Source code for sionna.sys.phy_abstraction

#
# SPDX-FileCopyrightText: Copyright (c) 2021-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
# SPDX-License-Identifier: Apache-2.0
#
"""Physical layer abstraction for Sionna SYS"""

import datetime
import json
import logging
import os
import time
import warnings
from typing import Any, Callable, Dict, List, Optional, Tuple, Union

import matplotlib.pyplot as plt
import numpy as np
import torch

from sionna.phy import Block, config, dtypes
from sionna.phy.config import Precision
from sionna.phy.nr.utils import CodedAWGNChannelNR, MCSDecoderNR, TransportBlockNR
from sionna.phy.utils import (
    DeepUpdateDict,
    Interpolate,
    MCSDecoder,
    SingleLinkChannel,
    SplineGriddataInterpolation,
    TransportBlock,
    dict_keys_to_int,
    gather_from_batched_indices,
    lin_to_db,
    scalar_to_shaped_tensor,
    sim_ber,
    to_list,
)
from sionna.sys.effective_sinr import EESM, EffectiveSINR

__all__ = ["PHYAbstraction"]


[docs] class PHYAbstraction(Block): r"""Class for physical layer abstraction. For a given signal-to-interference-plus-noise-ratio (SINR) provided on a per-stream basis, and for a given modulation order, coderate, and number of coded bits specified for each user, it produces the corresponding number of successfully decoded bits, HARQ feedback, effective SINR, block error rate (BLER), and transport BLER (TBLER). At object instantiation, precomputed BLER tables are loaded and interpolated on a fine (SINR, code block size) grid for each modulation and coding scheme (MCS) index. When the object is called, the post-equalization SINR is first converted to an effective SINR. Then, the effective SINR is used to retrieve the BLER from pre-computed and interpolated tables. Finally, the BLER determines the TBLER, which represents the probability that at least one code block is incorrectly received. :param interp_fun: Function for interpolating data defined on rectangular or unstructured grids, used for BLER and SINR interpolation. If `None`, it is set to an instance of :class:`~sionna.phy.utils.SplineGriddataInterpolation`. :param mcs_decoder_fun: Function mapping MCS indices to modulation order and coderate. If `None`, it is set to an instance of :class:`~sionna.phy.nr.utils.MCSDecoderNR`. :param transport_block_fun: Function computing the number and size (measured in bits) of code blocks within a transport block. If `None`, it is set to an instance of :class:`~sionna.phy.nr.utils.TransportBlockNR`. :param sinr_effective_fun: Function computing the effective SINR. If `None`, it is set to an instance of :class:`~sionna.sys.EESM`. :param load_bler_tables_from: Name of file(s) containing pre-computed SINR-to-BLER tables for different categories, table indices, MCS indices, SINR and code block sizes. If "default", then the pre-computed tables stored in "bler_tables/" folder are loaded. :param snr_db_interp_min_max_delta: Tuple of (`min`, `max`, `delta`) values [dB] defining the list of SINR [dB] values at which the BLER is interpolated, as `min, min+delta, min+2*delta,...,` up until `max`. :param cbs_interp_min_max_delta: Tuple of (`min`, `max`, `delta`) values defining the list of code block size values at which the BLER and SINR are interpolated, as `min, min+delta, min+2*delta,...,max`. :param bler_interp_delta: Spacing of the BLER grid at which SINR is interpolated. :param precision: Precision used for internal calculations and outputs. If set to `None`, :attr:`~sionna.phy.config.Config.precision` is used. :param device: Device for computation. If `None`, :attr:`~sionna.phy.config.Config.device` is used. :input mcs_index: [..., num_ut], `torch.int32`. MCS index for each user. :input sinr: [..., num_ofdm_symbols, num_subcarriers, num_ut, num_streams_per_ut], `torch.float` | `None`. Post-equalization SINR in linear scale for each OFDM symbol, subcarrier, user and stream. If `None`, then ``sinr_eff`` and ``num_allocated_re`` are both required. :input sinr_eff: [..., num_ut], `torch.float` | `None`. Effective SINR in linear scale for each user. If `None`, then ``sinr`` is required. :input num_allocated_re: [..., num_ut], `torch.int32` | `None`. Number of allocated resources in a slot, computed across OFDM symbols, subcarriers and streams, for each user. If `None`, then ``sinr`` is required. :input mcs_table_index: [..., num_ut], `torch.int32` | `int`. MCS table index. Defaults to 1. For further details, refer to the :ref:`mcs_table_cat_note`. :input mcs_category: [..., num_ut], `torch.int32` | `int`. MCS table category. Defaults to 0. For further details, refer to the :ref:`mcs_table_cat_note`. :input