Learning-to-Transition for Large-scale and High-Order MIMO Detection
2026-08-14 • Information Theory
Information TheoryArtificial Intelligence
AI summaryⓘ
The authors address the challenge of detecting signals in complex wireless communications using multiple antennas (MIMO). They propose a method called learning-to-transition (L2T), which treats signal detection as a step-by-step process and uses a Transformer model to improve accuracy while keeping computation manageable. Their approach first learns to detect signals reliably and then adapts to support soft-output detection, integrating well with error-correcting decoders. This method is trained in stages to ensure stable performance during detection and decoding.
MIMO detectionTransformersoft-output detectionLDPC decodingautoregressive factorizationiterative detection and decoding (IDD)sampling policyresidual-to-BER curriculumBayes' rule
Authors
Yubo Zhang, Yiyao Liu, Xiaodong Wang
Abstract
High-order multiple-input multiple-output (MIMO) detection requires efficient search over a large discrete symbol space while producing reliable soft information for channel decoding. This paper develops a learning-to-transition (L2T) framework that formulates MIMO detection as a stochastic sequence of complete-vector transitions. At each transition, a channel-coupled Transformer updates both the instance embedding and the sampling policy, while a blockwise autoregressive factorization captures inter-stream dependence with moderate sequential complexity. For hard-output detection, a transition network is applied recursively and trained through a residual-to-BER curriculum, which first learns the MIMO search geometry from the exact residual metric and then aligns the policy with transmitted-bit accuracy. For soft-output reception, the well-trained hard policy is cloned at the parameter level into every layer of an untied soft-input soft-output iterative detection and decoding (IDD) receiver. This tied-to-untied transfer preserves the learned zero-prior search dynamics while enabling layer- and round-specific specialization under decoder feedback. Within each IDD round, decoder priors tilt candidate generation according to Bayes' rule, and likelihood-weighted terminal hypotheses produce posterior and extrinsic log-likelihood ratios for LDPC decoding. A multi-stage training strategy further stabilizes the hard-to-soft transfer by progressively exposing the receiver to synthetic and in-loop decoder-generated priors.