Low Complexity Binarized 2D-CNN Classifier for Wearable Edge AI Devices.

Wearable Artificial Intelligence-of-Things (AIoT) devices exhibit the need to be resource and energy-efficient. In this paper, we introduced a quantized multilayer perceptron (qMLP) for converting ECG signals to binary image, which can be combined with binary convolutional neural network (bCNN) for classification. We deploy our model into a low-power and low-resource field programmable gate array (FPGA) fabric. The model requires 5.8× lesser multiply and accumulate (MAC) operations than known wearable CNN models. Our model also achieves a classification accuracy of 98.5%, sensitivity of 85.4%, specificity of 99.5%, precision of 93.3%, and F1-score of 89.2%, along with dynamic power dissipation of 34.9 µW.

An Energy Efficient ECG Ventricular Ectopic Beat Classifier Using Binarized CNN for Edge AI Devices.

Wearable Artificial Intelligence-of-Things (AIoT) requires edge devices to be resource and energy-efficient. In this paper, we design and implement an efficient binary convolutional neural network (bCNN) algorithm utilizing function-merging and block-reuse techniques to classify between Ventricular and non-Ventricular Ectopic Beat images. We deploy our model into a low-resource low-power field programmable gate array (FPGA) fabric. Our model achieves a classification accuracy of 97.3%, sensitivity of 91.3%, specificity of 98.1%, precision of 86.7%, and F1-score of 88.9%, along with dynamic power dissipation of only 10.5-µW.

Hybrid-Flexible Bimodal Sensing Wearable Glove System for Complex Hand Gesture Recognition.

As 5G communication technology allows for speedier access to extended information and knowledge, a more sophisticated human-machine interface beyond touchscreens and keyboards is necessary to improve the communication bandwidth and overcome the interfacing barrier. However, the full extent of human interaction beyond operation dexterity, spatial awareness, sensory feedback, and collaborative capability to be replicated completely remains a challenge. Here, we demonstrate a hybrid-flexible wearable system, consisting of simple bimodal capacitive sensors and a customized low power interface circuit integrated with machine learning algorithms, to accurately recognize complex gestures. The 16 channel sensor array extracts spatial and temporal information of the finger movement (deformation) and hand location (proximity) simultaneously. Using machine learning, over 99 and 91% accuracy are achieved for user-independent static and dynamic gesture recognition, respectively. Our approach proves that an extremely simple bimodal sensing platform that identifies local interactions and perceives spatial context concurrently, is crucial in the field of sign communication, remote robotics, and smart manufacturing.

A 13-Channel 1.53-mW 11.28-mm2 Electrical Impedance Tomography SoC Based on Frequency Division Multiplexing for Lung Physiological Imaging.

An electrical impedance tomography (EIT) system based on frequency division multiplexing (FDM) is proposed for real-time lung physiological imaging. The FDM technique allows the integration of 13 dedicated voltage sensing channels by combining data on-chip and sharing of ADC to alleviate area penalty caused by multi-channel. The EIT system-on-chip (SoC) is of the following features. 1) Early I/Q demodulation to relax the bandwidth requirement of analog front end and minimize the impact of motion artifacts and dc electrode offset. 2) Eliminates the need of adaptive gain control with constant inverted "U-shape" gain configuration to compensate amplitude variations across all channels. 3) FDM to combine 13 pairs of I/Q signals into two data streams for quantization using only two ΔΣ modulators. 4) Batch data recovery by Blackman window corrected fast Fourier transform without any digital filtering involved. 5) Lowest power consumption and smallest area occupation per channel reported to date. The EIT SoC occupies an area of 11.28 mm2 in 130-nm CMOS technology with a total power consumption of 1.53 mW under 1-V power supply. As a result, it generates lung EIT images at up to five frames per second.

A 74-µW 11-Mb/s Wireless Vital Signs Monitoring SoC for Three-Lead ECG, Respiration Rate, and Body Temperature.

