mokka.e2e

Module with blocks for end-to-end simulations.

Module with blocks for end-to-end simulations in PyTorch.

class mokka.e2e.torch.BitwiseAutoEncoder(m, demapper_width=127, demapper_depth=3, channel=None, mod_extra_params=(0,), demod_extra_params=(0,), mapper=None, demapper=None)

Bases: Module

Bitwise Auto-Encoder implementation.

Parameters:

• m – bits per symbol and input and output width of the Autoencoder

• demapper_width – Width of the hidden layers in the demapper

• demapper_depth – Number of hidden layers in the demapper

__init__(m, demapper_width=127, demapper_depth=3, channel=None, mod_extra_params=(0,), demod_extra_params=(0,), mapper=None, demapper=None)

Construct BitwiseAutoEncoder.

forward(b, *args)

Perform end-to-end simulation with the BitWiseAutoEncoder.

Take input-bits and output LLRs, then outside of the auto-encoder the loss can be calculated.

Parameters:

b – tensor of bit-strings.

Returns:

tensor of LLRs.

static load_model(model_dict)

Load model weights from a dictionary.

Parameters:

model_dict – model weights from a file, opened with torch.load.

Returns:

BitwiseAutoEncoder model initialized with saved weights.