TinySense+: Entropy-Aware CSI Compression for Scalable and Adaptive Wi-Fi Sensing

Phuc H. Nguyen; Toan Gian; Cuong D. Do; Symeon Chatzinotas; Van-Dinh Nguyen

cs.CVJan 22, 2026

TinySense+: Entropy-Aware CSI Compression for Scalable and Adaptive Wi-Fi Sensing

Authors: Phuc H. Nguyen, Toan Gian, Cuong D. Do, Symeon Chatzinotas, Van-Dinh Nguyen

Organizations: Smart Green Transformation Center (GREEN-X), VinUniversity, Hanoi, Vietnam · School of Computer Science and Statistics, Trinity College Dublin, Ireland

Abstract

Wi-Fi human pose estimation requires high-rate channel state information (CSI), making communication a bottleneck when sensing is offloaded from resource-constrained devices. Effective compression must therefore preserve pose information while accommodating different communication budgets. We present TinySense+, an efficient compression framework for scalable Wi-Fi-based human sensing with entropy-aware transmission and test-time rate adjustment. The framework uses a vector-quantized generative adversarial network to encode CSI into discrete latent indices shared between an edge device and a server. A causal Transformer predicts conditional index probabilities for arithmetic coding, while a differentiable rate-distortion objective balances expected code length, CSI reconstruction, and pose accuracy. Prefix-conditioned latent completion reuses the entropy model to infer an omitted suffix, allowing the transmitted prefix length to change without retraining the encoder. We evaluate the framework on MM-Fi and WiPose datasets and prototype it on Raspberry Pi and Jetson Nano devices connected to an RTX 8000 server. On MM-Fi, entropy coding reduces the baseline cost from 8.00 to 4.82 bits per index without changing pose accuracy. Rate-aware training further reduces this cost to 3.21 bits per index and improves PCK20 from 41.21% to 44.01%. Results also show that transmitting half of the latent sequence retains a PCK20 of 40.36%. The evaluated operating range reaches a compression ratio of 5840 with PCK20 = 37.92%. These results show that optimizing latent predictability alongside sensing fidelity provides a useful communication-accuracy trade-off for edge-server Wi-Fi sensing.

Figures & tables

Fig. 1: Edge–server Wi-Fi sensing: local devices compress CSI into latent indices for transmission and server-side pose estimation.

Aspect	$\mathtt{TinySense}$	$\mathtt{TinySense}+$
Training	Task-aware VQGAN	Adds rate regularization
Index coding	Fixed-length binary	Causal entropy model with arithmetic coding
Rate control	Codebook resizing	Adds adjustable prefix length
Recovery	Predicts erased indices	Entropy model to complete omitted suffix
Evaluation	Compression, sensing, and testbed costs	Adds rate and prefix ablations

TABLE I: TinySense and TinySense+: Key Differences

Fig. 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.

Fig. 3: Recovery Transformer for completing missing VQ indices.

Fig. 4: The

\mathtt{TinySense+}

pipeline: causal entropy modeling for arithmetic encoding (AE) and decoding (AD), followed by prefix-conditioned suffix completion using the same model.

Method	$\boldsymbol{\eta}$	MM-Fi				WiPose
Method	$\boldsymbol{\eta}$	NMSE (dB) $\downarrow$	$\textbf{PCK}_{30}\uparrow$	$\textbf{PCK}_{20}\uparrow$	MPJPE $\downarrow$	NMSE (dB) $\downarrow$	$\textbf{PCK}_{10}\uparrow$	$\textbf{PCK}_{5}\uparrow$	MPJPE $\downarrow$
LASSO	4	-8.42	N/A	N/A	N/A	-15.64	N/A	N/A	N/A
DeepCMC	4	-9.58	N/A	N/A	N/A	-16.86	N/A	N/A	N/A
GLC	4	-13.18	N/A	N/A	N/A	-20.75	N/A	N/A	N/A
CsiNet	4	-4.75	N/A	N/A	N/A	-8.16	N/A	N/A	N/A
	16	-4.25				-7.93

TABLE II: Compression Performance Comparison across Different Schemes on both MM-Fi and WiPose Datasets ( “N/A” indicates methods not applicable to the task )

Fig. 5: Testbed for measuring edge–server latency, local energy, and transmitted payload.

