Multipath Mitigation for Bridge Deformation Monitoring

Transcription

1 Journl of Globl Positioning Systems (22) Vol. 1, No. 1: Multipth Mitigtion for Bridge Deformtion Monitoring G. W. Roberts, X. Meng, A. H. Dodson, E. Cosser Institute of Engineering Surveying nd Spce Geodesy, The University of Nottinghm, University Prk, Nottinghm. NG7 2RD. UK e-mil: Tel: +44() ; Fx: +44() Received: 24 June 22 / Accepted: 1 July 22 Abstrct. GPS crrier phse multipth with vrying mplitudes of up to severl centimetres nd periods of couple of minutes is mor error source, which ffects the correct interprettion of bridge deformtion. In this pper, recursive dptive filtering (AF) lgorithm hs been employed to mitigte multipth signture in the coordinte time series of GPS solutions. In order to mximise the suppression of the multipth signture, exct lignment of the input time series into the AF system is crucil. An lgorithm using the crosscorreltion coefficient of dy-to-dy GPS coordinte time series is presented to lign GPS coordintes. To isolte multipth from the contminted GPS coordinte time series, reltive displcements clculted from the ccelertions sensed simultneously with GPS receiver by trixil ccelerometer housed in specilly designed cge is used s the reference signl sequence within the AF system. Associted lgorithm for the reltive velocity nd displcement clcultion is lso introduced in the pper. It demonstrtes tht it is possible to chieve millimetre positioning ccurcy by the AF pproch nd n integrted sensor system of GPS receiver nd trixil ccelerometer. Key words: Multipth mitigtion, Structurl deformtion monitoring, Adptive filtering, GPS nd ccelerometer integrtion, Cross-correltion. 1 Introduction With the development of rel-time kinemtic (RTK) GPS receiver nd ntenn technology, GPS is currently used in res where high mesurement precision is required within high dynmic environment. Attempts hve been mde in recent yers to investigte the fesibility of pplying GPS technology to monitor structurl deformtions nd dynmic responses to ctive lodings, such s wind, temperture chnge, trffic, nd even erthqukes (Duff et l. 1998; Og et l. 21; Roberts et l. 1999; Roberts et l. 2). The dvnce of dt processing lgorithms further mkes detiled component nlysis nd dignosis of structurl deflection possible (Meng 22). It provides the opportunity for GPS reserchers to conduct subtle studies into the potentil error sources nd their impcts on the dt qulity. Furthermore, it mkes correct error modelling possible from such component nlysis. Of mny error sources relevnt to stellite geometry nd signl propgtion medi, multipth is still reserch focus for precise GPS positioning (Rizos 1999). The proposed pproches for mitigting multipth effects re fr from prcticl nd effective, especilly when GPS is pplied to monitor dynmic structurl deformtion. Ge et l. (22) present simple pproch using n dptive finite impulse response (FIR) filtering technique to reduce the impcts of multipth on continuous GPS (CGPS) sites. Since 1993, the Institute of Engineering Surveying nd Spce Geodesy (IESSG) t the University of Nottinghm hs initilised studies of the pplictions of kinemtic GPS on structurl deformtion monitoring (Ashkenzi et l. 1997). The reserch emphses in the pst were on the implementtions of GPS-bsed monitoring system, dt collection, nd simple nlysis. Further work in this re discloses tht n integrted system with other sensors is necessry in providing precise mesurements nd hence robust monitoring system. It ws lso recognised tht sophisticted signl identifiction, dt processing, nd deformtion nlysis techniques re essentil. Recent reserch focuses of the IESSG in this field re on sensor integrtion of trixil ccelerometers with dul frequency geodetic GPS receivers (Roberts et l. 21) nd even more recently with single frequency geodetic GPS receivers. A new multipth mitigtion lgorithm bsed on recursive dptive FIR filtering (AF) in the dynmic environment of bridge deflection monitoring scenrio is

