Fig. 1. Flow chart of the proposed scheme

Fig. 2. Structure of trainable BPS

Fig. 3. CReLU, photonic Sigmoid, and standard Sigmoid function curves

Fig. 4. Hidden layer outputs of Rx-NN with different activation functions. (a) ReLU function; (b) photonic Sigmoid function

Fig. 5. Schematic of simulation system

Fig. 6. Results of proposed GCS scheme under different conditions. (a) RSN=17 dB, Δv=100 kHz; (b) RSN=13 dB, Δv=600 kHz

Fig. 7. GMI curves of various shemes at different laser linewidths

Fig. 8. BER curves of QAM, proposed Geo-E2E (with CD), and BTB at different OSNRs

Fig. 9. GMI curves of trainable BPS and regular BPS at different linewidths

Fig. 10. At 1200 km, Q-factor curves of 64QAM and proposed GeoPCS-E2E (with CD) at different input powers

Fig. 11. GMI curves of various schemes at 800 km, 1040 km, and 1200 km under different laser linewidths. (a) 800 km; (b) 1040 km; (c) 1200 km

Fig. 12. Training epochs and GMI performance for systems with different numbers of hidden layers

Fig. 13. Analysis of key parameters for photonic Sigmoid functions. (a) Analysis of A₁ and A₂; (b) analysis of d

Fig. 14. GMI performance of different models