check_mcs_index_validity: `bool`. If `True`, a ValueError is thrown if the input MCS indices are not valid for the given configuration. Defaults to `True`. :output num_decoded_bits: [..., num_ut], `torch.int32`. Number of successfully decoded bits for each user. :output harq_feedback: [..., num_ut], -1 | 0 | 1. If 0 (1, resp.), then a NACK (ACK, resp.) is received. If -1, feedback is missing since the user is not scheduled for transmission. :output sinr_eff: [..., num_ut], `torch.float`. Effective SINR in linear scale for each user. :output tbler: [..., num_ut], `torch.float`. Transport block error rate (BLER) for each user. :output bler: [..., num_ut], `torch.float`. Block error rate (BLER) for each user. .. rubric:: Notes In this class, the terms SNR (signal-to-noise ratio) and SINR (signal-to-interference-plus-noise ratio) can be used interchangeably. This is because the equivalent AWGN model used for BLER mapping does not explicitly account for interference. .. rubric:: Examples .. code-block:: python import numpy as np from sionna.sys import PHYAbstraction, EESM from sionna.phy.nr.utils import MCSDecoderNR, TransportBlockNR from sionna.phy.utils import SplineGriddataInterpolation # Instantiate the class for BLER and SINR interpolation bler_snr_interp_fun = SplineGriddataInterpolation() # Instantiate the class for mapping MCS to modulation order and coderate # in 5G NR mcs_decoder_fun = MCSDecoderNR() # Instantiate the class for computing the number and size of code blocks # within a transport block in 5G NR transport_block_fun = TransportBlockNR() # Instantiate the class for computing the effective SINR sinr_effective_fun = EESM() # By instantiating a PHYAbstraction object, precomputed BLER tables are # loaded and interpolated on a fine (SINR, code block size) grid for each MCS phy_abs = PHYAbstraction( interp_fun=bler_snr_interp_fun, mcs_decoder_fun=mcs_decoder_fun, transport_block_fun=transport_block_fun, sinr_effective_fun=sinr_effective_fun) # Plot a BLER table phy_abs.plot(plot_subset={'category': {0: {'index': {1: {'MCS': 14}}}}}, show=True); .. figure:: ../figures/category0_table1_mcs14.png :align: center :width: 70% .. code-block:: python # One can also compute new BLER tables # SINR values and code block sizes @ new simulations are performed snr_dbs = np.linspace(-5, 25, 5) cb_sizes = np.arange(24, 8448, 1000) # MCS values @ new simulations are performed sim_set = {'category': { 0: {'index': { 1: {'MCS': [15]} }}}} # Compute new tables new_table = phy_abs.new_bler_table( snr_dbs, cb_sizes, sim_set, max_mc_iter=15, batch_size=10, verbose=True) """ def __init__( self, interp_fun: Optional[Interpolate] = None, mcs_decoder_fun: Optional[MCSDecoder] = None, transport_block_fun: Optional[TransportBlock] = None, sinr_effective_fun: Optional[EffectiveSINR] = None, load_bler_tables_from: Union[str, List[str]] = "default", snr_db_interp_min_max_delta: Tuple[float, float, float] = (-5, 30.01, 0.1), cbs_interp_min_max_delta: Tuple[int, int, int] = (24, 8448, 100), bler_interp_delta: float = 0.01, precision: Optional[Precision] = None, device: Optional[str] = None, **kwargs: Any, ) -> None: super().__init__(precision=precision, device=device) # Set default values if interp_fun is None: interp_fun = SplineGriddataInterpolation() if mcs_decoder_fun is None: mcs_decoder_fun = MCSDecoderNR(precision=precision, device=device) if transport_block_fun is None: transport_block_fun = TransportBlockNR(precision=precision, device=device) if sinr_effective_fun is None: sinr_effective_fun = EESM(precision=precision, device=device) # Check inputs if not isinstance(interp_fun, Interpolate): raise ValueError( "interp_fun must be an instance of sionna.phy.utils.Interpolate" ) if not isinstance(mcs_decoder_fun, MCSDecoder): raise ValueError( "mcs_decoder_fun must be an instance of sionna.phy.utils.MCSDecoder" ) if not isinstance(transport_block_fun, TransportBlock): raise ValueError( "transport_block_fun must be a subclass of sionna.phy.utils.TransportBlock" ) if not isinstance(sinr_effective_fun, EffectiveSINR): raise ValueError( "sinr_effective_fun must be a subclass of sionna.sys.EffectiveSINR" ) # -------------- # # Initialization # # -------------- # self._kwargs = kwargs self._bler_table: Optional[DeepUpdateDict] = None self._bler_table_interp: Optional[torch.Tensor] = None self._snr_table_interp: Optional[torch.Tensor] = None # ------------- # # Instantiation # # ------------- # # Function interpolating (CBS, SNR) -> BLER self._bler_interp_fun = interp_fun.struct # Function interpolating (CBS, BLER) -> SNR self._snr_interp_fun = interp_fun.unstruct # Function mapping MCS index to