2 26 Journl of Globl Positioning Systems proposed by Dodson et l. (21) nd further studied by Meng et l. (21) nd Meng (22). In this pper, the fundmentl of the AF technique is briefly reviewed. GPS dt collected on the turret of the IESSG building with Leic SR51 single frequency GPS receivers nd ssocited ntenns re employed to investigte the dy-to-dy multipth signture. A crosscorreltion lgorithm is used to estimte the time shift of the dy-to-dy position time series. The multipth pttern isolted from normlly time shifted coordinte sequences of two consecutive dys (four minutes) is compred with the exctly ligned coordinte time series, ccording to the estimted time shift of ech individul direction in WGS84 during the two dys. It ttempts to emphsise the importnce of ligning the input signls to n AF system. As n lterntive to mitigte multipth, simple lgorithm is presented in the pper to clculte the reltive displcements from the ccelertions sensed by trixil ccelerometer, which is physiclly connected with the GPS ntenn by specilly designed cge. Spectrl nlysis pproch is pplied to the input nd output time series to evlute the efficiency of dptive filtering in suppressing multipth in dynmic environment. Through the bove geodetic signl dignosis nd nlysis, millimetre positioning ccurcy cn be chieved which further confirms tht with the proposed procedure even in multipth hostile environment, n integrted sensor system of GPS receivers nd ccelerometers cn still provide robust nd highly ccurte positioning solution. 2 Alignment of Dy-to-dy GPS Position Solutions environment is the min purpose of pplying AF pproch in the dt processing. Figure 1 is simple schemtic of n dptive system tht consists of n FIR filter (processor) nd n dptive lgorithm. d is the ppliction provided input signl or desired sequence. It cn be the rel mesurement time series for specific process tht cn be compred with the FIR filter predicted output y. x is the reference signl which is used to output prediction y with Eq. 1. In Eq. 1, reference sequence X including the current input x with sequence length Q, which is the filter length, is employed to estimte instntneous prediction t epoch n. Prmeter b is the filter coefficient. Q 1 q q= y ( n) = b x( n q) (1) In the ctul clcultion, it is possible to strt with rbitrry initil vlues of filter coefficient sequence v T B = b, b, b,..., ). Then s ech new input smple x(n) ( 1 2 enters the dptive filter, the corresponding output or filter prediction b Q y(n) is mde nd compred with d(n). The error signl e(n ) = d( n) y( n) is formed, nd used to updte the filter coefficients bsed on the method of steepest descent using Eq. 2 (Hykin 21). µ is prmeter tht controls the rte of convergence of n AF pproch. The updted filter coefficients re then employed in the AF processor to predict the next signl output with the coming of new reference signl by n dptive lgorithm. v v r Bn+ 1 = Bn + 2µ e( n) X n (2) 2.1 Fundmentl of Adptive Filtering (AF) Multipth, receiver rndom noise, nd unmodelled reltive tropospheric dely re mor error sources under the scenrio of bridge deformtion monitoring s nlysed by Meng (22). Multipth impct on the positioning solution is relted to the repetition of the stellite constelltion between two siderel dys. This chrcteristic cn be exploited to extrct multipth signture from the positioning time series. If closely setup reference sttions re employed nd the height differences between the sttions re within severl tens of metres, the unmodelled tropospheric dely cn be resonbly neglected. Otherwise meteorologicl dt should be collected to cope with these residuls. Receiver rndom noise chrcterises pure white noise if systemtic errors re properly modelled. So, in this cse, three potentil components form the positioning solution sequence, which re the rel bridge movement, receiver rndom noise, nd multipth signture. Isoltion of these components in dynmic bridge deformtion Fig. 1 Schemtic of n AF system In the ppliction of AF pproch, filter length nd convergence prmeter re two fctors tht ffect the performnce of n AF system. The determintions of the optiml vlues of these prmeters re discussed by Meng (22).

