This article was originally published in a journal published by Elsevier, and the attached copy is provided by Elsevier for the author s benefit and for the benefit of the author s institution, for non-commercial research and educational use including without limitation use in instruction at your institution, sending it to specific colleagues that you know, and providing a copy to your institution s administrator. All other uses, reproduction and distribution, including without limitation commercial reprints, selling or licensing copies or access, or posting on open internet sites, your personal or institution s website or repository, are prohibited. For exceptions, permission may be sought for such use through Elsevier s permissions site at: http://www.elsevier.com/locate/permissionusematerial

Journal of Applied Geophysics 61 (2007) 217 226 www.elsevier.com/locate/jappgeo Abstract Coaxial coil towed EMI sensor array for UXO detection and characterization Haoping Huang a,, Bill SanFilipo b, Alex Oren b, I.J. Won b a Formerly Geophex, Ltd., presently Geo-EM, LLC, 2001 Waterbrook Ct. Raleigh, NC 27603-5180, USA b Geophex, Ltd., 605 Mercury St., Raleigh, NC 27603-2343, USA Received 1 November 2005; accepted 5 June 2006 A new broadband electromagnetic induction (EMI) array sensor, GEM-5, for detecting and characterizing Unexploded Ordnance (UXO) has been developed in order to provide high production rates for EMI surveys. The sensor consists of a single rectangular loop transmitter around a linear array of seven pairs of coaxial receiver coils, with each coil in a pair located at the same vertical distance above and below the loop transmitter. The coil pairs are wired in an inverted series so that the signal consists of the difference between the voltage induced in the upper and lower coils. This particular configuration provides a high degree of primary field cancellation, dense spatial sampling rate due to simultaneous and continuous operation of all sensors, suppression of motion-induced and environmental noise, and strong source fields at typical UXO burial depths providing deep detection range. Our prototype tests indicate that the array yields a lower static and motion-induced noise over the critical low frequencies than that of existing sensors, and in particular, the signal-to-noise ratio at 90 Hz is 32 db higher. Environmental noise can be largely removed from the difference measurements. The field test results from UXO test sites show that the prototype sensor has smoother background and appears to detect more seeded targets than the GEM-3 concentric sensor, however some of that gain can be attributed to higher power transmitter electronics. 2006 Elsevier B.V. All rights reserved. Keywords: Electromagnetic induction; Coaxial coil configuration; Sensor array; UXO 1. Introduction Cleanup of Unexploded Ordnance (UXO) at former military weapons test sites is an ongoing environmental task that is costly and requires a large manpower effort. The problem is important as many sites contaminated with dangerous UXO are returned to non-military use. In order to increase the cost effectiveness of UXO cleanup operations, advanced technology is needed for locating buried UXO efficiently with the means for reducing the digging of false targets (clutter). One effort in sensor advancement is to develop multi-sensor systems, including an electromagnetic induction (EMI) sensor combined with a magnetometer or ground penetration radar (GPR), multi-axis EMI sensors, and EMI sensor arrays (e. g., Nelson et al., 1996; Barrow and Nelson, 2001; Nelson et al., 2003; Becker et al., 2005; Morgenstjeme et al., 2005; Rogers et al., 2005). Corresponding author. Tel./fax: +1 919 773 0119. E-mail address: haoping_huang@hotmail.com (H. Huang). 0926-9851/$ - see front matter 2006 Elsevier B.V. All rights reserved. doi:10.1016/j.jappgeo.2006.06.005

218 H. Huang et al. / Journal of Applied Geophysics 61 (2007) 217 226 We developed a multi-sensor EMI vehicle-towed system in order to provide high production rates for EMI surveys. The array sensor, designated as the GEM-5, has multiple vertical coaxial receiver coil pairs (typically 6 or 7) and a single loop transmitter surrounding the midpoints of the receiver coil pairs. Since the receiver coils of each pair are located at the same distance from the transmitter coil, the difference in output between the coils in each receiver pair would be zero in free space. When the sensor is brought near a UXO object, the balance breaks, producing a non-zero output that represents the magnetic field perturbation from the UXO object. This is a gradient measurement, i.e., the balanced cancellation of the primary field using coaxial coils. The Geonics EM61 time-domain EMI sensor has an optional gradient measurement output mode using upper and lower coils. Gradient measurements had been tried by Barringer, Ltd. many years ago, but this work never appeared in the literature. Recently, more studies on this technology have been performed for EM sensors (Sattel and Macnae, 2001; Won, 2001; Morrison et al., 2004; Huang et al., 2005). In this paper, we first introduce a prototype of the sensor array and present the results from tests on motioninduced and environmental noise to illustrate the merits of gradient measurements. Finally, we demonstrate some field examples from UXO test sites. 