Skip to content

v2.1.0

Choose a tag to compare

@gmarcusm gmarcusm released this 09 Sep 11:36
· 1 commit to main since this release

Highlights

Sionna 2.1 adds calibrated 3GPP channel models, redesigned channel estimation, and improved PyTorch compatibility. Sionna RT now supports Mitsuba 3.9.1 and Dr.Jit 1.5.0, GPU acceleration on macOS through Metal, and deterministic path generation and radio-map diffraction. The release also includes many smaller bug fixes and improvements identified through an in-depth review of the codebase.

Sionna PHY

  • Updated and calibrated the 3GPP TR 38.901 channel models for V16.1 and V19.2, including Indoor Hotspot (InH) and Indoor Factory (InF) models, blockage, static spatial consistency, handheld user-terminal arrays, and calibration metrics.
  • (Breaking change) TR 38.901 V19.2 is now the default. Use spec_version="16.1" to select the V16.1 parameter tables.
  • Added fixed-delay TDL profiles from 3GPP TS 38.101-4 V19.2.2.
  • Fixed #1111: Corrected UMa environment-height sampling.
  • Fixed #1108: Corrected indoor-distance sampling.
  • Fixed #1107: Corrected a TR 38.901 ray offset.
  • Redesigned OFDM channel estimation and interpolation, and added LMMSEChannelEstimator and PUSCHLMMSEChannelEstimator.
  • (Breaking change) Removed LMMSEInterpolator, LMMSEInterpolator1D, and SpatialChannelFilter. Use LMMSEChannelEstimator instead. Custom pilot-domain estimators must now inherit from BasePilotChannelEstimator.
  • Reworked PolarSCLDecoder for efficient torch.compile execution and CUDA graph capture.
  • (Breaking change) Removed the use_hybrid_sc, use_fast_scl, cpu_only, and use_scatter arguments from PolarSCLDecoder.
  • Improved torch.compile compatibility, device placement, and numerical-precision handling throughout Sionna PHY.
  • Fixed #1149: Reduced peak GPU memory usage in channel conversion and precoding.
  • Fixed #1157: Corrected a channel frequency-response plot label (PR #1160).
  • Fixed #949: Clarified the noise-variance convention used by GaussianPriorSource.
  • Fixed LDPC rate matching and several issues related to 5G NR PUSCH, MCS selection, and transport-block sizing.
  • (Breaking change) MCSDecoderNR now uses transform_precoding=False by default, and high-level PUSCH configurations reject transform precoding.

Sionna SYS

  • Added 3GPP TR 38.901 topology generators and calibration metrics for multi-cell, indoor-office, and indoor-factory deployments.
  • Fixed issues in PHY abstraction, link adaptation, power control, hexagonal-grid updates, and path-loss association.
  • Adjusted link-adaptation settings in the End-to-End SYS tutorial to meet its BLER target.