Thesis
Semantic communication for video coding and transmission
- Creator
- Rights statement
- Awarding institution
- University of Strathclyde
- Date of award
- 2026
- Thesis identifier
- T18156
- Person Identifier (Local)
- 202391694
- Qualification Level
- Qualification Name
- Department, School or Faculty
- Abstract
- ideo traffic dominates global internet bandwidth, and new applications such as extended reality, autonomous vehicle telemetry, and sixth-generation (6G) wireless networks are increasing the demands on video communication in bitrate, quality, robustness, content diversity, and dynamic range. Conventional standards such as H.266/Versatile Video Coding (VVC) are approaching the limits of manually designed coding tools, while end-to-end learned video codecs still suffer from unstable training, limited generalisation, and severe quality loss under channel impairments. Both approaches are semantically unaware because they aim for pixel-accurate reconstruction rather than preserving the meaning of the transmitted signal. This thesis argues that semantic communication, which sends task-relevant meaning instead of bit-accurate signal representations, offers a practical way forward, but its value for real video coding has not yet been developed in a single end-to-end framework. This thesis develops that framework through original contributions in four areas: hybrid coding, content adaptation, HDR extension, and robust wireless transmission, supported by methods for Deep Neural Network (DNN) parameter efficiency. For hybrid coding, the thesis introduces a hybrid semantic–VVC video codec, extends it with hierarchical bidirectional frame prediction, and further develops it into a context conditioned architecture based on a shared pretrained model with motion-aligned context generation. This system gives consistent rate–distortion gains over VVC and competitive performance against leading learned codecs. For content adaptation, the thesis restores per-Group-of-Pictures (GOP) specialisation by adjusting only a small part of the decoder parameters, which gives further gains across several datasets. For HDR extension, the thesis introduces an automatic and invertible tone-mapping step that bridges the HDR-to-SDR gap, allowing an SDR-trained codec to handle HDR content without changing the model architecture. For robust wireless transmission, the thesis develops a Semantic Multiple Description Coding (SMDC) framework that removes cliff effects across fading channels for both image and video transmission. Two further contributions on DNN parameter efficiency, temporal prediction of decoder parameters and VVC-based weight compression, reduce parameter transmission overhead and support future work toward a fully semantic video codec. Taken together, these contributions show that semantic communication can give consistent and reproducible gains over conventional and learned baselines in rate–distortion efficiency, channel robustness, content adaptation, and parameter-transmission efficiency. They provide a clear technical basis for semantic-driven video communication in next-generation wireless networks.
- Advisor / supervisor
- Fernando, Anil
- Dong, Feng
- Resource Type
- DOI
Relations
Items
| Thumbnail | Title | Date Uploaded | Visibility | Actions |
|---|---|---|---|---|
|
|
PDF of thesis T18156 | 2026-09-18 | Public | Download |