Abstract
Nuclear Magnetic Resonance (NMR) and Nuclear Quadrupole Resonance (NQR) are spectroscopic techniques that offer the ability to characterise samples non-destructively in-situ for quality control in industrial processes (e.g. water content in food) and the detection of explosives and narcotics in defence and security sensing applications. Current NMR and NQR based sensing technologies can achieve good performance in a controlled laboratory environment where the effect of external Radio Frequency Interference (RFI) can be mitigated by RFI shielding. However, for in-situ sensing applications outside the laboratory, complete physical shielding is often not possible or not practical (heavy, bulky) and therefore, alternative RFI mitigation strategies are needed for NMR and NQR based sensing technologies to be helpful.
This thesis presents work undertaken to develop signal processing methods of digital burst mode RFI suppression for NMR and NQR technology to improve the Signal-to-Noise Ratio (SNR) of the final NMR and NQR results. This is to enable in-situ applications to be undertaken with greater accuracy and reliability without RFI shielding. The primary suppression technique developed in this research utilises a decision tree (DT) model to identify the RFI and a process termed ‘direct removal’ (DR) to suppress the identified RFI. An alternative RFI identification process has also been developed that uses a smart thresholding process to identify the RFI instead of using a machine learning model. Both methods show potential for improving NQR results in a security type application.
An experimental testbed to produce NQR data was assembled, using commercial off the shelf (COTS) equipment to allow NQR experiments to be undertaken with or without pseudo-RFI. This testbed provided a Faraday shielded enclosure to record NQR experimental data with or without transmitted external real world RFI. The data produced was used to train the DT model and test the effectiveness of the RFI suppression techniques developed.
The performance of the DT and DR suppression technique was validated using human operator performance data, generated by volunteers rating the degree to which they can confidently declare an NQR signal present or not in RFI-polluted data. The performance of the DT and DR process was quantified as a Receiver Operating Characteristic (ROC) curve. For the overall results, the DT and DR process improved (relative to no RFI removed) the area under curve (AUC) value from 0.58 to 0.91, where AUC = 1 means a 100% detection rate with a 0% false alarm rate. Additional work showed that processing and using just four measurements produced a substantial result (AUC = 0.86) compared to using 32 measurements (AUC = 0.9).
The smart thresholding technique, designed to suppress longer duration RFI without needing machine learning, has been shown to provide some improvements to NQR signal clarity by removing a prescribed percentage of the echo signal with the most significant amplitude, which is most likely to be RFI. The impact of this process is not as substantial as the DT and DR method but could be used alongside other RFI suppression techniques.
An additional investigation was undertaken to determine if ‘Symmetric Projected Attractor Reconstruction’ (SPAR) can be used to visualise NQR results in a more accessible novel manner for human operators to interpret. The results have shown that SPAR can be effectively used to visualise a binary classification (Sample or No Sample) and visually represent the relaxation processes of NQR signals from samples of sodium nitrite.