Research
Wireless Sensing + Machine Learning + Signal Processing + Wireless Communication
Prospects and applications of incoherent light in wireless sensing
Wireless sensing utilizes wireless technologies to detect and quantify physical characteristics of an object or environment without physical contact, and it has become a vital aspect of human life. Due to the shortage of frequencies for traditional radio frequency-based wireless systems, incoherent light sources like light emitting diodes (LED) with light sensors (photodetectors) have been explored as an alternative option. In contrast to coherent light or lasers, which can pose risks to human eyes and skin, incoherent visible and infrared light is safe for human use since it has low intensity. LED-based incoherent light has the potential to provide additional benefits over other wireless sensing technologies, such as lower cost, longer lifespan, and energy efficiency. Part of my research focus is to explore the potential of incoherent light in wireless sensing by investigating various applications such as vehicle speed estimation, human vitals monitoring, blood glucose sensing, gesture recognition, occupancy estimation, and structural health monitoring.
Incoherent light-wave sensing based non-contact respiration monitoring
The respiration rate conveys important information about the subject’s physical and psychological state that can be used to assess the subject's health condition and raise health awareness in general. Currently available contact-based or wearable breathing monitoring devices can interrupt the user and affect the regular breathing rate; hence non-contact methods are preferred. Light-wave sensing technology uses infrared or visible light which can be used to detect variation in breathing by using the light signal reflected back from the chest. This is safer, simpler, faster and more private than other technology counterparts like radio frequency and imaging. Everyone’s breathing pattern is unique, hence the system needs to learn the usual breathing pattern through machine learning to detect breathing anomaly.
Figure 1: Functional representation of the system model with applications
Machine learning was an essential part of this research work for enabling the system to smartly detect breathing anomaly. Following are the machine learning based works that I am doing in this project:
Different types of breathing data were collected using an artificial breathing robot, handcrafted features extraction were performed and the data were classified using machine learning models like Decision Tree, Random Forest, KNN, SVM and Gaussian Naive Bayes. Scikit-learn library was used to implement the models.
Model performances were optimized by hyper-parameter tuning and evaluated through cross-validation and confusion matrix.
Deep learning was also applied in terms of 1 dimensional convolutional neural network (CNN) to see how automatically extracted CNN-based features perform on the data-set. Keras API was used for implementing 1D-CNN.
Dimensionality reduction techniques like Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA) were tested with the collected data
Figure 2: Normal breathing data and it's spectrum
Figure 3: Hyper-parameter tuning using Decision Tree model
Figure 4: Cross-validation based confusion matrices using Random Forest model
Gesture recognition using light-wave sensing and machine learning
Gesture recognition is another promising application of visible light or infrared sensing. Data related to 8 different finger gestures performed by humans were collected using light-wave sensing system. A series of signal processing and standardization algorithms (see Fig. 6: Flow diagram) were applied on the data to make it ready for classification. Processed data itself were used as features and classification was done using machine learning based KNN and SVM models. Cross-validation accuracies were evaluated using confusion matrices.
Fig 5: Different gestures and corresponding collected and processed signal
Figure 6: Flow diagram
Figure 7: 10-fold CV confusion matrix for infrared sensing at 20 cm distance (ambient light was on)
Design and Simulation of RF-based Lunar Communication System Using GNU Radio
NASA is focusing on Radio Frequency (RF) and Optical frequency based hybrid lunar communication to ensure high speed data rate and uninterrupted connectivity between Earth and Moon through satellite constellation around the earth and a lunar gateway near the Moon. The RF communication channel has slower data rate but high reliability while optical channel has high data rate but prone to interruption due to pointing error, undue atmospheric condition etc. Because of high data rate, optical link can be the main link of the communication system while the RF link can perform supporting role in communicating control information for the optical link when the optical link is live and in transmitting the main data when the optical link is down. Poor signal to noise ratio and/or high bit error rate can be switching criteria between optical and RF link.
The RF channel for lunar communication has been emulated using software defined radio (NI USRP-2901) and GNU radio in a Raspberry Pi. Interagency Operations Advisory Group (IOAF) standard for lunar communication was followed while simulating the RF communication system. A text file containing some message has been successfully transferred through the RF channel using USRP.
Figure 8: Flow diagram of RF communications system in GNU radio
Fig 9: Software defined radio (NI USRP)
Broadband Internet through TVWS based Dynamic Spectrum Access
Frequency band used for television broadcasting has numerous gaps now-a-days known as Television White Space (TVWS) which can be leveraged through dynamic spectrum access to provide broadband internet connectivity in remote areas. We conducted feasibility analysis of broadband internet connectivity using TVWS through laboratory and outdoor testing of TVWS equipment purchased from Adaptrum Inc. Network performance, effect of elevation and azimuth of client antenna and path loss were analyzed for outdoor testing. Then, we successfully provided 16 Mbps live internet connection through the developed setup at Tillman county, Oklahoma, USA.
Figure 10: TVWS channel spectrum
Figure 11: Setup diagram
Figure 12: TVWS path loss model
