- South Korean researchers created ConcreteSC, a new way to improve wireless communication
- ConcreteSC avoids large codebooks, improves image transmission, and reduces errors
- The team believes it could help power 6G networks, smart factories, and healthcare devices
Researchers in South Korea have developed a new approach to semantic communication that could make future wireless systems faster and more efficient.
The new ConcreteSC method was created by a team led by Dr. Dong Jin Ji, Associate Professor at Seoul National University of Science and Technology, and was published on 19 June 2025 in IEEE Wireless Communications Letters.
Semantic communication is a shift in wireless technology where meaning is sent rather than raw data. For example, when transmitting an image, the system prioritizes what the picture represents instead of sending every pixel exactly. This saves both time and bandwidth and could be particularly useful for artificial intelligence and connected devices.
Potential for 6G
Existing systems often depend on vector quantization, a process that uses giant “codebooks” to store possible signal patterns. These codebooks are not only heavy to manage but struggle with errors and noise.
ConcreteSC solves this with a different mathematical idea.
Despite its name, it has nothing to do with the material used in buildings. Instead, “concrete” here refers to a special probability distribution in machine learning.
This tool makes it possible to turn continuous information into digital signals more smoothly, allowing the system to generate the bitstreams it needs directly, without the burden of managing large codebooks.
“Unlike vector quantization (VQ) – a state-of-the-art digitization technique that suffers from channel noise and codebook divergence during training – our framework offers a fully differentiable solution to quantization, allowing end-to-end training even under channel noise,” Dr. Ji said.
“Notably, due to the nature of the ConcreteSC that directly generates the required bitstream, it is possible to train a multi-feedback-length model pair with a relatively simple masking scheme,” he added.
In simulations, the researchers say ConcreteSC outperformed VQ-based methods in both structural similarity and peak signal-to-noise ratio. It also reduced complexity, as its operations grow only with bit length rather than expanding exponentially with codebook size.
The researchers believe this framework could play a worthwhile role in next-generation wireless systems, such as 6G.
They cite other potential uses such as in smart factories with ultra-dense machine communications, as well as healthcare and lifestyle monitoring systems powered by small AI-enabled devices.
