CQILAB/GenSC-6G-Segmentation

GenSC-6G - Scalable Semantic Communication Framework and Dataset

This repository contains the first semantic communication dataset and playground, designed to be scalable, reproducible, and adaptable for a wide range of applications. The dataset and framework are tailored for semantic decoding, classification, and localization tasks in 6G applications, integrating generative AI and semantic communication. Implementation of GenSC-6G: A Prototype Testbed for Integrated Generative AI, Quantum, and Semantic Communication.

---

Citation

The paper can be found at arXiv. If you use this dataset or framework in your research, please cite:

@article{gensc6g,
      title={GenSC-6G: A Prototype Testbed for Integrated Generative AI, Quantum, and Semantic Communication}, 
      author={Brian E. Arfeto and Shehbaz Tariq and Uman Khalid and Trung Q. Duong and Hyundong Shin},
      year={2025},
      eprint={2501.09918},
      archivePrefix={arXiv},
      primaryClass={cs.AI},
      url={https://arxiv.org/abs/2501.09918}, 
}

---

Features of the GenSC-6G Dataset

🔧 Adaptable SC Framework

A flexible prototype that supports modifications to baseline models, communication modules, and decoders, enabling customization for diverse communication needs.

🤖 Generative AI-Driven SC

The integration of generative AI for synthetic data generation, enriching the Knowledge Base (KB) and leveraging large language model (LLM) capabilities for enhanced semantic tasks.

📊 Noise-Augmented Dataset

A labeled dataset with injected noise, specifically optimized for semantic tasks such as target recognition, localization, and recovery. The dataset comprises 4,829 training and 1,320 testing instances across 15 classes of military and civilian vehicle types. It incorporates Additive White Gaussian Noise (AWGN) and Radio Frequency (RF) interference at varying Signal-to-Noise Ratios (SNRs) to evaluate model robustness under realistic channel conditions.

📥 Dataset Download and Overview

Main Dataset

Download the main dataset here

Segmentation Dataset

Download the segmentation dataset here

Setup Instructions

Can be found in Official Repository: CQILAB/GenSC-6G

Reproducibility

🗃️ Dataset

Labeled dataset with ground-truth data, noise features, and extracted semantic features. Uploaded to HuggingFace🤗

Dataset Columns and Descriptions

• image: Raw image data used for training and evaluation.
• image_path: Path to the corresponding image file.
• classification_class: Integer label corresponding to the classification category (0-15).
• classification_{basemodel}_features: Extracted feature embeddings from {basemodel}'s encoder, consisting of 1000 float32 tensors.
• classification_awgn10dB_{basemodel}_features: Feature embeddings extracted from {basemodel} encoder with Additive White Gaussian Noise (AWGN) at 10dB SNR.
• classification_awgn30dB_{basemodel}_features: Feature embeddings extracted from {basemodel} encoder with AWGN at 30dB SNR.
• upsampling_{basemodel}_features: Extracted feature embeddings for upsampling tasks using {basemodel} encoder, consisting of 1000 float32 tensors.
• upsampling_awgn10dB_{basemodel}_features: Upsampling features with AWGN at 10dB SNR for {basemodel}.
• upsampling_awgn30dB_{basemodel}_features: Upsampling features with AWGN at 30dB SNR for {basemodel}.

🏗️ Testbed

To experiment with real-world semantic communication, you can use the GNURadio and HackRF.

1. Install Dependencies:
   • Install GNU Radio
   • Install HackRF tools: sudo apt install hackrf
2. Configure Transceiver:
   • Transmitter config: GNURadio/transmitter.grc
   • Outputs a streaming binary file
3. Run Transmitter:
   • Open GNURadio/transmitter.grc in GNU Radio Companion
   • Set SDR parameters (frequency, gain, bandwidth)
   • Execute to start transmission
4. Run Receiver:
   • Modify GNURadio/receiver.grc settings
   • Run to capture and process signals By following these steps, you can replicate real-world transmission experiments using the testbed and analyze its performance.

📊 Performance Metrics & Flexible Code

Can be found in Official Repository: CQILAB/GenSC-6G

Others

• Official Repository: CQILAB/GenSC-6G
• Dataset: HuggingFace

License

MIT License