An LDPC-coded orbital angular momentum shift-keying (OAM-SK) free space optical (FSO) communication system with a dual-pattern convolutional neural network (CNN) demodulator is put forward to resist the adverse effect of atmospheric turbulence (AT). Diverging from the standard single-pattern CNN demodulator, the proposed method inputs a dual OAM light pattern into the demodulator. The first pattern stems from the Laguerre-Gaussian OAM-SK light focused by a convex lens, while the second is acquired post-cylindrical lens modulation. The dual-pattern CNN demodulator can extract richer features from the incoming light, enhancing the recognition accuracy for OAM-SK modes. This advancement notably enables the recognition of OAM-SK modes with opposite orbital quantum numbers, a challenging task for standard single-pattern CNN demodulators. We have refined communication reliability by integrating LDPC coding with OAM-SK, and LDPC decoding has also been explicitly designed for OAM-SK channels. Simulations validate our proposed method, achieving a recognition accuracy 0.869 under strong AT (Cn2=5×10−14m−2/3). We use the image transmission as a benchmark; the dual-pattern CNN demodulator enhances the LDPC-coded OAM-SK FSO link, achieving a 30 dB boost in the image’s PSNR compared to the traditional CNN demodulation. The bit error rate (BER) drops to 3.8e−5, achieving significant advancement in comparison to the single-pattern CNN demodulators.
Read full abstract