Security-Aware Spectral Amplitude Coding for High-Speed OCDMA Networks: Full-Resolution SSFM Validation and Direct Operating-Boundary Optimization
DOI:
https://doi.org/10.70917/ijcisim-2026-4305Keywords:
SAC-OCDMA, physical-layer security, information leakage rate, split-step Fourier method, fiber nonlinearities, region-aware calibration, direct boundary validation, power optimizationAbstract
Security in spectral-amplitude-coding optical code-division multiple access (SAC-OCDMA) is often inferred from code obscurity, even though a code that suppresses multiuser interference may still be learnable by an unauthorized receiver. We examined this issue through a security-aware SAC (SA-SAC) link in which the wavelength assignment changes with a secret key and time index while the users retain disjoint spectral supports. The reference system used L = 64, K = 16, w = 4, and 10 Gb/s per user. Its analytical Bob–Eve model was checked against a 27-point, full-resolution split-step Fourier method (SSFM) grid covering three fiber conditions, three distances, and launch powers of −6, 4, and 12 dBm/user. Every point was rerun with dz ≤ 0.125 km, 64 bits, and spectral oversampling of 3, the receiver also compensated the wavelength-dependent chip delays before detection. The 12 dBm/user rows behaved as a separate nonlinear-stress regime and were not used to fit the operating-region correction.
For the remaining 18 rows, ridge calibration reduced Q-RMSE from 10.421 to 2.5624 and changed Q − R² from −0.3904 to 0.9159. A quadratic-ridge model (λ = 0.001) gave a leave-one-distance-out Q-RMSE of 4.3427, but its uncertainty interval was too broad for a defensible boundary claim. We therefore used the model only to choose 14 additional distance cases and six power cases for direct SSFM evaluation. With QSSFM ≥ 6 and ILR ≤ 0.05 bit/symbol, the last tested feasible points were 65 km at 4 dBm/user for standard SMF, 50 km at 3 dBm/user for dense WDM, and 70 km at 4 dBm/user for the low-dispersion case. Their Q-factors were 6.3786, 6.2473, and 6.8981, and the modeled ILR was zero in each case. The MATLAB workflow was checked by 132 unit and integration tests. The main outcome is a directly verified set of operating brackets, together with an explicit distinction between the region that was calibrated and the high-power region in which the analytical approximation failed.
