Figure 1.
CGAN model architecture.
Figure 1.
CGAN model architecture.
Figure 2.
Flowchart of the OFDM baseband communication system.
Figure 2.
Flowchart of the OFDM baseband communication system.
Figure 3.
Signal reconstruction model.
Figure 3.
Signal reconstruction model.
Figure 4.
Improved CGAN model architecture.
Figure 4.
Improved CGAN model architecture.
Figure 5.
Discriminator network model.
Figure 5.
Discriminator network model.
Figure 6.
Generator network model.
Figure 6.
Generator network model.
Figure 7.
Reconstruction results of the time-domain waveform map and constellation diagram when SNR was 20 dB. (Left to right modulation style is BPSK, QPSK, 16QAM). (a) Original signal time-domain waveform; (b) Reconstruction of signal time-domain waveform; (c) Original signal constellation diagram; (d) Reconstruction of signal constellation diagram.
Figure 7.
Reconstruction results of the time-domain waveform map and constellation diagram when SNR was 20 dB. (Left to right modulation style is BPSK, QPSK, 16QAM). (a) Original signal time-domain waveform; (b) Reconstruction of signal time-domain waveform; (c) Original signal constellation diagram; (d) Reconstruction of signal constellation diagram.
Figure 8.
Reconstruction results of time-domain waveform map and constellation diagram when SNR was 15 dB. (Left to right modulation style is BPSK, QPSK, 16QAM). (a) Original signal time-domain waveform; (b) Reconstruction of signal time-domain waveform; (c) Original signal constellation diagram; (d) Reconstruction of signal constellation diagram.
Figure 8.
Reconstruction results of time-domain waveform map and constellation diagram when SNR was 15 dB. (Left to right modulation style is BPSK, QPSK, 16QAM). (a) Original signal time-domain waveform; (b) Reconstruction of signal time-domain waveform; (c) Original signal constellation diagram; (d) Reconstruction of signal constellation diagram.
Figure 9.
Reconstruction results of time-domain waveform map and constellation diagram when SNR was 10 dB. (Left to right modulation style is BPSK, QPSK, 16QAM). (a) Original signal time-domain waveform; (b) Reconstruction of signal time-domain waveform; (c) Original signal constellation diagram; (d) Reconstruction of signal constellation diagram.
Figure 9.
Reconstruction results of time-domain waveform map and constellation diagram when SNR was 10 dB. (Left to right modulation style is BPSK, QPSK, 16QAM). (a) Original signal time-domain waveform; (b) Reconstruction of signal time-domain waveform; (c) Original signal constellation diagram; (d) Reconstruction of signal constellation diagram.
Figure 10.
Visual comparison of time-domain waveform of OFDM symbol sequence when the SNR was 20 dB. (Left to right modulation style is BPSK, QPSK, 16QAM). (a) Real part; (b) Imaginary part.
Figure 10.
Visual comparison of time-domain waveform of OFDM symbol sequence when the SNR was 20 dB. (Left to right modulation style is BPSK, QPSK, 16QAM). (a) Real part; (b) Imaginary part.
Figure 11.
Visual comparison of OFDM signal spectrogram when the SNR was 20 dB. (Left to right modulation style is BPSK, QPSK, 16QAM).
Figure 11.
Visual comparison of OFDM signal spectrogram when the SNR was 20 dB. (Left to right modulation style is BPSK, QPSK, 16QAM).
Figure 12.
Visual comparison of time-domain waveform of OFDM symbol sequence when the SNR was 15 dB. (Left to right modulation style is BPSK, QPSK, 16QAM). (a) Real part; (b) Imaginary part.
Figure 12.
Visual comparison of time-domain waveform of OFDM symbol sequence when the SNR was 15 dB. (Left to right modulation style is BPSK, QPSK, 16QAM). (a) Real part; (b) Imaginary part.
Figure 13.
Visual comparison of OFDM signal spectrogram when the SNR was 15 dB. (Left to right modulation style is BPSK, QPSK, 16QAM).
Figure 13.
Visual comparison of OFDM signal spectrogram when the SNR was 15 dB. (Left to right modulation style is BPSK, QPSK, 16QAM).
Figure 14.
Visual comparison of time-domain waveform of OFDM symbol sequence when the SNR was 10 dB. (Left to right modulation style is BPSK, QPSK, 16QAM). (a) Real part; (b) Imaginary part.
Figure 14.
Visual comparison of time-domain waveform of OFDM symbol sequence when the SNR was 10 dB. (Left to right modulation style is BPSK, QPSK, 16QAM). (a) Real part; (b) Imaginary part.
Figure 15.
Visual comparison of OFDM signal spectrogram when the SNR was 10 dB. (Left to right modulation style is BPSK, QPSK, 16QAM).
Figure 15.
Visual comparison of OFDM signal spectrogram when the SNR was 10 dB. (Left to right modulation style is BPSK, QPSK, 16QAM).
Figure 16.
Probability density distribution when the SNR was 20 dB. (Left to right modulation style is BPSK, QPSK, 16QAM). (a) Real part; (b) Imaginary part.
Figure 16.
Probability density distribution when the SNR was 20 dB. (Left to right modulation style is BPSK, QPSK, 16QAM). (a) Real part; (b) Imaginary part.
Figure 17.
Probability density distribution when the SNR was 15 dB. (Left to right modulation style is BPSK, QPSK, 16QAM). (a) Real part; (b) Imaginary part.
Figure 17.
Probability density distribution when the SNR was 15 dB. (Left to right modulation style is BPSK, QPSK, 16QAM). (a) Real part; (b) Imaginary part.
Figure 18.
Probability density distribution when the SNR was 10 dB. (Left to right modulation style is BPSK, QPSK, 16QAM). (a) Real part; (b) Imaginary part.
Figure 18.
Probability density distribution when the SNR was 10 dB. (Left to right modulation style is BPSK, QPSK, 16QAM). (a) Real part; (b) Imaginary part.
Table 1.
Dataset information.
Table 1.
Dataset information.
Details |
---|
Modulation mode | 16QAM, QPSK, BPSK |
Sample length | 256 |
SNR range | (10 dB, 15 dB, 20 dB) |
Signal sample size | 18,000 |
Table 2.
Network structure of the first discriminator.
Table 2.
Network structure of the first discriminator.
Types | Paraments | Output Size |
---|
IQ + AP | - | 4 × 256 |
BiLSTM | Num_layers = 1 | 256 × Embed_size |
Positional_Encoding | - | 256 × Embed_size |
Table 3.
Network structure of the second discriminator.
Table 3.
Network structure of the second discriminator.
Types | Size/Step | Output Size |
---|
2D Convolution | 2 × 2/2 | 128 × Embed_size |
Positional_Encoding | - | 128 × Embed_size |
Table 4.
Network structure of the discriminator Transformer encoder layer.
Table 4.
Network structure of the discriminator Transformer encoder layer.
Types | Paraments | Output Size |
---|
BatchNorm 1D | Embed_size | 256/128 × Embed_size |
Multi-Head Attention | Num_heads = 16 | 256/128 × Embed_size |
Dropout | Probability = 0.3 | 256/128 × Embed_size |
Layer Normalization | Embed_size | 256/128 × Embed_size |
Linear | Expansion = 4 | 256/128 × 4 × Embed_size |
GELU | - | 256/128 × 4 × Embed_size |
Linear | Expansion = 4 | 256/128 × Embed_size |
Dropout | Probability = 0.3 | 256/128 × Embed_size |
Table 5.
Network structure of the discriminant module.
Table 5.
Network structure of the discriminant module.
Types | Paraments | Output Size |
---|
Reduce | Reduction = mean | Embed_size |
Layer Normalization | Embed_size | Embed_size |
Linear | - | 1 |
Table 6.
Network structure of embedding.
Table 6.
Network structure of embedding.
Types | Paraments | Output Size |
---|
Linear | Laten_dim | 2 × 256 × Embed_size |
Positionl_Encoding | - | 2 × 256 × Embed_size |
Table 7.
Network structure of the generator Transformer encoder layer.
Table 7.
Network structure of the generator Transformer encoder layer.
Types | Paraments | Output Size |
---|
BatchNorm 1D | Embed_size | 512 × Embed_size |
Prob Attention | Num_heads = 16 | 512 × Embed_size |
Layer Normalization | Embed_size | 128 × Embed_size |
Dropout | Probability = 0.3 | 512 × Embed_size |
Linear | Expansion = 4 | 512 × 4 × Embed_size |
GELU |
-
| 512 × 4 × Embed_size |
Linear | Expansion = 4 | 512 × 4 × Embed_size |
Dropout | Probability = 0.3 | 512 × Embed_size |
2D Convolution | Kernel_size = 1 | 2 × 256 |
Table 8.
Network model training parameters.
Table 8.
Network model training parameters.
Model Hyperparameters |
---|
Generator Embed_dim | 160 |
Discriminator Embed_dim | 160 |
LSTM Hiddem_dim | 320 |
Batch size | 64 |
Epochs | 75 |
Learning rate | 0.001 |
β1, β2 | 0.9, 0.999 |
Optimizer | Adam |
Table 9.
Hardware and software environments.
Table 9.
Hardware and software environments.
Environment | Technical Parameters |
---|
OS | Windows 10 |
CPU | Intel Xeon Silver 4212R |
GPU | NVIDIA GeForce 4090 |
Memory | 128 G |
Python | Python 3.8.8 |
Pytorch | Pytorch 1.8.1 |
Table 10.
Comparison of model parameters.
Table 10.
Comparison of model parameters.
Model Name | Generator | Discriminator |
---|
Parameters /B | Time Complexity /FLOPs | Parameters /B | Time Complexity /FLOPs |
---|
LSTM&Transformer Based CGAN | 1.65 × 107 | 3.42 × 108 | 4.85 × 106 | 1.97 × 108 |
TOR-GAN [15] | 1.69 × 107 | 3.45 × 108 | 5.37 × 106 | 1.82 × 108 |
Pattern-Constellation dual GAN [14] | 1.65 × 107 | 1.78 × 108 | 2.04 × 107 | 2.34 × 108 |
DRAGAN [13] | 1.43 × 106 | 8.53 × 107 | 6.99 × 107 | 1.93 × 109 |
Table 11.
Reconstruction similarity evaluation.
Table 11.
Reconstruction similarity evaluation.
Model Name | Similarity Analysis | BPSK | QPSK | 16QAM |
---|
10 dB | 15 dB | 20 dB | 10 dB | 15 dB | 20 dB | 10 dB | 15 dB | 20 dB |
---|
LSTM&Transformer Based CGAN | MSE | 0.3715 | 0.3129 | 0.1303 | 0.3008 | 0.2960 | 0.1520 | 0.3903 | 0.3224 | 0.2140 |
MAE | 0.3606 | 0.2731 | 0.1096 | 0.2975 | 0.2608 | 0.1403 | 0.4120 | 0.3055 | 0.1948 |
EVM | 1.1057 | 0.9635 | 0.7202 | 1.0359 | 0.9986 | 0.8134 | 1.3538 | 1.1489 | 1.0673 |
TOR-GAN | MSE | 0.3970 | 0.3083 | 0.1460 | 0.4307 | 0.3086 | 0.1981 | 0.4624 | 0.3531 | 0.2529 |
MAE | 0.3413 | 0.2849 | 0.1284 | 0.4142 | 0.2981 | 0.1660 | 0.4546 | 0.2164 | 0.2502 |
EVM | 1.1395 | 0.9588 | 0.7911 | 1.2415 | 0.9935 | 0.8303 | 1.4145 | 0.9303 | 0.9137 |
DRAGAN | MSE | 0.7051 | 0.5319 | 0.4197 | 0.6424 | 0.5740 | 0.4200 | 0.7135 | 0.6501 | 0.5265 |
MAE | 0.6376 | 0.4945 | 0.4086 | 0.5359 | 0.5542 | 0.3447 | 0.6995 | 0.5914 | 0.5209 |
EVM | 2.7872 | 1.5969 | 1.2761 | 2.4858 | 1.3715 | 1.1097 | 2.5895 | 1.4049 | 1.2388 |