modulation order and coderate self._mcs_decoder_fun = mcs_decoder_fun # Function computing number and size of code blocks self._transport_block_fun = transport_block_fun # Function computing the effective SINR self._sinr_effective_fun = sinr_effective_fun # Interpolation grid self._cbs_interp: Optional[np.ndarray] = None self._snr_dbs_interp: Optional[np.ndarray] = None self._blers_interp: Optional[np.ndarray] = None if load_bler_tables_from == "default": filenames = [ "bler_tables/PUSCH_table1.json", "bler_tables/PUSCH_table2.json", "bler_tables/PDSCH_table1.json", "bler_tables/PDSCH_table2.json", "bler_tables/PDSCH_table3.json", "bler_tables/PDSCH_table4.json", ] self.bler_table_filenames = [ os.path.join(os.path.dirname(os.path.abspath(__file__)), f) for f in filenames ] else: self.bler_table_filenames = load_bler_tables_from # SNR/BLER interpolation grid self.snr_db_interp_min_max_delta = snr_db_interp_min_max_delta self.cbs_interp_min_max_delta = cbs_interp_min_max_delta self.bler_interp_delta = bler_interp_delta
[docs] @staticmethod def load_table(filename: str) -> Dict: r"""Loads a table stored in JSON file. :param filename: Name of the JSON file containing the table :output table: Table loaded from file """ with open(filename, "r", encoding="utf-8") as f: table = json.load(f, object_hook=dict_keys_to_int) return table
# ----------- # # BLER tables # # ----------- # @property def bler_table_filenames(self) -> List[str]: r"""`str` | list of `str`: Get/set the absolute path name of the files containing BLER tables. """ return self._bler_table_filenames @bler_table_filenames.setter def bler_table_filenames(self, value: Union[str, List[str]]) -> None: self._bler_table_filenames = to_list(value) # Load the table self._bler_table = DeepUpdateDict({"category": {}}) for f in self._bler_table_filenames: try: with open(f, "r", encoding="utf-8") as file: bler_subtable = json.load(file, object_hook=dict_keys_to_int) # Merge with the existing one self._bler_table.deep_update( bler_subtable, stop_at_keys=("CBS", "SNR_db") ) except FileNotFoundError: warnings.warn(f"BLER table file '{f}' does not exist. Skipping...") if self._bler_table == {}: warnings.warn( "No BLER table found. You can generate them via " "PHYAbstraction.new_bler_table method." ) # Check table validity self.validate_bler_table() def _get_batch_size_interp_mat(self) -> List[int]: """Compute the batch size of interpolation tensors.""" categories = list(self._bler_table["category"].keys()) max_table_idx_list, max_mcs_list = [], [] for ch in categories: table_idx_list = list(self._bler_table["category"][ch]["index"].keys()) max_table_idx_list.append(max(table_idx_list)) for table_idx in table_idx_list: mcs_list = list( self._bler_table["category"][ch]["index"][table_idx]["MCS"].keys() ) max_mcs_list.append(max(mcs_list)) if ( len(categories) > 0 and len(max_table_idx_list) > 0 and len(max_mcs_list) > 0 ): return [max(categories) + 1, max(max_table_idx_list), max(max_mcs_list) + 1] return [0, 0, 0] @property def bler_table(self) -> Dict: r"""`dict` (read-only): Collection of tables containing BLER values for different values of SNR, MCS table, MCS index and CB size. ``bler_table['category'][cat]['index'][mcs_table_index]['MCS'][mcs]['CBS'][cb_size]`` contains the lists of BLER values. ``bler_table['category'][cat]['index'][mcs_table_index]['MCS'][mcs]['SNR_db']`` contains the list of SNR values. ``bler_table['category'][cat]['index'][mcs_table_index]['MCS'][mcs]['EbN0_db']`` contains the list of :math:`E_b/N_0` values. """ return self._bler_table # ------------------- # # Interpolated tables # # ------------------- # @property def bler_table_interp(self) -> torch.Tensor: r"""[n_categories, n_tables, n_mcs, n_cbs_index, n_snr], `torch.float` (read-only): Tensor containing BLER values interpolated across SINR and CBS values, for different categories and MCS table indices. The first axis accounts for the category, e.g., 'PDSCH' or 'PUSCH' in 5G-NR, the second axis corresponds to the 38.214 MCS table index while the third axis carries the MCS index. """ return self._bler_table_interp @property def snr_table_interp(self) -> torch.Tensor: r"""[n_categories, n_tables, n_mcs, n_cbs_index, n_bler], `torch.float` (read-only): Tensor containing SINR values interpolated across BLER and CBS values, for different categories and MCS table indices. The first axis accounts for the category, e.g., 'PDSCH' or 'PUSCH' in 5G-NR, the second axis corresponds to the 38.214 MCS table index and the third axis accounts for the MCS index. """ return self._snr_table_interp # ------------------ # # Interpolation grid # # ------------------ # @property def snr_db_interp_min_max_delta(self) -> Tuple[float, float, float]: r"""[3], `tuple`: Get/set the tuple of (`min`, `max`, `delta`) values [dB] defining the list of SINR values at which the BLER is interpolated, as `min, min+delta, min+2*delta,...,` up until `max`. """ return self._snr_db_interp_min_max_delta @snr_db_interp_min_max_delta.setter def snr_db_interp_min_max_delta(self, value: Tuple[float, float, float]) -> None: if hasattr(value, "__len__") and len(value) == 3: self._snr_db_interp_min_max_delta = value else: raise ValueError("snr_db_interp_min_max_delta must have length 3") self._snr_dbs_interp = np.arange( self._snr_db_interp_min_max_delta[0], self._snr_db_interp_min_max_delta[1], self._snr_db_interp_min_max_delta[2], ) if (self._bler_table is not None) and (self._cbs_interp is not None): # Interpolate BLER self._interpolate_bler() @property def cbs_interp_min_max_delta(self) -> Tuple[int, int, int]: r"""[3], `tuple`: Get/set the tuple of (`min`, `max`, `delta`) values defining the list of code block size values at which the BLER and SINR are interpolated, as `min, min+delta, min+2*delta,...,` up until `max`. """ return self._cbs_interp_min_max_delta @cbs_interp_min_max_delta.setter def cbs_interp_min_max_delta(self, value: Tuple[int, int, int]) -> None: if hasattr(value, "__len__") and len(value) == 3: self._cbs_interp_min_max_delta = value else: raise ValueError("cbs_interp_min_max_delta must have length 3") self._cbs_interp = np.arange( self._cbs_interp_min_max_delta[0], self._cbs_interp_min_max_delta[1], self._cbs_interp_min_max_delta[2], ) if self._bler_table is not None: if self._blers_interp is not None: # Interpolate SNR self._interpolate_snr() if self._snr_dbs_interp is not None: # Interpolate BLER self._interpolate_bler() @property def bler_interp_delta(self) -> float: r"""`float`: Get/set the spacing of the BLER grid at which SINR is interpolated.""" return self._bler_interp_delta @bler_interp_delta.setter def bler_interp_delta(self, value: float) -> None: self._bler_interp_delta = value self._blers_interp = np.arange(0, 1, self._bler_interp_delta) if (self._bler_table is not None) and (self._cbs_interp is not None): # Interpolate SNR self._interpolate_snr()
[docs] def get_idx_from_grid(self, val: torch.Tensor, which: str) -> torch.Tensor: r"""Retrieves the index of a SINR or CBS value in the interpolation grid. :param val: Values to be quantized :param which: Whether the values are SNR (equivalent to SINR) or CBS. Must be "snr" or "cbs". :output idx: Index of the values in the interpolation grid """ assert which in ["snr", "cbs"], "which must be 'snr' or 'cbs'" if which == "snr": len_grid = len(self._snr_dbs_interp) min_max_delta = self._snr_db_interp_min_max_delta else: len_grid = len(self._cbs_interp) min_max_delta = self._cbs_interp_min_max_delta min_grid = min_max_delta[0] delta_grid = min_max_delta[2] if not isinstance(val, torch.Tensor): val = torch.tensor(val, dtype=torch.float32) idx = torch.round((val - min_grid) / delta_grid).to(torch.int32) idx = torch.clamp(idx, 0, len_grid - 1) return idx
# ------------- # # Retrieve BLER # # ------------- #
[docs] def get_bler( self, mcs_index: Union[int, torch.Tensor], mcs_table_index: Union[int, torch.Tensor], mcs_category: Union[int, torch.Tensor], cb_size: Union[int, torch.Tensor], snr_eff: torch.Tensor, ) -> torch.Tensor: r"""Retrieves from interpolated tables the BLER corresponding to a certain table index, MCS, CB size, and SINR values provided as input. If the corresponding interpolated table is not available, it returns `Inf`. :param mcs_index: MCS index for each user :param mcs_table_index: MCS table index for each user. For further details, refer to the :ref:`mcs_table_cat_note`. :param mcs_category: MCS table category for each user. For further details, refer to the :ref:`mcs_table_cat_note`. :param cb_size: Code block size for each user :param snr_eff: Effective SINR for each user :output bler: BLER corresponding to the input channel type, table index, MCS, CB size and SINR, retrieved from internal interpolation tables """ # Cast inputs to appropriate type and shape if not isinstance(snr_eff, torch.Tensor): snr_eff = torch.tensor(snr_eff, dtype=self.dtype, device=self.device) snr_eff = snr_eff.to(self.dtype) shape = list(snr_eff.shape) mcs_category = scalar_to_shaped_tensor( mcs_category, torch.int32, shape, device=self.device ) mcs_index = scalar_to_shaped_tensor( mcs_index, torch.int32, shape, device=self.device ) mcs_table_index = scalar_to_shaped_tensor( mcs_table_index, torch.int32, shape, device=self.device ) cb_size = scalar_to_shaped_tensor(cb_size, torch.int32, shape, device=self.device) # Convert SNR to dB snr_eff_db = lin_to_db(snr_eff, precision=self.precision) # Quantize the SNR [dB] and CBS to the corresponding interpolation index snr_db_idx = self.get_idx_from_grid(snr_eff_db, "snr") cbs_idx = self.get_idx_from_grid(cb_size.to(self.dtype), "cbs") # Stack indices to extract BLER from interpolated table idx = torch.stack( [mcs_category, mcs_table_index - 1, mcs_index, cbs_idx, snr_db_idx], dim=-1 ) # Compute BLER bler = gather_from_batched_indices(self._bler_table_interp, idx) bler = bler.to(self.dtype) return bler
