GPS Multpath Change Detecton n Permanent GPS Statons Lnln Ge, Shaowe Han, and Chrs Rzos School of Geomatc Engneerng The Unversty of New South Wales Sydney, NSW 2052, AUSTRALIA Yuk Hatanaka Geographcal Survey Insttute Mnstry of Constructon, Japan Abstract A technque based on an adaptve flter usng the least-mean-square algorthm s proposed for detecton of multpath change n permanent, contnuously operatng GPS (CGPS) statons. The technque has been tested on some expermental data, ndcatng that the multpath change above the recever nose level can be detected. Further tests have been conducted wth some CGPS data from the Japanese GEONET when there was a snowfall. The results show that when there s a change n the antenna envronment t wll ndeed be detected n both the pseudo-range and carrer phase data. Ths technque provdes an easy-to-mplement, qualty assurance tool for CGPS antenna envronmental sensng after events such as typhoons, cyclones, snowfalls, volcano eruptons, earthquakes, etc. Other possble applcatons nclude the montorng of slope stablty and ground subsdence. 1. Introducton No matter how well contnuously operatng GPS (CGPS) networks are desgned, multpath s a sgnfcant concern as t mpacts on the qualty of the CGPS outputs or 'products'. Moreover, t may change slowly on a seasonal bass, or abruptly due to events such as a snowfall. As s well known, multpath effects occur when GPS sgnals arrve at a recever ste va multple paths due to reflectons from nearby objects, such as the ground and water surfaces, buldngs, vehcles, hlls, trees, etc. Multpath dstorts the C/A-code and P-code modulatons, as well as the carrer phase observatons. However, multpath sgnals are always delayed compared to lne-of-sght sgnals because of the longer travel paths caused by reflectons. Although use of choke-rng antennas and the careful selecton of antenna ste can effectvely mtgate multpath, t cannot always be elmnated and sometmes the resdual multpath dsturbance remans a major contrbutor of error n contnuous GPS results. There are also some applcatons, such as volcano and opencut mne slope montorng, for whch t s often mpossble to dentfy antenna stes whch are not vulnerable to multpath. In the case of volcano montorng, all the GPS recevers have to be placed on the slope or at the foot of the mountan. The only antenna ste whch may be free of multpath s the one on the summt, where there s often a great reluctance to nstall a recever! The mportance of multpath has long been recognzed by the GPS communty. Many methods, both onrecever and post-recepton processng, have been developed. Much effort has been nvested n refnng onrecever processng algorthms, both to reduce the threshold for multpath detecton and rejecton, and to smultaneously mprove the measurement accuracy. The artcle by Fenton et al. (1991) descrbed one of the frst low-cost recevers employng the narrow correlaton trackng technque, as well as provdng useful background on the theory behnd achevable pseudo-range measurement accuracy. Other on-recever processng methods to reduce carrer phase and pseudo-range multpath have also been explored.