MM-Fi
Method	$\eta$	$\textbf{PCK}_{20}\uparrow$	$\textbf{MPJPE$$}\downarrow$	$\textbf{Params}\downarrow$
WiSPPN	N/A	45.41	166.59	26.78M
PerUnet	N/A	52.12	154.66	34.51M
MetaFi++	N/A	48.64	162.45	26.42M
PiW-3D	N/A	52.18	155.69	16.47M
AdaPose	N/A	50.58	161.75	14.62M

TABLE III: HPE Performance on MM-Fi and WiPose (Parameter counts are in millions (M); N/A indicates an unreported compression ratio)

	$\mathtt{TinySense}$			$\begin{array}[]{c}\textbf{$ \mathtt{TinySense} $}\\ \textbf{Without Transformer}\end{array}$
$\epsilon$	NMSE $\downarrow$	$\textbf{PCK}_{20}$ $\uparrow$	MPJPE $\downarrow$	NMSE	$\textbf{PCK}_{20}$	MPJPE
0.1	-12.02	39.16	171.45	-10.14	31.64	190.26
0.3	-11.24	36.12	178.63	-9.26	26.63	198.74
0.5	-10.88	32.69	186.54	-8.56	21.16	218.65
0.7	-10.04	30.14	192.68	-7.94	18.63	227.98
0.9	-9.65	26.89	198.86	-7.01	14.23	242.12

TABLE IV: Effect of the recovery Transformer under synthetic index erasure on MM-Fi

Model	$\boldsymbol{\epsilon}$	$\boldsymbol{\eta}\uparrow$	$\textbf{NMSE}\downarrow$	$\textbf{PCK}_{20}\uparrow$	$\textbf{MPJPE}\downarrow$
$\mathtt{TinySense}$	0.1	215	-6.29	36.51	148.54
	0.3	176	-6.44	36.78	148.52
	0.5	96	-6.21	35.51	149.58
	0.7	64	-5.51	35.78	149.73
	0.9	16	-5.29	35.01	157.03
$\mathtt{TinySense+}$	0.1	1949	-8.05	46.09	165.53

TABLE V: Adaptation Performance Comparison: Base

\mathtt{TinySense}

vs.

\mathtt{TinySense+}

under varying loss rates (

\epsilon

), aiming to maintain

\text{PCK}_{20}>35\%

Fig. 6: t-distributed stochastic neighbor embedding (t-SNE) visualizations of CSI representations: (a) raw CSI with action labels, (b) quantized features with action labels, (c) raw CSI with subject labels, and (d) quantized features with subject labels.

Fig. 7: Visualization of the human pose landmarks generated by the vision model (red) and Wi-Fi model (blue) on MM-Fi at different compression ratios.

Weight $\lambda$	$K=64$	$K=128$	$K=256$	$K=512$
0.10	36.8	41.8	46.5	44.1
0.25	40.5	42.1	44.3	43.6
0.50	41.8	45.7	41.5	39.3
0.75	39.7	42.6	39.6	42.6
1.00	37.5	42.5	41.5	43.3

TABLE VI: PCK 20 (%) versus

K

and

\lambda

on MM-Fi (Best in bold ; second best underlined)

Configuration	NMSE $\downarrow$	$\textbf{PCK}_{20}\uparrow$	$\textbf{MPJPE}\downarrow$
$\mathtt{TinySense+}$	-12.43	44.01	170.17
$\mathtt{TinySense}$	-13.74	41.21	170.98
w/o $\mathcal{L}_{\mathtt{GAN}}$	-13.12	38.50	179.45
w/o $\mathcal{L}_{\mathtt{keypoint}}$	-14.50	29.15	205.10

TABLE VII: Ablation Study of Loss Function Components

$\boldsymbol{\gamma}$	Bits/Index $\downarrow$	$\boldsymbol{\eta}\uparrow$	$\textbf{PCK}_{20}\uparrow$	NMSE $\downarrow$
0	4.82	356	41.21	-13.74
0.01	4.15	413	42.56	-13.31
0.05	3.62	474	43.72	-12.91
0.10	3.21	535	44.01	-12.43
0.25	2.63	653	41.82	-11.36
0.50	2.11	814	37.95	-10.17

TABLE VIII: Effect of the Rate Regularization Weight

\gamma

on TinySense+

Configuration	Bits/Index $\downarrow$	$\boldsymbol{\eta}\uparrow$	NMSE $\downarrow$	$\textbf{PCK}_{20}\uparrow$	MPJPE $\downarrow$
TinySense	8.00	215	-13.74	41.21	170.98
TinySense + entropy model	4.82	356	-13.74	41.21	170.98
TinySense+ w/o $\mathcal{L}_{\mathrm{rate}}$	4.82	356	-13.74	41.21	170.98
TinySense+ w/ $\mathcal{L}_{\mathrm{rate}}$	3.21	535	-12.43	44.01	170.17
TinySense+ + PCLC ( $\rho=0.75$ )	2.42	710	-11.86	42.73	173.91
TinySense+ + PCLC ( $\rho=0.50$ )	1.69	1015	-10.92	40.36	178.42

TABLE IX: Ablation Study of the Training and Transmission Components of TinySense+ on MM-Fi

$\text{P}\times\text{P}$	NMSE $\downarrow$	$\textbf{PCK}_{30}\uparrow$	$\textbf{PCK}_{20}\uparrow$	MPJPE $\downarrow$
$2\times 2$	-13.782	60.152	39.086	170.45
$3\times 3$	-14.422	63.158	46.516	166.87
$4\times 4$	-13.582	59.238	40.562	172.69
$5\times 5$	-14.082	62.881	44.163	168.14

TABLE X: Analysis of the Impact of the Transformer’s Sliding Window on

\mathtt{TinySense}

Performance

Fig. 8: Codebook utilization histogram of

\mathtt{TinySense+}

(

K=256

Fig. 9: Rate–accuracy operating points of

\mathtt{TinySense+}

under a controlled prefix-ratio sweep (Lines connect evaluated settings and

\eta

is shown on a logarithmic axis).