This paper presents a wireless vital signs monitoring system-on-chip including three-lead ECG, bio-impedance (Bio-Z) and body temperature. A Bio-Z readout channel with early demodulation is introduced to detect small body impedance change below 300 mΩ under large baseline resistance while consuming power of 9.8 µW. A direct temperature-to-digital converter is also incorporated, which achieves an absolute resolution of 0.023 °C and inaccuracy less than ±0.13 °C over temperature range of 32-42 °C. The quantized vital sign signal is transmitted through a multi-channel reconfigurable QPSK/BFSK transmitter (TX) to accommodate different power-constraint scenarios. Fabricated in 130-nm CMOS technology with a total die area of 6.25 mm2, the proposed SoC only consumes 74 µW under 0.9-V supply.

A Soft Polydimethylsiloxane Liquid Metal Interdigitated Capacitor Sensor and Its Integration in a Flexible Hybrid System for On-Body Respiratory Sensing.

We report on the dual mechanical and proximity sensing effect of soft-matter interdigitated (IDE) capacitor sensors, together with its modelling using finite element (FE) simulation to elucidate the sensing mechanism. The IDE capacitor is based on liquid-phase GaInSn alloy (Galinstan) embedded in a polydimethylsiloxane (PDMS) microfludics channel. The use of liquid-metal as a material for soft sensors allows theoretically infinite deformation without breaking electrical connections. The capacitance sensing is a result of E-field line disturbances from electrode deformation (mechanical effect), as well as floating electrodes in the form of human skin (proximity effect). Using the proximity effect, we show that spatial detection as large as 28 cm can be achieved. As a demonstration of a hybrid electronic system, we show that by integrating the IDE capacitors with a capacitance sensing chip, respiration rate due to a human's chest motion can be captured, showing potential in its implementation for wearable health-monitoring.

A Wearable, Patient-Adaptive Freezing of Gait Detection System for Biofeedback Cueing in Parkinson's Disease.

Freezing of Gait (FoG) is a common motor-related impairment among Parkinson's disease patients, which substantially reduces their quality of life and puts them at risk of falls. These patients benefit from wearable FoG detection systems that provide timely biofeedback cues and hence help them regain control over their gait. Unfortunately, the systems proposed thus far are bulky and obtrusive when worn. The objective of this paper is to demonstrate the first integration of an FoG detection system into a single sensor node. To achieve such an integration, features with low computational load are selected and dedicated hardware is designed that limits area and memory utilization. Classification is achieved with a neural network that is capable of learning in real time and thus allows the system to adapt to a patient during run-time. A small form factor FPGA implements the feature extraction and classification, whereas a custom PCB integrates the system into a single node. The system fits into a 4.5 × 3.5 × 1.5 cm 3 housing case, weighs 32 g, and achieves 95.6% sensitivity and 90.2% specificity when adapted to a patient. Biofeedback cues are provided either through auditory or somatosensory means and the system can remain operational for longer than 9 h while providing cues. The proposed system is highly competitive in terms of classification performance and excels with respect to wearability and real-time patient adaptivity.

Continuous ECG Monitoring Trial for Outpatient - Patient Receptiveness and Signal Accuracy.

We have conducted a clinical trial to investigate the receptiveness and signal accuracy of our in-house light-weight single lead wearable wireless ECG device with 20 outpatients, who were suspected of cardiac rhythm issues. The receptiveness was measured via a survey score sheet while the signal accuracy was evaluated by comparing the Holter's hourly heart rate report (the gold-standard) against the ones from our device. In terms of receptiveness, a score of 8.6 indicates good patient compliance. In terms of accuracy, the mean absolute error is 1.4 bpm (beats per minute) with precision of ±1.6 bpm. In addition, measurements from both devices were found to be linearly related with coefficient of determination, r2, of 0.97. Furthermore, the limits of agreement were calculated to be +3.54 and -4.71 based on the Altman and Bland technique, which indicated good agreement for most of our measurements against the Holter device. In addition, this paper also discusses the unexpected challenges of conducting a trial on actual outpatients which can be used as a reference for similar subsequent studies.

Regression analysis of gait parameters and mobility measures in a healthy cohort for subject-specific normative values.

