[논문 리뷰] TinySense: Effective CSI Compression for Scalable and Accurate Wi-Fi Sensing

Toan Gian, Dung Thi Phuong Tran외 3명|arXiv (Cornell University)|2026. 01. 22.

Indoor and Outdoor Localization Technologies인용 수 0

한 줄 요약

TinySense는 적응형 코드북과 Transformer 기반 복구 모듈을 갖춘 VQGAN 기반 CSI 압축 프레임워크를 도입하여 대역폭과 지연을 줄인 상태에서 확장 가능하고 높은 정확도의 Wi-Fi sensing을 가능하게 한다.

ABSTRACT

With the growing demand for device-free and privacy-preserving sensing solutions, Wi-Fi sensing has emerged as a promising approach for human pose estimation (HPE). However, existing methods often process vast amounts of channel state information (CSI) data directly, ultimately straining networking resources. This paper introduces TinySense, an efficient compression framework that enhances the scalability of Wi-Fi-based human sensing. Our approach is based on a new vector quantization-based generative adversarial network (VQGAN). Specifically, by leveraging a VQGAN-learned codebook, TinySense significantly reduces CSI data while maintaining the accuracy required for reliable HPE. To optimize compression, we employ the K-means algorithm to dynamically adjust compression bitrates to cluster a large-scale pre-trained codebook into smaller subsets. Furthermore, a Transformer model is incorporated to mitigate bitrate loss, enhancing robustness in unreliable networking conditions. We prototype TinySense on an experimental testbed using Jetson Nano and Raspberry Pi to measure latency and network resource use. Extensive results demonstrate that TinySense significantly outperforms state-of-the-art compression schemes, achieving up to 1.5x higher HPE accuracy score (PCK20) under the same compression rate. It also reduces latency and networking overhead, respectively, by up to 5x and 2.5x. The code repository is available online at here.

연구 동기 및 목표

• Motivate scalable, privacy-preserving Wi-Fi sensing via efficient CSI transmission.
• Develop a task-aware compression framework that preserves HPE accuracy under high compression.
• Enable adaptive bitrate control through codebook resizing and clustering.
• Improve robustness to packet loss with a Transformer-based missing-index recovery mechanism.

제안 방법

• Encode CSI on-device using a VQGAN-based latent representation to produce VQ indices.
• Compress the VQ indices by mapping to a shared codebook and transmitting a compact bitstream.
• Dynamically resize the codebook with K-means to achieve variable bitrate and quality trade-offs.
• Decode on the server to reconstruct CSI and perform HPE via an estimator and a decoder.
• Use a second-stage Transformer to predict missing VQ indices under transmission losses for robustness.

Figure 1 : Overview of Cloud-based Wi-Fi Sensing.

실험 결과

연구 질문

• RQ1 How can CSI data be compressed to minimize bandwidth while preserving HPE accuracy?
• RQ2 Can a shared learned codebook enable flexible Bitrate control and robust reconstruction under network conditions?
• RQ3 Does introducing a Transformer-based missing-index prediction improve robustness to packet loss?
• RQ4 What are the end-to-end latency and networking overhead benefits of TinySense compared to state-of-the-art methods?

주요 결과

• TinySense achieves up to 1.5x higher PCK20 than EfficientFi and RSCNet at the same compression rate.
• It reduces latency by up to 5x and networking overhead by up to 2.5x compared to EfficientFi.
• Adaptive codebook resizing via K-means enables variable bitrate while maintaining reconstruction quality.
• In presence of lost indices, the Transformer-based recovery maintains performance much better than non-transformer variants.
• Experiments on MM-Fi and Wi-Pose datasets show consistent gains across NMSE and HPE metrics.

Figure 2 : Illustration of the proposed $\mathtt{TinySense}$ framework: (a) the overall compression pipeline, (b) the K-means clustering algorithm to adjust the codebook size for variable bitrates, and (c) a transformer approach for predicting lost VQ indices.