FSR sensors network: performance and parameters. Edgbaston, Birmingham, B15 2TT, UK

Transcription

1 FSR sensors network: performance and parameters V. Sizov (1), M. Gashinova (2), N.E.A. Rashid (2), N.A. Zakaria (2), P. Jancovic (2) and M. Cherniakov (2) (1) Department of Radio Electronics, Moscow State Institute of Electronic Technology, Moscow, , Russia (2) School of EECE, Microwave Integrated System Laboratory, University of Birmingham, Edgbaston, Birmingham, B15 2TT, UK Abstract The performance and parameters of ground-based short-range forward scatter radar (FSR) micro-sensors network are discussed: power budget for radar and communication channels; target detection capability; possibilities of target classification and communication with the command centre. The network structure and possible node hardware realisation are discussed. Introduction Research on a ground-based micro-sensor FSR network for situational awareness [1, 2] has been conducted at the University of Birmingham during the last few years. The network has been developed to detect and classify ground targets such as personnel on foot and vehicles entering the network coverage area. In this ad hoc wireless network each sensor (node) has a separate receiver and transmitter with omni-directional antennas, so that the transmitted signals are received by neighbouring nodes, creating the netted forward scatter radar configuration (Fig. 1). The FSR sensors can give advantages in difficult (small, low speed, stealth, etc.) target detection as well as rapid network deployment [3]. In contrast to networks currently used for situational awareness, no other specific sensors (acoustic, magnetic, seismic, etc) are used in the system, except the proposed FSR. The same radio frequency (RF) technology and hardware are used for targets sensing and for the internode communications. Thus, the network hardware is compact, light-weight and can be battery powered for a long period of time. The system operates in VHF band and is built from low-cost COTS components and subsystems. The sensors could be dropped from the air vehicle without the need for a specific ground-based orientation or configuration. To reduce the landing shock and to avoid the sensors from rolling over an unpredictable area when touching the ground, a light parachuting subsystem is recommended. FSR sensors are capable of operating in any weather conditions (rain, snow, sand, dust, etc.), in complex terrain and when the environment includes vegetation, foliage, fog, etc. Communication with the command centre is based on the IRIDIUM satellite service that guaranties global 24/7 coverage. Figure 1: The concept of the FSR network 7 th EMRS DTC Technical Conference Edinburgh 21 A13

2 In this paper, results of theoretical [4-7] and experimental study [8-1] of the system are summarised. The network structure is proposed and estimated FSR micro-sensor performance is discussed. FSR network subsystems The network includes four major segments: forward scatter radar that provides the target sensing capability (i.e., detection and parameters estimation); inter-nodes communication to transfer data within the network; navigation which operates at the initial stage to register node positions, and finally global communication for data exchange between the network and the command centre. To achieve maximum network reliability, it is proposed that all the nodes have the same hardware and software, and their functions are determined and redetermined from the command centre after the network deployments. A. Radar segment A pair of neighbour nodes, one working as a transmitter (TX) and another as a receiver (RX), creates one FSR (Fig.2). RX x φ y o z Figure 2: The FSR geometry and the target signature y Baseline TX If a target is crossing the baseline, the scattered signal interferes with the direct path signal, and introduces phase and amplitude modulation (Doppler shift) in the resulting waveform at the receiver input [5]. This target Doppler signature depends on the target size and silhouette, speed and trajectory (Fig.2). The target parameters (duration, magnitude, envelope, spectrum, etc.) are extracted from the received waveform by the proper signal processing [6, 1] and are used for automatic target detection and parameter estimation. B. Communication with the command centre A Short Burst Data (SBD) IRIDIUM modem (Fig.5) in each node