BACKGROUND: Deviation in gait performance from normative data of healthy cohorts is used to quantify gait ability. However, normative data is influenced by anthropometry and such differences among subjects impede accurate assessment. De-correlation of anthropometry from gait parameters and mobility measures is therefore desirable. METHODS: 87 (42 male) healthy subjects varying form 21 to 84 years of age were assessed on gait parameters (cadence, ankle velocity, stride time, stride length) and mobility measures (the 3-meter/7-meter Timed Up-and-Go, 10-meter Walk Test). Multiple linear regression models were derived for each gait parameter and mobility measure, with anthropometric measurements (age, height, body mass, gender) and self-selected walking speed as independent variables. The resulting models were used to normalize the gait parameters and mobility measures. The normalization's capability in de-correlating data and reducing data dispersion were evaluated. RESULTS: Gait parameters were predominantly influenced by height and walking speed, while mobility measures were affected by age and walking speed. Normalization de-correlated data from anthropometric measurements from |rs| < 0.74 to |rs| < 0.23, and reduced data dispersion by up to 69%. CONCLUSION: Normalization of gait parameters and mobility measures through linear regression models augment the capability to compare subjects with varying anthropometric measurements.

A 400 MHz Wireless Neural Signal Processing IC With 625 $\times$ On-Chip Data Reduction and Reconfigurable BFSK/QPSK Transmitter Based on Sequential Injection Locking.

An 8-channel wireless neural signal processing IC, which can perform real-time spike detection, alignment, and feature extraction, and wireless data transmission is proposed. A reconfigurable BFSK/QPSK transmitter (TX) at MICS/MedRadio band is incorporated to support different data rate requirement. By using an Exponential Component-Polynomial Component (EC-PC) spike processing unit with an incremental principal component analysis (IPCA) engine, the detection of neural spikes with poor SNR is possible while achieving 625× data reduction. For the TX, a dual-channel at 401 MHz and 403.8 MHz are supported by applying sequential injection locked techniques while attaining phase noise of -102 dBc/Hz at 100 kHz offset. From the measurement, error vector magnitude (EVM) of 4.60%/9.55% with power amplifier (PA) output power of -15 dBm is achieved for the QPSK at 8 Mbps and the BFSK at 12.5 kbps. Fabricated in 65 nm CMOS with an active area of 1 mm 2, the design consumes a total current of 5  âˆ¼ 5.6 mA with a maximum energy efficiency of 0.7 nJ/b.

Broadband Energy Harvester Using Non-linear Polymer Spring and Electromagnetic/Triboelectric Hybrid Mechanism.

Over the years, several approaches have been devised to widen the operating bandwidth, but most of them can only be triggered at high accelerations. In this work, we investigate a broadband energy harvester based on combination of non-linear stiffening effect and multimodal energy harvesting to obtain high bandwidth over wide range of accelerations (0.1 g-2.0 g). In order to achieve broadband behavior, a polymer based spring exhibiting multimodal energy harvesting is used. Besides, non-linear stiffening effect is introduced by using mechanical stoppers. At low accelerations (<0.5 g), the nearby mode frequencies of polymer spring contribute to broadening characteristics, while proof mass engages with mechanical stoppers to introduce broadening by non-linear stiffening at higher accelerations. The electromagnetic mechanism is employed in this design to enhance its output at low accelerations when triboelectric output is negligible. Our device displays bandwidth of 40 Hz even at low acceleration of 0.1 g and it is increased up to 68 Hz at 2 g. When non-linear stiffening is used along with multimodal energy-harvesting, the obtained bandwidth increases from 23 Hz to 68 Hz with percentage increment of 295% at 1.8 g. Further, we have demonstrated the triboelectric output measured as acceleration sensing signals in terms of voltage and current sensitivity of 4.7 Vg-1 and 19.7 nAg-1, respectively.

A Hybrid Data Compression Scheme for Power Reduction in Wireless Sensors for IoT.

