Adaptive Channel Coding with Semantic-Aware Error Protection for 6G Non-Terrestrial Networks

Abstract Non-terrestrial networks (NTNs) — comprising low-Earth-orbit (LEO) satellites, high-altitude platform stations (HAPS), and unmanned aerial vehicles (UAVs) — are increasingly regarded as a foundational layer of sixth-generation (6G) wireless systems, extending connectivity to rural, maritime, and disaster-affected regions that remain beyond the economic reach of terrestrial infrastructure. NTN links, however, present a uniquely hostile channel environment: long and time-varying propagation delay, large Doppler shifts induced by high relative platform velocity, deep fades from atmospheric and rain attenuation, and severe constraints on onboard power and computation. Conventional fixed-rate or classically-adaptive channel coding schemes, which react only to instantaneous signal-to-noise ratio (SNR), are poorly matched to this environment: they either over-provision redundancy, wasting scarce bandwidth and energy, or under-provision it and trigger the so-called cliff effect, in which a small SNR drop causes catastrophic decoding failure. In this paper we propose semantic-aware adaptive channel coding (SA-ACC), a framework that couples semantic importance estimation with rate- and redundancy-adaptive coding so that error-protection resources are allocated according to the task-relevant meaning of transmitted content rather than uniformly across all bits. We formalize the system model for a representative NTN scenario — a LEO downlink relayed through a HAPS node — define a semantic importance metric derived from task-gradient sensitivity, and design a two-stage adaptation control loop that jointly selects code rate, modulation order, and per-feature protection level under explicit delay and energy budgets. We further specify a complete evaluation methodology, comparing SA-ACC against fixed-rate LDPC/polar coding, conventional adaptive coding and modulation (ACM), and non-semantic deep joint source-channel coding (JSCC) across bit error rate, block error rate, semantic fidelity, throughput, latency, and energy metrics, and use this methodology to reason about SA-ACC’s expected behavior ahead of full numerical validation. Our architectural analysis indicates that SA-ACC should be able to sustain task-relevant reconstruction quality at lower effective SNR than baselines optimized purely at the bit level, at the cost of additional encoder and decoder complexity that we discuss explicitly. We close by outlining the simulation and ablation study required to convert this design-and-methodology contribution into a fully validated result.