Low-cost far-field and near-field radio frequency sensors for human sensing and liquid characterization

This dissertation presents the theoretical analysis, system design, and experimental evaluation of low-cost far-field and near-field radio frequency sensors for human sensing and liquid characterization. The electromagnetic field characteristics of a radiating element are dependent of the measurement distance. The region furthest away from the radiating element is dominated by radiated electromagnetic fields and is called the far-field region. In this region, the radiation pattern is independent of the distance and the electric and magnetic fields are orthogonal to each other and to the direction of propagation, similar to plane waves. Portable far-field sensors are being widely studied and used due to their versatility in various applications such as vital signs detection, human tracking, gesture recognition, speech sensing, etc. Among far-field sensors, microwave radar systems are the most widely used. For instance, the deployment of these sensor has exponentially increased in recent years due to their high demand in the automotive sector. However, low-cost far-field sensors present special conditions and limitations. For instance, the most widely used architecture in microwave Doppler radar systems is the quadrature direct-conversion architecture (homodyne), since it provides lower hardware cost, less complexity, easier chip integration, and straightforward range correlation effect that mitigates the oscillator phase noise effect. Unfortunately, the quadrature direct-conversion receiver presents I/Q channels phase and amplitude imbalance that destroys the orthogonality of the received signal, causing a mirrored signal to appear superimposed to the desired signal. Additionally, the spectral components of slow motion and physiological signals are around DC, which makes them susceptible to 1/f noise in homodyne receivers. To address these issues, this dissertation analyzes the noise contribution and receiver link budget for the small-motion case based on circuits theory and signal processing methods used in small-motion detection. It is shown that the conventional link budget analysis used in communication systems is a pessimistic estimation for the small-motion interferometric case. In addition, it is shown that the radar range equation is not suitable for directly estimating the desired signal received power because of the generated harmonic tones in the detected signal. For the first time, the effect of the window functions on the spectrogram is discussed and introduced to the sensitivity calculation for the interferometric case. Additionally, a thorough distortion analysis is performed, showing the negligible effect of phase imbalance for small movements (i.e., compared with wavelength). A distortion correction method is proposed including a novel cross-correlation based phase imbalance correction technique. Furthermore, the advantages of the Welch’s estimator for small motion detection are analyzed experimentally and in simulation. Moreover, a sinusoidally modulated continuous wave (SMCW) far-field sensor is proposed. By periodically changing the sensor’s transmitted frequency, the target’s motion is successfully up-converted, effectively reducing the impact of the 1/f noise without affecting the range correlation effect (coherence). Finally, the feasibility of electronically spoofing far-field sensors with portable devices is studied. First, a simple binary phase-shift keying (BPSK) system is used to generate artificial spectral components in the reflected demodulated signal. Additionally, an analog phase shifter is driven by an arbitrary signal generator to mimic the human cardiorespiratory motion. The interest on low-power far- and near-field sensors with small form factor have increased in the last few years to fulfill the internet of things (IoT) requirements. Additionally, in order to make feasible the deployment of the high number of sensors demanded in the IoT era, it is imperative to pursue the design of ultra-low-cost on-board electronics. Therefore, it is important to reduce the number of radio frequency active devices in portable far- and near-field sensors. Therefore, a novel, low-cost, low-power 2-port monostatic far-field sensor suitable for massive fabrication and array implementation is proposed. Unlike conventional radar systems, this architecture does not use separate TX/RX signal chains, just counts on a single antenna for TX/RX and does not use a hybrid or circulator in the RF front-end. Besides the local oscillator, there is no other active device consuming any DC power, greatly reducing the power consumption and cost. Experimental results have verified the functionalities and demonstrated the potential for the proposed system to be utilized in human tracking and small motion detection. The region next to a radiating element is called the near-field region and it can be divided into two sub-regions called reactive and radiative near-field regions. The region closest to the radiating element forms the reactive near-field region. In this region, the radiation is negligible, and the electric and magnetic fields are 90 degrees out of phase, turning the field reactive. On the other hand, the radiative near-field region is where the transition of the electromagnetic field from reactive to radiative begins. To this end, this dissertation investigates a near-field sensor, which uses already existing hardware to realize new sensing functions with very few or no hardware added. The proposed system leverages the wireless power transfer (WPT) or near field communication (NFC) coil integrated in phones for liquid characterization. A circuit model for the beverage-sensor interaction is developed and the performance of features from different nature (e.g., magnitude, amplitude, phase) for machine learning based classification is analyzed and tested. The computational cost and radio frequency bandwidth needed for classification is reduced using box plot analysis, singular value decomposition, and histogram analysis without affecting the classification accuracy. Several experiments were carried out to verify the effectiveness of the proposed system. Experiments with fresh and spoiled milk showed that the proposed system can perform non-invasive beverage freshness detection, which is usually very difficult for the current existing methods (RFID tags, electrophysical and electrochemical measurements). Additionally, the non-invasive classification of liquid solutions with different concentrations is studied. Experiments with milk adulterated with different water volumes were carried out, milk concentrations of 100%, 80%, 60%, and 40% were used to train the machine learning classifiers, obtaining high classification accuracies.

Restricted until 06/2023. To request the author grant access, click on the PDF link to the left.