Learning-to-Transition Framework for High-Order MIMO Detection
A new study shared on arXiv (2608.14511) introduces a framework called learning-to-transition (L2T) for advanced multiple-input multiple-output (MIMO) detection. It views MIMO detection as a stochastic sequence with complete-vector transitions, using a channel-coupled Transformer to refine the instance embedding and sampling policy with each step. To handle inter-stream dependence, it utilizes a blockwise autoregressive factorization, which helps keep sequential complexity manageable. For hard-output detection, a transition network is used and trained based on a residual-to-BER curriculum, focusing first on MIMO search geometry and then aligning with transmitted-bit accuracy. In soft-output scenarios, the established hard policy is applied uniformly across the layers of a soft-input soft-output decoder. This work seems geared for a conference presentation and aims to boost efficiency in MIMO detection, crucial for modern wireless systems.
Key facts
- Paper arXiv:2608.14511 introduces a learning-to-transition (L2T) framework for MIMO detection.
- The framework models MIMO detection as a stochastic sequence of complete-vector transitions.
- A channel-coupled Transformer updates instance embedding and sampling policy at each transition.
- Blockwise autoregressive factorization handles inter-stream dependence.
- Hard-output detection uses a recursive transition network trained with a residual-to-BER curriculum.
- Soft-output reception clones the hard policy into every layer of an untied soft-input soft-output decoder.
- The paper is a cross-type announcement on arXiv.
- The method targets high-order MIMO detection, improving search efficiency and soft information reliability.
Entities
Institutions
- arXiv