def call( self, mcs_index: torch.Tensor, sinr: Optional[torch.Tensor] = None, sinr_eff: Optional[torch.Tensor] = None, num_allocated_re: Optional[torch.Tensor] = None, mcs_table_index: Union[int, torch.Tensor] = 1, mcs_category: Union[int, torch.Tensor] = 0, check_mcs_index_validity: bool = True, **kwargs: Any, ) -> Tuple[torch.Tensor, torch.Tensor, torch.Tensor, torch.Tensor, torch.Tensor]: """Compute PHY abstraction outputs. Refer to the class docstring for the Input/Output specification. """ assert (sinr is not None) ^ ( (sinr_eff is not None) and (num_allocated_re is not None) ), "Either 'sinr' or ('sinr_eff','num_allocated_re') is required as input" if sinr is not None: # Total number of allocated streams across all resource elements # [..., num_ut] num_allocated_re = (sinr > 0).to(torch.int32).sum(dim=(-4, -3, -1)) # Effective SINR # [..., num_ut] sinr_eff = self._sinr_effective_fun( sinr, mcs_index=mcs_index, mcs_table_index=mcs_table_index, mcs_category=mcs_category, per_stream=False, **kwargs, ) else: sinr_eff = sinr_eff.to(self.dtype) num_allocated_re = num_allocated_re.to(torch.int32) # Whether a user is scheduled # [..., num_ut] ut_is_scheduled = num_allocated_re > 0 # Convert MCS index to modulation order and coderate # [..., num_ut] modulation_order, target_coderate = self._mcs_decoder_fun( mcs_index, mcs_table_index, mcs_category, check_index_validity=check_mcs_index_validity, **kwargs, ) # Compute the number of coded bits num_coded_bits = modulation_order * num_allocated_re # Compute n. and size of Code Blocks (CBs) in a Transport Block # [..., num_ut] cb_size, num_cb = self._transport_block_fun( modulation_order, target_coderate, num_coded_bits, **kwargs ) # Retrieve the BLER from the stored tables # [..., num_ut] bler = self.get_bler( mcs_index, mcs_table_index, mcs_category, cb_size, sinr_eff ) # Compute TBLER = Pr(at least a CB is incorrectly received) # [..., num_ut] one = torch.tensor(1.0, dtype=bler.dtype, device=self.device) tbler = one - torch.pow(one - bler, num_cb.to(bler.dtype)) # Set BLER=-1 and TBLER=-1 for non-scheduled UTs minus_one = torch.tensor(-1.0, dtype=bler.dtype, device=self.device) bler = torch.where(ut_is_scheduled, bler, minus_one) tbler = torch.where(ut_is_scheduled, tbler, minus_one) # HARQ feedback generator = None if torch.compiler.is_compiling() else self.torch_rng rnd = torch.rand( tbler.shape, dtype=self.dtype, device=self.device, generator=generator ) harq_feedback = torch.where( rnd < tbler, torch.tensor(0, dtype=torch.int32, device=self.device), torch.tensor(1, dtype=torch.int32, device=self.device), ) # Successfully decoded bits for each user # [..., num_ut] num_decoded_bits = harq_feedback * num_cb * cb_size num_decoded_bits = torch.where( ut_is_scheduled, num_decoded_bits, torch.tensor(0, dtype=torch.int32, device=self.device), ) # Assign HARQ=-1 for the non-scheduled UTs harq_feedback = torch.where( ut_is_scheduled, harq_feedback, torch.tensor(-1, dtype=torch.int32, device=self.device), ) return num_decoded_bits, harq_feedback, sinr_eff, tbler, bler # -------------------- # # Interpolation method # # -------------------- # def _interpolate_bler(self) -> None: """Interpolates the BLER over a fine (CBS, SINR) grid.""" if self._bler_table is None: raise ValueError( "BLER table is not provided; Interpolation cannot be performed" ) if self._cbs_interp is None: raise ValueError( "CBS interpolation grid is not provided; Interpolation cannot be performed" ) interp_batch_size = self._get_batch_size_interp_mat() # [num_category, num_table_idx, num_mcs_index, num_cbs_interp, num_snr_interp] self._bler_table_interp_np = np.full( interp_batch_size + [len(self._cbs_interp), len(self._snr_dbs_interp)], np.inf, ) for category in self._bler_table["category"]: for table_idx in self._bler_table["category"][category]["index"]: table_mcs = self._bler_table["category"][category]["index"][table_idx][ "MCS" ] for mcs in table_mcs: cbs_vec = list(table_mcs[mcs]["CBS"].keys()) snr_vec = table_mcs[mcs]["SNR_db"] # Collect BLER values bler_val = np.zeros((len(cbs_vec), len(snr_vec))) for idx_cbs, cbs in enumerate(cbs_vec): bler_val[idx_cbs, :] = table_mcs[mcs]["CBS"][cbs]["BLER"] # Interpolate BLER try: bler_interp = self._bler_interp_fun( bler_val, cbs_vec, snr_vec, self._cbs_interp, self._snr_dbs_interp, **self._kwargs, ) except ValueError as e: warnings.warn( f"SINR-to-BLER interpolation failed for " f"category {category}, " f"index {table_idx}, MCS {mcs}.