Channel-Adaptive Region Adjacency Graph Carriers for Semantic Image Communication

Karim Abdallah; Maria Slim; Mariette Awad; Hadi Sarieddeen

eess.SPSep 13, 2026

Channel-Adaptive Region Adjacency Graph Carriers for Semantic Image Communication

Authors: Karim Abdallah, Maria Slim, Mariette Awad, Hadi Sarieddeen

Organizations: Department of Electrical and Computer Engineering American University of Beirut, Beirut, Lebanon

Abstract

Semantic image communication seeks to preserve task-relevant scene structure under limited channel resources, but carriers are often dense latent tensors or grid-aligned semantic layouts that do not explicitly encode region-level relations. This work introduces a segmentation-derived region adjacency graph (RAG) carrier, termed channel-adaptive RAG (CA-RAG), for joint source-channel coding-style image communication. Nodes store interpretable region attributes, edges preserve adjacency, channel-adaptive graph simplification (CGS) controls the node budget, and semantic belief propagation refines noisy graph embeddings before diffusion-based reconstruction. On Cityscapes, pre-channel RAG payloads are several times smaller than compressed class-index layouts in a 2,000-image study. Under additive white Gaussian noise at signal-to-noise ratios from 0 to 15 dB, CA-RAG reports higher semantic consistency than deep joint source-channel coding and a same-decoder layout baseline, with comparable perceptual quality. At 10 dB, the full-budget rate-sweep point reaches mean intersection over union (mIoU) = 0.329 at approximately 3.3 x 10^3 channel uses, while the default adaptive-CGS setting reports mIoU = 0.294 at approximately 2.6 x 10^3 channel uses.

Figures & tables

	Layout (PNG)	RAG (fp16+zlib)
Mean bits	45 969.8	8 170.5
Median bits	45 256.0	8 200.0
RAG smaller on	2,000 / 2,000 images

TABLE I: Pre-channel carrier compactness on 2,000 Cityscapes images. PNG and zlib are lossless; fp16 can quantize. Not rate-matched.

SNR (dB)	$\overline{M}$	Ch. uses	Bandwidth ratio $\rho$	Storage (16-bit, kbit)
0	27.2	1 742	$3.3\times 10^{-3}$	27.9
5	34.0	2 175	$4.1\times 10^{-3}$	34.8
10	40.8	2 610	$5.0\times 10^{-3}$	41.8
15	47.6	3 046	$5.8\times 10^{-3}$	48.7
20	54.4	3 480	$6.6\times 10^{-3}$	55.7

TABLE II: Transmission cost on 2,000 Cityscapes images; channel uses are real symbols, 16-bit sizes are storage only, and

\overline{M}

is the mean retained-node count.

	SNR (dB)	0	5	10	15
LPIPS $\downarrow$	DeepJSCC	0.673 (.021)	0.663 (.022)	0.657 (.024)	0.653 (.025)
	Layout	0.587 (.024)	0.561 (.025)	0.561 (.025)	0.559 (.025)
	Ours	0.571 (.025)	0.563 (.024)	0.555 (.025)	0.551 (.023)
mIoU $\uparrow$	DeepJSCC	0.014 (.011)	0.016 (.011)	0.017 (.011)	0.017 (.011)
	Layout	0.142 (.037)	0.257 (.054)	0.255 (.055)	0.251 (.049)
	Ours	0.233 (.062)	0.264 (.069)	0.294 (.071)	0.312 (.070)

TABLE III: End-to-end AWGN comparison; 500 images; four channel-noise realizations per image. Layout is the GESCO-based baseline. Bold : best per SNR; parentheses: standard deviations.

Fig. 1: End-to-end AWGN comparison over SNR.

Fig. 2: Carrier-level rate–quality sweep at 10 dB, excluding CA-RAG metadata. RAG has lower LPIPS and higher mIoU than the layout baseline.

Fig. 3: Qualitative comparison at 0/15 dB. Bandwidth ratios: DeepJSCC $\rho\!\approx\!0.021$ , layout carrier $\rho\!\approx\!3.0$ , and CA-RAG $\rho\!=\!3.3\times 10^{-3}$ / $5.8\!\times\!10^{-3}$ .

	Adaptive CGS (ours)				Fixed-max (no adaptation)
SNR	LPIPS	mIoU	Rate	Nodes	LPIPS	mIoU	Rate	Nodes
0	0.571	0.233	1 774	27.7	0.550	0.300	3 548	55.4
5	0.563	0.264	2 219	34.7	0.547	0.312	3 548	55.4
10	0.555	0.294	2 661	41.6	0.548	0.321	3 548	55.4
15	0.551	0.312	3 105	48.5	0.548	0.318	3 548	55.4

TABLE IV: Adaptive CGS versus fixed-max graph budget on 500 validation images with four channel-noise realizations per image.