Derendonck et al. (1992) descrbes the theory and performance of narrow correlator technology. Most state-of-the-art recevers now employ technques smlar to the Multpath Elmnaton Technology and the Multpath Elmnaton Delay Lock Loop (Townsend & Fenton, 1994) to elmnate, as far as possble, multpath at the recever sgnal processng stage. But multpath cannot be removed completely, and the resdual may stll be too sgnfcant to gnore when compared to, for example, the crustal deformaton sgnal beng sought n the GPS results. Hence, t s stll necessary to nvestgate post-recepton data processng technques for mtgatng multpath. Fortunately, snce the GPS satellte ground tracks repeat every sdereal day, the multpath errors experenced today, for example, from 1:00pm 2:00pm wll be very smlar to those experenced tomorrow, but wll occur from 12:56pm 1:56pm, provde the antenna envronment remans unchanged. In recent years several post-recepton methods to reduce multpath have been proposed. For example, one suggeston s to map the multpath envronment surroundng a GPS antenna so that the multpath correctons for each satellte sgnal, as a functon of ts azmuth and elevaton, can be determned (Haj, 1990; Cohen & Parknson, 1991). One of the lmtatons of such an approach s that the antenna envronment has to be mapped repeatedly f there are any changes to t. In 1998 a group of researchers, headed by Jm Davs of the Harvard-Smthsonan Center for Astrophyscs, was funded to buld an Antenna and Multpath Calbraton System (AMCS) (Meertens, 2000). In ther desgn, a parabolc reflector rado antenna wth hgh drectvty wll be connected to a standard geodetc recever, provdng a sgnal free of multpath effects. By comparson wth the sgnal receved through the nomnal GPS antenna, one can assess the multpath envroment. When completed, the AMCS wll provde mportant nformaton about the multpath characterstcs of many CGPS stes. Nevertheless, t s unlkely that many AMCS wll be bult due to ther hgh cost. Hence a few AMCS have to be shpped from ste to ste, and they wll certanly not be avalable on a daly bass at each of the thousands of CGPS stes around the world. Thus any multpath error that s affected by changes n the weather (and thus reflectvty of the nearby surfaces), changes n vegetaton, and skyvew, cannot be calbrated by the AMCS. To dentfy and elmnate multpath sources Axelrad et al. (1994) has suggested analysng the sgnal-tonose rato (SNR) of GPS sgnals. Multpath reflectors are dentfed by solatng segments of the SNR data wth strong spectral peaks. A model of the phase errors s generated based on the frequences of these peaks, whch s then used to reduce multpath. The shortcomng of such a method s that t cannot be mplemented n real-tme, and may need some changes to be made n the CGPS operaton. The use of a multpath template for the purpose of multpath mtgaton was frst proposed by Bshop et al. (1994). An deal complment to these technques would be a multpath calbraton method that could estmate and remove the multpath error from the CGPS data. Han & Rzos (1997) proposed the use of bandpass fnte mpulse response (FIR) flters to extract and elmnate multpath based on the CGPS observatons. These do not only sgnfcantly mtgate multpath but can also generate a multpath model that can be used for realtme applcatons. However, the lmtaton of such an approach s that sgnals (for example, crustal deformaton nduced by an earthquake) fallng n the same frequency band as the FIR flters wll be fltered out. Instead of focusng on multpath mtgaton, n ths paper GPS multpath s used as a sgnal to sense the change n the GPS antenna envronment. Ths capablty s very useful n the mantenance and qualty assurrance of CGPS networks such as the GEONET, where the GPS montorng envronment can be changed not only by natural phenomenon such as snow accummulaton and tree growth, but also by dsasters such as volcanc erupton, typhoon and earthquake. Snce the GPS multpath, and other noses, as well as crustal deformaton sgnals (partcularly n the case of slent/slow sesmc events), tend to fall n the same range of frequences, and the noses are changng contnuously (Han & Rzos, 1997; Ln & Rzos, 1997), the unknown flter parameters must be estmated or "tuned" n real-tme, and altered contnuously, n order to extract multpath. Therefore, t s preferable that an adaptve flter rather than a fxed flter be used for extractng multpath. In ths paper, the adaptve flter based on the least-mean-square algorthm s descrbed and used as the mathematcal bass for multpath extracton and change detecton. The flter s employed to process the multpath resdual seres of both pseudo-range and carrer phase measurements. 2. MULTIPATH COMBINATION