This paper presents a novel data compression and transmission scheme for power reduction in Internet-of-Things (IoT) enabled wireless sensors. In the proposed scheme, data is compressed with both lossy and lossless techniques, so as to enable hybrid transmission mode, support adaptive data rate selection and save power in wireless transmission. Applying the method to electrocardiogram (ECG), the data is first compressed using a lossy compression technique with a high compression ratio (CR). The residual error between the original data and the decompressed lossy data is preserved using entropy coding, enabling a lossless restoration of the original data when required. Average CR of 2.1 × and 7.8 × were achieved for lossless and lossy compression respectively with MIT/BIH database. The power reduction is demonstrated using a Bluetooth transceiver and is found to be reduced to 18% for lossy and 53% for lossless transmission respectively. Options for hybrid transmission mode, adaptive rate selection and system level power reduction make the proposed scheme attractive for IoT wireless sensors in healthcare applications.

Large Scale Triboelectric Nanogenerator and Self-Powered Pressure Sensor Array Using Low Cost Roll-to-Roll UV Embossing.

Triboelectric nanogenerators (TENGs) have emerged as a potential solution for mechanical energy harvesting over conventional mechanisms such as piezoelectric and electromagnetic, due to easy fabrication, high efficiency and wider choice of materials. Traditional fabrication techniques used to realize TENGs involve plasma etching, soft lithography and nanoparticle deposition for higher performance. But lack of truly scalable fabrication processes still remains a critical challenge and bottleneck in the path of bringing TENGs to commercial production. In this paper, we demonstrate fabrication of large scale triboelectric nanogenerator (LS-TENG) using roll-to-roll ultraviolet embossing to pattern polyethylene terephthalate sheets. These LS-TENGs can be used to harvest energy from human motion and vehicle motion from embedded devices in floors and roads, respectively. LS-TENG generated a power density of 62.5 mW m(-2). Using roll-to-roll processing technique, we also demonstrate a large scale triboelectric pressure sensor array with pressure detection sensitivity of 1.33 V kPa(-1). The large scale pressure sensor array has applications in self-powered motion tracking, posture monitoring and electronic skin applications. This work demonstrates scalable fabrication of TENGs and self-powered pressure sensor arrays, which will lead to extremely low cost and bring them closer to commercial production.

A 2.4 GHz ULP reconfigurable asymmetric transceiver for single-chip wireless neural recording IC.

This paper presents a 2.4 GHz ultra-low-power (ULP) reconfigurable asymmetric transceiver and demonstrates its application in wireless neural recording. Fabricated in 0.13 µm CMOS technology, the transceiver is optimized for sensor-gateway communications within a star-shaped network, and supports both the sensor and gateway operation modes. Binary phase-shift keying (BPSK) modulation with high data rate (DR) of 1 to 8 Mbps is used in the uplink from sensor to gateway, while on-off keying (OOK) modulation with low DR of 100 kbps is adopted in the downlink. A fully integrated Class-E PA with moderate output power has also been proposed and achieves power efficiency of 53%. To minimize area usage, inductor reuse is adopted between PA and LNA, and eliminates the need of lossy T/R switch in the RF signal path. When used as sensor, the transceiver with frequency locked phase-locked loop (PLL) achieves TX (BPSK) power efficiency of 28% @ 0 dBm output power, and RX (OOK) sensitivity of -80 dBm @ 100 kbps while consuming only 780 µW . When configured as gateway, the transceiver achieves sensitivity levels of -92, -84.5, and -77 dBm for 1, 5, and 8 Mbps BPSK, respectively. The transceiver is integrated with an 8-channel neural recording front-end, and neural signals from a rat are captured to verify the system functionality.

Low-power ultrawideband wireless telemetry transceiver for medical sensor applications.

An integrated CMOS ultrawideband wireless telemetry transceiver for wearable and implantable medical sensor applications is reported in this letter. This high duty cycled, noncoherent transceiver supports scalable data rate up to 10 Mb/s with energy efficiency of 0.35 nJ/bit and 6.2 nJ/bit for transmitter and receiver, respectively. A prototype wireless capsule endoscopy using the proposed transceiver demonstrated in vivo image transmission of 640 × 480 resolution at a frame rate of 2.5 frames/s with 10 Mb/s data rate.