Fixes related to theoretical bounds

• Fix REP bound equation
• Use real simulated SNR from Planewalker for plotting
• Capture Planewalker output of max code energy

Comparisons to theoretical bounds

In this version, some theoretical background of performance bounds was added. The results obtained with the NN are compared to the bounds. In addition, a short listing of future development ideas was initiated in this release.

Training plots

At the end of the training process, average values (BLER and losses) of episodes are plotted.

Fix errors

This is a bug-squashing version. Actor loss gradients are no longer backpropagated through critic NN. Some other smaller fixes are also included.

Actor NN improvements

This version has a few major improvements.

• Use actor NN to learn the best action distribution covariance matrix.
• Backpropagate policy loss gradients also through critic network.
• Add a linear and activation layer to the common part of actor network.

Introduce plots

In this version, tracking training activity by BLER and loss plots is introduced.

Critic NN improvements

Added SNR as input to critic neural network to improve accuracy of value function estimation.

Improvements and mathematical descriptions

This version has only minor improvements to the model, but more documentation added.

• Critic loss is calculated using a distribution object
• Description about the mathematical background of the reinforcement learning task

Lattice matrix normalization

This version fixes the exploding base matrices by normalizing them during training. Also, some code style improvements are included.

More key features

In this version, the following features are added:

• Complex number support for lattice matrices
• Using different SNR for different learning episodes

First "release version" of this project

The working features include:

• A reinforcement learning model with actor-critic paradigm
• BLER calculation with Planewalker simulator
• Controlling Planewalker by giving it lattice base matrices and other settings