To estmate multpath, t s not possble to smply compare the measured pseudo-range or carrer phase to the true geometrc range snce the error s a combnaton of many effects n addton to multpath. Therefore, multpath solaton or the formaton of a multpath 'data combnaton' s essental. The followng pseudo-range combnaton s acheved through judcous combnaton of pseudo-range and carrer phase measurements, takng advantage of the fact that the nose and multpath effects on the carrer phase are neglgble compared to those of the pseudo-range, although most other error sources are the same: 9529 7200 MPP 1 = P1 φ 1 + φ 2 + K1 2329 2329 (1) 11858 9529 MPP 2 = P2 φ 1 + φ 2 + K2 2329 2329 (2) where MP and MP P2 P 1 are the pseudo-range multpath effects on L1 and L2; P 1 and P 2 are pseudo-range data on L1 and L2; φ 1 and φ 2 are carrer phase data on L1 and L2; and K 1 and K 2 are functons of the recever nose, the multpath on carrer phase and nclude the unknown nteger ambgutes (whch can be assumed constant f there s no cycle slp n the carrer phase data). All terms are expressed n meters. After the combnaton, the result s the pseudo-range multpath mxed wth the recever nose, whch s referred to as the 'pseudo-range multpath resdual seres' later on. For the estmaton of carrer phase multpath, start wth the one-way carrer phase equaton: L = ρ + dρ+ dtrop don +. + ( dt dt) λ N+ MC υc (3) where = subscrpt ndcates a certan frequency of sgnal (=1 or 2); L = phase measurement n range unt; ρ = geometrc range between the antenna and satellte; dρ = ephemers (orbtal) error; dtrop = tropospherc bas; don = onospherc bas; dt,dt = recever and satellte clock errors; λ = the sgnal wavelength; N = phase ambguty (nteger); MC = phase multpath; and νc = phase nose. If the two GPS statons are very close to each other, after the between-satellte and between-recever dfferencng (double-dfferencng: DD), Eq. (3) reduces to: L = ρ λ. N + MC + υc (4) where s the DD operator. In Eq. (4), although L can be drectly calculated from the RINEX fles of the two statons, ρ and N have to be estmated before the DD phase multpath MC can be derved. Therefore, the combnaton result s also the carrer phase multpath mxed wth the phase nose, whch s referred to as the 'carrer phase multpath resdual seres' later on. When one of the two statons s relatvely multpath-free, the DD multpath can, therefore, be consdered as the carrer phase multpath sgnal for the other staton. 3. Adaptve Flter
An adaptve flter s a dual-nput, dual-output, and closed-loop adaptve feedback system (Haykn, 1996; Ge, 1999). The operaton of such an adaptve flter nvolves two basc processes: 1) a flterng process to produce an output n response to an nput sequence, and 2) an adaptve process for the control of adjustable parameters used n the flterng process. The two nputs are: Prmary nput: d(n) = s(n) + x (n) Reference nput: x(n) The prmary nput d(n) conssts of the desred sgnal of nterest s(n) (n ths applcaton, the multpath) bured n (or contamnated by) nose x (n). The reference nput x(n) supples nose alone. In order to extract the multpath usng the adaptve flter output, s(n), x (n), and x(n) have to satsfy the followng two condtons: 1) the sgnal s(n) and nose x (n) n the prmary nput are uncorrelated wth each other; and 2) the nose n the reference nput x(n) s uncorrelated wth the sgnal s(n) but s correlated wth the nose component of the prmary nput x (n). The two outputs are: M 1 Flter output: y( n) = w ( n) x( n ) = 0 Estmaton error: e( n) = d( n) y( n) The least-mean-square adaptve algorthm s: w ( n) = w ( n) + e( n) x( n ) 1 µ As mentoned above, a true nose seres has to be used as the reference sgnal x(n) for the flter n order to extract multpath from the prmary sgnal. However, to derve true nose