Adaptive conformal finite-blocklength link adaptation for burst-mode coherent passive optical networks

Abstract In upstream coherent passive optical networks (PONs), the challenge of coping with short bursts arises due to variations in received powers, carrier offsets, phase states, and digital signal processing residuals. While point-estimate adaptation may not provide calibrated reliability, fixed margins have the drawback of imposing constraints on power and spectral efficiency. To address this problem, the paper presents an approach called adaptive conformal finite block length (AC-FBLA), which consists of an optical line terminal controller that transforms recent signed residual estimates into the context-dependent and recency-weighted one-sided lower bound for burst SNR. Adaptive miscoverage feedback helps handle the issues of undercover, while the scheme uses a finite block length feasibility sieve for selecting the levels of launch power, modulation formats, and coding rates for each grant effectively. Using Monte Carlo simulations with 32 ONUs, 25 GBd transmissions, 1,024 symbol payload cases, six coded modes, four powers of launch, and four regimes of channel estimation enabled the teams to reach a high 97.67–97.77 % coverage compared to the target of 98 %. Therefore, one can see that the offered solution can improve the block error rate by 87.3–97.8 % compared to point estimate adaptation and reach improvements in spectral efficiency of 45.3–79.6 %.