is used for communication to the remote central post (CP), i.e. command centre. The networking process is initiated by the nodes transferring their GPS coordinates to the CP and is continued by receiving from the CP the nodes respective identification numbers, carrier frequencies plan and some other information relevant to the network structure. In case of target detection the appropriate target information is sent to the CP. The SBD modem has been tested and its reliability has been confirmed as well as its robustness within target environments (open space, forest, urban, etc.). Normally, the SBD modem is in sleep mode. The information is sent and received by short messages (~34 bytes) with estimated rate of four times per hour. C. Inter-node communication segment This segment is a miscellaneous network communication function, used to circulate data between network nodes. It can also be activated if a satellite channel in a particular node is absent or a battery loses charge, so that information could be transferred to neighbouring nodes and then transmitted via satellite to the central post. The inter-nodes communication subsystem is fully integrated with the radar transceivers and operates at the same carrier frequencies. Due to the small Doppler frequency bandwidth (up to 2Hz) the target signature can be easily filtered from data rate frequency of 3-5 khz, so that the communication channel can simultaneously use the same transmitter as the FSR channel 7 th EMRS DTC Technical Conference Edinburgh 21 A13

3 and does not introduce extra power consumption or interference. D. Navigation segment At the initial stage after deployment, each node must determine its position. A GPS receiver (see Fig. 5) has been tested in the experimental network with sufficient accuracy. The receiver operates only for about the first 2 min (cold start) after deployment. Its power consumption has little effect on the system total power consumption. Network frequency layout We shall consider here the simplest FSR network in which only one frequency band for radar channels is used. Each transmitter has it own carrier frequency within the band. Receivers in the node are tuned to the frequencies of neighbour transmitters. The proposed network layout is shown in Fig. 3. There was specifically investigated for mobile telephony, that cellular hexagonal structure is optimal for area coverage. Frequency repetition TX RX Figure 3: Possible FSR network layout Of course, following deployment, nodes will not create the perfect geometry as shown in the figure. The position of each node will be random and it can expected that the dispersion of the sensors will be about.25 of the average baseline, e.g. σ=25m for 1m baseline. In most cases it is enough to use a two-line network, but where the terrain is especially complex the three-line layout may be more suitable. So, assuming a 1m baseline and two-line network layout, 1 sensors would be required to protect 5km of territory, It is sufficient for the nodes to have one TX channel and four RX channels to cover any Frequency repetition complex area. TX and RX channels should be tuned by carrier frequency in a set of 2-3 frequencies, separated within the band in 12-2 KHz. The frequency plan is defined by the central post (depending on the nodes GPS positions) and transferred to the nodes by the satellite communication channel. If the length of the area to be covered is large enough, TX node frequencies can be repeated periodically without significant interference due to very fast decreasing of the received power with the baseline distance (in 1 D ratio). Similar to the cellular 4 radio, in order to minimize the interference level and to save battery life, the transmitting power should be reduced for the nodes positioned at a shorter distances. Thus, total bandwidth occupied by the system could be less than.5mhz. The frequency separation of 12.5 khz gives a data rate of 5kbps, which has been tested with COTS multichannel transceivers made by Radiometrix (see Fig. 5). Node hardware An artistic image of a proposed node assembly is shown in Fig. 4. The node has a tetrahedral shape which is ideal for a freestanding device with unknown orientation after the deployment. Radome FSR antenna GPS/Iridium antenna Figure 4: Possible FSR node assembly Four antenna systems protrude from the node, three touch the ground and the fourth vertical one is the FSRare used, but only one of them (closest to vertical orientation) is active, and the others three are grounded. The radar antennas are vertically polarized monopole antennas. The GPS/Iridium 7 th EMRS DTC Technical Conference Edinburgh 21 A13

