Finite-State Uncertainty-Aware Widely-Linear Equalization for Aeronautical GMSK Links

Abstract

Aeronautical and unmanned aerial vehicle (UAV) receivers must equalize doubly selective channels under synchronization uncertainty and limited onboard computation and memory. We propose a finite-state, uncertainty-aware widely-linear (WL) equalization algorithm for Gaussian minimum-shift keying (GMSK) links. The algorithm uses synchronization estimates and their reliability descriptors to select one of a small number of precomputed filters, avoiding repeated online coefficient synthesis. Using a finite-memory Laurent model, we design the stored affine WL filters from conditional physical-state moments. We derive an exact characterization of excess mean-square error (MSE) relative to adaptation to continuous side information, Lloyd-type conditions for finite-state design, local timing--frequency mismatch certificates, and an orthogonal decomposition of design and information losses. On a reference channel at $E_b/N_0=8$~dB under high synchronization-estimation uncertainty, a $16$-state bank reduces pseudo-symbol MSE by $51.9\%$ relative to point-estimate WL time-varying MMSE equalization and by $6.6\%$ relative to posterior-mean (PM) substitution. With identical partitioning, routing, and storage, conditional-moment filters reduce MSE by $10.1\%$ and full-waveform average bit error rate (ABER) by $15.6\%$ relative to representative-state filters. Across $18$ synthetic three-path channels with individually fitted banks, MSE and ABER decrease by $5.3\%$ and $12.9\%$, respectively, relative to PM substitution. The $16$-state receiver stores $2048$ real coefficients. Posterior integration, robustness sweeps, and software timing measurements quantify the tradeoff between equalization accuracy, online computational cost, and bank refresh.