3 Roberts: Multipth Mitigtion for Bridge Deformtion Monitoring Result Comprison of Different Alignment Approches As explined by Roberts et l. (22), finding the exct mtch point of two time series is very importnt in order to mitigte multipth nd isolte deformtion effectively. Suppose two time series d nd i x, which could be the coordinte time series t one observtion site on two consecutive dys, or the rw pseudornge mesurements. d nd x could be sme length vectors or vectors of i i different length. In the AF pproch, only the sme length vectors re used s desired nd reference input signls. Dt from the first time series, of pproximtely 2 minute period, is chopped out to be compred with the second time series with longer period for the clcultion of ech epoch s cross-correltion coefficient. In this pper the dt used re the finl coordinte time series vi postprocessing. The coordinte sequence strting t epoch of GPS time with 2-minute intervl on the first dy t dy1 is designted s d i dy 1 i. The dt used on the second dy is t dy1 t dy1 coordinte sequence of four minutes within the intervl of [ +864, +8628]. The coefficient of crosscorreltion t ech epoch ( σ ) d x cn be estimted by dx ρ = (3) where ( σ ) ( σ ) (σ ) d dx (σ ) (σ ) x = N = = = 1 N = 1 N = 1 ( d d )( x x )) ( d d )( d d )) (( x x )( x x )) re the covrince nd stndrd devitions of the vrinces of two time series, respectively. ρ The epoch with mximum is then used s the mtch point of the two time series. Using this pproch nd rw pseudornge mesurement, Roberts et l. (22) illustrte the dy-to-dy time shifts of ech PRN stellite by using pseudornge mesurements. The results revel there re differences in the time shifts for ech individul stellite. These estimted time shifts could be employed to mitigte multipth in pseudornge mesurements. A new RINEX dt file could be formed by the clened pseudornge mesurements with the potentil ppliction in shortening mbiguity serch time. Since it is very difficult to correct the multipth impct directly from the crrier phse mesurements by using dy-to-dy pproch, it limits the pplictions of GPS in mny res where high ccurte positioning solutions re required in rel-time mode. Therefore, the finl positioning solutions vi post-processing of rw mesurements re usully used to estimte the time shifts in three dimensions in specific coordinte system for mitigting coordinte multipth. To evlute the performnce of Leic single frequency SR51 receivers, zero bseline nd short bseline tests using the IESSG geodetic fcilities were conducted over period of two-weeks. Three kinds of Leic GPS ntenns were used for two consecutive dys in the short bseline tests with smpling rte of 1 Hz. The clculted time shifts re listed in Tble 1. The differences of time shifts on X nd Y coordintes in WGS84 re ll within 1 second rnge for three types of Leic ntenns, but the time shift differences between Z with X or Y could rech severl seconds. Also the results show tht the time shifts bsed on the dy-to-dy position solutions re not necessrily 4 minutes. The clculted mximum cross-correltion coefficients of X, Y nd Z using SR51 single frequency receivers nd n AT51 single frequency ntenns re.38,.35 nd.74, which re different from those of AT53 lightweight chokering ntenn, which re.25,.65, nd.51, due to the Tb. 1 Time shifts for ech Leic ntenn in three directions (WGS84) X Y Z AT51 (Smple 1) AT51 (Smple 2) AT AT differences in signl reception nd filtering. Further studies hve been mde on the Z coordintes in WGS84 to evlute the impct of misligned dt on the efficiency of the AF pproch. The pproch nd lgorithm used here re similr to those illustrted by Dodson et l. (21). Figure 2 nd 3 re the results from two dys normlly (4 minutes) nd exctly (4 minutes nd 7.8 seconds) ligned dt sets using the SR51 receiver nd the AT51 ntenn (Smple 1 in Tble 1). The first rows in both grphs re the desired signls of the AF lgorithm for bout 2, second period with 1 Hz smple rte, which represent the Z coordinte time series t the sme loction. The second rows re the time shifted reference signls of the following dy s Z coordinte time series. The third rows in Figure 2 nd 3 re the uncorrelted signl outputs, which contin minly receiver noise for this short bseline nd unfiltered multipth residuls. The finl rows re the correlted components, which represent the multipth signtures. The residuls re clculted by compring the corresponding output signls to evlute the impct of

