaton can be used for observaton weghtng or for selectng optmal combnatons of the L1 and L observables. However, ths nformaton can only be generated for satelltes that have correcton nformaton. When combnng uncorrected observatons, the algorthm must also account for the dstance dependent errors affectng these measurements, whch can have dfferent stochastc propertes compared to corrected observatons. The SmartRTK atmospherc decorrelator treats resdual dstance dependent errors usng optmal combnatons of the L1 and L observables and onospherc resdual stochastc modellng. The result of the SmartRTK Soluton σ Hz σ Ht Standard Net RTK SmartRTK SmartRTK reduces the standard devaton of horzontal and vertcal poston errors by a factor of two. The soluton s acheved wthout swtchng to an L3 soluton. The result s more homogeneous poston accuracy throughout the network even n dsturbed atmospherc condtons. 7. APPLICABILITY TO OTHER NETWORK RTK APPROACHES The advantage of combnng raw and corrected observatons has been demonstrated usng the MAC network RTK approach. Theoretcally, the dea of combnng observatons can also be appled to other types of network RTK solutons. However, there are several factors that lmt the general applcaton of ths approach n practce. Presented at ENC GNSS 008, Aprl -5, Toulouse, France

10 For non-physcal network RTK solutons such as MAX and VRS, t s the network servce and not the rover software that apples ndvdualsed network nformaton to the raw reference data. Ths s problematc n practce because the processes appled by the network software are not fully descrbed and the rover software has no way of dentfyng uncorrected measurements. Mxng these observatons n the poston soluton could bas the results and degrade performance. To llustrate the problem, consder the vrtual observaton for a non-physcal reference staton v to satellte p as gven by: p p p p v, = φm, + ρmv + d mv, φ (7) where: p φ m, s the raw observaton from a physcal reference staton m to satellte p p ρ mv s the geometrc dsplacement between statons m and v. In the formaton of the vrtual observaton, a computed sngle-dfference correcton term d s appled to the undfferenced observaton of the physcal reference staton. The dstance dependent term n (7) s gven by: p p d mv, mv, + bmv, = κ (8) The correcton conssts of the true value κ relatve to a physcal reference staton m and a sngle-dfference bas term b, whch s descrbed n secton 4. Snce the bas term s not satellte dependent, t cancels n the doubledfference f only corrected observatons are used. If uncorrected observatons (d = 0) are mxed wth corrected observatons n the double-dfference, then only the sngle-dfference correcton s appled and the bas term wll not cancel n conventonal processng schemes. A second problem arses for uncorrected observatons of non-physcal reference statons. The geometry of the observatons s related to some vrtual pont n the network due to the dsplacement of the physcal reference staton. However, these observatons no longer have any physcal meanng because the dstance dependent errors stll refer to the locaton of the physcal reference staton. Data processng algorthms often use baselne length for observaton weghtng and buldng optmal combnatons of L1 and L observatons. It would be nvald to apply these algorthms to the uncorrected observatons. Ths partcular problem does not affect the MAX soluton snce the non-physcal reference staton s not dsplaced (ρ = 0). In the case of FKP, the rover receves the raw observatons of the physcal reference staton and network nformaton separately. The problems dentfed for non-physcal reference statons do not apply. Therefore, t s also possble to mx corrected and uncorrected observatons n FKP mode. The SmartRTK soluton also conssts of an atmospherc decorrelator to deal wth resdual model errors as descrbed n secton 6. Ths technology s employed when only corrected observatons are processed and also n combnaton wth uncorrected measurements. Therefore, the soluton s applcable to all types of network RTK solutons. 8. CONCLUSION Dspersve and non-dspersve observaton errors nduced by the onosphere and troposphere lmt the operatng range of conventonal RTK. The goal of Network RTK s to model the dstance dependent effects n order to provde homogeneous poston accuracy wthn the regon bounded by the reference statons. Despte the proven benefts of ths technology, Network RTK stll has lmtatons. If the approxmaton models used for the dstances dependent effects do not match the spatal shape of the actual errors, the effectveness of network RTK wll dmnsh to the pont where at best t no longer provdes any beneft over conventonal RTK or, worse, degrades the sngle base soluton. To meet the accuracy demands of hgh-precson RTK applcatons, estmates of the dspersve and nondspersve errors are derved from a fxed ambguty soluton. Therefore, correcton nformaton mght only be avalable for a subset of satelltes observed n the network. Tradtonally, raw observatons wthout correctons are not ncluded n the poston soluton. In many cases, these observatons stll contan valuable nformaton for postonng. A SmartRTK soluton was presented n ths paper that combnes all of the avalable observaton nformaton n the poston soluton. The results demonstrate that uncorrected observatons can be effectvely ncluded n order to mprove the precson of the poston soluton. An atmospherc decorrelator, whch uses optmal combnatons of the L1 and L observatons and atmospherc stochastc modellng, was effectve at mtgatng the effects of resdual modellng errors. The result of SmartRTK s more homogeneous postonng accuracy throughout the network. The SmartRTK soluton was demonstrated usng the MAC network RTK approach. MAC s realsed n the RTCM v3.1 standard for dfferental servces. There s also provson for non-physcal reference staton observatons n the standard; however, the methods appled n the generaton of the network nformaton are not clearly descrbed. Ths lmts the general applcaton of SmartRTK to non-physcal network RTK technology. The SmartRTK soluton s mplemented n the latest release of the Leca System 100 frmware. Presented at ENC GNSS 008, Aprl -5, Toulouse, France

