arXiv cs.AI· Lei Zhai, Zhihao Chang, Shuyuan Yang, Zhixi Feng·· 9 天前AI 评分15
BATok：面向指令化电磁频谱理解的预算自适应信号分词器
Instruction-Conditioned Electromagnetic Spectrum Understanding via Budget-Adaptive Signal Tokenization
AI 导读
研究者提出预算自适应信号分词器 BATok，可根据输入长度调整 token 容量并按信号内容分配，token 数量随长度自适应且严格有界，再投影到 VLM 的语言嵌入空间。同时发布 EMSpec-Instruct 多模态指令数据集，对齐原始 I/Q 信号、瀑布图与语言监督，覆盖调制识别、结构化检测和语言条件信号定位。实验显示 BATok 学到有效信号表示，在所有任务上取得有竞争力的表现。
正文 · 原文
Abstract:Electromagnetic spectrum monitoring increasingly requires flexible analysis beyond task-specific recognition and detection. Multimodal large language models offer a unified interface, but extending vision-language models (VLMs) to raw I/Q signals requires tokenization that balances fidelity against a strict budget. For signals, dense encoding causes token costs to grow with observation length, whereas fixed-resolution compression may discard short-duration or localized signal evidence. Thus, we propose \textbf{BATok}, a budget-adaptive signal tokenizer that adjusts token capacity to the input length while allocating that capacity according to the signal content. BATok constructs candidate representations from signal-derived features using lightweight multi-resolution branches, then combines a local energy prior with learnable queries to resample these representations into compact signal tokens. The number of tokens adapts to the input length while remaining strictly bounded. The resulting tokens are projected into the language embedding space of VLMs. We further introduce \textbf{EMSpec-Instruct}, a multimodal instruction dataset aligning raw I/Q signals, waterfall images, and language supervision for modulation recognition, structured detection, and language-conditioned signal grounding. Experiments show that BATok learns effective signal representations and achieves competitive performance across all tasks.
| Subjects: | Artificial Intelligence (cs.AI); Signal Processing (eess.SP) |
| Cite as: | arXiv:2610.12142 [cs.AI] |
| (or arXiv:2610.12142v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2610.12142 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Lei Zhai [view email]
[v1]
Thu, 8 Oct 2026 15:28:37 UTC (14,845 KB)
来源：arXiv cs.AI · arxiv.org