kaira.models.DeepJSCCModel

Inheritance diagram of DeepJSCCModel — Inheritance diagram for DeepJSCCModel

class kaira.models.DeepJSCCModel(encoder: BaseModel, constraint: BaseConstraint, channel: BaseChannel, decoder: BaseModel, *args: Any, **kwargs: Any)[source]

Bases: SequentialModel

Deep Joint Source-Channel Coding model.

This model implements end-to-end joint source-channel coding using deep neural networks. It consists of: - An encoder that compresses and encodes the source signal - A power constraint that normalizes the encoded signal - A channel that simulates the transmission medium - A decoder that reconstructs the original signal

Key features: - End-to-end differentiable architecture - Automatic adaptation to channel conditions - No separate source/channel coding - Graceful degradation with channel quality - Support for various source types (images, audio, etc.)

Example

>>> # Create a DeepJSCC model for image transmission
>>> model = DeepJSCCModel(
...     encoder=image_encoder,
...     constraint=power_constraint,
...     channel=awgn_channel,
...     decoder=image_decoder
... )
>>> # Transmit image through noisy channel
>>> received = model(image, snr=10.0)

encoder

Neural network that compresses the input

Type:: BaseModel

constraint

Module that applies power constraints to the encoded signal

Type:: BaseConstraint

channel

Simulates the communication channel effects

Type:: BaseChannel

decoder

Neural network that reconstructs the input from the received signal

Type:: BaseModel

Methods

`__init__`	Initialize the DeepJSCC model.
`add_step`	Add a processing step to the model.
`forward`	Execute the model sequentially on the input data.
`remove_step`	Remove a processing step from the model.

__init__(encoder: BaseModel, constraint: BaseConstraint, channel: BaseChannel, decoder: BaseModel, *args: Any, **kwargs: Any)[source]

Initialize the DeepJSCC model.

Parameters:

encoder (BaseModel) – Neural network model for encoding/compressing the input
constraint (BaseConstraint) – Module for applying power constraints to the encoded signal
channel (BaseChannel) – Module simulating the communication channel
decoder (BaseModel) – Neural network model for decoding/reconstructing the input
*args – Variable positional arguments passed to the base class.
**kwargs – Variable keyword arguments passed to the base class.

add_step(step: Callable) → ConfigurableModel

Add a processing step to the model.

Parameters:: step – A callable that will be added to the processing pipeline. Must accept and return tensor-like objects.
Returns:: Self for method chaining

forward(input_data: Any, *args: Any, **kwargs: Any) → Any

Execute the model sequentially on the input data.

Parameters:

input_data – The initial data to process
*args – Additional positional arguments passed to each step.
**kwargs – Additional keyword arguments passed to each step.

Returns:

The final result after passing through all steps

remove_step(index: int) → ConfigurableModel

Remove a processing step from the model.

Parameters:: index – The index of the step to remove
Returns:: Self for method chaining
Raises:: IndexError – If the index is out of range