Conventional bio-impedance analysis to estimate the body composition has been used for many bio applications. It has been commonly applied by a sinusoidal wave or wide-band frequency signal in current domain on the target tissue, and its measured response is processed for providing the impedances. These techniques, however, have common trade-off between detectable frequency range and data rate. For some applications both wide-band frequency detection and high-speed update rate are required. This paper presents a pulse stimulation technique for the bio-impedance analysis, which is proper to the high-speed and wide-band applications. The proposed method is proved by theoretical analysis. It is also verified with the simulation and measurement results.

Time-mode digital pixel sensors have several advantages in Internet-of-Things applications, which require a compact circuit and low-power operation under poorly illuminated environments. Although the time-mode digitization technique can theoretically achieve a wide dynamic range by overcoming the supply voltage limitation, its practical dynamic range is limited by the maximum clock frequency and device leakage. This study proposes an extended time-mode digitization technique and a low-leakage pixel circuit to accommodate a wide range of light intensities with a small number of digital bits. The prototype sensor was fabricated in a 0.18 μm standard CMOS process, and the measurement results demonstrate its capability to accommodate a 0.03 lx minimum light intensity, providing a dynamic range figure-of-merit of 1.6 and a power figure-of-merit of 37 pJ/frame·pixel.

This paper proposes a low-offset balanced inverter that forms a push-pull structure using two composite transistors. Each composite transistor operates as either a current-sourcing or a current-sinking device, and can replace both the PMOS and NMOS transistors in a CMOS inverter. The symmetry between the pull-up and pull-down paths achieves a very low offset, enabling the implementation of a single-ended, lossless integrator without requiring any offset cancellation. The proposed balanced inverter is applied to realize a second-order delta-sigma ADC that digitizes the integration of extremely weak input currents from the resistor. Experimental results from the fabricated chip demonstrate that the resistor measuring ADC achieved a 0.4-Ω resolution over 106-dB dynamic range, confirming its practical applicability despite the process and temperature variations.

Lack of large-scale note-level labeled data is the major obstacle to singing transcription from polyphonic music. We address the issue by using pseudo labels from vocal pitch estimation models given unlabeled data. The proposed method first converts the frame-level pseudo labels to note-level through pitch and rhythm quantization steps. Then, it further improves the label quality through self-training in a teacher-student framework. To validate the method, we conduct various experiment settings by investigating two vocal pitch estimation models as pseudo-label generators, two setups of teacher-student frameworks, and the number of iterations in self-training. The results show that the proposed method can effectively leverage large-scale unlabeled audio data and self-training with the noisy student model helps to improve performance. Finally, we show that the model trained with only unlabeled data has comparable performance to previous works and the model trained with additional labeled data achieves higher accuracy than the model trained with only labeled data.