2. The prototype sensor The prototype sensor array can be sled-mounted and vehicular towed, or front-mounted on a specialized robotic vehicle. Fig. 1 shows the coil configuration and photos of the prototype in use. Geometry parameters for the prototype are 2a = 2.1 m, 2b = 0.3 m and 2s=0.4 m, and the receiver coil separation is 0.3 m. Our approach for a frequency-domain (continuouswave) sensor tailored for arrays (simultaneous operation of all sensors with no mutual interference) is to utilize a coaxial coil configuration that maintains mutual primary field cancellation for adjacent sensors. There is a single central transmitter coil surrounding balanced coaxial receiver coil pairs, i.e., reference channel RX1 and signal channel RX2, connected in a differential configuration (in series but opposite sense). The reference and signal coils are made exactly the same and each receiver pair is symmetric with respect to the transmitter. Thus, the difference signal measures nominally zero voltage from the primary field. Special attention has been made to the mechanical rigidity of the array to ensure flexure does not result in noise from modulation of the primary fields asingleintegrated structure is used rather than assembling individual sensors onto a platform. Advanced electronics provide state-ofthe-art digital transmitter waveform generation for Fig. 1. (a) The coil configuration, (b) tractor-towed and (c) remote-controlled array sensors.

H. Huang et al. / Journal of Applied Geophysics 61 (2007) 217 226 219 simultaneous multi-frequency operation, enhanced resolution and dynamic range (24 bit A/D), and capability for additional real-time processing such as digital highpass filtering. The digital, programmable sensor operates in the frequency-domain at all user-selected frequencies (typically ten) continuously and simultaneously (Won et al., 1997, 2003). Each channel, denoted as RX1 i (reference) and RX2 i RX1 i (signal), receives, amplifies, and digitizes its output into a time-series lasting over a base period (1/30 s). The sensor has an analogto-digital converter (ADC) rate of 192 khz and, therefore, each time-series has 6400 points per base period of 1/30th of a second (i.e., 192,000/30). Both the signal and reference channels produce such a timeseries at every base period. A digital signal processor (DSP) performs a sine and a cosine convolution at each frequency. The results from the signal channel are then normalized against those from the reference channel to produce the real or in-phase (I) response and the imaginary or quadrature (Q) response in a dimensionless unit of parts-per-million, or ppm, as defined by ppm i ¼ 10 6 RX 2 i RX 1 i RX 1 i ; i ¼ 1; :::; N; ð1þ where N is the number of sensors in the array and the subscript i indicates ith sensor. The system stores a set of I- and Q-data at all frequencies as the raw ppm data. Afterwards, the two time-series are discarded. While the time-series processing goes on, the sensor now works for the next base period without interruption, and the process repeats at every base period. Thus, the overall data rate is 30 Hz for each receiver regardless of the number of frequencies used for the survey. Responses to a UXO object under a sensor slightly depend upon the sensor's location within the array since the primary field differs from the center to the end of the array. Thus, we need to scale each sensor to obtain the identical response for a target at typical depths (unity scaling for central receivers and a software correction for the others). A closed-loop test coil (Q-coil) can be used to test the integrity of the scaling. Fig. 2a shows all sensors' responses from a Q-coil placed over each sensor sequentially. As can be seen, all sensors yield an almost identical response. Errors can be estimated by standard deviation at each frequency as illustrated in Fig. 2b. The largest Fig. 2. (a) The Q-coil responses at 9 frequencies (30, 90, 150, 450, 1050, 2610, 6330, 15,450 and 37,710 Hz) for all sensors, (b) standard deviation. standard deviation is about 18 ppm against a signal level of 1400 ppm. 3. Advantages of the coaxial sensor array By operating simultaneously and continuously, data sampling rates for each sensor do not need to be reduced for additional sensors in the array, and the number of sensors is limited only by practical constraints on the size and structural considerations of the array. Incorporating additional sensors allow reduction in line spacing for denser coverage without a trade-off in array width (and production rate) while maintaining maximum sample density along lines. This is superior to a concentric coil sensor that must, in an array, operate sequentially, one sensor each time, because of mutual interference among sensors (Nelson et al., 2003). A sequentially fired array results in reduction of individual sample rates as sensors are added. Since the coaxial configuration is symmetric and balanced, this sensor will be immune to ambient electromagnetic field noise from distant sources, such that the fields are essentially plane waves at the sensor, including manmade sources such as power lines, radio transmitters, and industrial electric machinery, as well as natural EM noise (sferics from distant lightning, geomagnetic storms). The balanced receiver will cancel such noise.

