Would Learning Help? Adaptive CRC-QC-LDPC Selection for Integrity in 5G-NR V2X
This addresses physical-layer integrity for vehicle-to-everything communications, but it is incremental as it applies a known contextual bandit method to a specific configuration problem.
The paper tackled the problem of reducing silent errors in 5G-NR V2X communications by adaptively selecting CRC polynomials and QC-LDPC coding rates, finding that learning-assisted adaptation reduces undetected error probability by up to 50-70% at low to moderate mobility in low-SNR regimes.
Vehicle-to-everything (V2X) communications impose stringent physical-layer integrity requirements, particularly under short-packet transmission and mobility-induced channel variation. This paper studies whether standard-compliant online selection of Cyclic Redundancy Check (CRC) polynomials and Quasi-Cyclic Low-Density Parity-Check (QC-LDPC) coding rates can reduce silent (undetected) errors in 5G New Radio (5G-NR) V2X links. The joint configuration problem is formulated as a lightweight Contextual Bandit (CB) with a small, discrete action space, and a discounted LinUCB policy is evaluated against greedy online adaptation and a conservative fixed baseline. A 5G-NR-compliant physical-layer simulation is developed using Sionna, modeling mobility through time-correlated Rayleigh fading, where vehicle speed governs channel correlation, and non-stationary interference via a two-state Markov process. The learning agent operates on coarse receiver feedback, including a noisy Signal-to-Noise Ratio (SNR) estimate and indicators of burst interference and deep fades, and targets minimization of the Undetected Error Probability ((P{UE})) while accounting for the Detected Error Probability ((P{DE})). Overall, our objective is to delineate the mobility regimes in which learning-assisted CRC-QC-LDPC configuration improves physical-layer integrity in 5G-NR V2X systems. Our results indicate that learning-assisted adaptation is most effective at low to moderate mobility, reducing (P_UE) by up to 50-70% relative to greedy selection in the low-SNR regime ((-5) to 5~dB) and approaching the best fixed configuration at higher (E_b/N_0). At high mobility (>= 180~km/h), fast channel decorrelation weakens temporal predictability, limiting the effectiveness of online learning and reducing performance differences across policies.