s as dffcult as dervng a true sgnal n practcal applcatons. In fact, there s no fundamental dfference between sgnal and nose, except that a sgnal s a desred output and a nose an undesred one. To address practcal applcatons of the adaptve flter t s necessary to ntroduce the terms coherent component and ncoherent component when dscussng the compostons of the prmary and reference sgnals. A coherent component s a composton n the prmary sgnal whch s correlated wth a composton n the reference sgnal, whle an ncoherent component s a composton n the prmary sgnal whch s uncorrelated wth any composton n the reference sgnal. Hence, when the flter converges the coherent component s the drect output of the FIR flter y(n), and the ncoherent component s the output of the whole adaptve flterng system e(n). Based on the above analyss, n the case of multpath extracton the multpath resdual sequence derved usng the multpath combnatons on a selected day can be expressed as: D t) = MP + N ( ) (5) 1( 1 1 t where N 1( t ) and MP 1 are the nose free of multpath and the nose contrbuton from multpath respectvely. The multpath sequence on the next day s expressed as: D 2 = MP2 + N 2 (6)
where N ( ) and MP ( ) have smlar defntons as N ( ) and MP ( ). 2 t 2 t Accordng to prevous studes (e.g. Han & Rzos, 1997), MP 1( t ) and MP ( t 2 ) are hghly correlated. But N 1( t ) and N 2 are uncorrelated, whch means that the two condtons mentoned above are only partly satsfed. In ths case, as demonstrated n a numercal smulaton study (Ge et al., 2000), the adaptve flter can stll be used to extract multpath from the combnaton results of the second day by employng those of the frst day as a reference sgnal. And N 2 ( t ) and MP 2 wll be outputs from the flter as the ncoherent component e(n) and coherent component y(n) respectvely. 1 t 1 t 4. Multpath extracton In order to extract multpath, M days worth of GPS data are dvded nto M-2 sets, each of whch conssts of data from three consecutve days, as follows: Set Consecutve days 1 1, 2, 3 2 2, 3, 4 3 3, 4, 5. M-2 M-2, M-1, M That s, there are 3 days n each set and among them there are 2 days overlappng n two consecutve sets. After the multpath combnaton, the multpath resdual sequences can be expressed as: Day 1 n Set, D t) = MP N ( ) 1 ( 1 + 1 t 2 = MP 2 + N 2 ( t 3 = MP 3 + N 3( t Dj t) = MPj + j Day 2 n Set, D ) Day 3 n Set, D ) Or for Day j n Set, ( N Where, =1, 2,, M-2 j=1, 2, and 3 We defne @ as the adaptve flterng operator and A@B as the adaptve flterng process usng A as the prmary nput and B as the reference nput. Then for the Day 1 and Day 2 par: D 2 1 = t @ D [ MP 2 N 2 ( )] (7) where MP 2 s the adaptve flter estmated multpath for Day 2 n Set (coherent output) and N 2 s the estmated recever nose (ncoherent output). Smlarly for the Day 2 and Day 3 par: D 3 2 = t @ D [ MP 3 N 3 ( )] The estmated multpath for Day j n Set can be further expressed as: MP j = MP j + δ MP j (8)
where δ MP 2 ( t ) s the multpath estmaton error. If there s no change n the antenna envronment then: MP j MP 1( t + t) = j (9) t s typcally 246 sec although t may be dfferent for dfferent satelltes (Wannnger & May, 2000). 5. Multpath change detecton If there s a change of the antenna envronment between Days 2 and 3, then: MP 3 = MP 2 ( t + t) + MP (10) and MP 3 @ MP 2 = [ MP 2 ( t + t) MP ] where MP (t) s the multpath change between Days 2 and 3 n Set. (11) Ths approach to multpath change detecton resembles the 3-pass dfferental radar nterferometry (InSAR) technque (Gabrel et al., 1989), n whch the elevaton change s extracted by InSAR from the terran model that s also constructed by InSAR. To summarze, as llustrated n Fg. 1, there are three man steps n mplementng multpath change detecton based on adaptve flterng. The frst step s to derve the pseudo-range or carrer phase multpath resdual tme seres through multpath combnaton of the GPS measurements. Snce the pseudo-range multpath combnaton (Eqs. (1) and (2)) s formed from one-way, dual-frequency phase and pseudorange data, whch are avalable from the one GPS recever, pseudo-range multpath mtgaton can be mplemented on a sngle recever bass. CGPS data RINEX data Multpath combnaton Day 1 n Set MP1+N1 Day 2 n Set MP2+N2 Day 3 n Set MP3+N3 Multpath extracton Multpath change detecton MP2 MP3 DMP2,3 Fgure 1 Multpath change detecton procedures. On the other hand, the carrer phase multpath seres are obtaned by double-dfferencng (Eq. (4)) the phase measurements from two relatvely closely spaced recevers (close enough to make the assumpton of effectve cancellaton of spatally-correlated atmospherc bases), one subject to multpath dsturbance, and
