1 INTRODUCTION: PERFORMANCE MONITORING OF STRUCTURES Performance monitoring of as-built, full-scale struc- tures has become increasingly popular due to devel- opments in design philosophy, changes in perform- ance requirements and developments in structural health monitoring (SHM). Performance is measured either as displacement or acceleration resulting from wind or seismic actions and design aims to provide adequate resistance against various degrees of fail- ure. Full-scale monitoring has been used to calibrate wind codes (Lam & Lam 1979, Littler & Ellis, 1990), to provide information about structural and foundation system performance that can be used for other designs and to provide indication of changes to structural state, including damage (Jeary et al., 2001, Stewart et al., 2005). There has been long experience of performance monitoring of tall buildings, but generally long span bridges have been the subjects of the greatest ad- vances in monitoring technology. They are mostly public owned, accessible, visible and high profile. Their designs have usually resulted from elaborate experimental and numerical simulation exercises and there is still much to be learnt about load and re- sponse mechanisms. They also have large mainte- nance requirements and performance monitoring via SHM systems is now seen as the way forward in as- sisting lifetime management of these major assets. Bridge SHM systems have become very elaborate (Wong, 2003, Le Diourion, 2005) and also provide opportunities to evaluate newly developed sensors and conduct research. A major target for bridge monitoring is the mo- tion of the deck structure, which for long span sus- pension bridges can be of the order of a metre. Ver- tical, lateral, torsional and even longitudinal response is driven by a combination of wind, traffic and thermal actions Brownjohn et al. 1994) and measurement of ‘static’ and ‘dynamic’ components provides information on structural and loading mechanisms. ‘Dynamic’ implies modal response while ‘static’ relates to non-modal response in the frequency range up to fundamental modes. For long span bridge, with first mode frequencies so far as low as 0.05Hz, recovery of displacements from ac- celerometers is prone to a range of errors. Primarily, broadband signal noise can result, after double inte- gration, in large artificial low frequency displace- ments. This can be minimized with high specifica- Experience with RTK-GPS system for monitoring wind and seismic effects on a tall building J.M.W. Brownjohn University of Sheffield, United Kingdom M. Stringer, G. H. Tan, Y.K. Poh Syseng(S) Pte Ltd, Singapore L. Ge University of New South Wales, Australia T. C. Pan Nanyang Technological University, Singapore ABSTRACT: A dual-rover monitoring system using GPS to track displacements was installed on a tall building in 2000 and integrated with an existing vibration monitoring system. The original purpose of the sys- tem was to identify the slowly varying quasi-dynamic component of wind induced response in order to com- pare with seismic effects on the building. The system operated in real time, collecting data synchronous with acceleration signals, as well as off-line, through post processing of large raw data files. There were numerous technical problems with the system, partly due to the inherent limitations on capability for displacement reso- lution and the nature of the signal noise, leading to questions about the validity of the larger non-dynamic movements. Nevertheless, the system has been able to detect and resolved dynamic response to both wind and seismic effects. The paper described the practical implementation of the system, the problems encountered and some of the results. tion DC (servo) accelerometers, noise free cables and highly stable signal conditioning but there is no way to avoid kinematic effects due to cross-coupling between motion in different direction. In particular, a DC accelerometer mounted horizontally to meas- ure lateral vibrations will sense rotation such that a lateral mode involving swinging (variation of angle of attack with lateral deflection) will be unable to distinguish the lateral component of deflection. Hy- draulic level sensors have been used to measure ver- tical deflection (Wong, 2003, List, 2004) but are not favoured. Faced with the difficulties of low fre- quency multi-component motion, an obvious solu- tion is to use an optical system (Marecos et al., 1969). Apart from adaptations of surveying equip- ment, solutions have been available