Millimetre Wave Radar for Tracking and Imaging Applications

Transcription

1 Abstract Millimetre Wave Radar for Tracking and Imaging Applications Graham M Brooker Australian Centre for Field Robotics, University of Sydney, Sydney, Australia gbrooker@acfr.usyd.edu.au This paper discusses the design and implementation of some millimetre wave radar systems for tracking and imaging applications. It demonstrates the versatility of this technology and its application as the sensor of choice for autonomous guidance and navigation applications. Keywords: Millimetre wave radar, tracking, imaging, 94GHz, 77GHz, autonomous guidance, navigation 1 Introduction Over the past two decades we have been building millimetre wave radars for both military and commercial applications. This paper discusses some of the practical aspects of these developments with illustrations of both the systems and their outputs. 2 Tracking Radar A number of different 94GHz tracking radar systems have been developed to investigate various aspects of the technology. Probably the most simple is a single stage non-coherent 5W pulsed IMPATT transmitter with a conventional super heterodyne receiver shown in Figure 1. Because of the cost and poor performance of amplifiers at 94GHz it was decided to dispense with amplification and feed a mixer directly. The local oscillator (LO) is a conventional 2 nd harmonic Gunn device tuned to 93.7GHz to produce an IF frequency of 300MHz. Successive Detection Log Amp Pulse Generator Amplifier Matched 300MHz Pulsed IMPATT Oscillator 93.7GHz 94GHz Gunn Oscillator Circulator Horn-lens Antenna Figure 1: Pulsed millimetre wave radar configuration One of the issues with this configuration is the large intra-pulse chirp which is generated as the IMPATT heats due to the extremely high current density [1] which is typically greater than 100MHz for a 100ns rectangular pulse. This can be reduced by shaping the pulse current [2], but it does preclude using a filter matched to the pulsewidth, and hence the overall noise performance is not very good. It is however, much less complex than implementing an injection locked oscillator chain driven by a fixed frequency Gunn oscillator [3]. 2.1 Range Tracking An analog split-gate range tracker performs this function [4]. It comprises two fast sample-and-hold boards each with a 50ns aperture time followed by sum and difference circuits. The magnitude of the sum signal is used for target detection, and the difference signal drives a pair of cascaded integrators to produce a second order tracker. The range voltage controls a timing generator that produces two outputs separated by 50ns to trigger the early and late gate sample and hold circuits as shown in Figure 2. Radar Video Sample & Hold Sample & Hold VE VL Sum + Difference Early Gate Late gate Threshold Target Detection Range Designation Range & Velocity Estimator Timing Generator Velocity Range Figure 2: Analog implementation of a classical split gate range tracking loop An external range designation input to the estimator allows the range gates to be positioned manually using a joystick, or scanned in range during the automatic target detection sequence. If the gates are swept through a point target, the difference signal maps out the split gate error function as shown in Figure 3. In general an AGC is required to normalise this error function with respect to the sum channel signal amplitude, but because a successive detection log amplifier (SDLA) is used, the error voltage is independent of the size of the echo. It can be seen that there is a good (albeit slightly 158

