Dynamic Sensor Selection for Cognitive Radar Tracking

Transcription

1 Dynamic Sensor Selection for Cognitive Radar Tracing Fulvio Gini, Pietro Stinco, Maria S. Greco Dipartimento di Ingegneria dell Informazione, Università di Pisa This wor has been funded by SELEX Sistemi Integrati, Italy 1

2 2 of 28 Why Cognitive Radar? There are many variations of the definition of cognition. One which is relevant to cognitive radar is that of the National Institutes of Health (NIH), National Institute of Mental Health (NIMH): Cognition: Conscious mental activity that informs a person about his or her environment. Cognitive actions include perceiving, thining, reasoning, judging, problem solving, and remembering. While true thining machines are still very much the stuff of science fiction, it is still possible to be guided by the above principles and map each of the above cognitive properties into real engineering systems.

3 3 of 28 Why Cognitive Radar? Cognitive Property Perceiving Thining, Reasoning, Judging, Problem Solving Remembering Memory Cognitive Radar Equivalent Sensing Expert Systems, Rule-based Reasoning, Adaptive Algorithms, and Computation Memory, Environmental Database What exactly are the potential benefits of a radar possessing some manner of the cognitive abilities described in the table? This of course depends on the type of radar, its mission, and the environment in which it must operate.

4 4 of 28 Why Cognitive Radar? The concept of Cognitive Radar has been introduced for the first time by Prof. Simon Hayin in the seminal paper: S. Hayin, "Cognitive radar: a way of the future", IEEE Signal Processing Magazine, Vol. 23, No.1, pp , January, published in the Special Issue of the SPM: Special Issue on "Knowledge-based systems for adaptive radar", Guest Editor: F. Gini, IEEE Signal Processing Magazine, Vol. 23, No. 1, January 2006.

5 5 of 28 Why Cognitive Radar? A conventional adaptive radar Source: J. R. Guerci, Cognitive Radar, Artech House, A conventional radar is adapative, but adaptivity is usually confined to the receiver and is often very reactive to the receive data stream, i.e. adaptivity is based solely on the data stream to be interrogated for targets. There is very little provision for learning over time, feedbac to the transmitter, or integration of exogenous environmental information sources that can provide significant benefits, such as e.g. Digital Terrain Mapping (DTM).

6 6 of 28 Why Cognitive Radar? A cognitive radar Source: J. R. Guerci, Cognitive Radar, Artech House, A cognitive radar exhibits a number of advanced elements that could be argued to better mimic biologically cognitive systems (such as e.g. echolocated mammals lie bats).

7 7 of 28 Why Cognitive Radar? A cognitive radar An environmental dynamic database (EDDB) that contains nowledge of the environment and/or targets of interest obtained from both onboard (endogenous) and off-board (exogenous) information sources. The EDDB is a primary component of Knowledge Based (KB), or Knowledge Aided (KA), algorithms.

8 8 of 28 Why Cognitive Radar? A cognitive radar In addition to an adaptive receiver, a cognitive radar includes an adaptive transmitter, and thus it provides a feedbac from the receiver chain. The presence of this feedbac has been identified by Simon Hayin in his SPM paper as an essential ingredient in any cognitive radar.

9 Why Cognitive Radar? A cognitive radar A cognitive tracing radar implements in the KA processor a cognitive waveform selection (CWS) algorithm that can optimally pic the transmit waveform from a prescribed library, in response to information fed bac from the receiver to the transmitter. In accordance with dynamic programming, the CWS algorithm sees to minimize the expected tracing error over a temporal horizon of prescribed length. S. Hayin; A. Zia; I. Arasaratnam; Yanbo Xue; Cognitive tracing radar, 2010 IEEE Radar Conference, pp , Washington D.C., USA, May of 28

10 10 of 28 To Learn More Intelligent Adaptive Signal Processing and Knowledge Based Systems (KBS) for Radar: Knowledge based radar detection, tracing and classification, F. Gini and M. Rangaswamy editors, John Wiley & Sons, Inc., Hoboen, New Jersey, Adaptivity of Transmission Waveform Diversity and Design (WDD): Waveform Design and Diversity for Advanced Radar Systems, F. Gini, A. De Maio, L. K. Patton editors, IET, Radar Sonar and Navigation Series 22, in press, 2012.

