opt.n_layers_D = 3
opt.label_smooth = 1 # Label smoothing factor (for lsgan and vanilla gan only)
opt.n_downsample = 3 # Downsample times
opt.n_blocks = 2 # Numebr of residual blocks
opt.first_kernel = 5 # The filter size of the first convolutional layer in encoder
opt.batchSize = 64
opt.n_epochs = 30 # # of epochs without lr decay
opt.n_epochs_decay = 30 # # of epochs with lr decay
opt.lr_policy = 'linear' # decay policy. Availability: see options/train_options.py
opt.beta1 = 0.5 # parameter for ADAM
opt.lr = 5e-4
############################ Things recommanded to be changed ##########################################
# Set up the training procedure
opt.C_channel = 12
opt.SNR = 20
opt.is_feedback = False
opt.feedforward = 'EXPLICIT-RES'
opt.N_pilot = 2 # Number of pilots for chanenl estimation
opt.CE = 'MMSE' # Channel Estimation Method
opt.EQ = 'MMSE' # Equalization Method
opt.pilot = 'ZadoffChu' # QPSK or ZadoffChu
opt.is_clip = False
opt.CR = 0 if not opt.is_clip else 1
opt.is_regu_PAPR = True
opt.lam_PAPR = 0.5
opt.is_regu_sigma = False
opt.lam_sigma = 100
opt.crop_size = 64
dataset = create_dataset(opt) # create a dataset given opt.dataset_mode and other options
dataset_size = len(dataset)
print('#training images = %d' % dataset_size)
else:
raise Exception('Not implemented yet')
if opt.channel == 'bsc':
opt.ber = 0.1 # Set the bit flip rate for bsc channel
opt.enc_type = 'hard'
opt.sample_type = 'gumbel_softmax_hard'
opt.temp = 0.5
channel_name = '_BER'+str(opt.ber)+'_'+opt.enc_type+'_'+opt.sample_type
elif opt.channel == 'awgn':
opt.SNR = 20 # Set the SNR for awgn channel
channel_name = '_SNR'+str(opt.SNR)
elif opt.channel == 'ofdm':
pass
else:
raise Exception('Not implemented yet')
# Display setting
opt.checkpoints_dir = './Checkpoints/'+ opt.dataset_mode + '_' + opt.channel
opt.name = opt.gan_mode + '_C' + str(opt.C_channel) + channel_name
opt.display_env = opt.dataset_mode + '_' + opt.channel + '_' + opt.name
# Choose the neural network model
opt.model = 'StoGAN'
opt.label_smooth = 1 # Label smoothing factor (for lsgan and vanilla gan only)
opt.n_downsample = 3 # Downsample times
opt.n_blocks = 4 # Numebr of residual blocks
opt.first_kernel = 5 # The filter size of the first convolutional layer in encoder
opt.batch_size = 16
opt.n_epochs = 20 # # of epochs without lr decay
opt.n_epochs_decay = 20 # # of epochs with lr decay
opt.lr_policy = 'linear' # decay policy. Availability: see options/train_options.py
opt.beta1 = 0.5 # parameter for ADAM
opt.lr = 5e-4
############################ Things recommanded to be changed ##########################################
# Set up the training procedure
opt.C_channel = 24
opt.SNR = 5
opt.is_feedback = False
opt.feedforward = 'EXPLICIT-RES'
opt.N_pilot = 1 # Number of pilots for chanenl estimation
opt.CE = 'MMSE' # Channel Estimation Method
opt.EQ = 'MMSE' # Equalization Method
opt.pilot = 'ZadoffChu' # QPSK or ZadoffChu
opt.is_clip = False
opt.CR = 0 if not opt.is_clip else 1.2
opt.is_regu_PAPR = False
opt.lam_PAPR = 0.1
opt.lam_h = 0.5
opt.first_kernel = 5 # The filter size of the first convolutional layer in encoder
# Set the input dataset
opt.dataset_mode = 'CIFAR10' # Current dataset: CIFAR10, CelebA
# Set up the training procedure
opt.batchSize = 64 # batch size
opt.n_epochs = 200 # # of epochs without lr decay
opt.n_epochs_decay = 200 # # of epochs with lr decay
opt.lr = 1e-3 # Initial learning rate
opt.lr_policy = 'linear' # decay policy. Availability: see options/train_options.py
opt.beta1 = 0.5 # parameter for ADAM
opt.L = 8
opt.decay = 4
opt.SNR = 0
# Set up the loss function
opt.lambda_L2 = 128 # The weight for L2 loss
opt.is_Feat = False # Whether to use feature matching loss or not
opt.lambda_feat = 1
##############################################################################################################
if opt.gan_mode == 'wgangp':
opt.norm_D = 'instance' # Use instance normalization when using WGAN. Available: 'instance', 'batch', 'none'
else:
opt.norm_D = 'batch' # Used batch normalization otherwise
opt.activation = 'sigmoid' # The output activation function at the last layer in the decoder
opt.norm_EG = 'batch'