using light beams (Ahola & Tervaskanto, 1991) or image proc- essing (Brownjohn et al., 1994). Optical systems are challenged by adverse weather and long lines of sight so the promise of GPS to provide absolute displacement signals unaffected by weather, electri- cal noise or complex kinematics has been welcomed by a number of bridge operators. So far, however, only a few experiences have been reported (Ashken- zia & Roberts, 1997, Nakamura, 2000, Guo et al., 2005, Wong et al., 2000, Kashima et al., 2001). GPS units were temporarily installed on the deck and towers of Humber Bridge (Ashkenazi & Rob- erts, 1997), and they showed the same characteristic response obtained using optical systems a few years ealier (Brownjohn et al., 1994), thus demonstrating the capability of the system to track displacements. For tall buildings the problems are more acute. While the frequencies are higher, the action- response patterns are more complex, movements are smaller then for a flexible long-span bridge, and op- tical systems are problematic due to lack of reliable line of sight, particular in built-up city areas. There is, however, as much demand for the ability to measure an absolute deflection. While accelerome- ters are completely capable to measure sway vibra- tion response, they cannot detect changes in building tilt resulting from bending in wind or due to thermal effects, or even subtle shifts in foundation. The ex- tent of quasi-static wind-induced building tilt is han- dled semi-empirically by loading codes and repre- sents a large potential source of design error, while permanent set or tilt resulting from varying ground movement can be used as a measure of structural in- tegrity. There have been a limited number of build- ing studies (Li et al., 2004, Pagnini et al., 2002, Lo- yse et al., 1995) but so far the most promising would be the Chicago Experiment (Kijewski-Correa & Ka- reem, 2003) involving instrumentation of three Towers with GPS, accelerometers and anemometers. A GPS system was installed on the 280m Repub- lic Plaza office tower in Singapore (Figure 1) be- tween November 2001 and January 2005 with the aim of providing direct measurement of building de- flections to calibrate loading code predictions of the relative contributions of dynamic and quasi-static wind-induced deflections. The experiment was as much to test the capability of GPS to resolve the relatively small deflections expected as to character- ize the loading and response. Figure 1 View of completed building 2 BUILDING CONFIGURATION A full description of the building is provided elsewhere (Brownjohn et al., 1998). In summary, the tower has a frame-tube structural system with an in- ternal reinforced concrete core wall connected to a ring of 16 external steel columns by horizontal steel framing system at every floor, all supported by a deep caisson system, and with diagonal outriggers provided at only two floor levels. The fundamental natural frequency for the building was predicted by the architect to be 0.14Hz, and as well as meting lo- cal design requirements on resistance to design winds (up to 35m/sec gust at 10m height) and acci- dental eccentricity (equivalent to 1.5% of weight as a lateral force), a vibration serviceability require- ment of 100mm/sec2 maximum acceleration was used. 3 PEFORMANCE MONITORING The building had been instrumented from 1996 us- ing a set of four accelerometers, two at basement level, two at roof, and a pair of three-component anemometers perched just above the parapet on each corner of the building. Signals were continually digi- tized at 60Hz per channel before decimation at 8x into blocks of 4096 data samples. Mean and variance values for these 540-second blocks were saved and the raw time history also collected when trigger con- ditions of strong wind or strong acceleration were met. From these data it was possible to determine the correspondence of dynamic response with loading conditions, specifically wind and tremor ex- citation. The only other source of strong response was the roof-mounted crane used for window- cleaning. From the observations it became clear that re- sponse to regional earthquake, occurring at distances of up to 3000km, were the most significant vibration source. The strongest roof level acceleration re- sponse, reaching +/-56mm/sec2, resulted from the June 2000 magnitude 8 earthquake. This response exceeded even that caused by the 2004 Boxing Day earthquake, also epicentred in the Sumatran subduc- tion zone. Compared to these