11 11 of 28 Subject of the Tal Dynamical sensor (i.e. waveform) selection in a passive radar system for ship tracing in the context of a maritime border control scenario Outline: Passive Coherent Location (PCL) Systems FM radio station UMTS base station Measurement Error Transmitters of opportunity Bistatic Ambiguity Function (BAF) Bistatic CRLB on range and velocity estimation Posterior Cramér-Rao Lower Bounds (PCRLB) for target state estimation Sensor Selection based on Bistatic PCRLB Numerical results & conclusions

12 Passive Coherent Location (PCL) Systems 12 of 28 Passive radars (PCL) do not emit e.m. waves, but exploit those produced by sources of opportunity (FM radio, DVB-T, DAB, UMTS, etc.) The different illuminator systems are characterised by individual strengths and weanesses with respect to radar requirements fields of application have been identified for the different sensor types. This ability to exploit the available illuminators as source of the e.m. power for target detection is one of the main strengths of passive radar. At the same time it is one of the main challenges in system design: the transmitted waveform is not within the control of the radar engineer, so that, unlie in the conventional radar systems, it has not been properly designed to achieve an ambiguity function with desirable properties (e.g. high resolution in range and Doppler frequency and low sidelobes level).

13 13 of 28 Passive Coherent Location (PCL) Systems As regard coastal surveillance, PCL systems are expected to offer a number of advantages in terms of eco-compatibility and sustainability. In fact, they can be installed even in protected and populated areas, reasonably without providing additional e.m. pollution. [ For more details on PCL systems, see the forthcoming Special Issue of the IEEE AES Systems Magazine on PCL Systems, Oct.-Nov ]

14 14 of 28 Analyzed Scenario Receiver UMTS Tx FM Tx Monte Serra Multistatic System: One Receiver: In via Michelangelo near Leghorn harbour. Antenna gain of 10 db. HPBW of 3. Two Transmitters: UMTS Base Station: In Piazzale Marmi, 453 m away the receiver in the South-East direction. FM commercial radio station: In Monte Serra, 36 m away the receiver in the North-East direction.

15 15 of 28 Transmitted Waveforms UMTS Base Station N 1 1 U u ( t) c g( t nt ) U n U NU n0 gt () c n T U N U Root Raised Cosine, α=0.22 QPSK symbols, independent identical distributed (i.i.d) Inverse of symbol rate (0.26 μsec) Number of integrated pulses (N U T U = 0.1 sec = coherent integration time) FM Radio Station 1 j sin2 f0t / 2 t uf () t e rect T T F F f 0 BC T F Modulation Index (5) Frequency of the modulating signal (Audible Frequency Range, 15Hz) 2 f0 Carson s bandwidth (150 Hz) coherent integration time (0.5 sec)

16 Complex Ambiguity Function (CAF) Definition of CAF: H, a, H, a u(t a )u* t H exp j 2 H a t dt true delay hypothesized delay hypothesized Doppler frequency true Doppler frequency UMTS Base Station: NU 1 1 NU 1 * * j 2 t, E c g ( t nt ) c g ( t nt ) e dt U U n 0 NU n 0 1 sin NU TU j TU ( NU 1) e g, NU sin TU N.B.: The expectation is taen w.r.t. the QPSK symbols (Average Complex Ambiguity Function ACAF) FM Radio Station sin n f 0 TF j n f, e j n / 2e j 2 f e J n J n f 0 TF n 0 where: H a H a 0 16 of 28

17 17 of 28 Bistatic Ambiguity Function (BAF) To obtain the BAF, which is in terms of target range r and bistatic velocity v, time delay τ and Doppler frequency ξ must be replaced by the following: r r Lr 2L r L sin 2r L sin ( r, (, rl,), L) c c f f 11 rrl Lsin sin ( r( r, v, v,,, L, L) ) 2 2 vv C C c c rr L 2r2 Lsin sin L r L Tx=(0, L) rt v Tg=(x, y) r Bistatic Geometry The relation between time delay τ and Doppler frequency ξ, and range r (from RX to target) and (bistatic) velocity v is not linear. -θ Rx=(0, 0) The BAF depends on the transmitted waveform and on the bistatic geometry (which changes with discrete-time when the targets moves along its trajectory).

18 Cramér-Rao Bounds for Monostatic and Bistatic Radars Monostatic FIM (Van Trees, Vol. III): J, 2SNR M, X, 2, X, X X 0 0 Bistatic FIM: Bistatic FIM of range and velocity is related to monostatic FIM of time-delay and Doppler shift: r, v, J P J P B T M UMTS Base Station:* J J M M 1,1 2, SNR 2 3T T T N 1 2SNR 4 3 J J M 0 1,2 M 2,1 0 J 2SNR 2 f T sin 2 f T J M M 1,1 2, f T 2 2 T SNR J J 2 cos sin M 1,2 M SNR f 2,1 0T f0t f0t Tf 0 2 where P r v r v 2 3 FM Radio Station: N0 fc 4 r rt r rt SNR GP c T P depends on the bistatic geometry only! - J M depends on the transmitted waveform (and on the bistatic geometry through SNR). 1 * We calculated the Modified FIM, obtained by taing the expectation of the conditional FIM (the FIM for a fixed code sequence) 18 of 28

19 19 of 28 UMTS Measurement Errors Range Velocity Better range estimation accuracy due to the transmitted waveform. Lower accuracy in the far range due to the lower transmitted power.

