Information & Communications Engineering, HUFS. I build embedded systems end to end — power and circuit design, firmware, and the Linux side.
Edge AI in a hard real-time loop · thesis-edgeai-adaptive-kalman-embedded Designed and built a 2WD robot platform — two isolated supply domains (2S LiPo → LM2596 buck for logic, unregulated VM for motors, common ground), TB6612FNG H-bridge drive, ToF and quadrature encoders — then wrote all of its STM32F446RE firmware. An INT8 MLP (257 parameters, 3.2 KB) predicts the Kalman filter's measurement-noise covariance inside a 200 Hz control loop: 35.32 µs mean inference over 242,992 runs, zero overruns, zero watchdog resets. Phase 0–7 hardware verification covers rail stability, motor-induced sensor noise, and EMI on the telemetry link.
Linux device drivers on NVIDIA Jetson Xavier NX · jetson-gpio-driver-lab
The same switch/LED I/O built at three layers — /dev/mem register access, character device drivers
with request_irq() and a wait queue, then measured against each other. The interrupt-driven design
used one eighth the CPU time of polling.
Network degradation detection · kics2026 Inferring link degradation from encrypted traffic metadata and switching controllers accordingly. Paper in preparation for KICS 2026 Fall.
Tools — C, Embedded C, Python · STM32CubeIDE, ST Edge AI Core, KiCad · Linux kernel modules, device drivers, I2C/SPI/UART/DMA · TensorFlow Lite, TinyML quantization