Channel	Mode	SNR = 0 dB		SNR = 15 dB
		LPIPS $\downarrow$	mIoU $\uparrow$	LPIPS $\downarrow$	mIoU $\uparrow$
AWGN	–	0.571	0.237	0.550	0.309
Rayleigh	node-block	0.572	0.229	0.550	0.310
Rayleigh	graph-block	0.570	0.233	0.549	0.307

TABLE V: Idealized Rayleigh fading vs. AWGN on 500 validation images with four channel realizations per image.

Fig. 4: Structured node erasure at AWGN 10 dB on 500 validation images with four channel-noise realizations per image.

Fig. 5: Encoder ablation on 500 validation images with four channel-noise realizations per image: GAT vs. per-node MLP.

$(r_{\min},r_{\max})$	$\overline{M}$	LPIPS $\downarrow$	mIoU $\uparrow$
$(0.30,0.80)$	28.5	0.561	0.259
$(0.30,1.00)$	33.6	0.556	0.284
$(0.40,0.90)$	33.6	0.556	0.278
$(0.50,1.00)^{\ast}$	38.8	0.550	0.301
$(0.60,1.00)$	41.5	0.545	0.318
$(0.70,1.00)$	44.0	0.545	0.313

TABLE VI: CGS

(r_{\min},r_{\max})

sensitivity at 10 dB on 500 validation images with four channel-noise realizations per image. The ∗ denotes the default configuration.

Variant	LPIPS $\downarrow$	mIoU $\uparrow$
Full system	0.551 (.024)	0.290 (.063)
No CGS (full graph)	0.546 (.026)	0.319 (.074)
No SBP	0.550 (.025)	0.291 (.069)
No CGS + No SBP	0.550 (.027)	0.305 (.062)

TABLE VII: CGS/SBP ablation (10 dB; 500 images; four channel-noise realizations per image. “No CGS” transmits the full graph.

Joint Source-Channel Coding

See all 14 papers

Adaptive Role Assignment and Controlled Feature Reconstruction for Federated Image Semantic Communication
Oct 11, 2026Yuxuan Chen, Jiao Bie, Hang Qi +1Semantic Communication Federated Learning
Guangdong Prov./Zhuhai Key Lab of IRADS, Beijing Normal-Hong Kong Baptist University, Guangdong, P. R. China
Doubly Adaptive Channel and Spatial Attention for Semantic Image Communication by IoT Devices
Feb 26, 2026Soroosh Miri, Sepehr Abolhasani, Shahrokh Farahmand +2Deep Joint Source-Channel Coding Spatial Attention
School of Electrical Engineering, Iran University of Science and Technology (IUST), Tehran, Iran · Great Bay University, Dongguan 523000, China
Adaptive Semantic Communication for Wireless Image Transmission Leveraging Mixture-of-Experts Mechanism
Apr 3, 2026Haowen Wan, Qianqian YangAdaptive Model Routing Sparse Mixture-of-Experts
Polytechnic Institute, Zhejiang University, Hangzhou, China · College of Information Science and Electronic Engineering, Zhejiang University, Hangzhou, China

Oct 11, 2026J/K move · Enter open · S save

Adaptive Role Assignment and Controlled Feature Reconstruction for Federated Image Semantic Communication

Yuxuan Chen, Jiao Bie, Hang Qi, Jingjin Wu

Guangdong Prov./Zhuhai Key Lab of IRADS, Beijing Normal-Hong Kong Baptist University, Guangdong, P. R. China

Semantic Communication Federated Learning

Open paper AskPDF arXiv

Hybrid federated semantic communication enables user equipment (UEs) to contribute model updates or semantic features under a shared training framework. However, heterogeneous role costs and the interaction between model aggregation and feature reconstruction (FR) remain key challenges in practical radio access network (RAN) deployment. To address these issues, we propose ARC-FedSC, a RAN-aware framework that coordinates UE role assignment and server-side controlled FR under a fixed communication budget. Within ARC-FedSC, RAN Aware Role-Bandit learns the historical contribution of UEs in different roles together with current uplink conditions, while Temporal Consistency Alignment (TCA) regulates FR according to recent model aggregation directions. Experiments on CIFAR-10 with time-varying RAN channels show that ARC-FedSC outperforms baselines across all five tested role configurations in the later training stage, with the best configurations achieving about 24.17 dB average PSNR over the last 20 rounds. Component level results further show that Role-Bandit maintains an average rolling UE coverage of 85.3%, compared with about 65.7% for baseline, while TCA reduces the probability of harmful FR events from 0.245 to 0.093, corresponding to a 62.0% reduction. Ablation and sensitivity studies further support the effectiveness and robustness of the proposed design.

Abstract

Joint Source-Channel Coding

Explore similar work

Adaptive Role Assignment and Controlled Feature Reconstruction for Federated Image Semantic Communication

Doubly Adaptive Channel and Spatial Attention for Semantic Image Communication by IoT Devices

Adaptive Semantic Communication for Wireless Image Transmission Leveraging Mixture-of-Experts Mechanism