Next generation miniature radars: performance enhancement and machine learning applications

Miniature radar sensors have already started conquering the next generation sensing technology in nearly every sector of consumer application. These applications extend from autonomous driving to touchless controlling of household appliances, electronic gadgets and machineries to non-invasive monitoring in medical, infrastructure, agriculture, and irrigation fields. A couple of decades back, no-one could possibly imagine radar sensors to be a part of our daily lives, the reasons being the bulky size and expensive components of the traditional radar, which always hindered its deployment in consumer applications. However, the innovations in semiconductor technology have resulted in the development of low-cost, smaller, and more sophisticated miniature radar sensors that can fit in our finger tip, offering a diverse range of consumer applications. In addition, the ability of these radar sensors to work reliably irrespective of the environmental condition makes them the most favourite candidate for smart sensing technologies. This overwhelming demand of miniature radars for smart sensing applications, however, brings new development challenges in performance evaluation, design, and manufacturing of highly efficient radar sensors. To tackle the challenges in miniature radars in consumer applications, there are many studies aimed at evaluating and enhancing the reliability of the radar. Considering the rapid growth of the global market volume of the miniature radars, it is expected that there will be millions of radars all over the world serving every domain in the society.  This enormous amount of radar sensors will be sharing the same electromagnetic spectrum. The shared spectrum is allocated and regulated by the spectrum management authorities to ensure efficient spectrum usage and avoid interference with other wireless services. Despite this spectrum allocation, one cannot rule out the obvious possibility of mutual interference between the radar sensors, which will eventually affect its reliable performance. Interference between radars can degrade the detection performance. Therefore, the harmonious coexistence of the radar sensors in the shared spectrum is an indispensable task that needs to be analysed and solved by well-structured interference mitigation techniques. Radar is also envisioned as one of the primary sensors for the future non-invasive controlling and monitoring of several critical applications such as infection control in the medical field. Therefore, another major aspect that needs to be inspected is the performance of the miniature radar in terms of target classification accuracy. The consequences of mere false positives or false negatives in the target classification can be significant; therefore, it is vital to ensure that the radar is consistently giving an accurate classification performance. As a matter of fact, there has been a paramount interest in integrating machine learning with radar signal processing to enhance the state-of-art radar detection process and consistent result interpretation for target classification. Although machine learning is efficient in establishing insights from the inputs, the key to gaining advancements in the target classification process is in the representation or pre-processing of the signals captured by the radar, which are fed to the machine learning technique. Similarly, the choices of the utilised machine learning algorithm and the type of radar with respect to the application requirements significantly impact the quality of the achieved performance. In this research, we aim to investigate different methods for improving the performance of miniature radars in consumer applications. We explore various methods for interference analysis and investigate corresponding mitigation techniques in radar sensors for automotive scenario. The automotive radar performance is evaluated in terms of the signal-to-interference ratio; this analysis is formulated as a mathematical framework that characterises the statistical parameters of the mutual interference. The performance enhancement of the sensors for target classification is studied in the field of hand-gesture recognition and material identification. In the hand-gesture recognition application, the radar performance is evaluated in terms of the classification accuracy of different hand-gestures. In this research, we also explore the application of two popular radar architectures capable of capturing fine gesture signatures through both velocity signatures and range-velocity signatures of the hand. The classification enhancement is achieved by incorporating deep-learning techniques augmented with several traditional machine learning classifiers. Furthermore, the investigation on gesture recognition is expanded to include the use of multiple sensors to enrich the gesture signatures, which in turn enhance the classification performance. As one potential application, in this research, we also study the performance of miniature radar sensors in classifying different materials. We explore the application of one of the widely used radar architecture with multiple receiving channels to capture the signals reflected from different materials. The variations in the characteristics of the signals reflected from different materials are utilised for classifying the materials. These variations of the signals which reflects the material properties are analysed and classified employing multiple machine learning techniques to enhance the classification performance. The results of the performance evaluation of the miniature radars in the above-mentioned applications demonstrate promising results, which has the potential for real-world applications as well as it opens new possibilities for expansion. The majority of the analysis presented in this dissertation is based on experimental observations on practical miniature radars. We also rely on computer simulations using Monte-Carlo analysis to verify the analytic models that are presented in this research. The technical analysis and experimental results presented in the dissertation are either published in peer-reviewed journals, conference proceedings, or undergoing a review process. Finally, we wish the reader a joyful time reading the dissertation and sincerely hope our work will be useful for the future of miniature radars in consumer applications.