Risk-aware link adaptation architecture for Ka-band satellite leo downlinks under ionospheric disturbances

Abstract

Ka-band LEO downlinks can experience second-scale reliability degradations during flare-driven ionospheric disturbances. We present a global navigation satellite system-free, link-internal predictive controller that turns the downlink into its own estimator. A 10-Hz geometry-free dual-carrier phase observable is filtered and tracked with a four-state nearly-constant-velocity Kalman filter to estimate AVTEC and its rate; a 60-s look-ahead maps the state to a baseline-aligned endpoint-outage risk proxy P-out(t; H) that gates discrete one-step modulation and coding scheme down-switching and (optionally) pilot-time updates with hysteresis under an explicit risk threshold. Using a frozen-calibration replay protocol driven by one-minute geostationary operational environmental satellite (GOES) Xray sensor (XRS) flare profiles, all global thresholds and the shared block error rate (BLER)-margin scoring layer are fixed on disjoint logs; a short causal startup window (excluded from metrics) only sets a per-trace baseline offset for the risk proxy. Across held-out flare families, higher risk-score buckets consistently correspond to larger future shortfall and higher future BLER, with median event-level Spearman correlations of 0.80 and 0.97, respectively; under the shared frozen anchor, the controller reduces crest BLER relative to no adaptation while avoiding the extreme low-rate / high-switch behavior of a reactive ACM baseline. The controller runs in O(1) time on a single thread and takes 0.042 ms per 0.1 s epoch.