Dynamic Data Driven Informa0on Fusion for Situa0onal Awareness

Transcription

1 Dynamic Data Driven Informa0on Fusion for Situa0onal Awareness Biao Chen Syracuse University Co-PIs: Yingbin Liang, Jiang Tang, and Pramod Varshney (Syracuse) Venugopal Veeravalli (UIUC)

2 Background: Airborne MIMO (with AFRL) Selected Publica-ons/Patent B. Chen and M.J. Gans, "MIMO communicajons in ad hoc networks," IEEE Trans. Signal Processing, July X. Shang, B. Chen, and M.J. Gans, "On the achievable sum rate for MIMO interference channels," IEEE Trans. Informa4on Theory, Sept X. Shang, B. Chen, and J. Matyjas, "Sum capacity opjmality of orthogonal transmissions in vector Gaussian muljple access channels," IEEE Trans. Wireless communica4ons, Nov M. J. Gans, K. Borle, B. Chen, T. Freeland, D. McCarthy, R. Nelson, D. Overrocker and P. Oleski, "Enhancing connecjvity of unmanned vehicles through MIMO communicajons," Proc. of IEEE VTC, Sept B. Chen, M.J. Gans, and J. Matyjas, Efficient channel esjmajon and symbol detecjon for massive MIMO- OFDM, provisional patent filed, Oct Air Force Applica-ons

3 Observa0ons MIMO channels for airborne plaaorms are rather `structured The usual i.i.d. entry assumpjon is not valid due to lack of scadering Structure depends highly on the plaaorm geometry and RF environment Digression: why MIMO? Large aperture compensates for the lack of scadering

4 H1

5 H2 H1 H2

6 Synopsis Mode A (Comm & Low-power Sensing) UAV Platform MoJvaJon: Dual use of RF assets OperaJng modes Passive (low-power) sensing AcJve (high-power) sensing Real-Jme toggling driven by the dynamic input data From MIMO communicajons to controlled sensing High dimensional data/low latency constraint Scalable and real-jme streamprocessing computajonal plaaorm Computation Platform UAV Platform for Dynamic Communication Sensing Tx Adaptation Data Proc D 3 Local Situation Assessment Target Indication D 3 Global Situation Assessment Target Confirmation Yes UAV Platform for Dynamic Communication Sensing Mode B ( Comm & High-Power Sensing) Target Classification/Training Command Control Center No Stay in Mode A Comm & Sensing Adatptation

7 Challenges Conformal array instead of uniform linear array Angle domain decomposijon of MIMO matrix does not work RunJme adaptajon between and within modes require real-jme processing of high dimensional data MulJ-mission RF Sensing and situajonal awareness mission should not compromise communicajon and connecjvity

8 Three Thrusts Thrust 1 MulJ-mission RF for Local and Global SituaJon Assessment Thrust 2 Dynamic Data Driven InformaJon Processing and Fusion Thrust 3 Scalable ComputaJonal Plaaorm for Dynamic Data Driven InformaJon Processing

9 Thrust 1 Mul0-mission RF for Local and Global Situa0on Assessment Dual-Use MIMO Systems for Airborne Networks The presence of airborne targets alters the channel characterisjcs in a non-scadering airborne environment D3 Local SituaJon Assessment using Quickest Change DetecJon Quickest change detecjon at MIMO receiver D3 Global SituaJon Assessment via Controlled SequenJal Outlier DetecJon Universal sequenjal outlier detecjon ComputaJonally challenging with distributed airborne plaaorms

10 Thrust 2 Dynamic Data Driven (D3) Informa0on Processing and Fusion Dependence across muljple plaaorms with unknown targets Copula-based Target ClassificaJon Kernel-based target classificajon Data-driven kernel selecjon Dynamic classificajon design Data driven target tracking involving distributed plaaorms Non-parametric adapjve feature matching Scaling up Copula based approach to that involve muljple plaaorms

11 Thrust 3 Scalable Computa0onal PlaQorm for D3 Informa0on Processing GPU-accelerated Storm (G-Storm) for Stream Processing Architectural design Resource opjmizajon Addressing computajonal challenges in D3 fusion using G-Storm Universal muljple hypotheses sequenjal outlier detecjon RunJme copula selecjon for informajon fusion

12 Illustra0on G-Storm for Copula based Likelihood Es0ma0on

13 Impact and Technology Transi0on MulJ-mission RF to maximize the payoff of available airborne RF assets Closed-loop design of all situajonal awareness components A truly dynamic and semi-autonomous informajon system Leveraging on-going collaborajon with AFRL for technology transijon In-house: Large aperture airborne MIMO with distributed ground arrays Newport and Stockbridge AFRL facilijes Access to AFRL HPC A pending DURIP proposal (PI Prof J. Tang) for GPU clusters