2 asymmetrical) linear region of +/-2m and a total capture distance, in which the sense of the error is correct, of +/-8m. Figure 3: Measured range error function for the split gate tracker 2.2 Angle Tracking To determine the target offset from the antenna boresight, conical-scan radars locate the centre of the target image by nutation of a single feed that is displaced from the axis as shown in Figure 4 [4, 5]. When the target is centred on the axis, the feed receives equal amounts of power from all angles. When the image is offset, more power is received by the feed on the one side which results in amplitude modulation of the received beam. This modulation is compared to the in-phase and quadrature reference signals generated by the nutation mechanism to determine the sense of the error, while the amplitude determines its magnitude. It should be made clear that the received radar signal that is demodulated by the reference signals will have been gated in range to include only the target being tracked. As with the range tracking case, the use of an SDLA removes the requirement for an AGC loop. Over the past two decades, techniques to implement this function for a wide range of W-band antenna types including a horn-lens, a twist Cassegrain and a standard Cassegrain antenna have been developed by us for various applications [6]. For this application an antenna comprising a standard rectangular horn illuminating a 110mm diameter plano-convex lens was manufactured [7] as shown in Figure 5. A disk of cross-linked polystyrene with a relative dielectric constant ε r = 2.6 was inserted at an angle between the lens and the horn to displace the beam off the boresight. The disk was spun to produce the appropriate beam nutation for conical-scan operation. Teflon Lens Spinning Polystyrene Disc Reflective Transducers Horn Figure 5: Schematic diagram of the horn-lens antenna showing the position of a spinning polystyrene disc designed to produce the beam nutation For a 3dB beamwidth of 2.1 and a relative squint angle of 0.3 [4], the actual squint angle is 0.6. This equates to a physical displacement of the beam focal point of only 1mm for the focal length of 100mm which was achieved using a 7.5mm thick disk tilted at 20. This radar uses an error demodulation technique based on analog multiplier integrated circuits with reference signals generated using a simple but elegant technique shown in the following block diagram. Range Gated Radar Video Reflective Encoders Index Counter Tube housing motor and modulator plate Signal Conditioner count index Flip-Flop State Machine I Q Bandpass Bandpass Demodulation Lowpass Lowpass Azim Error Elev Error Figure 4: Conical-scan concept showing the pencil beam offset from the radar boresight by a small squint angle and the resultant amplitude modulation of the received echo pulses Figure 6: Reflective encoders detect markers on the spinning cylinder to produce the I and Q reference signals used to demodulate the radar signal. The spinning cylinder that houses the angled disc is marked with four black bars equally spaced 90 apart in a ring on its outer surface. These bars produce pulses which then generate two quadrature square waves that are filtered to produce the reference signals. Four quadrant analog multiplier ICs perform the angle error demodulation function. 159

3 The following photographs show the radar with its associated antenna alongside a view of the conscan mechanism obtained by removing the lens. Figure 7: Photographs of the pulsed radar seen from the front and the radar with lens removed showing conscan mechanism The angle error transfer function shown in Figure 8 was determined by scanning the beam across a distant point reflector and plotting the amplitude of the error signal as a function of angle. Figure 8: Measured angle error transfer function obtained by scanning across a corner reflector As with the range error function, the use of the SDLA ensures that the angle error function is also independent of the amplitude of the received signal. This transfer function shows the classical shape with a linear tracking region spanning +/-0.7 with a total capture angle of +/ Ground Target Tracking Some experimental work was undertaken to evaluate the tracking performance of the pulsed radar in cluttered environments. The radar was mounted along with a boresight TV camera on a two axis pedestal developed for the application. A joystick input or a raster search pattern scanned the beam in azimuth and elevation until a target was detected. After range tracking was established the angle errors generated by the conscan closed the angle tracking loops. The system was capable of tracking ground vehicles in clutter out to more than 1km as shown in Figure 9. Figure 9: Target tracking. Measured and filtered position estimates of the tracked vehicle Photograph of the unsealed road in a moderately cluttered tracking environment Figure 9a shows the measured position of a vehicle driving slowly ( 10km/h) along a section of dirt road in a moderately cluttered environment as shown in. The target was lost for a short period as the vehicle disappeared behind a spur, but was reacquired about 10s later and tracked continuously for the remainder of the 140s run. The closed loop tracking function was performed using analog processes discussed earlier, and the range voltage and angle transducer data was digitised at 20Hz and stored for processing and analysis. The racking error variances for this run are σ 2 x = 1.9m 2 and σ 2 y = 3.15m 2 which is good considering the clutter level. 3 Imaging Two different modalities can be used to generate radar images. Range gate limited imaging relies on the ability of a radar to obtain returns from a large number of contiguous range gates as the beam is scanned in azimuth to produce a 2D backscatter intensity plot of the surface. Beamwidth limited imaging relies on the use of a pencil beam scanned in two dimensions to build up a 3D image of the surface. These mechanisms are illustrated in Figure 10 [8]. 160