4 antenna has circular polarisation and is integrated into the node as a patch antenna. Antennas are protected by shock absorbing radomes, which prevent antennas from mechanical damage and at the same time minimise the node rolling over after the landing. The node should have some kind of parachuting system to decrease the speed and to reduce the bump. The parachute design has not been considered in the research. The node block-diagram is proposed in Fig.5. The nodes consist of one TX and four RX FSR channels with antennas, a GPS receiver and IRIDIUM modem, a microcontroller unit (MCU) and a digital signal processor (DSP) and memory. GPS/Iridium antennas Orientation sensor Antenna switch Control Switch Control GPS receiver GPS receiver Iridium modem SBD modem 4-channel receiver/adc Data Data, control Data MCU EPROM/ RAM DSP FSR antennas Antenna switch Splitter Receiver TX Transmitter Data, control Target pre-detection, communications Permanetly ON Target detection, speed estimation, target recognition Occasionally ON Figure 5: FSR node block-diagram At the final stage of the development all hardware can be packed into two silicon chips, one of which is a RF subsystem and another is a digital processing and control subsystem. The antennas are combined into two groups by four FSR and four satellite antennas. Only one pair of antennas is switched to be active, controlled by an orientation sensor. Inactive FSR antennas are grounded. RX and TX channels use frequency synthesizer to set any of the carrier frequency in the working band. Micro-power MCU performs the control of all hardware and simple data processing algorithms, like Doppler filtering and target pre-detection as well as communication. If a target (or false alarm) is detected, more complex data processing can be made in a more powerful DSP unit. These algorithms include the estimation of target size, speed and trajectory and target classification. The transmitter, the receiver and microcontroller are switched on permanently and are the main sources of the node s power consumption. A fragment of the experimental network is shown in Fig. 6. TX1 TX2 TX3 FSR antennas RX GPS antenna Figure 6: Network fragment tested Modem 7 th EMRS DTC Technical Conference Edinburgh 21 A13

5 It contains one receiver (RX) and three transmitter (TX) nodes, each operating at different set of frequencies in the VHF and UHF bands (64; one of 135, 144, 151 or 173; and 434 MHz). Receiver and transmitter utilise omni-directional, verticallypolarised monopole antennas, separately for each carrier frequency. This experimental equipment was not optimised from the point of view of weight, size and power consumption. It was developed using off-the-shelf components (see Fig.6), which can be easily replaced and tuned for the required carrier frequencies set. The measurements for all frequency bands were made simultaneously and the results were compared to determine the best one. Each node has a GPS receiver. The central post computer communicates to the nodes by the terrestrial modem (in Fig. 7) operating at the same frequency as one of the channels (in trials this was at 151MHz). Power budget As a part of the study it was experimentally confirmed in [5] that wave propagation above flat ground with minimal antenna heights is well described by a two-ray path (TRP) propagation model even though the antennas were positioned directly on the ground with "zero" elevation. The ground is considered as a dielectric material with some values of permittivity and conductivity. In this model the propagation loss essentially depends on the carrier frequency (in approximately square power for VHF band). A. Inter-node communication channel In Fig. 8 the signal-to-noise ratio (SNR) graphs are shown for different carrier frequencies in the assumption of dbm radiated power and dbi gain of TX and RX antennas. The db value on the graph's vertical axis corresponds to received noise power (the sum of thermal noise at the receiver noise figure (NF) 3dB and ambient noise for a quiet rural environment) in 1kHz receiver bandwidth (BW). The figure shows that for 12dB SINAD ratio the communication