\nError: {e}" ) continue # Ensure BLER is within 0 and 1 bler_interp = np.clip(bler_interp, 0, 1) # Store it self._bler_table_interp_np[ category, table_idx - 1, mcs, :, : ] = bler_interp # Convert to tensor self._bler_table_interp = torch.tensor( self._bler_table_interp_np, dtype=self.dtype, device=self.device ) def _interpolate_snr(self) -> None: """Interpolates the SINR table over a fine (CBS, BLER) grid.""" if self._bler_table is None: raise ValueError( "BLER table is not provided; Interpolation cannot be performed" ) if self._blers_interp is None: raise ValueError( "BLER interpolation grid is not provided; Interpolation cannot be performed" ) interp_batch_size = self._get_batch_size_interp_mat() self._snr_table_interp_np = np.full( interp_batch_size + [len(self._cbs_interp), len(self._blers_interp)], np.inf ) for category in self._bler_table["category"]: for table_index in self._bler_table["category"][category]["index"]: table_mcs = self._bler_table["category"][category]["index"][table_index][ "MCS" ] for mcs in table_mcs: # Collect values of SNR, CBS and BLER in arrays snr_vec = table_mcs[mcs]["SNR_db"] cbs_vec = list(table_mcs[mcs]["CBS"].keys()) snr_vec_tile = np.tile(snr_vec, len(cbs_vec)) cbs_vec_rep = np.repeat(cbs_vec, len(snr_vec)) bler_vec = [ bler for cbs in cbs_vec for bler in table_mcs[mcs]["CBS"][cbs]["BLER"] ] try: # Interpolate the SNR as a function of CBS and BLER snr_interp = self._snr_interp_fun( snr_vec_tile, cbs_vec_rep, bler_vec, self._cbs_interp, self._blers_interp, **self._kwargs, ) except ValueError as e: warnings.warn( f"BLER-to-SINR interpolation failed for " f"category {category}, " f"index {table_index}, MCS {mcs}.\n" f"Error message: {e}" ) continue self._snr_table_interp_np[ category, table_index - 1, mcs, :, : ] = snr_interp # Convert to tensor self._snr_table_interp = torch.tensor( self._snr_table_interp_np, dtype=self.dtype, device=self.device )
[docs] def validate_bler_table(self) -> bool: r"""Validates the dictionary structure of ``self.bler_table``. :output is_valid: `True` if ``self.bler_table`` has a valid structure :raises ValueError: If the structure is invalid """ if not isinstance(self._bler_table, dict): raise ValueError("Must be a dictionary") if np.any(np.array(list(self._bler_table["category"].keys())) < 0): raise ValueError("Categories must be nonnegative integers") for _, bler_table_tmp in self._bler_table["category"].items(): if set(bler_table_tmp.keys()) != {"index"}: raise ValueError("Key must be 'index'") if np.any(np.array(list(bler_table_tmp["index"].keys())) < 1): raise ValueError("Table indices must be positive integers") for table_index in bler_table_tmp["index"]: if set(bler_table_tmp["index"][table_index].keys()) != {"MCS"}: raise ValueError("Key must be 'MCS'") if np.any( np.array( list(bler_table_tmp["index"][table_index]["MCS"].keys()) ) < 0 ): raise ValueError("MCS indices must be nonnegative integers") for mcs in bler_table_tmp["index"][table_index]["MCS"]: if set( bler_table_tmp["index"][table_index]["MCS"][mcs].keys() ) != {"CBS", "SNR_db"}: raise ValueError("Keys must be ['CBS', 'SNR_db']") for cbs in bler_table_tmp["index"][table_index]["MCS"][mcs]["CBS"]: if ( set( bler_table_tmp["index"][table_index]["MCS"][mcs]["CBS"][ cbs ].keys() ) != {"BLER"} ): raise ValueError("Keys must be 'BLER'") return True
[docs] def plot( self, plot_subset: Union[str, Dict] = "all", show: bool = True, save_path: Optional[str] = None, ) -> List[str]: r"""Visualizes and/or saves to file the SINR-to-BLER tables. :param plot_subset: Dictionary containing the list of MCS indices to consider, stored at ``plot_subset['category'][category]['index'][mcs_table_index]['MCS']``. If "all", then plots are produced for all available BLER tables. :param show: If `True`, then plots are visualized. Defaults to `True`. :param save_path: Folder path where BLER plots are saved. If `None`, then plots are not saved. :output fignames: List of names of files containing BLER plots """ if self._bler_table is None: raise ValueError( "Plots cannot be produced as self.bler_table has not been loaded or computed" ) # Create folder if it does not exist if save_path is not None: if not os.path.exists(save_path): os.makedirs(save_path) logging.info("\nCreated folder %s\n", save_path) fignames = [] if plot_subset == "all": plot_subset = self._bler_table for category, plot_subset_cat in plot_subset["category"].items(): for table_index, plot_subset_cat_tab in plot_subset_cat["index"].items(): for mcs in to_list(plot_subset_cat_tab["MCS"]): try: num_bits_per_symbol, coderate = self._mcs_decoder_fun( mcs, table_index, category, **self._kwargs ) if isinstance(num_bits_per_symbol, torch.Tensor): num_bits_per_symbol = num_bits_per_symbol.item() if isinstance(coderate, torch.Tensor): coderate = coderate.item() except (ValueError, AssertionError) as e: print( f"Invalid (category={category}, " f"index={table_index}, " f"MCS={mcs}) combination. \n" f"Error message: {e}\n" f"Skipping...\n" ) continue # SNR value at the channel capacity for the spectral # efficiency associated to the current MCS snr_shannon = 2 ** (num_bits_per_symbol * coderate) - 1 snr_shannon_db = 10 * np.log10(snr_shannon) try: snr_dbs = self._bler_table["category"][category]["index"][ table_index ]["MCS"][mcs]["SNR_db"] fig, ax = plt.subplots() for cbs in self._bler_table["category"][category]["index"][ table_index ]["MCS"][mcs]["CBS"]: bler = self._bler_table["category"][category]["index"][ table_index ]["MCS"][mcs]["CBS"][cbs]["BLER"] ax.semilogy(snr_dbs, bler, label=f"code block size={cbs}") ax.plot( [snr_shannon_db] * 2, ax.get_ylim(), "--k", label="SNR @capacity", ) ax.set_title( f"MCS index {mcs} (table category {category}, " f"index {table_index})" ) ax.legend() ax.grid(True) ax.set_xlabel("SNR [dB]") ax.set_ylabel("BLER") # Save to file if save_path is not None: figname = ( f"category{category}_table" f"{table_index}_mcs{mcs}.png" ) figname = os.path.join(save_path, figname) fig.savefig(figname) fignames.append(figname) if show: plt.show() plt.close(fig) except KeyError as e: print( f"\nBLER for (category={category}, index=" f"{table_index}, MCS={mcs}) not available " f"for plotting. Error message: {e}." "\nSkipping..." ) return fignames
[docs] def new_bler_table( self, snr_dbs: Union[List[float], np.ndarray], cb_sizes: Union[List[int], np.ndarray], sim_set: Dict, channel: Optional[SingleLinkChannel] = None, filename: Optional[str] = None, write_mode: str = "w", batch_size: int = 1000, max_mc_iter: int = 100, target_bler: Optional[float] = None, compile_mode: Optional[str] = None, early_stop: bool = True, filename_log: Optional[str] = None, verbose: bool = True, ) -> Dict: r"""Computes static tables mapping SNR values of an AWGN channel to the corresponding block error rate (BLER) via Monte-Carlo simulations for different MCS indices, code block sizes and channel types. Note that the newly computed table is merged with the internal ``self.bler_table``. The simulation continues with the next SNR point after ``max_mc_iter`` batches of size ``batch_size`` have been simulated. Early stopping allows to stop the simulation after the first error-free SNR point or after reaching a certain ``target_bler``. For more details, please see :func:`~sionna.phy.utils.misc.sim_ber`. :param snr_dbs: List of SNR [dB] value(s) at which the BLER is computed :param cb_sizes: List of code block (CB) size(s) at which the BLER is computed :param sim_set: Dictionary containing the list of MCS indices at which the BLER is computed via simulation. The dictionary structure is of the kind: ``sim_set['category'][category]['index'][mcs_table_index]['MCS'][mcs_list]``. :param channel: Object for simulating single-link, i.e., single-carrier and single-stream, channels. If `None`, it is set to an instance of :class:`~sionna.phy.nr.utils.CodedAWGNChannelNR`. :param filename: Name of JSON file where the BLER tables are saved. If `None`, results are not saved. :param write_mode: If 'w', then ``filename`` is rewritten. If 'a', then the produced results are appended to ``filename``. :param batch_size: Batch size for Monte-Carlo BLER simulations :param max_mc_iter: Maximum number of Monte-Carlo iterations per SNR point :param target_bler: The simulation stops after the first SNR point which achieves a lower block error rate as specified by ``target_bler``. This requires ``early_stop`` to be `True`. :param compile_mode: Execution mode of channel call method. If `None`, channel is executed as is. :param early_stop: If `True`, the simulation stops after the first error-free SNR point. :param filename_log: Name of logging file. If `None`, logs are not produced. :param verbose: If `True`, the simulation progress is visualized, as well as the names of files of results and figures. :output new_table: Newly computed BLER table """ def _log(msg: str, level: str = "info") -> None: """Logging and printing simulation