Cognitive Radar Experiments At The Ohio State University Graeme E. Smith The OSU ElectroScience Lab
All Radar Systems Are Cognitive Consider an air traffic control radar The turn-and-burn sensor is not considered cognitive The overall air traffic control system is cognitive Air traffic controllers and pilots provide the cognition 2
Phase Array Radar Case High Angle Search/Track Midcourse Guidance Terminal Homing MULTIPLE TARGET TRACK Targets Jamming Missile Launch Jamming High Speed Horizon Search Surface Picture SURVEILLANCE Parameter update rate is faster than human can provide inputs: radar level cognition needed for best performance 3
The Case For Cognition A human operators currently provides cognition Humans operate on slow timescales Already there is latent capacity in radar system - Electronic scanning allows parameter changing each pulse Change role of operator to be editorial: operator sets goal, radar determines how to achieve it For next generation of radar we need cognition to achieve maximum capability 4
Cognitive Radar Concept 5
Cognitive Radar Framework Sensor model now includes adjustable sensor parameters and sensor cost/constraint function R Q ( θ k ) Controller determines next sensor parameters by minimizing a loss function that balances state estimation cost and sensor cost θ = arg min L θ Z ; Θ = arg min L R θ Z ; Θ, R θ 6 { } ( ) ( ) ( ) k C, Q k-1 k-1 C k-1 k-1 Q θ θ θ k
Cognitive Radar Framework State estimation cost characterized by predicted conditional Bayes risk: Bayes risk for next estimate given past data ( ) ˆ ( ) ( ) ( ) R θ Z ; Θ = ò C x z, Z, x f x, z Z ; θ, Θ dx dz C k-1 k-1 k k-1 k k k k-1 k-1 k k ( x, z ) º ( x, z Z ; Q ) = ( z x ; θ ) ( x ) f f f f - k k k k k-1 k k k k k 7
Example: Single Target Tracker Initialization f ( x ) º q( x ) 0 0 Recursion + { } k ( ; ) E ln (, ) k f B θ Z Θ x z k k k-1 k- 1 =- D x x k k ( ) = ( ) =ò ( ) ( ) - f ( zk xk) f ( xk) ( xk) = f ( xk Zk) = - ò f ( z x ) f ( x ) dx f x f x Z q x x f x dx f - + k k k-1 k k-1 k-1 k-1 + { { - ( ) 1 } ( )} -1-1 RQ θ = arg min L tr B θ; Θ, Z, θ k k k k θ Obtain z k using θ ( ) = + = E + { } ˆ k k k xz µ x k k k k k PC-BIM Controller Optimization Sensor measurement Motion update Information update State estimation from posterior 8
The CREW Systems At OSU Cognitive Radar Engineering Workspace (CREW) Dedicated experimental system for cognitive radar research - Operates at W-band - Four transmit & four receive channels - Arbitrary waveform generation for each Tx channel - Independent digitization for each Rx channel - All channels coherent - Capable of real-time parameter adaption Funded from an AFOSR DURIP award (~$650k) System design came from Cognitive Sensing Lab - Digital hardware came from Keysight - RF hardware came from Millitech 9
System Block Diagram Local Oscillator 1 (6.3 GHz) Baseband Convertor Module 10 MHz Reference Local Oscillator 2 (16.33-16.83 GHz) Digital Signal Generation AWG 1 AWG 2 AWG 3 AWG 4 0.3 to 1.3 GHz LO 2 Distribution (8 Way Split) LO 1 Distribution (8 Way Split) Upconvert to IF (4 of) Tx Head (4 of) 6 x LO Mult Upconvert W-Band Tx Signal Digitization Data Bus ADC 1 ADC 2 ADC 3 0.3 to 1.3 GHz Downconvert to IF (4 of) Rx Head (4 of) 6 x LO Mult Downconvert W-Band Rx ADC 4 Control PC 10
System Photographs Built by Agilent/Keysight and Millitech Solve clock & trigger challenges by centralized rack based solution Can position Tx/Rx heads around lab to 11 achieve multistatic operation
The FARMS Environment Fully Adaptive Radar Modelling & Simulation (FARMS) Needed to test algorithms before experiments Simplified environment simulation Same interface to radar data as the CREW Software switch to say experiment or simulation Same code used for simulation and experiments 12
Cognitive Tracking Experiment CREW ~11 m Lab Bench Lab Bench Lab Bench Lab Bench Lab Bench Lab Bench ~15 m 13
Fixed Param: Simulation 14
Fixed Param: Experiment 15
Bandwidth Adapt: Simulation 16
Bandwidth Adapt: Experiment 17
Adapt All Params: Simulation 18
Adapt All Params: Experiment 19
Summary Presented a case for needing cognition in radar Introduced the FAR framework Introduced the CREW at The Ohio State University Reported on results of experiments in cognitive radar - Tracking a single target - Can use adaption to optimize performance against set goals - Close match between simulation and experiment - Can adapt multiple parameters at once 20
Cognitive Radar Experiments At The Ohio State University Graeme E. Smith smith.8347@osu.edu