the other n a multpath-free envronment. Ths may be a severe lmtaton n a sparse CGPS array. However, n an array such as the GEONET of Japan, t s always possble to dentfy some reference statons free of multpath, and then generate the apparent RINEX data for a Vrtual Reference Staton (VRS) (Bücherl et al., 2000) located close enough to the CGPS staton targeted for multpath mtgaton. In the second step, the adaptve flter (ndcated by the star n Fg. 1) s used frst to extract the multpath from the resdual seres from two consecutve days, obtaned from the multpath combnaton (Eqs. (7) and (8)). Two multpath seres wll be obtaned for each 3 day set as a result. In the thrd step, the adaptve flter s used agan to detect the multpath change between the two multpath seres derved n the second step (Eq. (11)). In the followng two sectons ths technque s frst used for multpath change detecton n a statonary antenna envronment to test f the detected change s zero when there s no change n the antenna envronment. Then t s appled for multpath change detecton n a changed antenna envronment to test f the technque can dentfy the change when there s ndeed a change n the multpath envronment caused by snowfall. 6. Multpath change detecton n a statonary antenna envronment Two data sets were used n ths test, whch were the same as those used n the Han & Rzos (1997) study. In the frst data set, the pseudo-range data over a perod of nearly 3 hours for four successve days, collected on the roof of the Geography and Surveyng (GAS) buldng, at the Unversty of New South Wales (UNSW), from 30 September to 3 October 1997, usng an Ashtech Z12 GPS recever, were used to compute the pseudo-range multpath tme seres accordng to Eq. (1). (In ths study only the 'pseudo-range multpath tme seres' for L1 for satellte PRN 9 are used.) In the second data set, multpath seres of double-dfferenced carrer phase measurements over a perod of nearly 2 hours on four successve days for two satelltes (PRNs 1 and 21) were calculated from data collected n an experment carred out on the roof of the GAS buldng at UNSW, from 28 Aprl to 1 May 1997, usng two Leca SR299 GPS recevers wth a baselne length of about 6 meters. Here the resdual seres of double-dfferenced carrer phase observatons are used as the 'carrer phase multpath tme seres' because they reflect the multpath dsturbance and observaton nose (t s assumed that all the other errors and bases are neglgble because the baselne length s only of the order of 6 meters). Fg. 2 shows the results of the pseudo-range multpath combnaton. From top to bottom the plots represent the pseudo-range multpath resdual sequences from Day 1 to Day 4. It can be seen that although they look very smlar there are dfferences between them due to the change of uncorrelated nose from day to day. Also the t (about 246 sec) advance between two consecutve days can be clearly seen.
Fgure 2 Pseudo-range Multpath Combnaton Results. Fg. 3 shows the multpath extracton result for Day 2 usng the multpath resdual sequence of Day 1 as reference nput. The extracted multpath on Day 2 (MP2) s dsplayed n the thrd plot and the fourth one gves the uncorrelated nose. Smlar extracton s done on pars of days: Day 2 & Day 3, and Day 3 & Day 4 so that MP2, MP3 and MP4 are all obtaned. Fgure 3 Multpath Extracton Result.
Fgs. 4 and 5 dsplay the multpath change detecton results for pars of days Day 2 & Day 3 and Day 3 & Day 4. The multpath change s gven n the fourth plot. Compare the plots for MP2, MP3, and MP4 n these fgures, they look much more smlar to each other than the multpath resdual sequences for Day 1 to Day 4, whch hghlghts the mportance of multpath extracton for relable multpath change detecton. Fgure 4 Multpath Change Detecton Result (Day 2-3). Fgure 5 Multpath Change Detecton Result (Day 3-4).