20 20 of 28 FM Measurement Errors Range Velocity Better velocity accuracy due to the higher integration time. Better accuracy in the far range due to the higher transmitted power.

21 21 of 28 Best Channel Map Range Velocity Better range accuracy for the UMTS channel (Ch.1) Better velocity accuracy for the FM channel (Ch.2) Better accuracy in the near range for the UMTS channel Better accuracy in the far range for the FM channel

22 Channel Selection based on Bistatic PCRLB We propose to select dynamically, at time +1, the measurements coming from the best bistatic channel selected according to the lowest value of the measurement information term of the Bistatic Posterior CRLB (PCRLB) along the trajectory of the traced target. The idea of utilizing some sort of CRLB for sensor selection in distributed radar networs has already been described in: M. S. Greco, F. Gini, P. Stinco, and A. Farina, Cramér-Rao bounds and selection of bistatic channels for multistatic radar systems, IEEE Trans. on Aerospace and Electronic Systems, Vol. 47, No. 4, pp , October H. Godrich, A. P. Petropulu, and V. H. Poor, Sensor Selection in Distributed Multiple-Radar Architectures for Localization: a Knapsac Problem Formulation, IEEE Trans. on Signal Processing, Vol. 60, No. 1, pp , January (In another paper the same authors proposed the use of the CRLB for power allocation for target localization in distributed multiple-radar architectures, IEEE-SP July 2011). 22 of 28

23 Analyzed Case: - Target approaching Leghorn harbour - Deterministic target trajectory (n =0) - Probability of Detection (P D ) < 1 State Equation: x Fx Ga n where: x 1 Measurements Equation: T hx h x, h x w ~ N 0, R z h x w where: x, x, y, y r v T Range Bistatic Posterior Cramér-Rao Lower Bounds 1 T F, G T Single Channel Velocity Measurements at time are available only if the target has been detected (P D <1). R J P J P 1 B T M 2x2 non-diagonal, time-varing, covariance matrix 23 of 28

24 Bistatic Posterior Cramér-Rao Lower Bounds 24 of 28 PCRLB on target state estimation: ˆ ˆ J E x x x x 1 T Single Channel 1 T 1 T 1 1 PD J F J F H R H A-priori Information (previous target state) Dynamical channel selection: Select the channel (Ch.1: UMTS, Ch.2: FM) with the lowest global measurement error, i.e. the channel with the higher value of the measurement information index. PCRLB with channel selection: 1 T 1 CH 1 1 J F J F J P CH ( i) ( i) T ( i) 1 ( i) 1 max det D i 1,2 J H R H Measurement Information For each point of the surveillance area it is possible to evaluate the best channel, i.e. the channel with the higher value of the measurement information index.

25 25 of 28 Bistatic Posterior Cramér-Rao Lower Bounds As a result, the PCRLB depends on target trajectory, sensor accuracy, transmitted waveform and the bistatic geometry. PCRLB: RPCRLB of x [m] x UMTS FM SELECTED RPCRLB of V x [m/sec] 1 0,1 0,01 x UMTS FM SELECTED 0, , There is a gain with respect to each bistatic channel. The performance we get are equal or better than the performance of the channel with the lowest PCRLB.

26 26 of 28 Bistatic Posterior Cramér-Rao Lower Bounds PCRLB: y y RPCRLB of y [m] UMTS FM SELECTED RPCRLB of V y [m/sec] 1 0,1 0,01 UMTS FM SELECTED 0, , Initially the performance are the same as that of the FM channel, because in the far range this channel has the highest SNR. When the target approaches the harbour, exploiting the UMTS channel, which has better range estimation accuracy, the proposed receiver is able to increase the performance in estimating the target trajectory.

27 Conclusions 27 of 28 We analyzed a multisensor radar system in the context of a maritime border control scenario. The system is composed by two PCL channels: UMTS base station and FM radio station. We evaluated the measurement errors for each channel in terms of Bistatic CRLB. We proposed an algorithm based on the Bistatic Posterior CRLB for dynamically selecting the transmitter for the target tracing. We compared the performance of the proposed receiver with the performance of each bistatic channel performance improvement w.r.t. the channel with the lowest PCRLB.

28 ICASSP IEEE Signal Processing Society 28 of 28