Deep Variable-Length Feedback Codes
Yu Ding, Yulin Shao
TL;DR
DeepVLF introduces a neural, variable-length feedback coding framework that dynamically adapts transmission length through learned receiver feedback (DeepVLF-R) or transmitter-driven termination (DeepVLF-T). By partitioning information into bit groups and employing transformer-based encoders/decoders with threshold-based decisions, the approach achieves superior spectral efficiency and dramatically lower error floors, especially at high code rates, on both AWGN and 5G-NR fading channels. A key finding is the emergent two-phase encoding dynamic—an information-rich initial transmission followed by noise-cancellation refinements—resembling Schalkwijk–Kailath coding, which provides interpretability and alignment with information-theoretic principles. The paper also introduces a hybrid framework that blends per-group receiver termination with global transmitter verification, offering design guidance across reliable and noisy feedback regimes.
Abstract
Deep learning has enabled significant advances in feedback-based channel coding, yet existing learned schemes remain fundamentally limited: they employ fixed block lengths, suffer degraded performance at high rates, and cannot fully exploit the adaptive potential of feedback. This paper introduces Deep Variable-Length Feedback (DeepVLF) coding, a flexible coding framework that dynamically adjusts transmission length via learned feedback. We propose two complementary architectures: DeepVLF-R, where termination is receiver-driven, and DeepVLF-T, where the transmitter controls termination. Both architectures leverage bit-group partitioning and transformer-based encoder-decoder networks to enable fine-grained rate adaptation in response to feedback. Evaluations over AWGN and 5G-NR fading channels demonstrate that DeepVLF substantially outperforms state-of-the-art learned feedback codes. It achieves the same block error rate with 20%-55% fewer channel uses and lowers error floors by orders of magnitude, particularly in high-rate regimes. Encoding dynamics analysis further reveals that the models autonomously learn a two-phase strategy analogous to classical Schalkwijk-Kailath coding: an initial information-carrying phase followed by a noise-cancellation refinement phase. This emergent behavior underscores the interpretability and information-theoretic alignment of the learned codes.