distance is increased with carrier frequency decreasing from 3m for 434MHz to 63m for 64MHz. SNR, db P G G T T R dbm Quiet rural environment, 1kHz BW 64 MHz 135 MHz 173 MHz 434 MHz 12dB SINAD Distance, m Figure 8: Communication channel power budget B. Radar channel The same TRP propagation model is used for the prediction of target received power in FSR channel. FSR power budget is given by the known equation [5]: P P 4 L R GRGT 2 T T tg L tgr, (1) where P R and P T are received and transmitter power, respectively; G R and G T are the RX and TX antenna gains; is target RCS defined for free space; is wavelength; and LT tg, Ltg R are TRP loss on the path from transmitter to target and from target to receiver, respectively. Received power, dbm Ambient noise Optimal band Thermal noise Receiver IF BW=1kHz, NF=3dB** Frequency, MHz Strong wind Medium wind Figure 9: Radar channel power budget Target Residental Rural Quiet rural Light wind 7 th EMRS DTC Technical Conference Edinburgh 21 A13

6 In Fig. 9 the target received power (calculated for human target) vs. the carrier frequency is shown in comparison to received noise power (thermal and ambient noise for different environments) and vegetation clutter. All FSR parameters are calculated for a 1m baseline distance, unity antenna gains and dbm TX power. The target received power has the maximum at 7MHz due to resonance scattering when target height (1.75m) is close to half of wavelength / 2 [4]. The receiver thermal noise, calculated for NF=3dB and intermediate frequency (IF) BW=1kHz is small enough compared to target power, but the presence of ambient (man-made and galactic) noise increases the total received noise power at lower frequencies. The signal-to-noise ratio can be increased by TX power increasing. In rural and quiet rural landscapes, dbm radiated power gives more than 1dB SNR in optimal frequency band 6-8MHz. Clutter power depends on many factors: type and density of vegetation, wind speed, carrier frequency, baseline distance, etc. [7]-[9]. In the figure, average estimated values of clutter for medium vegetation density are shown for different wind speeds. Clutter is the main factor limiting the FSR detection range at higher frequencies. Even for medium wind speed and medium vegetation density clutter power may exceed target power at frequencies higher than 15MHz and make it difficult to detect target against the background of clutter. The primitive increase of the TX radiated power will not give any improvement in the system performance due to proportional increase in clutter power. To have reliable target detection, proper signal processing algorithms should be used. Signal processing algorithms A. Non-coherent target detection The non-coherent target detection algorithm is a real-time procedure which includes target signature filtering in corresponding Doppler band (1.5Hz for the human and 1Hz for the car at 64MHz carrier frequency), envelope (amplitude) detection and comparison with constant or adaptive threshold. The algorithm is schematically shown in Fig. 1 and 11. Target signature Band-pass filter Envelope detector Low-pass filter 1 Low-pass filter 2 Constant threshold Threshold detector Adaptive threshold Figure 1: Non-coherent target detection blockdiagram Non-coherent target detection is not optimal, but it does not require high processor power and can be realised in a microcontroller unit (Fig.6). It is also tolerant to the target's trajectory parameters. Experimentally tested target detection ranges for different carrier frequencies using this algorithm are presented in Table 1. The measured target detection range is limited by clutter and is much less than the one defined from the receiver s sensitivity only. TX Amplitude,V A, V v 6 km/h Doppler signal 64 MHz Target trajectory RX Signal envelope -1 1Adaptive 2 threshold Constant Threshold Time, min Time, mins Figure 11: Non-coherent target detection Detection 7 th EMRS DTC Technical Conference Edinburgh 21 A13