220 H. Huang et al. / Journal of Applied Geophysics 61 (2007) 217 226 Similarly, noise induced by motion in the earth's field will be cancelled by virtue of the balanced receiver coils, whereas this is the primary source of lowfrequency noise for concentric coil sensors operating on a moving platform. Since the sampling rate is 30 Hz for each sensor in most surveys with a coaxial array, relatively rapid survey speeds will provide ample spatial sampling, and the associated increase in platform rocking will not diminish signal-to-noise ratio (SNR) in the lowest operational frequencies, as is the case, based on our history of testing towed and push-cart GEM-3 systems, for unsymmetrical coil configurations or transmitter bucked systems. Although the areas of the transmitters are nearly the same, the primary magnetic field decays more slowly with depth for the rectangular transmitter loop than a circular loop (the ratio of primary fields at deep to shallow depths is greater for the rectangular loop). In addition, transmitter power is four times larger for the array than for the standard concentric coil sensor (0.96 m diameter). These yield the stronger primary fields for the deepest targets that are difficult to detect. Fig. 3 compares the primary fields with depth for a concentric coil sensor (0.96 m diameter GEM-3) and the prototype array (GEM-5). Distance from ground surface to sensor bottom is 0.3 m for both sensors. The primary field falls off more slowly with depth for the array than for the other sensors using a small ( dipole) source. We have built concentric coil sensors with the high power transmitter (the Naval Research Laboratory GEMTADS), which showed significant improvement in SNR, particularly related to motion-induced noise and depth of detection, but not matching the intrinsic advantages of the balanced receiver coils. We have also found that the ratio of response of deep ( 1m) to small shallow targets is greater with the coaxial array, thereby enhancing the detection of deeper targets in the presence of shallow clutter and geologic noise. 4. Prototype tests 4.1. Static and motion noise We first performed tests against static and motion noise using the prototype and existing concentric coil sensors. EMI receiver coils are susceptible to motioninduced noise because any coil rapidly rotating in the ambient geomagnetic field due to bumps on the ground produces noise in the low frequencies (b200 Hz) that are particularly valuable for detecting and classifying metal objects. A vertical coaxial coil configuration will mitigate this noise by canceling out most of the motion-induced noise, which will yield a lower noise level and so higher SNR than a concentric coil sensor. Fig. 4 illustrates a segment of EMI data at 90 Hz obtained from test surveys for both sensor types. The data from x=0 to about x=800 were collected during the stop at the end of a survey line, i.e., static data, while the rest of the data were collected when the sensors were moving, i.e., motion data. The top panel shows the I and Q responses obtained by the concentric coils sensor, and the lower panels show those from four sensors in the array, in which sensor 1 is at the edge and sensor 4 at the center of the array. It is obvious that the difference between static and motion data noise envelopes is much more significant for the concentric sensor than for the array. Examining the array responses, it seems that both static and motion noises are larger for sensors at the edge of the array than those at the center of the array. To confirm this point, we calculate the standard deviations of static and motion noise for all 7 sensors as shown in Fig. 5, from which we see that the closer to the center of the array, the smaller the noise. For the static case, this can be attributed to the correction scale-factor (1.4 or 1 and 7) discussed above applied to the outer coils; the dynamic noise is likely related to acoustic vibrations in the structure modulating the primary field balance. Fig. 3. The primary magnetic fields, normalized by those at ground surface, with depth for two sensors, a concentric coil sensor with 0.96 m transmitter coil in diameter (C), assuming that the transmitter current is 1 A and number of turns is 6, and the prototype array (A), assuming that the transmitter current is 4 A and number of turns is 9.