progress.""" if verbose: print(msg) if filename_log is not None: if level == "info": logging.info(msg) elif level == "warning": logging.warning(msg) elif level == "error": logging.error(msg) else: raise ValueError("unrecognized 'level' input") if channel is None: channel = CodedAWGNChannelNR(precision=self.precision, device=self.device) # Check input validity if not isinstance(channel, SingleLinkChannel): raise ValueError( "'channel' must be an instance of sionna.phy.utils.SingleLinkChannel" ) if filename_log is not None: # Logging settings logging.basicConfig( filename=filename_log, filemode="w", level=logging.INFO, format="%(asctime)s - %(levelname)s - %(message)s", ) if write_mode not in ["a", "w"]: raise ValueError( "'write_mode' must be either 'a' for appending to file or 'w' for rewriting" ) snr_dbs = to_list(snr_dbs) cb_sizes = to_list(cb_sizes) if ( (filename is not None) and os.path.isfile(filename) and (write_mode == "a") ): # Append to file new_table = self.load_table(filename) else: # (Re)write to file new_table = {"category": {}} # Total number of points to simulate n_sims_tot = len(cb_sizes) * sum( len(sim_set["category"][cat]["index"][ti]["MCS"]) for cat in sim_set["category"] for ti in sim_set["category"][cat]["index"] ) _log( "\nBLER simulations started. " "\nTotal # (category, index, MCS, SINR) " f"points to simulate: {n_sims_tot}\n" ) # ---------------- # # BLER simulations # # ---------------- # n_sims_done = 0 time_start = time.time() for category, sim_set_cat in sim_set["category"].items(): if category not in new_table.keys(): new_table["category"][category] = {"index": {}} for table_index, sim_set_tab in sim_set_cat["index"].items(): if table_index not in new_table["category"][category]["index"].keys(): new_table["category"][category]["index"][table_index] = {"MCS": {}} for mcs in sim_set_tab["MCS"]: # Compute modulation order and code-rate associated with the # MCS index, table index and channel type try: num_bits_per_symbol, coderate = self._mcs_decoder_fun( mcs, table_index, category, **self._kwargs ) if isinstance(num_bits_per_symbol, torch.Tensor): num_bits_per_symbol = num_bits_per_symbol.item() if isinstance(coderate, torch.Tensor): coderate = coderate.item() except (ValueError, AssertionError) as e: _log( f"Invalid (category={category}, " f"index={table_index}, " f"MCS={mcs}) combination. \n" f"Error message: {e}\n" f"Skipping...\n", level="warning", ) continue # Eb/N0, where Eb=energy per information (uncoded) bit ebno_dbs = [ x - 10 * np.log10(num_bits_per_symbol * coderate) for x in snr_dbs ] # N. successful simulations for the MCS n_sims_mcs = 0 for cbs in cb_sizes: _log( f"\nSimulating category={category}, " f"index={table_index}, " f"CBS={cbs}, MCS={mcs}...\n" ) try: # Instantiate the AWGN coded channel channel.num_bits_per_symbol = num_bits_per_symbol channel.num_info_bits = cbs channel.target_coderate = coderate # Compute BLER via Monte-Carlo simulations _, bler = sim_ber( channel, ebno_dbs, batch_size, max_mc_iter, soft_estimates=False, early_stop=early_stop, target_bler=target_bler, compile_mode=compile_mode, forward_keyboard_interrupt=True, verbose=verbose, precision=self.precision, device=self.device, ) n_sims_mcs += 1 if n_sims_mcs == 1: # Initialize dictionary new_table["category"][category]["index"][table_index][ "MCS" ][mcs] = {"CBS": {}, "SNR_db": snr_dbs} # Record results in bler_table new_table["category"][category]["index"][table_index][ "MCS" ][mcs]["CBS"][cbs] = {"BLER": bler.cpu().numpy().tolist()} # Write to JSON file if filename is not None: with open(filename, "w", encoding="utf-8") as file: json.dump(new_table, file, indent=6) _log( f"\nResults written in file " f"{os.path.abspath(filename)}\n" ) except (ValueError, RuntimeError) as e: _log( f"\nBER/BLER simulations failed for " f"(category={category}, " f"index={table_index}, " f"CBS={cbs}, MCS={mcs})\n" f"Error message: {e}\n" f"Skipping...\n", level="error", ) n_sims_done += 1 # Compute simulation progress and remaining execution time progress = n_sims_done / n_sims_tot * 100 time_s = (time.time() - time_start) * (100 - progress) / progress time_s = datetime.timedelta(seconds=time_s) time_hms = str(time_s).split(".", maxsplit=1)[0] _log( f"\nPROGRESS: {progress:.2f}%\n" f"Estimated remaining time: " f"{time_hms} [h:m:s]\n" ) _log("\nSimulations completed!\n") # Merge the newly computed table with self.bler_table self._bler_table.deep_update(new_table, stop_at_keys=("CBS", "SNR_db")) self.validate_bler_table() # Append file name to the internal list of loaded BLER files if filename is not None: self._bler_table_filenames.append(filename) return new_table