4 28 Journl of Globl Positioning Systems Residul of Receiver Noise from AF Dy-to-Dy Templtes (From Exctly nd Normlly Aligned Dt Sets with AT51).15.1 Noise Residul (m) misligned time series on the efficiency of the AF pproch. Figure 4 nd 5 re these residuls for the uncorrelted components (receiver noise plus multipth residul) nd correlted prts (multipth) of both time series. To further investigte the impcts due to the different ntenn types, the sme procedure is pplied to the coordinte solution from AT54 nd AT53 choke ring ntenns with SR51 receivers No of Smples Fig. 5 Residul receiver noise due to mislignment (ntenn AT51) Figure 6 nd 7 re the residuls of multipth nd receiver noise for AT54 chokering ntenn. It is obvious even with single frequency receiver nd misligned time series tht if chokering ntenn is dopted for dt collection, significnt improvement in position solutions cn be expected. Residul of Receiver Noise from AF Dy to Dy Templtes (From Exctly Aligned nd Normlly Aligned Dt Sets with AT54).15 Fig. 2 AF results from normlly ligned Z coordintes (AT51) Noise Residul (m) No of Smples Fig. 6 Residul multipth due to mislignment (ntenn AT54) Multipth Residul from AF Dy to Dy Templtes (From Exctly nd Normlly Aligned Dt Sets with AT 54).15 Multipth Residul (m).1 Fig. 3 AF results from exctly ligned Z coordintes (AT51) No of Smples Multipth Residul Clculted from AF Dy-to-Dy Templtes (from Exctly nd Normlly Aligned Dt Sets with AT51) Fig. 7 Residul receiver noise due to mislignment (ntenn AT54).15 Multipth Residul (m).1 3 Estimting Reltive Displcements from Accelertions No of Smples Fig. 4 Residul multipth due to mislignment (ntenn AT51) The rw mesurements from trixil ccelerometer re the discrete outputs in voltges on ech xis vi n nlogue to digitl converter. These dt cn be further converted into ccelertions using the zero bises nd scle fctors corresponding to ech xis. To reduce the computtion burden in coordinte trnsformtions between different systems, specilly designed cge is used to house GPS ntenn nd ccelerometer (Roberts

5 Roberts: Multipth Mitigtion for Bridge Deformtion Monitoring 29 et l. 21). Figure 8 is the schemtic of this device. It consists of two rotry pltes connected with three bolts. The GPS ntenn is mounted on the upper plte, which cn be orientted to north. The trixil ccelerometer is fixed on the second plte with four screws. By rotting the second plte one of the three xes of trixil ccelerometer cn be ligned into the sme direction of bridge min xis (longitudinl direction). Fig. 8 A cge used to house GPS ntenn nd trixil ccelerometer Even though the two kinds of sensors re physiclly ttched together, the direct comprison of GPS coordintes nd ccelertions using the AF pproch is impossible due to different mesurement units. The ccelertions need to be converted into reltive displcements by double integrl. The ccelerometer errors such s zero bises nd scle fctor errors will ccumulte into distnce errors ccording to the eqution of motion (Lwrence 1998) S + 2 = vt 1/ 2t (4) In Eq. 4, v is the initil velocity; is the ccelertion mesurement; nd S is the distnce trvelled in time period t. Eq. 5 is used to clculte velocity from ccelertion, n pproch to pproximte ccelertion integrl, t v ( t) = v( t 1) + ( ( t) + ( t 1)) (5) 2 where v( t), v( t 1), ( t), ( t 1) re the velocities nd ccelertions t time t nd t 1, respectively. t is the time intervl of dt smpling. Eq. 5 cn lso be pplied to velocities clculted previously to output reltive displcements. Since the interest of the reserch is on the reltive displcements, the initil velocity cn be set to zero. A moving verging (MA) filter is pplied to cope with the drift problems of the velocities nd displcements in the pproximtion procedure using Eq. 5. In principle, the MA pproch is low-pss filter. It cn be used to isolte longer period movements from the time series of the integrted velocity nd displcement, which re contminted by the systemtic errors of the ccelerometer. The residul between the originl time series nd the smoothed output from the MA constitutes the locl higher frequency vrition (high-pss filter), which minly represents rel structurl vibrtion. Simultion revels tht the selection of smple number used for verging is crucil. This requires considerble experience plus knowledge of the frequency distribution of the time series. Spectrl nlysis is used for the purpose of frequency identifiction. In the spectrl nlysis of the bridge dt gthered from GPS nd the ccelerometer, the frequency of this reserch interest is below 2 Hz. In order to use the reltive displcements clculted from ccelerometer dt s reference signl to the AF pproch, the ccelertions re resmpled from 2 Hz down to 1 Hz. In this resmpling procedure, due to the low-pss nture of resmpling s n MA filter, ccelerometer noise or outliers with high frequency re filtered out, which cn be noticed from the following dt processing. 