Table 1 summarzes the standard devatons (STD) of the pseudo-range multpath extracton (the frst three rows) and change detecton (the last two rows) results. Note that n the multpath extracton step, the ncoherent output s the uncorrelated nose and the coherent output s multpath, whle n the multpath change detecton step the ncoherent output s the change of multpath and the coherent output s correlated multpath. Note that the changes of multpath detected are not zero as they should be, but are 0.044 and 0.046m n the pars Day 2 & Day 3 and Day 3 & Day 4 respectvely, whch are nevertheless smaller than the derved uncorrelated nose. Therefore, a tentatve concluson s that any multpath change smaller than the uncorrelated nose (manly the recever nose) wll go undetected. Table 1 Multpath Extracton and Change Detecton: Pseudo-Range Result. Day Incoherent Output STD (m) Coherent Output STD (m) 2 0.046 0.210 3 0.047 0.205 4 0.052 0.212 2-3 0.044 0.201 3-4 0.046 0.207 Now the technque s used to detect the multpath change of the carrer phase. Fg. 6 shows the results of the carrer phase multpath combnaton. From top to bottom the plots represent the carrer phase multpath resdual sequences from Day 1 to Day 4. As n the case of pseudo-range, t can be seen that although they look very smlar there are dfferences between them due to the change of uncorrelated nose from day to day. Also the t (about 246 sec) advance between two consecutve days can be clearly seen. Fgure 6 Carrer phase multpath combnaton result. Fg. 7 shows the carrer phase multpath extracton result for Day 2 usng the multpath resdual sequence of Day 1 as reference nput. The extracted multpath on Day 2 (MP2) s dsplayed n the thrd plot and the fourth one gves the uncorrelated nose. Smlar extracton s done on pars of days: Day 2 & Day 3, and Day 3 & Day 4 so that MP2, MP3 and MP4 are all obtaned.
Fgure 7 Multpath Extracton Result. Fgs. 8 and 9 dsplay the carrer phase multpath change detecton results for pars of days Day 2 & Day 3 and Day 3 & Day 4. The multpath change s gven n the fourth plot. Compare the plots for MP2, MP3, and MP4 n these fgures, they look much more smlar to each other than the multpath resdual sequences for Day 1 to Day 4, whch hghlghts the mportance of carrer phase multpath extracton for relable multpath change detecton. Fgure 8 Multpath Change Detecton Result (Day 2-3).
Fgure 9 Multpath Change Detecton Result (Day 3-4). Table 2 summarzes the standard devatons (STD) of the carrer phase multpath extracton (the frst three rows) and change detecton (the last two rows) results. Note that n the multpath extracton step, the ncoherent output s the uncorrelated nose and the coherent output s multpath whle n the multpath change detecton step the ncoherent output s the change of multpath and the coherent output s correlated multpath. Note that the changes of multpath detected are not zero as they should be, but are 0.966 and 1.010m n the pars Day 2 & Day 3 and Day 3 & Day 4 respectvely, whch are nevertheless smaller than the derved uncorrelated nose. Therefore, a tentatve concluson, smlar to the pseudo-range case, s that any carrer phase multpath change smaller than the uncorrelated nose (manly the recever phase nose) wll go undetected. Table 2 Carrer Phase Multpath Extracton and Change Detecton Result. Day Incoherent Output STD (mm) Coherent Output STD (mm) 2 1.543 2.367 3 1.409 1.866 4 1.457 1.969 2-3 0.966 1.624 3-4 1.010 1.718 Therefore, a concluson for ths secton s that multpath change (ether n pseudo-range or carrer phase) above the recever nose level (n the case of ths study: code: 0.05m; phase: 1.5mm) could be detected usng ths technque.
7. Multpath change detecton n a changed antenna envronment To test the capablty of ths technque n detectng multpath change, contnuous GPS data from two Japanese GEONET statons 960627 and 92110 located n the Tsukuba Cty were used. In both Staton 960627 and Staton 92110, a Trmble 4000SSI recever and a Permanent L1/L2 antenna are nstalled at each staton. The approxmate coordnates X, Y and Z for Staton 960627 are -3957244.2254m, 3310369.6968m, and 3737538.8800m (WGS84). Data from DOY 60 to 99 n the year 1998 are used n ths test. Whle at Staton 92110, the approxmate coordnates X, Y and Z are -3957164.7499m, 3310202.7523m and 3737759.8264m. Data used are from DOY 60 to 88 n the year 1998 (unfortunately the RINEX fles for DOY 89 to 99 were corrupted when wrtng to a CD). Hence the dstance between the two statons s about 286m. An abnormal change as large as 3cm n the tme seres of the baselne length between the two statons was detected by Hatanaka & Fujsaku (1999). In ther study, the elevaton-dependent antenna phase center was compared for both L1 and L2, calculated every 3 hours from 15:00 Japan Standard Tme (JST) on March 5 (DOY 64) to 9:00 JST on March 6. A bg offset was seen n the result for the perod 0:00 to 3:00 on March 6 compared to those of the other segments, whch corresponds to the tme of a snowfall, as llustrated n Fg. 10. Moreover, the phase change on L2 was more sgnfcant that on L1. Fgure 10 Precptaton and temperature n Tsukuba from 5 March (DOY 64) to 6 March (DOY 65) 1998. In order to test whether the adaptve flterng technque can detect ths change n the antenna envronment, the CGPS data were processed followng the procedures outlned n the prevous sectons. Fg. 11 s the pseudo-range multpath STD change on the C1 code at Staton 960627 for satellte PRN25. The total change (the bggest one n the fgure) n the multpath STD from DOY 64 to 65 s 0.19m, whle the second bggest change s 0.10m from DOY 87 to 88, whch s an almost 100% ncrease.
Fgure 11 Pseudo-range multpath change on C1 code at Staton 960627 for satellte PRN25. Fg. 12 s the pseudo-range multpath STD change on the C1 code at Staton 92110 for satellte PRN 22. The total change (the bggest one n the fgure) n the multpath STD from DOY 64 to 65 s almost 0.09m, whle the second bggest change s less than 0.07m from DOY 72 to 73, whch s not as sgnfcant as n Fg. 11. Ths mght ndcate that Staton 92110 s less vulnerable to multpath. Fgure 12 Pseudo-range multpath change on C1 code at Staton 92110 for satellte PRN 22.
Fgs. 13 and 14 are the results of carrer phase multpath change on L1 and L2 respectvely n DD between Statons 92110 and 960627, for satellte par PRNs 22 and 25. In the two results, the bggest changes all occur on DOY 64 to 65, although the L2 result gves a much better sgnal-to-nose rato, whch s n agreement wth the result of Hatanaka & Fujsaku (1999). Fgure 13 Carrer phase multpath change on L1 n DD between Statons 92110 and 960627 for satellte par PRNs 22 and 25. Fgure 14 Carrer phase multpath change on L2 n DD between Statons 92110 and 960627 for satellte par PRNs 22 and 25. Therefore, t seems that when there s ndeed a change n the antenna envronment t wll be detected usng ths technque.
8. Concludng remarks A technque based on an adaptve flter usng the least-mean-square algorthm s proposed for detecton of multpath change at permanent GPS statons. The technque was tested on some expermental data, ndcatng that the multpath change above the recever nose level can be detected. A further test was done wth some CGPS data from the Japanese GEONET when there was a snowfall. The results ndcate that when there s a change n the antenna envronment t wll ndeed be detected n both the pseudo-range and carrer phase data usng the proposed technque. For applcatons of pseudo-range multpath change detecton, data from a sngle dual-frequency recever s suffcent. For applcatons of carrer phase multpath change detecton, data from two closely-located recevers have to be used. Ths may become a lmtaton n a sparse CGPS array. However, n an array such as