H. Huang et al. / Journal of Applied Geophysics 61 (2007) 217 226 221 Fig. 4. The static (xb800) and motion (xn800) noise at 90 Hz obtained using both the array and concentric coil sensor. Signal-to-noise ratios (SNR) for the array and concentric sensor can be compared using the responses from a Q-coil placed at the same distance from the sensors. Fig. 6a and b illustrate the SNR as a function of frequency for static and motion cases, comparing the array (sensor 2) with the concentric coil sensor. The results from the static test show that a better SNR is obtained from the array when the frequency is less than about 600 Hz. As the frequency increases, the concentric coil sensor yields a better SNR. However, for motion tests, the SNR of the array is much better than that of the concentric coil sensor in the whole frequency band, especially at the lower frequencies. For example, SNR at 90 Hz is 32 db higher for the array than for the concentric coil sensor. 4.2. Environmental noise Ambient environmental EM noise, such as radio transmitters, power lines, industrial electrical machinery, Fig. 5. The static (open circles) and motion (solid circles) noises at 90 Hz for 7 sensors in the array. Sensor 4 is in the middle of the array. Fig. 6. (a) The static and (b) motion SNR as a function of frequency for sensor 2 in the array (solid circles) and the concentric sensor (open circles). The signal data are obtained from a Q-coil placed at the same distance from both sensors.

222 H. Huang et al. / Journal of Applied Geophysics 61 (2007) 217 226 Fig. 7. The EM data at 6 and 11 khz collected using the array and the concentric sensor when the noise source is off (xb450) and on (xn450). and sferics can be largely canceled out by the array coil configuration. We carried out tests against a manmade EM noise at the operating frequencies of the array and concentric coil sensor. Fig. 7 illustrates the typical data at 6 and 11 khz for the two sensors. The noise source is off for the first half of the data sequence (x b 450); it is on for the rest of the data series (x N 450). As expected, the interference noise distorts significant responses of the concentric sensor, while it is hardly seen from the responses of the array. The only indication of the manmade noise is on the in-phase responses when the noise starts. 4.3. Detection range Detection range is larger for the array than for dipolar-source sensors in part because of the stronger primary field, and also from reduced the motion and environmental noise. Note that the GEM-3 loses 15% of its primary field at large depths (more in the near field) by virtue of the transmitter bucking, but the coaxial array primary field is reduced by the extra transmitter height (also has greater relative impact for shallower targets). A small Q-coil is used to test the detection ranges for the array and a concentric coil sensor with 96 cm diameter transmitter. Responses to the Q-coil can be obtained at several distances to sensor bottom, and then SNR can be calculated. Fig. 8a and b show static SNR's at 1470 Hz as a function of distances from sensor bottom, and those at 0.8 m as a function of frequency. The array SNR is greater than the concentric sensor in the range from 0.3 m to 1 m, and the difference becomes slightly larger as the distance increases, indicating that the array is favorable for deep targets. Fig. 8. (a) Static SNR at 1470 Hz as a function of distances from sensor bottom, and (b) SNR at 0.8 m as a function of frequency.

H. Huang et al. / Journal of Applied Geophysics 61 (2007) 217 226 223 Fig. 9. (a) Geophex UXO test site in Raleigh, North Carolina. The 10 m 10 m site contains a total of 21 metal pipes of various diameters and lengths and a diabase boulder. Maps of Q-sum from (b) the array and (c) the concentric sensor, in which the white square indicates the seeded test site and circles indicate locations of the seeded pipes.

