NEURAL NETWORK CONTROLLED AUTOMATIC GAIN CONTROL FOR LOW COST IR SENSORS.

LOW-COST SMART SENSORS WILL BE AN INTEGRAL PART OF NEXTGENERATION WEAPONS SYSTEMS. WE PROPOSE TO RESEARCH AND DEVELOP ADVANCGD NEURAL NETWORK ARCHITECTURES SUITABLE FOR INTEGRATION WITH IR SENSORS, EITHER AT THE FOCAL PLANE ITSELF OR TIGHTLY COUPLED TO EXISTING IMAGERS, TO PROVIDE THE HIGHBANDWIDTH SIGNAL PROCESSING REQUIRED TO MEET THESE SYSTEM SPECIFICATIONS. PHASE I WILL INCLUDE THE DESIGN AND FABRICATION OF PROTOTYPE CUSTOM INTEGRATED CIRCUITS TO IMPLEMENT A NETWORK TO PERFORM AN AUTOMATIC GAIN CONTROL (AGC) TASK.OUR APPROACH WILL BE TO UTILIZE COMMERCIALLY AVAILABLE CMOS/BULK INTEGRATED CIRCUIT TECHNOLOGY TO IMPLEMENT THE NEURAL NETWORK, IN CONJUNCTION WITH AN EXISTING IR FOCAL PLANE ARRAY SENSOR. THE NETWORK WILL ACCEPT ANALOG INPUTS FROM THE SENSOR ARRAY, AND PRODUCE OUTPUT SIGNALS OF RELATIVELY CONSTANT DYNAMIC RANGE. THE SELECTION OF TECHNOLOGY IS INTENDED TO PERMIT ECONOMICAL AND RELIABLE FABRICATION OF THE NEURAL NETWORK, BY A NUMBER OF VENDORS. WE HAVE ALREADY DEMONSTRATED THE USE OF SUCH TECHNOLOGIES TO FABRICATE CONTINUOUS-TIME ANALOG NEURAL NETWORKS INTEGRATED WITH VISIBLE-LIGHT PHOTOSENSORS TO EXTRACT MOTION INFORMATION FROM A VISUAL SCENE. WE HAVE ALSODEMONSTRATED INTEGRATED SENSOR ARRAYS/NEURAL NETWORKS TO DO REAL-TIME FEATURE DETECTION OVER FIXED-SIZE RECEPTIVE FIELDS.