1 + s MA( ) = t s + q + (6) 1 r r = q MA( t ) is the smoothed output for time series time t. cn be ccelertion or filtered velocity. Bsed on the bove fundmentl, Mtlb script used for reltive displcement clcultion ws developed, with functionlities of MA smple length selection, velocity nd reltive displcement clcultions nd spectrl nlysis of output signls. Figure 9 is the comprison of originl verticl ccelertions smpled with 2 Hz nd the resmpled verticl ccelertions of 1 Hz. It is obvious tht most bnorml ccelertion mesurements hve been filtered out through the resmpling procedure. The dt set illustrted here is chopped from one-hour ccelerometer dt set, which ws collected on 2 Februry 21 on the Wilford suspension footbridge over the River Trent in Nottinghm, UK. The detiled experiment ws introduced by Dodson et l. (21). To investigte the bridge dynmic responses to the pedestrins, 8 people wlked repetedly over the bridge, which cused significnt ccelertion signtures. Figure 1 nd 11 re the time series of the reltive verticl velocity nd displcement by using the forementioned lgorithm. The first grph in Figure 1 is the integrl of the resmpled ccelertions, which re illustrted by second grph in Figure 9. The second grph in Figure 1 is the output of MA, which is the smoothed drift cused by integrl. The third grph is the clculted reltive velocity ctivted by pedestrins. Similr grphs t

6 3 Journl of Globl Positioning Systems for reltive displcement re demonstrted in Figure 11. The drift due to the double integrl of ccelertions hs been successfully overcome. integrted sensor system of GPS receivers nd trixil ccelerometers could be employed. With current GPS technology, 2 Hz smpling rte cn be immeditely vilble which cn be used to detect the vibrtion frequencies lower thn 1 Hz. Under this circumstnce, n integrted monitoring system of GPS receiver nd n ccelerometer cn provide broder detectble frequency bnd except for outputting more mesurements. Detiled nlysis to the frequency distribution is illustrted lter in the pper. Fig. 9 Originl ccelertion vs. resmpled ones Fig. 12 More signtures cn be identified with n integrted system 4 Accelertion Aided AF Approch for Multipth Mitigtion Fig. 1 MA pproch for velocity clcultion Fig. 11 MA pproch for displcement clcultion Figure 12 is the result of spectrl nlysis on the originl verticl ccelertion sensed by Kistler trixil ccelerometer. The min verticl vibrtion frequencies identified by the ccelerometer re 1.75 Hz, 2 Hz, 3 Hz, 5 Hz, 9 Hz, 11 Hz, nd 7Hz. By the principles of spectrl nlysis, the vibrtions with frequencies higher thn 5 Hz cnnot be detected by 1 Hz GPS receivers tking ccount of other error sources of GPS system lone (Jenkins nd Wtts 1968). To overcome this problem, either GPS receivers with higher smpling rte or n The reltive displcements clculted from the ccelertions by the proposed lgorithm cn be further used to mitigte GPS receiver noise nd isolte the rel bridge movements. Due to the high degree of multipth signture cused by the surroundings nd receiver noise, ccelerometer ided AF pproch cnnot be pplied to GPS position solutions directly. Creful tretment is required to filter the multipth nd receiver noise both t reference sttions nd rover sttions. Meng et l. (21) presented hierrchicl procedure to grdully reduce the impcts of multipth nd receiver noise, in ddition to modelling the reltive tropospheric dely between reference sttion network. Even with such procedure, the finl coordintes re still contminted by certin degree of the residul components forementioned. Since the GPS nd ccelerometer re independent instrument, the only correlted component in both systems is the sensed force/movement. So, it is n idel pproch to use ccelerometer ided AF technique to isolte reltive movement of the bridge. Figure 13 is the AF results using the verticl coordinte of two dys s the inputs signls t the sme observtion site. The two dys dt re exctly time shifted with the pproch proposed in this reserch. In the reference time series, there were signtures cused by csul pedestrins pssing over the bridge but in the desired signl there were vibrtions induced by orgnised crossing. Figure 14

