Answers are grounded in this paper and cite their sources.
In massive multiple-input multiple-output (MIMO) systems utilizing the frequency-division duplex (FDD), conventional methods for acquiring downlink channel state information (CSI) lead to high computational complexity and heavy feedback overheads. To tackle the aforementioned difficulties, we propose an end-to-end deep learning (DL)-based framework for CSI acquisition, called high-speed orthogonal probabilistic featurebased attention network (HOPFNet), which integrates pilot design and CSI feedback. Unlike conventional end-to-end network designs, HOPFNet ignores channel estimation at the user equipment (UE). Instead, it directly maps the pilot signals received at the UE into feedback codewords, which are then transmitted back to the base station (BS) to reconstruct the downlink channel. In recent years, Transformer-based networks have proven highly effective for CSI acquisition. However, the self-attention mechanism of Transformer-based networks introduces high computational complexity, posing challenges to actual deployment. To this end, we propose a lightweight Transformer, which is based on a highspeed orthogonal probabilistic feature-based attention (HOPFA) mechanism. The simulation results verify that the proposed HOPFNet can significantly reduce computation complexity while attaining lower normalized mean square error (NMSE) compared to the benchmark models. In addition, these results demonstrate superior efficiency in computing resources.
No takes yet. Share an insight, caveat, or question.
Sun et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: