Fig. 1. LSTM neural network model assisted FSO system model for polarization code decoding
Fig. 2. Basic unit of the LSTM network
Fig. 3. LSTM network prediction model framework
Fig. 4. Execution flow chart of LSTM-SCFlips decoding method
Fig. 5. Root mean square error distribution in a weak turbulence channel
Fig. 6. Training time in weak turbulence channel
Fig. 7. Performance of LSTM network in weak turbulence channel
Fig. 8. Correct recognition rate of optimal flip position in weak turbulence channel (σ0=0.2)
Fig. 9. Performance comparison chart of different bit flipping decoding schemes under weak turbulence channel (σ0=0.2)
Fig. 10. Comparison of decoding performance under different turbulence intensities
Parameter | Value |
---|
Length of polar code | 1024/2048 | Code rate | 0.25/0.5/0.75 | Turbulence intensity | 0.1/0.2/0.3 | Number of CRC bits | 8 | Wavelength /nm | 1550 | Modulation | 4-PPM | Transmission distance /km | 2 |
|
Table 1. Simulation parameters
Model | i'=5 | i'=6 | i'=7 | i'=8 | i'=9 | i'=10 |
---|
LSTM | 0.0893 | 0.0861 | 0.0842 | 0.0841 | 0.841 | 0.0838 | RNN | 0.1263 | 0.1237 | 0.1218 | 0.1217 | 0.1215 | 0.1215 | SVM | 0.1151 | 0.1126 | 0.1108 | 0.1104 | 0.1105 | 0.1107 |
|
Table 2. Comparison of RMSE of different models under weak turbulence channel (σ0=0.2)
Method | RSN=1 dB | RSN=1.5 dB | RSN=2 dB | RSN=2.5 dB | RSN=3 dB |
---|
D-SCFlips | 0.078 | 0.053 | 0.040 | 0.037 | 0.032 | LSTM-SCFlips | 0.052 | 0.046 | 0.038 | 0.035 | 0.031 | SCFlips | 0.048 | 0.039 | 0.033 | 0.031 | 0.029 |
|
Table 3. Average decoding time of each frame in different polarization code bit flipping decoding methodsunit: s