会議発表論文

会議発表論文

text

IEEE International Conference on Communications. 2026

Discrete-Valued Signal Estimation via GAMP under Finite-Sized Highly Correlated GLMs

Institute of Electrical and Electronics Engineers Inc.

英語

text

2026-07-14

https://doi.org/10.1109/ICC59461.2026.11587545

Furudoi T., Takahashi T., Ochiai H.. Discrete-Valued Signal Estimation via GAMP under Finite-Sized Highly Correlated GLMs. Proceedings of the IEEE International Conference on Communications (ICC 2026); https://doi.org/10.1109/ICC59461.2026.11587545.

This paper proposes a simple yet powerful strategy for generalized approximate message passing (GAMP) to achieve robust discrete-valued signal estimation under generalized linear models (GLMs) with model mismatches. Despite its minimal computational complexity, the design principle of GAMP critically relies on large-scale independent and identically distributed (i.i.d.) measurements, which are often violated in practice. The resulting statistical deviations in the propagating beliefs lead to severe performance degradation, particularly when discrete constraints on the unknown signal induce nonlinear denoising behavior. To address this issue, our strategy adaptively manipulates this nonlinearity through an embedded inverse temperature mechanism, effectively absorbing analytically intractable model mismatches. Numerical simulations demonstrate that, despite the combined challenges of GLM-induced belief distortions, finite-size effects, and highly correlated measurements, the proposed scheme robustly guides GAMP to approach the absolute performance limit while retaining its minimal computational complexity. Furthermore, we elucidate the underlying operating principle by analyzing the evolution of the message dynamics.

© 2026 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works.

106088

Discrete-Valued Signal Estimation via GAMP under Finite-Sized Highly Correlated GLMs

Institute of Electrical and Electronics Engineers Inc.

英語

2026-07-14

IEEE International Conference on Communications

https://doi.org/10.1109/ICC59461.2026.11587545

https://hdl.handle.net/11094/106088