levels, the maximum acceleration due to wind was +/-16mm/sec2. Based on experimentally obtained mode shapes and con- struction material weights provided by the contrac- tor, it has been possible to estimate the correspond- ing peak base shears; these come to 0.88MN for the tremor and 0.38MN for wind. Compared to the 711MN dead weight of the building these are insignificant, but the data lead to a question about the performance of lower rise build- ings responding more heavily to seismic loads. Re- public Plaza has relatively low response since it re- sponds primarily in second mode (around 0.7Hz) that has a lower participation factor than first mode (around 0.19Hz) and it is expected that typical resi- dential structures of around 30 storeys would be more vulnerable, particularly those on soft soil or reclaimed land sites. For tall buildings of this class, there is still a risk, given the sequence of tremors beginning in late 2004, that larger seismic motions could result, yet design only caters for wind load, and this is based on an archaic British design code (CP3). This code is based on an equivalent static load due to a 3-second gust, and it is generally believed that resulting lateral loads govern tall building design in Singapore. The monitoring exercise provided a means to check this belief, using a more rational design ap- proach (e.g. the new Australian code AS1170.2) which is based on a mean wind speed with a dy- namic amplification factor, typically at least 2, that accounts for turbulence, peak factor and resonance. Calculations for Republic Plaza showed that the base shear, and hence the lateral deflection, should comprise up to 50% from static effect of mean wind and at least 50% from background and resonant contributions. The accelerometer-based system could only detect accelerations so the ‘static’ contri- bution could not be measured. Despite the wide- spread use of the static+dynamic load approach, there have (apparently) been less than a handful of exercises to validate it at full-scale and the Republic Plaza exercise provided an opportunity to investigate the nature of the dynamic amplification factor from direct measurement of total displacement. 4 RECOVERY OF AND DYNAMIC DISPLACEMENTS FROM ACCELEROMETER SIGNALS To begin, recovery of displacements from acceler- ometer signals was investigated. The double- integration approach to recover displacements is subject to contamination by noise at very low fre- quencies which translates to high displacements. Experimentation with high pass filtering in the base- line correction process applied to tremor response data has shown that using a cut-off frequency of 0.02Hz or higher the lowest frequency components show the expected rigid body motion of the build- ing; Figure 2 shows time series of double-integrated accelerations resulting from a very distance tremor with a little mode 1 vibration response superposed on the roof level response (upper plot) compared to basement response (lower plot). Using lower fre- quencies generate random artifices (such as in the first part of the upper plot of Figure 2) and since the time taken for build up to peak wind levels even in sudden severe storms was observed to be much longer than 50 seconds, this method cannot be relied on to recover quasi-static response to wind. An alternative method based on acceleration sig- nals assumes that deflection is accompanied by rota- tion which would be evident as a shift in mean ac- celeration of the DC accelerometers. For example if the building behaves as a perfect cantilever with a point load at the tip, the rotation α is related to de- flection δ by 2 3 h δ α = where h is height and g is acceleration due to gravity. Since the accelerometers operate down to 0Hz (DC), a rotation of the accelerometer by a small an- gle α is sensed as a (mean) static acceleration x g α = ⋅  . Using a typical wind profile and an ex- perimentally validate finite element model, then 0.84 0.84 h h x g δ α = =  , a result that cold be used to infer the static drift; a 1mm/sec2 shift in mean accel- eration can be interpreted as an improbably large static deflection of 24mm. Displacement recovered this way is termed pseudo-displacement. Figure 2 Double integration of tremor acceleration signal This works if there is no temperature dependent bias in the instrument system so thermal correction factors have been applied to the signals. There re- mains a temperature dependence which cannot eas- ily be corrected for and given the issues of noise and thermal drift, this method appears to be unreliable. Hence, having observed the performance of GPS at Humber7,11, a system was designed and installed at Republic Plaza. 