224 H. Huang et al. / Journal of Applied Geophysics 61 (2007) 217 226 Table 1 Target description for Geophex UXO test site and results of target identification Target Description x (m) y (m) z (cm) L1 15.5 cm-id 50.8 cm steel pipe, horizontal, E W 7.25 7.25 100 L2 15.5 cm-id 50.8 cm steel pipe, 45 deg; SW(up)-NE 2.75 2.75 110 M1 7.9 cm-id 45.7 cm steel pipe, vertical 5.00 5.00 70 M2 7.9 cm-id 45.7 cm steel pipe, 45 deg, NW(up)-SE 8.25 4.75 80 M3 6.4 cm-id 30.5 cm steel pipe, horizontal E W 8.75 1.25 50 M4 6.4 cm-id 30.5 cm steel pipe, 45 deg, W(up)-E 5.25 1.75 60 M5 7.9 cm-id 45.7 cm steel pipe, horizontal, N S 1.75 5.25 70 M6 6.4 cm-id 30.5 cm aluminum pipe, horiz. SW NE 1.25 8.75 50 M7 6.4 cm-id 30.5 cm steel pipe, horizontal N S 4.75 8.25 60 S1 2 cm-id 10.2 cm steel pipe, horizontal N S 3.00 9.25 10 S2 2.3 cm-id 15.2 cm alumin. pipe, 45 deg, NW(up) SE 2.75 8.25 15 S3 4.1 cm-id 15.2 cm steel pipe, vertical 2.25 7.25 30 S4 4.1 cm-id 10.2 cm steel pipe, horizontal N S 0.75 7.00 30 S5 4.1 cm-id 15.2 cm steel pipe, 45 deg, SW(up) NE 3.25 6.25 30 S6 4.1 cm-id 15.2 cm steel pipe, horizontal E W 6.75 3.75 30 S7 4.1 cm-id 10.2 cm steel pipe, horizontal E W 9.25 3.00 30 S8 4.1 cm-id 15.2 cm steel pipe, 45 deg, SW(up) NE 7.75 2.75 20 S9 2.3 cm-id 15.2 cm aluminum pipe, horizontal E W 7.25 1.75 15 S10 2 cm-id 10.2 cm steel pipe, horizontal E W 7.00 0.75 10 S11 2.3 cm-id 15.2 cm copper pipe, 45 deg, NW(up) SE 0.75 0.75 20 S12 2.3 cm-id 15.2 cm copper pipe, horizontal N S 9.25 9.25 15 R1 30 30 33 cm diabase boulder 0.50 4.00 27 4.4. UXO detection We tested the prototype array at our simulated UXO Test Site at Geophex in Raleigh, North Carolina. The 10 m 10 m site, on a dense red clay soil, contains 21 metal pipes, ferrous and nonferrous, of various diameters and lengths buried at depths down to 110 cm. Fig. 9a shows the site plan and Table 1 describes each seeded target, its location (local coordinates), and the depth to the center of the pipes. The data are collected using the array and the concentric coil sensor at 9 frequencies between 90 Hz and 21 khz, using differential GPS positioning. Fig. 9b and c illustrate the maps of Q-sum, which is defined as sum of the Q responses at all frequencies (Huang and Won, 2003). The seeded area is indicated with the white square and seeded target locations are indicated with circles. As can be seen, both sensors detect all seeded pipes, but the array data have a quieter background. The linear feature on the array map is a chain placed on the ground for testing. The second example is from a government test facility. It is a seeded site for controlled testing and includes [1] calibration lanes for system training and target characterization; and [2] a blind test grid a 1600 m 2 rectangular grid including access lanes separating 400 discrete 1 m 1 m square interrogation points. The tests using the array sensor were performed in areas including the calibration lanes, blind test grid and some surrounding terrain. Fig. 10 presents the Q- sum maps for the array and concentric coil sensor, calculated for 9 frequencies from 90 Hz to 21 khz. The array data were obtained in 2005, while the concentric sensor data in 2003. Half of the calibration grid was not surveyed with the GEM-3 in 2003 targets included within the grid are scored as false alarms, so we intentionally stopped covering the calibration grid to reduce the number of targets to be manually removed from the automated target picks. The targets in the blind grid had been reseeded after 2003, and so the anomaly distribution looks different on the two maps. We can see, based on qualitative assessment of the clarity of anomalies, that the array not only detects more targets in the calibration lanes than the concentric sensor, but also has quieter background. 