kaira.models.image.Xie2023DTDeepJSCCDecoder

Inheritance diagram for Xie2023DTDeepJSCCDecoder

class kaira.models.image.Xie2023DTDeepJSCCDecoder(latent_channels, out_classes, architecture='cifar10', num_embeddings=None, *args, **kwargs)

Bases: BaseModel

Discrete Task-Oriented Deep JSCC decoder.

This implements the decoder part of the DT-DeepJSCC architecture as described in [Xie et al., 2023]. It maps discrete latent representations back to class predictions.

Parameters:

• architecture (str, optional) – Type of architecture to use. Defaults to ‘cifar10’. Options: ‘cifar10’ or ‘custom’.

• latent_channels (int) – Number of channels in the latent representation

• out_classes (int) – Number of output classes

• num_embeddings (int, optional) – Size of the discrete codebook. Defaults to None (automatically determined by architecture).

References

[Xie et al., 2023]

Methods

__init__
forward	Forward pass for the DT-DeepJSCC decoder.

__init__(latent_channels, out_classes, architecture='cifar10', num_embeddings=None, *args, **kwargs)

forward(received_bits)

Forward pass for the DT-DeepJSCC decoder.

Parameters:

received_bits (torch.Tensor) – Received bits from the channel [batch_size, h*w, bits_per_symbol]

Returns:

Class logits [batch_size, out_classes]

Return type:

torch.Tensor