Datasets#

Channel simulation dataset utilities for training and testing.

Overview#

The Datasets module provides tools for generating and managing channel simulation datasets using Sionna:

CDL Channel Generation - Generate 3GPP CDL channel models
Dataset Management - Setup and configure dataset pipelines
Configuration - Sionna CDL configuration utilities

API Reference#

Dataset utilities for channel simulation.

class ran.datasets.SionnaCDLConfig( num_sc_per_prb: int = 12, num_prb: int = 273, train_total: int = 20, test_total: int = 50, shard_size: int = 50, batch_tf: int = 256, prng_seed: int = 0, model_type: str = 'CDL', tdl_model: str = 'C', delay_spread_ns: float = 100.0, fc_ghz: float = 3.5, speed_min: float = 0.0, speed_max: float = 30.0, direction: str = 'uplink', bs_num_rows: int = 4, bs_num_cols: int = 1, bs_polarization: str = 'single', bs_polarization_type: str = 'V', bs_antenna_pattern: str = '38.901', ue_num_rows: int = 1, ue_num_cols: int = 1, ue_polarization: str = 'single', ue_polarization_type: str = 'V', ue_antenna_pattern: str = '38.901', )[source]#

Bases: object

Configuration class for CDL channel generation parameters.

Supports CDL (Clustered Delay Line) models with configurable antenna arrays.

__init__( num_sc_per_prb: int = 12, num_prb: int = 273, train_total: int = 20, test_total: int = 50, shard_size: int = 50, batch_tf: int = 256, prng_seed: int = 0, model_type: str = 'CDL', tdl_model: str = 'C', delay_spread_ns: float = 100.0, fc_ghz: float = 3.5, speed_min: float = 0.0, speed_max: float = 30.0, direction: str = 'uplink', bs_num_rows: int = 4, bs_num_cols: int = 1, bs_polarization: str = 'single', bs_polarization_type: str = 'V', bs_antenna_pattern: str = '38.901', ue_num_rows: int = 1, ue_num_cols: int = 1, ue_polarization: str = 'single', ue_polarization_type: str = 'V', ue_antenna_pattern: str = '38.901', ) → None#

batch_tf: int = 256#

bs_antenna_pattern: str = '38.901'#

bs_num_cols: int = 1#

bs_num_rows: int = 4#

bs_polarization: str = 'single'#

bs_polarization_type: str = 'V'#

delay_spread_ns: float = 100.0#

direction: str = 'uplink'#

fc_ghz: float = 3.5#

model_type: str = 'CDL'#

num_prb: int = 273#

num_sc_per_prb: int = 12#

prng_seed: int = 0#

shard_size: int = 50#

speed_max: float = 30.0#

speed_min: float = 0.0#

tdl_model: str = 'C'#

test_total: int = 50#

train_total: int = 20#

ue_antenna_pattern: str = '38.901'#

ue_num_cols: int = 1#

ue_num_rows: int = 1#

ue_polarization: str = 'single'#

ue_polarization_type: str = 'V'#

ran.datasets.gen_cdl_channels( config: SionnaCDLConfig, validation_frac: float = 0.15, out_dir: str = '/opt/nvidia/aerial-framework/out/sionna_dataset', ) → tuple[int, int, int][source]#

Generate CDL channel dataset using Sionna and save to safetensors.

This function generates channel realizations, splits them into train/val/test, and saves directly to safetensors format, eliminating intermediate files.

Parameters:

config – Configuration object containing all channel generation parameters.
validation_frac – Fraction of training data for validation.
out_dir – Output directory for generated files.

Returns:

Tuple of (num_train, num_val, num_test) samples.

ran.datasets.setup_datasets( train_glob: str, test_glob: str, num_sc: int, validation_frac: float, prng_seed: int = 0, num_symbols: int = 14, num_rx: int = 4, ) → tuple[SionnaChannelDataset, SionnaChannelDataset, SionnaChannelDataset][source]#

Setup train, validation, and test datasets.

Parameters:

train_glob – Glob pattern for training data files
test_glob – Glob pattern for test data files
num_sc – Number of subcarriers
validation_frac – Fraction of training data to use for validation
prng_seed – Random seed for dataset splitting
num_symbols – Number of OFDM symbols
num_rx – Number of RX antennas

Return type:

Tuple of (train_dataset, val_dataset, test_dataset)