5 GLOBAL POSITIONING SYSTEM The acceleration and wind recording system was up- graded in 2001 to include a dual-rover displacement tracking system using real time kinematic (RTK) differential GPS. Data sample rate was changed to 8Hz. For a single receiver (rover) as with a personal satellite navigation system, estimates of location are subject to errors that depend on modification to sig- nal transit time due to atmospheric and other effects. If a fixed base or reference (base) station is used nearby, given the known fixed location, the ‘differ- ential’ errors can be identified and used to adjust the rover position estimate. When the errors are trans- mitted from base station to rover and incorporated in rover position estimates in real time, position fixes accurate to the order of a centimeter or better are possible, at real-time rates of 10Hz or more. Real time operation uses software embedded in the re- ceivers, whereas by saving original signal transmis- sion times as ‘raw data’ for each receiver, subse- quent post-processing can merge the base station 2 4 6 8 10 12 14 16 18 -5 0 5 L65A mm Taliabu Lat -2.1 Long 124.9 (2372km) Ms 8.3 tremor 14:10:32 29/11/98 GMT 2 4 6 8 10 12 14 16 18 -5 0 5 B1A mm time /minutes since 22:12:19 29 Nov 1998 Figure 3 Schematic of GPS system of base station and two rovers and rover data sets to provide position fixes taking advantage of more sophisticated off-line processing and virtual base stations, potentially providing more accurate data. 5.1 GPS configuration The system at Republic Plaza as shown in Figure 3 and was designed to operate in both RTK and off- line post-processing modes. The system comprised Leica SR530 receivers for base station and rover, us- ing standard geodetic antennae. The base station antenna was mounted on the roof of Diethelm headquarters in Singapore (Figure 4), rover antennae were mounted on slender, rigid poles at each corner of the open roof top of the Republic Plaza (Figure 5). RTK corrections generated by the fixed base station at 1Hz data rate were transmitted by permanent leased line to both rovers, which gen- erated corrected solutions as 1Hz. Leica ‘raw data’ files were continuously acquired into ring buffers and these were saved on trigger command for each significant event. The data save process involved re- trieval of the base station data files over a separate dial-up line, during which time new trigger condi- tions were not recognized. For a single isolated trig- ger event, data files spanning 30 minutes either side would be stored in the host computer at Republic Plaza for each of the three rovers, for subsequent post-processing. As well as saving the raw data files, the corrected solutions for each rover were continuously output as NMEA format ASCII data. Since the data recorder PC used a primitive data acquisition programming environment, synchronization of ASCII input with analog signals was impossible and it was necessary to install digital to analog converter units to inter- face with the existing analog recorder. The 1Hz data rate was chosen as it was expected that displacement levels corresponding even to sig- nificant seismic response in building second mode would not reach the best estimate of GPS resolution of about 3mm and even at 1Hz there were consider- able overheads in transmitting the raw data over phone lines. The post-processing technique used was also semi-manual restricting the amount of data that could readily be processed. The antenna configuration at Republic Plaza (Figure 5) was not ideal; it was not possible to use larger, more expensive and more accurate choke ring antennae and the low level mounting was intended to reduce likelihood of lightning strikes. 