Sionna

Features

- Support for optical fiber simulations
- Split step Fourier method (SSFM)
- Erbium doped fiber amplifier
- New tutorial notebook on optical fiber simulations
- Refactorization of the channel documentation
- BCJR decoder for convolutional codes
- Support for Turbo codes
- Turbo encoding layer
- Turbo decoding layer
- New interleaver layer supporting the LTE Turbo code interleaver patterns (cf. 36.212)
- Utility functions for code generation and for puncturing

- Improved scrambling layer
- Support for custom scrambling sequences
- Utility function to support scrambling sequences as defined in for 38.211
- Adds LDPC rate-matching output interleaver to LDPC5GEncoder and LDPC5GDecoder
- Utility function to generate random regular LDPC parity-check matrices
- Zero-forcing (ZF) and matched filter (MF) MIMO equalizers
- Tensor utility function for pseudo inverse computation
- Several MIMO utility functions for channel whitening and complex-to-real valued representations
- (OFDM) MIMO maximum likelihood detector with prior
- Utility layer for computing logits on constellation points from LLRs
- New tutorial on using the DeepMIMO dataset with Sionna
- Restructured website
- Made with Sionna section
- Direct links to Github discussion/issues

Breaking Changes

- In LDPCBPDecoder and LDPC5GDecoder, `keep_state` has been replaced by `stateful` flag and the decoder now returns the states instead of keeping an internal reference
- TDL and CDL power delay profiles are now normalized to have unit total power (related to issue 12).

Fixes

- (38) Fixes issue with PanelArray()
- (43) Fixes issue with MLDetector
- Adds support for TensorFlow 2.10
- Bug fix in SymbolLogits2Moments
- Bug fix in mapping module related to double precision
- Adds missing reference of generate_dense_polar in `sionna.fec.polar.__init__`
- Improves stability of graph mode in Polar SCL decoding
- Issue of LDPC decoding when graph is not pruned
- Fixes missing attribute in Linear Encoder
- Improved selection of polynomials for convolutional codes
- Removes deprecated items in pylintrc due to updated version of Lynter
- Typos in the documentation
- Corrects the LoS path power calculation oft the 3GPP 38.901 channel models
- Fixes a bug in the effective noise variance computation of the lmmse_equalizer

New features

- Universal FEC encoder for binary linear codes
- Utility functions for FEC to
- Convert parity-check matrix to generator matrix
- Make generator matrix systematic
- Generate dense Polar parity-check and generator matrices
- Verify that generator and parity check matrix are orthogonal in GF(2)
- Linear interpolator for OFDM channel estimation (with optional time-averaging)
- MIMO maximum likelihood detector that computes either LLRs on bits or logits on constellation points
- SymbolSource layer for sampling arbitrary constellations
- PAMSource layer for sampling PAM constellations
- The SymbolSource, PAMSource, QAMSource, and Mapper optionally return the indices of the constellation points and/or the bit labels
- Utility layer to compute logits on constellation points to LLRs on bits (with optional priors)
- Utility layer to compute the mean and variance of a constellation given logits on the points
- Addition of integration tests

Fixes

- (25) XLA compatibility with TF2.9 for the BP decoder
- (29) SNR property in PlotBER
- Adds support for TF2.9
- Minor bug fixes
- Typos in the documentation

Fixes

- Drop support for TensorFlow 2.5 as it no longer receives updates
- Update minimum TensorFlow version requirements to address recent CVEs

Fixes

- Solves deprecated warning in utils.metrics (14)
- Issue related to explicit interleaver seed in graph mode
- Issues related to RessourceGridDemapper (19)
- Output shape of signal.utils.convolution()
- Corrects year of Gallager quote in channel coding example notebook
- Corrects Eb/No rate calculation in channel coding example notebook
- Typos in documentation
- Solves issue related to PanelArray.show() (23)
- Optimization of signal.utils.convolution() for complex-valued inputs

Features

New Signal Module

- Layers for up-/downsampling
- Layers for  (trainable) filters (e.g., pulse shaping)
- Layers for  (trainable) windowing functions
- Utility functions for convolution, (I)FFT, PSD, and ACLR calculation
- New example notebook on pulse-shaping

Fixes

- OFDMModulator issue
- Removes several warnings during unittests

Fixes

- OFDMDemodulator() issue for a cyclic_prefix equal to zero (closes github 6)
- Fixes issue with Jupyter notebook requesting a token
- Improve installation on macOS (closes github 2)
- Other small bug fixes and enhancements