HOLISTIC FRAMEWORK FOR NEXT GENERATION RFID

NEXTID IS FOCUSED ON SOLVING SOME OF THE EXISTING PROBLEMS IN UHF-RFID TECHNOLOGY THAT LIMIT ITS PERFORMANCE TO PROVIDE FURTHER PENETRATION INTO SOCIETY FROM AN INTERDISCIPLINARY POINT OF VIEW: BY COMBINING A HARDWARE-BASED APPROACH (ADVANCED ELECTROMAGNETIC DEVICES AND ANTENNAS) WITH A SOFTWARE-BASED APPROACH (AI DATA PROCESSING). IN ADDITION, NEW APPLICATIONS WIL BE EXPLORED SUCH AS SENSING APPLICATIONS IN AGRICULTURAL ENVIRONMENTS (WATER MANAGEMENT IN IRRIGATION PROCESSES), PRECISE ACCESS CONTROL AND ACCURATE ANTI-THEFT APPLICATIONS. ON THE HARDWARE SIDE, WE WILL USE TWO STRATEGIES TO ADDRESS THESE CHALLENGES. THE FIRST STRATEGY WILL FOCUS ON THE OPTIMIZATION OF RFID TAGS, WHILE THE SECOND STRATEGY WILL FOCUS ON THE CONTROL OF THE ELECTROMAGNETIC BEHAVIOR OF THE TRANSDUCERS CONNECTED TO THE READERS. IN THE CASE OF RFID TAG OPTIMIZATION, THE MAIN STRATEGY WILL BE TO IMPROVE THE FREQUENCY RESPONSE OF CURRENT DEVICES. OUR PROPOSAL IS TO CONTINUE THE PREVIOUS WORK ON BANDWIDTH LIMITS OF RFID IN SINGLE RESONANT TAGS AND EXTEND IT TO DUAL-RESONANCE RESPONSE OF RFID TAGS, TO ACHIEVE RFID TAGS WITH SECOND ORDER FREQUENCY CONTROLLABLE RESPONSES. FOR THE CASE OF RFID READERS, WE WILL CONSIDER TWO SCENARIOS. ONE IN WHICH THE OBJECT TO BE IDENTIFIED IS IN THE FAR FIELD REGION OF THE TRANSDUCER CONNECTED TO THE RFID READER, WHERE WE WILL USE HIGH DIRECTIVE AND LOW SIDELOBE ANTENNAS, AND THE OTHER WHEN THE OBJECT IS LOCATED IN THE NEAR FIELD REGION (AS MAY BE THE CASE IN A POINT OF SALE), WHERE FIELD CONFINEMENT DEVICES WILL BE USED. IN THIS PROJECT, PREVIOUS WORK FROM THE TEAM WILL BE EXTENDED TO DEVELOP NEW DEVICES WITH IMPROVED PERFORMANCE. FROM THE POINT OF VIEW OF SOFTWARE, OUR HYPOTHESIS IS THAT USING BOTH STATISTICAL AND PHYSICAL KNOWLEDGE INCREASES THE PERFORMANCE OF THE APPLICATIONS UNDER STUDY, WHICH USE RSSI MEASUREMENTS, AND WORK IN THE SMALL DATA REGIME. WE CONSIDER TWO SITUATIONS: THE STATIC CASE (RFID TAGS ARE NOT MOVING) AND THE DYNAMIC CASE. FOR THE STATIC CASE WE CONSIDER TWO RESEARCH LINES: I) BAYESIAN NETWORKS AND II) PHYSICS-INFORMED NEURAL NETWORKS. BAYESIAN NETWORKS OFFER THE POSSIBILITY TO MODEL THE STATISTICAL DEPENDENCIES AMONG VARIABLES AND THE GOAL IS TO DESIGN THE MODELS THAT BEST FIT TO THE APPLICATIONS CONSIDERED. PHYSICS-INFORMED NEURAL NETWORKS (PINN) IS A RECENT APPROACH THAT COMBINES PHYSICAL MODELS DESCRIBED BY MEANS OF PARTIAL DIFFERENTIAL EQUATIONS WITH NEURAL NETWORKS TO ENCOURAGE THESE TO SATISFY THE PHYSICAL LAWS UNDERNEATH. IN THE DYNAMIC CASE, WE NEED TO ADOPT SEQUENTIAL MODELS. IN THE CONTEXT OF THE BAYESIAN FRAMEWORK, HIDDEN MARKOV MODELS (HMM) AND LINEAR DYNAMICAL SYSTEMS (LDS) ARE THE MOST KNOWN TECHNIQUES. THE GOAL OF THESE MODELS IS TO PROPERLY ESTABLISH THE PROBABILISTIC RELATIONS OF THESE VARIABLES TO ESTIMATE THE HIDDEN VARIABLES FROM THE OBSERVATIONS. IN THIS PROJECT, WE WILL FOCUS ON HMM TO ADDRESS DYNAMICAL SCENARIOS WHERE THE PRESENCE/ABSENCE OF RFID DEVICES ARE THE HIDDEN (DISCRETE) VARIABLES. OUR HYPOTHESIS IS THAT A (PURE) BAYESIAN FRAMEWORK CAN BE PROPOSED TO SOLVE COMPLEX RFID-BASED DYNAMICAL SCENARIOS INVOLVING MULTI-TAG AND MULTI-READER ENVIRONMENTS IN THE SMALL DATA REGIME.