7 t,s ec t, sec t, se c t.,s 5 ec t,s ec The non-coherent algorithm can be used as pre-detection procedure; if the target (or a false alarm) is detected, more precision quasi-coherent processing can be applied to the target signature. A mplitu de, V Raw record 64 MHz Sliding window 4 sec min 64 MHz Partially overlapped sliding windows B. Quasi-coherent processing This algorithm was described in detail in [6] and [1]. This is a post-processing algorithm based on the convolution of a stored target signature with pre-calculated reference functions: R T / 2 St S t, v, y dt T / 2 S ref where S (t) is a received target signature with unknown parameters, and t, v, y is a set of reference functions calculated for a number of estimated targets with time duration T, vector of target speed v and baseline crossing point y. For one of the combinations the convolution set R has the maximum; and we can estimate the parameters of target trajectory: vector of speed and crossing point position corresponding to the max R. ref The algorithm realisation is shown schematically in Fig.12. To enable continuous processing, the real-time data flow is divided into a number of overlapping sliding windows (see the figure) in which signal realisation limited in time is correlated to the set of predefined reference functions. Maximum of convolution is compared to adaptive or constant threshold to make a decision of target presence. After the convolution of a stored target signal with reference function the target signal is compressed in time-domain (Fig.13-a). If there is more than one target in a convoy, the quasi-optimal signal processing provides the target s resolution (Fig. 13 b)., Amplitude,V : : Set of reference functions.1 : : : : Time, min max[r(t)] A mplitude, V Threshold detector Threshold Detection M ax corr s ignal Tim e, min Figure 12: Quasi-coherent signal processing Amplitude, V Filtered modelled signal std=.1*max(sr) Time, s Compressed signal (abs norm zoom) a) Time, sec b) Figure 13: Target signal compression As a result, signal-to-noise and signal-toclutter ratio are increased. The processing gain depends on a number of factors: target envelope time duration, target velocity and trajectory, carrier frequency, baseline distance, etc. In general, the processing gain is defined as the ratio between the bandwidth of the signal and appropriate filters at the input of the receiver and an effective signal processing bandwidth (inversion to the integration time). Thus, if the signature duration is about 3s, i.e. the effective band- Norm. magnitude C o m p r essio n o f M o d _s ig n al& C lu tter T im e, s e c 7 th EMRS DTC Technical Conference Edinburgh 21 A13

8 width is about.3hz and the possible Doppler spectrum bandwidth at the receiver input is 1.5Hz for 15MHz (running human), the processing gain is ~17dB. This algorithm requires a sufficient processing power and should be processed in the DSP unit in the node (Fig. 6). Probability of target detection and false alarms The performance of FSR in terms of probability of target detection (P TD ) and false alarm (P FA ) has been estimated in a series of two long-term experiments. The experiments were conducted over several days in areas controlled by videosurveillance in Horton Grange at the University of Birmingham (medium trees density and medium wind speed) and in the territory of the Roke Manor Research Ltd. in Romsey (low trees density and low wind speed). The baseline distance was 1m. Both FSR positions were surrounded by medium-dense vegetation. The records from several video cameras for all targets moving inside and outside FSR position were stored and then analysed and compared to the results of automatic target detection by FSR. The detection threshold was installed automatically to have constant value of false alarms caused by clutter and noise at 1-3 level. The results of around 12 hours surveillance are presented it Tables 2 and 3. Table 2: Roke Manor test site Frequency, MHz Video detections Non-coherent / Coherent detection Humans detected 21 2/ /2 16 Humans missed 1/ /1 5 P TD, % 95/1 1 1/95 76 False alarms / 2 15/3 9 P FA*1 3 / 1 11/2 6 In these measurements (Roke Manor test site) the number of false targets (humans, cars moving outside FSR position and small animals moving inside) was comparatively small. Table 3: Horton Grange test site Frequency, MHz Video detections Non-coherent / Coherent detection Humans detected 49 48/ Humans missed 1/ P TD, % 98/ False alarms 18/ P FA*1 3 2/ The measuring was done against a background with greater levels of vegetation clutter than in previous experiments, comparable to target signal magnitude, especially at higher frequencies. This meant that more targets were missed as the adaptive threshold in a non-coherent detection procedure was increased. There