RAG-CSI retrieves channel history to improve CSI prediction
Original title: RAG-CSI: Global Local Retrieval via Joint State and Dynamics Matching for CSI Prediction
On a multi-scene dataset, the method cut normalized mean squared error 15.42% and improved all five metrics by more than 10%.
RAG-CSI is a retrieval-augmented Transformer that predicts future wireless channel state information from history plus stored channel examples. A global-local knowledge store holds representative observations across scenes and history from the current serving scene. Joint state-dynamics retrieval matches channel state, phase dynamics, and amplitude dynamics, and cross-attention fuses the retrieved sequences into the predictor with few extra parameters. On a multi-scene CSI dataset, gains exceeded 10% on all five metrics versus the best baseline for each metric, including a 15.42% drop in normalized mean squared error.
- Retrieval covers channel state plus phase and amplitude dynamics.
- The store mixes cross-scene examples with the current scene's history.
- Normalized mean squared error fell 15.42% versus the best baseline.
- Improvements exceeded 10% on every one of five evaluation metrics.
- AI models
- RAG-CSI
From the original article
1 min read at arxiv.orgChannel state information (CSI) prediction forecasts future channel states from historical observations, supporting transceiver design and resource allocation with reduced measurement overhead. However, existing deep learning predictors mainly encode channel knowledge in model parameters, which can limit their ability to represent diverse patterns of channel dynamics across scenes.
By Xinyue Jiang · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.