E2E Learning Massive MIMO for Multimodal Semantic Non-Orthogonal Transmission and Fusion

2025-09-09

arXiv.org

AI Summary

To address the challenges of high-dimensional channel state information (CSI) acquisition, excessive feedback overhead, and limited precoding performance in massive MIMO systems, this paper proposes an end-to-end uplink–downlink CSI fusion precoding framework. The method introduces a multimodal semantic fusion architecture that jointly models uplink sounding reference signals (SRS) and downlink CSI-reference signals (CSI-RS); designs a MAXIM-based projection network to drive cross-domain CSI-RS optimization; integrates a quantized feedback encoder with dual-path SRS/feedback-driven precoding branches; and proposes a learnable fusion network along with a three-stage spectral efficiency-oriented training strategy. Simulation results demonstrate that, under constrained feedback budgets, the proposed scheme significantly improves spectral efficiency by effectively synergizing uplink channel priors and user equipment feedback, consistently outperforming conventional precoding approaches across all evaluated metrics.

Technology Category

• Machine Learning: Multimodal Learning
• Intelligent Robots: Multimodal Perception & Sensor Fusion
• Computer Vision: Multi-modal Vision

Application Category

• User Modeling, Personalization and Recommendation: Federated recommendation systems and personalization
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Massive multiple-input multiple-output (MIMO) promises high spectral efficiency but also leads to high-dimensional downlink channel state information (CSI), which complicates real-time channel acquisition and precoding. To address this, we propose an end-to-end (E2E) uplink-downlink CSI fusion precoding network that jointly models downlink CSI reference signal (CSI-RS) design, CSI feedback, and base-station (BS) precoding within a single E2E neural architecture. Concretely, a projection network built on the MAXIM architecture takes uplink sounding reference signals (SRS) as input and outputs frequency-, beam-, and port-domain projection matrices for designing downlink CSI-RS. User equipment (UE) then compresses/quantizes the resulting CSI-RS observations and feeds back a compact representation. At the base station (BS), two complementary branches produce candidate precoders: one is a feedback-only precoding network driven by quantized downlink observations, and the other is an SRS-only precoding network driven by uplink SRS. These candidate precoders are subsequently combined by a fusion precoding network to yield the final transmit precoder. All the modules are trained with a spectral-efficiency-oriented loss under a three-stage schedule. Simulation results show that the proposed approach effectively harnesses both SRS-derived information and UE feedback, achieving markedly better performance than conventional baselines.

Problem

• Optimizes multimodal semantic non-orthogonal transmission and fusion in massive MIMO systems.
• End-to-end trains cross-modal networks for joint physical and application-layer semantic processing.
• Enhances spectral efficiency and semantic task performance over traditional separated designs.

Innovation

• Transformer-based cross-modal semantic-aware network for joint optimization
• End-to-end training integrates physical and application layer tasks
• Five subnetworks enable multimodal semantic fusion and transmission