the GEONET t s always possble to dentfy some reference statons free of multpath, and to generate the RINEX data for a Vrtual Reference Staton (VRS) located close to the CGPS staton targeted for multpath change detecton. For relable multpath change detecton, the temporal resoluton of ths method s 24 hours. Ths technque provdes an easy-to-mplement qualty assurrance tool for CGPS antenna envronment sensng after dsasters such as typhoons, cyclones, and earthquakes. Other possble applcatons nclude the montorng of slope stablty and ground subsdence. ACKNOWLEDGEMENT The authors would lke to thank Mr Lwen Da at the School of Geomatc Engneerng, The Unversty of New South Wales, for processng the double-dfferenced carrer phase observaton resduals usng hs software. The frst author s supported by a SNAP scholarshp. References Axelrad, P., C.J. Comp and P.F. MacDoran, 1994. Use of sgnal-to-nose rato for multpath error correcton n GPS dfferental phase measurements: methodology and expermental results. 7th Int. Tech. Meetng of The Satellte Dvson of The U.S. Insttute of Navgaton, Salt Lake Cty, Utah, 20-23 September, 655-666. Bshop, G., et al, 1994. Studes and Performance of a New Technque for Mtgaton of Pseudorange Multpath Effects n GPS Ground Statons, Proceedngs of the 1994 ION Natonal Techncal Meetng, San Dego, CA, January, pp. 231-242. Bücherl, A., H. Landau, C. Pagels, U. Vollath, and B. Wagner, 2000. Long Range RTK Postonng Usng Vrtual Reference Statons. The Satellte Dvson of the Insttute of Navgaton 13th Internatonal Techncal Meetng (ION GPS 2000), September 19 22, Salt Lake Cty, Utah, USA. Cohen, C. and B. Parknson, 1991. Mtgatng multpath error n GPS based atttude determnaton. Gudance and Control, 74, Advances n the Astronautcal Scences. van Derendonck, A.J., P.C. Fenton and T.J. Ford, 1992. Theory and performance of narrow correlator spacng n a GPS recever. Navgaton, 39(3), 265-283. Fenton, P.C., W.H. Falkenberg, T.J. Ford, K.K. Ng, and A.J. van Derendonck, 1991. NovAtel s GPS recever: the hgh performance OEM sensor of the future. 4th Int. Tech. Meetng of the Satellte Dvson of The U.S. Insttute of Navgaton, Albuquerque, New Mexco, 11-13 September, 49-58. Gabrel, A.G., R.M. Goldsten, and H.A. Zebker, 1989. Mappng small elevaton changes over large areas: Dfferental radar nterferometry. J. Geophys. Res. 94, No. B7, 9183-91.
Ge, L., 1999. GPS Sesmometer and ts Sgnal Extracton. The Satellte Dvson of the Insttute of Navgaton 12th Internatonal Techncal Meetng (ION GPS '99), September 14-17, Nashvlle, Tennessee, USA. Ge, L., Han, S., & Rzos, C., 2000. Multpath mtgaton of contnuous GPS measurements usng an adaptve flter. GPS Solutons, 4(2), 19-30. Haj, G.A., 1990. The multpath smulator: a tool toward controllng multpath. 2nd Symp. on GPS Applcatons n Space, Hanscom AFB, February. Han, S. and C. Rzos, 1997. Multpath effects on GPS n mne envronments. Xth Int. Congress of the Int. Socety for Mne Surveyng, Fremantle, Australa, 2-6 November, 447-457. Hatanaka, Y. and J. Fujsaku, 1999. Error n baselne solutons of GEONET due to snow accumulaton. 92nd Meetng of the Geodetc Socety of Japan, November 9 11, Neo, Gfu, Japan. Haykn, S., 1996. Adaptve flter theory. Prentce Hall, New Jersey, 3rd ed, 989pp. Ln, L.S. and C. Rzos, 1997. Use of multpath template technque for mtgatng GPS pseudo-range multpath: methodology and test results. Frst Trans Tasman Surveyors Conf., Newcastle, Australa, 12-18 Aprl, Paper no. 22, 1-9. Meertens, C, 2000. The Antenna and Multpath Calbraton System webste: http://www.unavco.ucar.edu/projects/actve_projects/amcs. Townsend, B. & P.C. Fenton, 1994. A practcal approach to the reducton of pseudo-range multpath errors n an L1 GPS recever. 7th Int. Tech. Meetng of the Satellte Dvson of The U.S. Insttute of Navgaton, Salt Lake Cty, Utah, 20-23 September, 143-148. Wannnger, L., and M. May, 2000. Carrer Phase Multpath Calbraton of GPS Reference Statons. The Satellte Dvson of the Insttute of Navgaton 13th Internatonal Techncal Meetng (ION GPS 2000), September 19 22, Salt Lake Cty, Utah, USA.