Undergraduate, Electronic Information Science & Technology, School of Physics, Chongqing University. Expected graduation 2027. Interested in the intersection of signal processing, AI-RAN, and the physical layer of communication systems.
I am an undergraduate at Chongqing University, studying Electronic Information Science & Technology within the School of Physics. Housed within a physics department, my curriculum grounds electronic-information training in physical fundamentals — a foundation that proves especially useful for my interests in signal processing, AI-RAN, and on-device machine learning. I enjoy bringing algorithms onto real hardware, from FPGA and ESP32 implementations to hardware-aware compression methods. My guiding principle in engineering is simple: let applications empower scholarship, let knowledge permeate life.
教育经历Education
重庆大学 · 物理学院Chongqing University · School of Physics
电子信息科学与技术 · 学士学位(在读)B.S. in Electronic Information Science & Technology (ongoing)
2023.09 — 2027.06
南京市第九中学Nanjing No.9 High School
高中学业 · 学生会副主席High School · Vice President, Student Union
2020.09 — 2023.06
研究兴趣Research Interests
信号处理
Related Work载物智能音响 · FPGA 音频解码与 DSP 处理数字信号处理 · 课程与相关实验Signal Processing
信息处理
Related WorkCMeta · 元字典压缩算法Hans · 熵重排压缩算法载物融光 · 远距离能量信息同传Information Processing
AI-RAN
Related WorkHans · 面向边缘智能的 ML 浮点压缩信号处理与机器学习的双重基础研究生阶段拟深入的核心方向AI-RAN
人工智能
Related Work载物绘途 · Transformer + CNN 盲道路径规划Hans · 面向 ML 的浮点数据压缩机器学习 · 课程基础Artificial Intelligence
嵌入式系统
Related Work载物智能音响 · FPGA 系统整合情绪抚慰桌面机器人 · FPGA 实现ESP32 智能家居实验装置Embedded Systems
基于 FPGA 的多功能智能音响:集成音频解码、信号处理与交互控制于单一芯片平台。FPGA-based smart speaker integrating audio decoding, DSP, and interactive control on a single platform.
FPGAVerilogDSPAudio
02
Z-Work Series▶ Live Demo
载物绘途PathfinderPathfinder
基于 Transformer 及 CNN 架构的盲道路径规划系统A tactile-path planning system built on Transformer and CNN architectures
结合 Transformer 与 CNN 的视觉识别与路径规划系统,用于视障人士的盲道导航。A vision-based planning system combining Transformer and CNN architectures to assist visually-impaired pedestrians along tactile paving.
+
繪途是一套面向视障者的盲道导航辅助系统,由三个阶段构成:语义分割(采用 Transformer 架构的轻量级分割模型,以自注意力机制捕获长距离依赖,在自建盲道数据集上端到端训练,针对树影遮挡设计了定向数据增强)、几何推理(从分割结果提取盲道左右边界,经 RANSAC 鲁棒拟合消失点,据此估计地平线、中心线轨迹与前方距离)、障碍物决策(集成目标检测模块,判断行人/车辆是否位于盲道上,触发碰撞警告并动态改变中心线颜色)。系统在 GPU 上端到端实时推理,单帧约 40 ms,为后续语音引导预留了充足余量。
Pathfinder is an assistive navigation system for visually-impaired pedestrians, built as a three-stage pipeline: semantic segmentation (a lightweight Transformer-based model leveraging self-attention to capture long-range dependencies, trained end-to-end on a custom tactile-paving dataset with targeted data augmentation for tree-shadow occlusion), geometric reasoning (extracting tactile-path boundaries from the segmentation, using RANSAC to robustly fit a vanishing point, and thereby estimating the horizon line, centerline trajectory, and forward distance), and obstacle decision (integrating an object detector to determine whether pedestrians or vehicles occupy the tactile path, triggering a collision alert and dynamically recoloring the centerline). The full pipeline runs end-to-end on GPU at roughly 40 ms per frame, leaving ample headroom for downstream voice guidance.
Live Demo · inputscene · 01
Collision Warning
dist— m
pitch— °
模型在真实街景上的逐步推理过程——拖动进度条查看每个阶段。
拖动下方进度条可在五阶段间切换 · 双击恢复自动播放Drag the timeline to move between phases · double-click to resume autoplay
Simultaneous Lightwave Information and Power Transfer——以光电方式实现能量与信息同步远距离传输的实验装置。An optical-domain power-data coupling apparatus for simultaneous long-range transfer of energy and information.
OptoelectronicsSLIPTRF
04
▶ Live Demo
CMeta元字典压缩算法Codebook Meta · Compression
基于「元字典」机制的快速无损压缩算法A fast lossless compression algorithm driven by a persistent meta codebook
传统 Huffman 需对每份数据从零统计、重建码树;CMeta 引入随时间演化的「元字典」作为统计先验,跳过全量统计的同时让码本持续逼近真实分布——在 lzbench 标准测试中达到 29.84% 压缩率,优于 xz/brotli 最高级别。Classical Huffman re-scans every file and rebuilds the tree from scratch. CMeta introduces a time-evolving meta codebook as a statistical prior — skipping full scans while letting the codebook converge to the true distribution. On lzbench: 29.84% ratio, beating xz / brotli at their highest levels.
+
技术细节受保护
Protected Content
本项目算法细节涉及未发表的技术成果。如需查看完整内容,请输入访问密码,或联系作者。
This project's algorithmic details involve unpublished technical work. Please enter the access password to view full content, or contact the author.
基于比特重排的机器学习浮点数据压缩方法Bit-reordering floating-point compression for machine learning
针对机器学习权重与激活值的浮点压缩方法,通过比特重排揭示冗余结构以提升压缩率。A floating-point compression method for ML weights and activations, using bit-reordering to expose redundancy and improve compression ratio.
+
技术细节受保护
Protected Content
本项目涉及未发表的核心技术。如需查看完整内容,请输入访问密码,或联系作者。
This project involves unpublished core techniques. Please enter the access password to view full content, or contact the author.
以 ESP32 微控制器为核心的智能家居测试平台,集成多种传感与执行器件,构成完整的家居交互场景。An ESP32-based smart-home testbed integrating sensors and actuators to form a complete interactive living scenario.
以 FPGA 实现感知、表情生成与交互逻辑的小型桌面陪伴机器人,针对情绪识别与响应设计专用硬件电路。A small desktop companion robot with sensing, expression generation, and interaction logic on FPGA, featuring custom hardware for emotion recognition and response.
FPGAHDLHCI
奖项Awards
2025
全国大学生嵌入式芯片与系统设计竞赛National Embedded Chip & System Design Contest