4 continuous wave (FMCW), it is the method of choice for range gate limited imaging [11]. The most basic open loop FMCW radar was constructed around a broadband 10mW VCO, an Alpha mixer and various circulators and couplers as shown in Figure 12. Figure 10: The difference between beamwidth and range gate limited imaging 3.1 Range Gate Limited Imaging This imaging method is the most common for radar as it can produce photographic quality images of the earth s surface from a low grazing angles as seen in the strip map images produced by side looking airborne radar and synthetic aperture radar systems [9, 10]. In real aperture scanned applications the images are produced in polar space which is not ideal for further processing, particularly if the frame of reference is moving, and modern systems usually map the polar image pixels into Cartesian space. This is most often achieved by selecting a Cartesian grid with a resolution equal to the cross-range resolution at the smallest operational range and then over-sampling at longer ranges. The most simple method to allocate data to the Cartesian image plane is to transform the coordinates (R,θ) of the centroid of each polar pixel into Cartesian coordinates (x,y), and use those to load the nearest Cartesian pixel. However, this method often results in unfilled pixels in the Cartesian plane which are distracting, so the alternative, less efficient, method of clocking through all of the Cartesian pixels in the image and allocating to each the magnitude of the nearest polar pixel is performed instead. Ideally, partial pixel overlaps are integrated into their respective bins in proportion to the percentage overlap as shown in Figure 11. B&K Ramp Generator Personal Computer VCO HP Spectrum Analyzer Isolator Mini-Circuits Amp Coupler Alpha Circulator Cassegrain Antenna Figure 12: Block diagram of a prototype non linearised FMCW radar The unit was driven by the triangular output of a standard function generator and the output spectrum measured using an HP-8751 spectrum analyser. Subsequent to this, more sophisticated high-resolution closed-loop linearised FMCW radars were designed and built by us for both ground based and airborne imaging applications. As part of the research into autonomous navigation and landing, high resolution ground based images produced using the HP spectrum analyser have been analysed in minute detail with the specific goal of determining the differences in reflectivity of various forms of clutter at low grazing angles [12]. Polar Image Cartesian Image Magnified View A B A B D E D E Figure 11: Polar to Cartesian image transform shows an example where each Cartesian pixel becomes the weighted average of the polar pixels with which it intersects Because of the ease with which the range gating function is achieved using frequency modulated Figure 13: Image analysis of a section of runway Airborne measurements which followed could not use the spectrum analyser because it was both too heavy and too slow and so dedicated hardware based on the Texas Instruments TMS320C40 was designed and built. This hardware was capable of digitising the radar video signal at 10MHz and running a 1024 point FFT in near real time. Major problems with EMI generated by the helicopter avionics and power systems were encountered with the result that the images obtained were rather noisy as is shown in Figure

5 BIAS, PRE-AMP -10 BIAS, PRE-AMP -10 1st International Conference on Sensing Technology Figure 14: Airborne images of a section of runway showing composite millimetre wave radar and aerial photograph The imaging range of a typical 94GHz FMCW radar is limited to about 750m because of issues related to reflected power and oscillator phase noise [13]. To allow for longer range operation, a higher power Interrupted FMCW radar was developed as shown in Figure 15. Ramp Gen VCO Isolator Attenuator Isolator PIN Switch Circulator Loop DRO Harmonic PILO Fref Phase Det. Frequency Discrim. LNA Figure 16: Comparison between an aerial photograph and a radar image of an airfield In the last few years this imaging work has been expanded with the development of light weight stabilised mirror scanned radar that could be carried by one of the UAVs developed at the ACFR [15]. 3.2 Beamwidth Limited Imaging Other high resolution 77GHz radars with mirror scanners have also been developed to produce 3D images of the terrain. As can be seen in Figure 17, these radars were based on configurable modules developed for the application. IN OUT IN OUT Figure 15: FMICW radar details schematic block diagram and photograph of integrated front-end and antenna GUNN VCO MODULE (HIGH DRIVE MIXER) IF LO MODULE (LOW DRIVE MIXER) BIAS IF LO In this configuration, the transmitted signal is interrupted during the receive period. This eliminates the problems associated with reflected power and allows the transmit signal to be increased to 250mW, but it introduces a number of issues regarding the spectrum analysis of the received radar video [14]. These problems notwithstanding, the system is capable of producing excellent quality images out to a range of more than 3km from extremely low grazing angles as illustrated in Figure 16. RAMP GENERATOR DISCRIMINATOR GUNN LO Figure 17: Configuration of a closed loop linearised FMCW front end using two configurable modules The use of mirror scanners allows the beam to be directed both quickly and accurately as the servos need only drive an extremely lightweight honeycomb aluminium disc as can be seen in the photograph of one of the larger systems shown in Figure