were also a lot of false targets in the Horton Grange test site situated in an urban area. False targets are cars and humans which were not crossing the baseline but moving close to the FSR position. Another type of FA is caused by interferences specifically present in urban area. Thus, the P FA is also increased by the influence of all these source of false alarms. If a target crosses more than one baseline, the probability of detection increases. Assuming the same probability of false alarm rate and equal signal to clatter ratio, the probability of the detection can be estimated as follow: P 1 (1 ) N DN PD, where P DN is the probability of a detection target crossing N baselines, e.g. N=4, P D =.7, P D4 =.99. Target RCS estimation Target signal magnitude depends on the crossing point: the closer the target is to one of the nodes (symmetrically, TX or RX), the greater the target signal (see Fig.11 and 14). 7 th EMRS DTC Technical Conference Edinburgh 21 A13

9 Target-to-leakage ratio, db BL=1m F=64MHz Crossing point, m Figure 14: Measured and predicted target signal magnitude vs. crossing point position By measuring the target s received power and calculating the TRP loss for known crossing points, we can estimate target RCS using power budget equation (1). This presents the possibility of target classification being informed byusing the information of data regarding the target size and speed. Target classification Car(Cubic approximation) Human (Cubic approximation) Human 1 (measured) Human 2 (measured) Car (measured) Sheep (measured) Estimated car signal,17v Estimated human signal,7v Coherent processing gives 2m from the baseline middle In Table 4 the simulated forward scatter RCS values for different targets are shown. The simulation is made by the technique presented in [4]. It can be seen from Fig.14, that targets with different RCS can be separated effectively within an 8% zone of the total baseline length. Table 4: RCS and speed of different targets Target RCS, dbsm, at 64MHz Speed, km/h Small animals < -1-2 Humans Cars Tanks Trucks Taking into account possible target speed variance, the probability of true target classification may be increased as shown graphically in Fig. 15. The algorithm development of automatic target classification by target speed and size is still in progress. Target-to-leakage ratio, db Humans Target speed, km/h Big vehicles Cars Small animals 5 Cross-point, %BL Figure 15: The principle of target classification by speed and RCS A more accurate ATC algorithm was considered and described in [11]-[18]. It is based on the comparison of a target s measured signature power spectrum density with data held in a database containing spectra of previously collected different targets. Principal component analysis is used to reduce the number of parameters compared. The effectiveness of the algorithm was demonstrated first at higher frequencies (869 and 434MHz) and then confirmed for lower frequencies (151MHz) [18]. Processing the algorithm requires a lot of memory and great processor power and therefore it cannot be done in a node DSP. Thus, the detailed consideration of this method is outside the scope of the paper. Node hardware power consumption Estimated values of the node hardware blocks (see Fig. 6) power consumption at 3.6V supply are shown it Table 5. It can be seen, that all blocks which are occasionally On have negligible total power consumption, even though their peak current consumption is large. This is the same for the modem and DSP. Average current is defined mostly by permanently On blocks (FSR transmitter, receiver and microcontroller) and can be made as small as 2-25mA. A standard D-size lithium thionyl chloride battery with 19AH capacity at 3.6V will provide an estimated 3 days of battery lifetime. 7 th EMRS DTC Technical Conference Edinburgh 21 A13