11 9. REFERENCES Alves, P. R. S., (004), Development of Two Novel Carrer Phase-Based Methods for Multple Reference Staton Postonng, PhD Thess, Department of Geomatcs Engneerng, Unversty of Calgary, Calgary, Canada, December, 03pp. Chen, X., Landau, H. and Vollath, U., (003), New Tools for Networked RTK Integrty Montorng, n: Proc of the 16th Internatonal Techncal Meetng of the Satellte Dvson of The Insttute of Navgaton ION GPS 003, Portland, Oregon, September 9-1, pp Euler, H-J., Keenan, R. C., Zenhauser, B. E. and Wübbena, G., (001), Study of a Smplfed Approach Utlzng Informaton from Permanent Staton Arrays. n: Proc of the 14th Internatonal Techncal Meetng of the Satellte Dvson of The Insttute of Navgaton ION GPS 001, Salt Lake Cty, Utah, September Rado Techncal Commsson For Martme Servces (RTCM), (007), RTCM Standard For Dfferental GNSS Servces Verson 3 wth Amendment 1, RTCM Paper SC104-STD, Developed by the RTCM Specal Commttee No. 104, Amended May 1, 007. Takac, F. and Walford, J., (006), Leca System 100 Hgh Performance GNSS Technology for RTK Applcatons, n: Proc of the 19th Internatonal Techncal Meetng of the Satellte Dvson of The Insttute of Navgaton ION GNSS 006, Fort Worth, Texas, September 6-9. Wübbena, G., Schmtz, M. and Bagge, A., (005), PPP- RTK: Precse Pont Postonng Usng State-Space Representaton n RTK Networks, n: Proc of the 18th Internatonal Techncal Meetng of the Satellte Dvson of The Insttute of Navgaton ION GNSS 005, Long Beach, Calforna, September Wübbena, G. and Bagge, A., (006), RTCM Message Type 59-FKP For Transmsson of FKP verson 1.1, Geo++ Whte Paper Nr , Garbsen, Germany, 8pp. Presented at ENC GNSS 008, Aprl -5, Toulouse, France

12 Wth close to 00 years of poneerng solutons to measure the world, Leca Geosystems products and servces are trusted by professonals worldwde to help them capture, analyze, and present spatal nformaton. Leca Geosystems s best known for ts broad array of products that capture accurately, model uckly, analyze easly, and vsualze and present spatal nformaton. Those who use Leca Geosystems products every day trust them for ther dependablty, the value they delver, and the superor customer support. Based n Heerbrugg, Swtzerland, Leca Geosystems s a global company wth tens of thousands of customers supported by more than,400 employees n countres and hundreds of partners located n more than 10 countres around the world. Leca Geosystems s part of the Hexagon Group, Sweden. When t has to be rght. Illustratons, descrptons and techncal specfcatons are not bndng and may change. Prnted n Swtzerland Copyrght Leca Geosystems AG, Heerbrugg, Swtzerland, 008. Leca Geosystems AG Heerbrugg, Swtzerland