7 Roberts: Multipth Mitigtion for Bridge Deformtion Monitoring 31 is the frequecy distribution of desired signl. The power of multipth is much higher thn tht of the ctul bridge movement nd exhibits itself s very slow movement pttern in Figure 13. It is very difficult to identfy the excited movement without ny further dt tretment. From Figure 14 the dominnt multipth frequencies identified by spectrl nysis re.38 Hz (4 minutes 24 seconds) nd.75 Hz (2 minutes 13 seconds). Fig. 13 GPS AF results using two dy's position solution Fig. 16 Spectrl of isolted multipth signture Figure 17 is the output using the reltive multipth free GPS position solution (the third row in Figure 13 which constitutes receiver rndom noise nd rel movement) s the desired signl nd the clculted reltive bridge displcement from ccelerometer s the reference signl in the AF lgorithm. The synchronistion between the dt sets of GPS nd ccelerometer is relised in the softwre whose detil is introduced by Meng (22). The outputs of Figure 17 re the receiver noise time series (third row) nd reltive bridge movement (fourth row). Figure 18 illustrtes the results of using untreted positioning solution from GPS directly to compre with the reltive bridge movement sensed by the ccelerometer. It shows tht it is possible to isolte multipth from the rel bridge movement but the detected rel bridge movement is distorted t certin periods due to the error is too big to be dpted. Fig. 14 Spectrl distribution of desired signl Figure 15 is the spectrum of the reference signl. There is no signture of 1.75 Hz verticl movement since there ws no excittion cused by orgnised crossing, but the sme frequency signture of multipth is lmost filtered out through the AF procedure (Figure 16). Fig. 15 Spectrl distribution of reference signl Fig. 17 Accelertion ided AF result with multipth free GPS position Further nlysis hs been mde on the frequency distributions of the input nd output signls to nd from ccelertion ided AF system by spectrl nlysis. Figure 19 is the spectrum of the desired signl, which is the third row in Figure 13. It is evident tht significnt power reduction of multipth is relised through the AF procedure. Figure 2 is the spectrum of the reference signl of the reltive displcement from resmpled ccelertions. It is obvious tht the resmple procedure does not chnge the spectrum distribution of ccelertion.

8 32 Journl of Globl Positioning Systems resech is needed to identify the reson why the frequencies identified by ccelerometer nd within the meurble rnge of 1 Hz GPS receivers hve been filtered out. Fig. 18 Accelertion ided AF result with untreted position Fig. 21 Receiver rndom noise signture Fig. 19 Spectrum of desired signl (treted GPS dt) Fig. 22 Spectrum of the integrted system 5 Summry Fig. 2 Spectrum of reference signl (reltive displcement derived from ccelertion) GPS receiver noise is the uncorrelted output vi the AF procedure nd its spectrum is illustrted by Figure 21. It provides n opportunity to further nlyse the sttisticl nture of GPS receiver noise. Figure 22 is spectrum of the finl integrted reltive displcement (Figure 17 fourth row) from the two sensors. It illustrtes tht the instrument relted errors hve been removed successfully through the bove dt processing procedure. It needs to be pointed out tht by this proposed sensor system, the detectble frequency is limited to the smpling frequency of GPS receivers nd other signtures re left undetectble due to the unfiltered GPS noises nd the wek power of these rel vibrtion frequencies. Further Fundmentl of recursive dptive filtering (AF) pproch is reviewed first in the pper. Importnt issues such s filter length of n AF system, convergence of AF lgorithm, nd time series lignment re ddressed. The focus is on the time series lignment nd the consequence of mislignment of time series to the positioning solutions. The exct dy-to-dy time shifts of the whole stellite constelltion re estimted with the GPS positioning solutions. A cross-correltion lgorithm is used to detect the mtch point of two time series. The clculted results revel tht the time shifts between X, Y, nd Z re different. The AF result differences by ligning two dys time series using norml time shift (4 minutes) nd exct time shift hve been compred. Two centimetre differences in residul noise nd multipth re evident in the positioning solution from single frequency ntenn AT51 nd one centimetre for chokering ntenn AT 54. By using AF s n error suppression tool the uthors demonstrte n ccelertion ided AF pproch to isolte