Fig. 10: Latency (left) and transmitted payload (right) for

\mathtt{TinySense}

\mathtt{TinySense+}

, and the evaluated baselines.

Fig. 11: Pipeline latency breakdown (left) and local energy versus transmitted payload (right).

Wi-Fi Sensing

Wi-Fi sensing uses wireless signal changes to infer human motion, location, presence, and activity without dedicated wearable sensors.

See all 50 papers

WiFi-JEPA: Self-supervised Learning for WiFi-CSI 3D Human Pose Estimation
Jul 13, 2026Doeon Kim, Jungyoon Lee, Seongsin Kim +1Self-Supervised Learning Joint-Embedding Predictive Architecture
Department of Intelligent Semiconductors, Soongsil University, Republic of Korea · Department of AI Convergence Security, Soongsil University, Republic of Korea · School of AI Software, Soongsil University, Republic of Korea
Seeing Through WiFi: Lightweight Human Pose Estimation with Dynamic Kernel Attention
Jul 3, 2026Toan D. Gian, Van-Dinh Nguyen, Vo Phi Son +5Dynamic Convolution Channel Attention
Smart Green Transformation Center (GREEN-X), VinUniversity, Vietnam · Trinity College Dublin, Ireland · University of Oulu, Finland +3
Deep Learning-Enhanced Real-Time Wi-Fi Sensing Through Single Transceiver Pair
Oct 21, 2025Yuxuan Liu, Chiya Zhang, Yifeng Yuan +3Wi-Fi Sensing Wireless Localization
School of Electronic and Information Engineering, Harbin Institute of Technology, Shenzhen 518055, China. · School of Electronic and Information Engineering, Harbin Institute of Technology, Shenzhen 518055, China, and also with the Peng Cheng Laboratory (PCL), Shenzhen 518055, China. · Guangdong Key Laboratory of Intelligent Information Processing, Shenzhen University, Shenzhen 518060, China. +1

Jul 13, 2026J/K move · Enter open · S save

WiFi-JEPA: Self-supervised Learning for WiFi-CSI 3D Human Pose Estimation

Doeon Kim, Jungyoon Lee, Seongsin Kim, Seong-heum Kim

Department of Intelligent Semiconductors, Soongsil University, Republic of Korea · Department of AI Convergence Security, Soongsil University, Republic of Korea · School of AI Software, Soongsil University, Republic of Korea

Self-Supervised Learning Joint-Embedding Predictive Architecture 3D Human Pose Estimation Human Pose Estimation Wi-Fi Sensing

Open paper AskPDF arXiv

WiFi Channel State Information (CSI) enables privacy-preserving human pose sensing in camera-denied environments, but existing WiFi-based pose estimators often fail under environment shifts and rely on costly camera-based annotation pipelines that limit scale. We propose WiFi-JEPA, a self-supervised framework that learns CSI-native representations by predicting masked latent embeddings instead of reconstructing raw CSI signals that may contain hardware-specific artifacts. WiFi-JEPA makes three contributions: (i) CSI-specific tokenization and link masking tailored to the CSI tensor over channel, time, and link (C,T,L); masking entire Tx-Rx antenna links forces the model to predict one spatial link view from others, capturing cross-link correlations informative of 3D spatial structure. (ii) A ray-tracing CSI simulation pipeline that generates diverse unlabeled CSI from randomized geometric primitives, providing scalable pre-training data without pose annotations. (iii) State-of-the-art results on Person-in-WiFi-3D: WiFi-JEPA outperforms prior WiFi-CSI baselines on both single- and multi-person 3D pose estimation under the same evaluation protocol. We also show that simulated CSI provides complementary pre-training signal to real CSI, and that four vision-native SSL objectives degrade performance below training from scratch, whereas WiFi-JEPA consistently improves downstream pose estimation.

Abstract

Wi-Fi Sensing

Explore similar work

WiFi-JEPA: Self-supervised Learning for WiFi-CSI 3D Human Pose Estimation

Seeing Through WiFi: Lightweight Human Pose Estimation with Dynamic Kernel Attention

Deep Learning-Enhanced Real-Time Wi-Fi Sensing Through Single Transceiver Pair