5.2 Validation The initial GPS data appeared very noisy and not to correlate with other signals, hence it was difficult to believe what the data represented. In fact validating the GPS data was a major issue, as the signals were, in principle, subject to various forms of error such as multi-path, cycle-slip, random noise and systematic noise. Also, the total movement of the building was expected to be of the order of +/-0.1m as a combina- tion of all loads. Direct evidence of correct function- ing of the system was obtained first by physically moving the antenna during a recording and secondly by studying the signal during strong winds generat- ing first mode deflections at least 5mm amplitude. Figure 4 Base station antenna Figure 5 Rover 1 antenna flush with parapet Figure 6 shows the movement of the antenna in relation to the building configuration and direction, in a cycle of -220mm, +440mm then -220mm along the mounting beam. Figure 7 shows the recorded RTK signal. The first frame of data is corrupted by a periodic over-ranging of the system (at varying in- tervals of approximately 30 minutes) that appears in many of the earlier data and was never explained. Due to the orientation of the beam, the antenna moved +173mm, -346mm, +173mm in east direc- tion, at the same time -135mm, +270mm then - 135mm in north direction. The magnitudes shown in channel 13 (west rover -2, eastings) and channel 14 (west rover -2, northings) correspond, but the sign for the east direct is revered. Figure 6 Experiment moving antenna +/-200m along red line Figure 7 Recorded RTK signals fro experiment Only raw data sets for more interesting events were post-processed; Figures 8 shows results for two of the ‘best’ data sets; best in terms of clear signals and lack of sudden anomalous shifts. The raw data were able to provide information about confidence limits on each data point together with a wealth of other information, including details of satellite avail- ability but there were significant problems using it, specifically, it was not straightforward to syn- chronise the data with the analog recordings except by sliding one or other data set along the time axis to account for different start times. Because of the vari- ous practical difficulties with the raw data it was decided to first check that RTK and post-processed solutions concur and then rely on RTK for most pur- poses, with post-processing to be applied in collabo- ration exercises for occasional (special) events and to employ more accurate software. Figure 8 Building movements from raw data post-processing Figure 9 Comparison of RTK and post-processed GPS solu- tions: Top: RTK data Je_2_7 Bottom: Post-processed set 94. Last row is difference of post-processed signals -40 -30 -20 -10 0 10 20 30 -30 -20 -10 0 10 20 N/mm E/mm set73 GPS displacements -30 -20 -10 0 10 -15 -10 -5 0 5 10 15 20 N/mm E/mm set94 GPS displacements -200 0 200 ch 13 (mm) lt_4_7 18 20 22 24 26 28 30 32 -200 0 200 minutes ch 14 (mm) -40 -20 0 20 40 60 ch 11 (mm) je_2_7 -40 -20 0 20 40 60 ch 12 (mm) 0 500 1000 1500 2000 2500 3000 -40 -20 0 20 40 60 seconds ch 13 (mm) 0 500 1000 1500 2000 2500 3000 -40 -20 0 20 40 60 seconds ch 14 (mm) -50 0 50 ch 1 (mm) set94_diff -50 0 50 ch 2 (mm) -50 0 50 ch 3 (mm) -50 0 50 ch 4 (mm) 0 10 20 30 40 -50 0 50 minutes ch 5 (mm) 0 10 20 30 40 -50 0 50 minutes ch 6 (mm) Data set 94 and the concurrently acquired analog set JE, are compared in Figure 9. Although the se- quence of the sensors is slightly shifted, and apart from the periodic dropout, the signals match well enough for RTK to be used. Note that there are dif- ferences between the northings and easting between the two rovers; due to the fixed relative position, the only difference for a real signal should be due to ro- tation which is expected to be negligible. Hence the difference between the two signals provides one in- dication of possible errors. Figure 9 also shows, in the bottom row, for the raw data, that these differences are of the same order as the building deflections that the system is at- tempting to recover. Figure 10 shows response during what appears to be the windiest day of 2003. The building anemome- ters had failed but recorders elsewhere in Singapore had logged an unusually high (for Singapore) 30- minute mean wind speed over 20mph. There is a clear component of displacement at mode 1 fre- quency, also evident in the time series. 5.3 Continuous acquisition and further validation From December 2003, in addition to capture of trig- gered data files with 8Hz sample rate, continuous acquisition of a subset of six data channels (two level 65 acceleration and four GPS) at 1Hz was in operation, so that acceleration data were provided at the same rate as GPS. Only the response in the first mode and the quasi-static response are useful hence 1Hz is adequate to resolve these. Figure 11 shows one period during March 2004 when a storm excited relatively strong dynamic re- sponse. The enhanced dynamic response, in mode 1, is clear in the acceleration signals, which show sud- den increase in mean level of A-direction signal. There is a clear correlation between temperature and mean level. These storms