10 Table 5: RCS and speed of different targets Block name Current, ma Time of On state GPS receiver < 12 Max 2 minutes at start SBD modem TX: 8 RX: 8 1 burst of 8.3ms per hour 4 bursts of 4.5s per hour FSR transmitter 5-1 Permanently FSR receiver &ADC 15-2 Permanently Microcontroller 5 Permanently DSP <1 Few minutes, if target is detected Conclusions The FSR micro-sensor network is proposed for ground target detection and classification in situational awareness. Theoretical analysis and experimental testing of critical FSR sensors parameters show the possibility of producing the FSR micro-sensors and the network itself by using low-cost off-the-shelf hardware modules and current digital communication and signal processing technologies. The FSR micro-sensors operating at 6-8MHz carrier frequency with radiated power on dbm provide reliable human target (which is smallest target of interest) detection at a range of 1-15m under any weather conditions. The ad-hoc FSR network could be recommend for further development and for use in ground operations for situational awareness. Acknowledgments The work reported in this paper was funded by the Electro-magnetic Remote Sensing (EMRS) Defence Technology Centre, established by the UK Ministry of Defence and run by a consortium of SELEX Galileo, Thales UK and Roke Manor Research. References 1. M. Cherniakov and V. Sizov, Netted Forward Scattering Micro Radars for Ground Targets in Proc. of the 3d Annular Technical EMRS/DTC Conference, A27, Edinburgh, UK, M. Antoniou, V. Sizov, Cheng Hu, P. Jancovic, R. Abdullah, N.E.A. Rashid and M. Cherniakov, The Concept of a Forward Scattering Micro- Sensors Radar Network for Situational Awareness, in Proc. of the 28 International Conference on Radar, Adelaide, Australia, Sept. 2-5, 28, pp Bistatic Radar, Principles and Practice, M. Cherniakov, Ed.,: John Wiley & Sons, V. I. Sizov, M. Cherniakov and M. Antoniou, Forward Scatter RCS Estimation for Ground Targets, in Proc. of the EuRAD27, Munich, Germany, 1-12 Oct V. Sizov, M. Cherniakov and M. Antoniou, Forward scattering radar power budget analysis for ground targets, IET Radar, Sonar & Navigation, V.1, Issue 6, pp , Dec Cheng Hu, M. Antoniou, M. Cherniakov and V. Sizov, Quasi-Optimal Signal Processing in Ground Forward Scattering Radar in Proc. of the 28 IEEE Radar Conference, Rome, Italy, May 26-3, 28, pp V. Sizov, Cheng Hu, M. Antoniou and M. Cherniakov, Vegetation Clutter Spectral Properties in VHF/UHF Bistatic Doppler Radar in Proc. of the 28 IEEE Radar Conference, Rome, Italy, May 26-3, V. Sizov, M. Gashinova, N. E. Rashid, J. Chen and M. Cherniakov, Forward Scattering Micro Radar efficiency analysis for different landscapes, in Proc. of the 6 d Annular Technical EMRS/DTC Conference, A1, Edinburgh, UK, M. Gashinova, M. Cherniakov, V. Sizov and N.A. Zakaria Empirical Model of Vegetation Clutter in Forward Scatter Radar Micro-Sensors, in Proc. of the 21 IEEE International Radar Conference, Washington DC, USA, 9-15 May M. Gashinova,, V. Sizov, M. Antoniou and M. Cherniakov, Signature modelling and coherent target detection for Forward Scattering Radar (FSR) sensors, in Proc. of the 29International Radar Symposium, 9-11 Sept. 29, Hamburg, Germany. 11. M. Cherniakov,, V.V. Chapursky, R. Abdullah, Short-Range Forward Scattering Radar in Proc. of the International Radar Conference, Toulouse, France. October 18-24, th EMRS DTC Technical Conference Edinburgh 21 A13

11 12. R. Abdullah, M. Cherniakov, Forward Scattering Radar For Vehicles Classification, in Proc. of the First International Conference VehCom-23, Birmingham, UK, June 23, Proceeding pp R. Abdullah, M. Cherniakov and P. Jancovic, Automatic Vehicle Classification in Forward Scattering Radar, in the Proc. of 1 st International Workshop on Intelligent Transportation, 24, Hamburg, Germany, pp M. Cherniakov, R. Abdullah, P. Jancovic, M. Salous and V. Chapursky, Automatic ground target classification using forward scattering radar, IEE Proceedings Radar, Sonar and Navigation, Vol. 153, Issue: 5, pp M. Cherniakov, M. Salous, P. Jancovic, R. Abdullah and V. I. Kostylev, "Forward scattering radar for ground targets detection and recognition,'' in Proc. of the 2 d Annular Technical EMRS/DTC Conference, Edinburgh, UK, M. Cherniakov, R. Abdullah, P. Jancovic and M. Salous, Forward Scattering Micro Sensor For Vehicle Classification, in Proc. of the IEEE International Radar Conference, Washington DC, US, pp , R. Abdullah, M. Cherniakov, P. Jancovic, M. Salous, Progress On Using Principle Component Analysis In FSR For Vehicle Classification, in Proc. of the 2 d International Workshop On Intelligent Transportation, WIT 25, Hamburg, Germany 18. N.E.A. Rashid, M. Antoniou, P. Jancovic, V. Sizov, R. Abdullah and M. Cherniakov, Automatic Target Classification in a Low Frequency FSR Network in Proc. of the 5th European Radar Conference, Amsterdam, the Netherlands, Oct , 28, pp th EMRS DTC Technical Conference Edinburgh 21 A13