9 Roberts: Multipth Mitigtion for Bridge Deformtion Monitoring 33 rel bridge movement from noise polluted GPS positioning solutions. The moving verge (MA) lgorithm hs been pplied to the originl ccelertion to clculte the reltive displcement by double integrl nd cope with the long period drift cused by ccelertion double integrl. Spectrl nlysis is pplied s signl dignosis tool to check the efficiency of the AF lgorithm in the dt processing. The results illustrte tht through the AF procedure, the errors from both sensors hve been removed nd rel bridge movements hve been isolted. It is possible to chieve millimetre positioning ccurcy in 3D fter the removl of receiver noise nd residul multipth by the proposed pproch. However, the comprison of the frequencies detected from the originl ccelerometer dt which re within the detectble frequency bnd of 1 Hz GPS receivers nd those from the proposed integrted sensor system of dul frequency GPS receivers nd trixil ccelerometer demonstrtes tht some of the detectble components were filtered out. The possible reson for this is tht these detectble frequencies hve been removed during MA procedure which is illustrted very cler in the reltive displcement illustrted by Figure 2. Since there re no such components in the reltive displcements clculted from ccelertions with those from GPS mesurements, only the mode shpe chrterised the sme vibrtion of the first nturl frequency is detected by AF technique. Further reserch is needed to investigte the possible resons for this nd ssocite dt processing techniques. Acknowledgements Leic Geosystems Ltd. (UK) sponsored the initil work in this re through scholrship to support Xiolin s PhD reserch study t the IESSG. The UK s Engineering nd Physicl Sciences Reserch Council (EPSRC) is cknowledged for supplying three-yer grnt to the uthors to conduct reserch into using integrted GPS nd ccelerometers to evlute the integrity of structures. References Ashkenzi V., Dodson A., Moore T., nd Roberts G. W. (1997) Monitoring the Movements of Bridges by GPS. ION GPS'97, 1th Int. Tech. Meeting of the St. Div. of the U.S. Inst. of Nvigtion, September 1997, Knss City, USA, Dodson A. H., Meng X., nd Roberts G. W. (21) Adptive Method for Multipth Mitigtion nd Its Applictions for Structurl Deflection Monitoring. Interntionl Symposium on Kinemtic Systems in Geodesy, Geomtics nd Nvigtion, 5-8 June 21, Bnff, Albert, Cnd, Duff K., Hyzk M., nd Tucker D. (1998) Rel-time Deformtion Monitoring with GPS: Cpbilities nd Limittions. SPIE 333, Smrt Structures nd Mterils 1998: Sensory Phenomen nd Mesurement Instrumenttion for Smrt Structures nd Mterils, Hykin S. (21) Adptive Filter Theory. 4th ed. Prentice-Hll, Upper Sddle River, New Jersey. 92p. Ge L., Hn S., nd Rizos C. (22) GPS Multipth Chnge Detection in Permnent GPS Sttions. Survey Review, 36(283): Jenkins G. nd Wtts D. (1968) Spectrl Anlysis nd Its Applictions. Holden-Dy, Cliforni. 525p. Lwrence, A. (1998) Modern Inertil Technology: Nvigtion, Guidnce, nd Control. 2nd ed. Springer, New York. 278p. Meng X., Roberts G. W., nd Dodson A. H. (21) Error Anlysis nd Positioning Precision Improvement in GPS/Accelerometer Structurl Monitoring. Poster of IAG 21 Scientific Assembly, 2-7 September 21, Budpest, Hungry. Meng X. (22) Rel-time Deformtion Monitoring of Bridges Using GPS/Accelerometers. PhD Thesis, The University of Nottinghm. 239p. Og C., Rizos C., Wng J., nd Brownohn J. (21) Towrd the Implementtion of on-line Structurl Monitoring Using RTK-GPS nd Anlysis of Results Using the Wvelet Trnsform. Deformtion Mesurements nd Anlysis, 1th Interntionl Symposium on Deformtion Mesurements, Mrch 21, Ornge, Cliforni, USA, Rizos C. (1999) Qulity Issues in Rel-time GPS Positioning. IUGG Congress, Birminghm. Roberts G. W., Dodson A. H., nd Ashkenzi V. (1999) Twist nd Deflection: Monitoring Motion of Humber Bridge. GPS World, 1(1):