are associated with noticeable temperature drops as winds pick up and dynamic acceleration increases. There is, however, a lag before the accelerometer cools and the first row of Figure 10 shows accelerations after correction for thermal bias so instrument effects seem an unlikely explanation. On the other hand, if the shifts in mean acceleration are interpreted as pseudo-accelerations, the 2mm rise corresponds to a 50mm deflection, but it is always in the same direction, whatever the wind. The GPS displacement signal in this case is likely to be more reliable. The RTK data shown (channels 5 and 6) are corroborated by the second rover and appear to indicate a steady drift of the building with a kind of ratcheting, together with dy- namic response. 6 GPS RECORDING OF GROUND MOTION DURING 2004 BOXING DAY EARTHQUAKE The final result for the system before it was perma- nently shutdown in early 2005 was to provide a re- cord of the response during the Boxing Day (26th December) 2004 magnitude 9.0 earthquake. 100 200 300 400 500 -5 0 5 10 t3 JA001B 18/1/2003 13:40 100 200 300 400 500 0 5 10 15 20 t4 100 200 300 400 500 -40 -20 0 20 40 t13 100 200 300 400 500 -40 -30 -20 -10 0 10 t14 0.2 0.4 0 5 10 15 f3 F/Hz, t/sec 0.2 0.4 0 10 20 30 40 f4 F/Hz, t/sec 0.2 0.4 0 10 20 30 f13 F/Hz, t/sec 0.2 0.4 0 5 10 15 20 f14 F/Hz, t/sec Figure 10 Acceleration and GPS time series and FFT for on windy day Acceleration, RTK and raw GPS signals were re- corded and show some notable features. The re- sponse spectrum of the earthquake is unusually strong at lower frequencies; peaking below 0.7Hz and peak accelerations, displacements and base shears were all less than for the closer Bengkulu quake of 2004. Figure 12 shows a sample of displacements from GPS and double-integrated accelerometer signals. Since GPS provides signals relative to the base sta- tion, the accelerometer-derived signals are the val- ues relative to basement, also rotated to the GPS axes. The signals are band-passed in the range 0.1Hz-0.3Hz. The correspondence is very clear. Figure 13 shows untreated post-processed sig- nals, in the global XYZ Cartesian system (which is a frame of reference rotated w.r.t East/North/Vertical). Z is North and Y is approximately vertical. The vari- ance errors are also plotted; they are relatively low during the ‘strong motion’ part of the signal but val- ues are as large as the signal itself. Figure 14 is the spectrogram showing time-frequency variation. The enduring first mode response is clear, there is also some signal up to 0.08Hz. 7 CONCLUSIONS ON PERFORMANCE OF GPS Even after ten years of monitoring there are still mysteries about the building performance. For ex- ample the static deflections due to temperature have still not been determined and present a problem as temperature affects the instrumentation as well as the structure movements. The GPS data show no daily cycle and the pseudo displacements do not appear to be reliable given the strong correlation with temperature even after apply- ing the given correction factor. The way the building responds to mean and slowly varying winds is still not clear; so far there is not an obvious pattern, and even a suggestion of highly non-linear behaviour. The GPS did not quite live up to the high expec- tations; it was technically challenging to integrate the signals into the (rather archaic) existing logger, there were problems with unreliability for various reasons (communication issues from rover to base station, system overheating, antenna destruction, un- usual application). For the periods of uninterrupted RTK signals, the signal appeared to be very noisy with non-random behaviour difficult to interpret as building movements. Nevertheless, during sudden strong wind and seismic events with sustained large first-mode amplitude, there were clear and system- atic GPS displacement signals of quasi-static move- ment or dynamic response. These indications alone are enough to indicate that for (rare) building dy- namic response exceeding 10mm/sec2 amplitude GPS should be able to identify both static and dy- namic response. This includes response to large earthquakes, although only the relative motion of the building can be captured. Since the difficulty is to capture the very low fre- quency signals, it seems that a different approach should be to use a static (rather than dynamic) GPS solution with sample rate of 1 per minute and inte- grate this with accelerometer data. 4 6 8 10 12 ch 1 (mm/sec 2 ) nc1_T 0 5 10 15 ch 2 (mm/sec 2 ) -100 