6 developed for high resolution millimetre wave radar systems. It has demonstrated the versatility and performance capability of these radar types for short range navigation and guidance applications. Figure 18: 3D Mirror Scanner with the 77GHz radar Unlike laser scanners, these radars are capable of producing high resolution images in the rain or through dust [9, 16], and because they have some foliage penetrating capability, even through trees [17] as is shown in the following 3D perspective image of the quadrangle outside the ACFR. Figure 19: Point cloud image of the quadrangle outside the ACFR These images are being used to analyse terrain and foliage characteristics for autonomous navigation applications [18-20]. Figure 20: Comparison between a volume rendered perspective view using radar data and a photograph of the same row of trees 4 Conclusions This paper has outlined extremely briefly some of the tracking and imaging applications that we have 5 References [1] U. C. Ray and A. K. Gupta, "Intrapulse Frequency Variation in a W-Band Pulsed IMPATT Diode," in Microwave Journal, vol. April, 1994, pp [2] T. T. Fong and H. J. Kuno, "Millimeter-Wave Pulsed IMPATT Sources," IEEE Transactions on Microwave Theory and Techniques, vol. MTT-27, [3] H. J. Kuno, "Solid State Millimeter-Wave Power Sources and Combiners," in Microwave Journal, June 1981, pp [4] D. Barton, Modern Radar Systems Analysis: Artech House, [5] M. Skolnik, Radar Handbook: McGraw Hill, [6] G. Brooker, "Conical Scan Antennas for W-Band Applications," presented at Radar 2003, Australian International Conference on Radar, Adelaide, South Australia, [7] A. Celliers, "The Design of a 94GHz High Resolution Coherent Radar," University of Cape Town, [8] G. Brooker, S. Scheding, M. Bishop, and R. Hennessy, "Millimetre Wave Radar Sensors for Mining Applications," presented at Radar 2003, Australian International Conference on Radar, Adelaide, South Australia, [9] N. Currie and C. Brown, Principles and Applications of Millimeter-Wave Radar: Artech House, [10] F. Henderson and J. Lewis, "Principles and Applications of Imaging Radar," in Manual of Remote Sensing, vol. 2, 3 ed: John Wiley & Sons, [11] G. Brooker, "Long-Range Imaging Radar for Autonomous Navigation," vol. Ph.D: University of Sydney, [12] N. Currie, R. Hayes, and R. Trebits, Millimeter-Wave Radar Clutter: Artech House, [13] G. Brooker, D. Birch, and J. Solms, "A W-Band Airborne Interrupted Frequency Modulated Continuous Wave Radar," IEEE Transactions on Aerospace and Electronic Systems, vol. to be published. [14] G. Brooker, "A W-Band Interrupted FMCW Imaging Radar," presented at SPIE Aerosense Conference, Orlando Florida, [15] A. Goktogan, G. Brooker, and S. Sukkarieh, "A Compact Millimeter Wave Radar Sensor for Unmanned Air Vehicles," presented at 4th International Conference on Field and Service Robotics, Yamanashi, Japan, [16] S. Ghobrial and S. Sharief, "Microwave Attenuation and Cross Polarization in Dust Storms," IEEE Transactions on Antennas and Propagation, vol. AP-35, pp , [17] P. Bhartia and I. Bahl, Millimeter Wave Engineering and Applications: John Wiley & Sons, [18] D. Langer, "An Integrated MMW Radar System for Outdoor Navigation," Carnegie Mellon University, [19] A. Foessel, "Scene Modelling from Motion-Free Radar Sensing," Carnegie Mellon University, [20] S. Clark and H. F. Durrant-Whyte, "The Design of a High Performance MMW Radar System for Autonomous Land Vehicle Navigation," presented at International Conference on Field and Service Robotics, Australia,