5. Conclusions We have built a vertical coaxial EMI array sensor for UXO detection and characterization. The array coil configuration has four potential advantages over an array of concentric coil sensors, (1) denser spatial sampling rate due to simultaneous and continuous operation, (2) reducing the motion-induced noise and (3) the environmental noises, and (4) stronger primary field at depth that allows detection of deep targets. Our experiments have demonstrated these advantages. The field test results from UXO test sites show qualitatively

H. Huang et al. / Journal of Applied Geophysics 61 (2007) 217 226 225 Fig. 10. Maps of Q-sum for the calibration lanes and blind grids, (a) the array and (b) the concentric coil sensor. Note that the EM data are not available in the white area in panel (b). that the prototype sensor not only detects more clearly the seeded targets than the existing concentric sensor, but also has less noisy background. Acknowledgments This study has been partly funded by the US Army Test Center (ATC). References Barrow, B., Nelson, H.H., 2001. Model-based characterization of electromagnetic induction signatures obtained with the MTADS electromagnetic array. IEEE Transactions on Geoscience and Remote Sensing 39 (6), 1279 1285. Becker, A., Gasperikova, E., Morrison, F., Smith, T., 2005. A multisensor system for the detection and characterization of UXO. Proceedings of the International Symposium on the Application of

226 H. Huang et al. / Journal of Applied Geophysics 61 (2007) 217 226 Geophysics to Engineering and Environmental Problems. Environmental and Engineering Geophysical Society, Denver, CO, pp. 1236 1243. Huang, H., Won, I.J., 2003. Automated anomaly picking from broadband electromagnetic data in UXO survey. Geophysics 68, 1870 1876. Huang, H., SanFilipo, B., Won, I.J., 2005. Planetary exploration using a small electromagnetic sensor. IEEE Transactions on Geoscience and Remote Sensing 43 (7), 1499 1506. Morgenstjeme, A., Karlsen, B., Larsen, J., Sorensen, H.B.D., Jakobsen, K.B., 2005. A comparative and combined study of EMIS and GPR detectors by the use of independent component analysis, detection and remediation technologies for mines and minelike targets X. Proceedings of SPIE, The International Society for Optical Engineering 5794, 988 999. Morrison, H.F., Becker, A., Gasperikova, E., Smith, J.T., 2004. A multisensor system for the detection and characterization of UXO. Partners in Environmental Technology Technical Symposium and Workshop, Washington D.C. Nelson, H.H., Barrow, B.J., Bell, T.H., SanFilipo, B., Won, I.J., 2003. Characterization of a GEM-3 array for UXO classification. Detection and Remediation Technologies for Mines and Minelike Targets VIII: Proceedings of SPIE, vol. 5089, pp. 940 949. Nelson, H.H., McDonald, J.R., Robertson, R., 1996, MTADS TECHEVAL Demonstration, October 1996, NRL/PU/6110-97-348. Rogers, N., Sandberg, S., Bennett Jr., H., 2005. Multi-sensor UXO detection system. Proceedings of the International Symposium on the Application of Geophysics to Engineering and Environmental Problems. Environmental and Engineering Geophysical Society, Denver, CO, pp. 1254 1260. Sattel, D., Macnae, J., 2001. The feasibility of electromagnetic gradiometer measurements. Geophysical Prospecting 49, 309 320. Won, I.J., 2001, Electromagnetic Gradiometer, United States Patents No. 6, 204, 667, March 2001. Won, I.J., Keiswetter, D., Hanson, D., Novikova, E., Hall, T., 1997. GEM-3: a monostatic broadband electromagnetic induction sensor. Journal of Environmental & Engineering Geophysics 2 (1), 53 64. Won, I.J., Oren, A., Funak, F., 2003. GEM-2A: a programmable broadband helicopter-towed electromagnetic sensor. Geophysics 68, 1888 1895.