0 100 ch 5 (mm) -100 0 100 ch 6 (mm) 5300 5350 5400 5450 302 304 306 308 310 minutes ch 7 () 5300 5350 5400 5450 302 304 306 308 310 312 minutes ch 8 () Figure 11 Acceleration, GPS and temperature (degrees Kelvin) signals during storm REFERENCES Lam LCH, Lam RP 1979. An experimental study of the dy- namic behaviour of a multi-storey steel-framed building. Proceedings, Institution of Civil Engineers pt2 67: 707- 720. Littler, JD, Ellis, BR 1990. Interim findings from full-scale measurements on Hume Point. Journal of Wind Engineer- ing and Industrial Aerodynamics 36: 1181-1190. Jeary AP, Chiu GC, Wong JCK 2001 Wholistic structural ap- praisal. In Proc. 8th Int. conf. and structural safety and reli- ability: ICOSSAR 2001 Stewart JP, Whang D, Wallace JW, and Nigbor RL 2005. Field Testing Capabilities of nees@UCLA Equipment Site for Soil-Structure Interaction Applications. Proceeding, Geo- Frontiers 2005, Austin Texas, ASCE Geotechnical Special Publication 130-142. Wong KY 2003. Instrumentation and health monitoring of ca- ble-supported bridges. Structural Control and Health Moni- toring 11(2): 91-124. Le Diourion T 2005. The health monitoring system of Rion Antirion Bridge. Proceedings IMAC XXIII, Orlando, Fla. Brownjohn JMW, Bocciolone M, Curami A, Falco, M, Zasso A 1994. Humber Bridge full scale measure campaigns 1990-1991. Wind Engineering and Industrial Aerodynam- ics 52: 185-218. List D 2004. Rejuvenating the Tamar Bridge. A review of the strengthening and widening project and its effect on opera- tions. Proceedings, 4th International cable-supported bridge operators’ conference, Copenhagen Marecos J, Castenheta M, Trigo JT 1969. Field observation of Tagus River suspension bridge. ASCE Journal of Structural Engineering 95. Ahola R, Tervaskanto M 1991. A position sensing laser system for analyzing dynamic deflection measurements. Field Measurements in Geotechnics :129-137. Ashkenazi V, and Roberts GW 1997. Experimental monitoring of the Humber Bridge using GPS. Civil Engineering, Pro- ceedings, Institution of Civil Engineers 120: 177-182. Nakamura S 2000. GPS measurements of wind-induced sus- pension bridge girder displacements. ASCE Journal of Structural Engineering 126(12): 1413-1419. Guo J, Xu L, Dai L, McDonald M, Wu J, Li Y 2005. Applica- tion of the real-time kinematic global positioning system in bridge safety monitoring. ASCE Journal of Bridge engi- neering, 10(2): 163-168. Wong KY, Man KL, Chan WY 2000. Application of global positioning system to structural health monitoring of cable- supported bridges. Proceedings, SPIE 4337: 390-401. Kashima S, Yanaka S, Suzuki S, Mori K 2001. Monitoring the Akashi-Kaikyo Bridge: first experience. Structural Engi- neering International 11(2): 120-123. Li X, Ge L, Peng GD, Rizos C, Tamura Y, and Yoshida A 2004. Seismic response of a tower as measured by an inte- grated RTK-GPS system. Proceedings 1st FIG Interna- tional Symposium on Engineering Surveys for Construction Works and Structural Engineering, Nottingham UK. Pagnini LC, Tanura Y, Matsui M, Ishibashi R, Yoshida A 2002. Measurement of wind-indiced response of buildings using GPS. Journal of Wind engineering and Industrial Aerodynamics 90(12-15): 1783-1793. Lovse JW, Teskey WF, Lachapelle G, Cannon ME 1995. Dy- namic deformation montoring of tall structure using GPS technology. ASCE Journal of Surveying Engineering 121(1): 35-40. Kijewski-Correa T and Kareem A 2003. The Height of Preci- sion, GPS World, September: 20-34 Brownjohn JMW, Pan TC, Cheong HK 1998. Dynamic re- sponse of Republic Plaza , Singapore. The Structural Engi- neer 76(11): 221-226. -20 0 20 ch 1 (mm) xv_aceh_disps_rel_rotated -20 0 20 ch 2 (mm) -20 0 20 ch 3 (mm) -20 0 20 ch 4 (mm) 600 700 800 900 -20 0 20 seconds ch 5 (mm) 600 700 800 900 -20 0 20 seconds ch 6 (mm) Figure 12 Narrow-band (0.1-0.3Hz) filtered displacement signals during Boxing Day 2004 earthquake: Ch1/2 are rover 1, ch3/4 are rover 2, ch1/3 are eastings, ch2/4 are northings. Ch5/6 are difference or roof and basement accelerometer signals rotated to north/east axes. Figure 13 Post-processed rover 2 data in Cartesian global co- ordinates, ch4-6 are variance Figure 14 Time frequency plot (spectrogram) of rover 2 signal -50 0 50 ch 1 (mm) 26dec_session3_R2_0mean -50 0 50 ch 2 (mm) 0 10 20 30 -50 0 50 minutes ch 3 (mm) 0 5 10 15 ch 4 (mm) 0 10 20 30 ch 5 (mm) 0 10 20 30 0 10 20 30 minutes ch 6 (mm) ch 1 f/Hz 26dec_session3_R2_0mean 0.1 0.2 0.3 0.4 0.5 ch 2 f/Hz 0.1 0.2 0.3 0.4 0.5 minutes ch 3 f/Hz 0 5 10 15 20 25 0.1 0.2 0.3 0.4 0.5