在 Aerial Data Lake 数据上使用 pyAerial 进行信道估计

此示例展示了如何使用 pyAerial 绑定来运行 cuPHY GPU 加速的 5G NR PUSCH 信道估计。5G NR PUSCH 数据从示例空中捕获的 PUSCH 数据集中读取，该数据集使用 Aerial Data Lake 收集和存储，并且信道是使用 pyAerial 和基于相应 PUSCH 参数的 cuPHY 估计的。

注意： 此示例需要 clickhouse 服务器正在运行，并且示例数据已存储在数据库中。请参阅 Aerial Data Lake 文档，了解如何执行此操作。

# Check platform.
import platform
if platform.machine() not in ['x86_64', 'aarch64']:
    raise SystemExit("Unsupported platform!")

导入

import math
import os
os.environ["CUDA_VISIBLE_DEVICES"] = "0"

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt

# Connecting to clickhouse on remote server
import clickhouse_connect

# Import the channel estimator and some utilities for converting
# the DMRS fields in the right format from the SCF FAPI format that the dataset follows.
from aerial.phy5g.algorithms import ChannelEstimator
from aerial.util.fapi import dmrs_fapi_to_bit_array

数据

我们使用从真实空中 PUSCH 传输中捕获的示例数据集。数据库中的“fapi”表包含每个 PUSCH 传输的元数据，而“fh”表包含该时隙的所有样本。

# Create the pyAerial (cuPHY) channel estimator.
num_ues = 1
num_rx_ant = 4
channel_estimator = ChannelEstimator(num_rx_ant=num_rx_ant)

# Connect to the local database
client = clickhouse_connect.get_client(host='localhost')

# Pick some pusch records from the database
pusch_records = client.query_df('select * from fapi order by TsTaiNs limit 10')

运行信道估计

从 PUSCH 记录中，我们提取 PUSCH DMRS 参数，并使用 TAI 时间条目选择该时隙的 IQ 样本。然后使用提取的参数运行信道估计，并将估计信道的绝对值绘制在同一图中。

for index,pusch_record in pusch_records.iterrows():
    query = f"""select TsTaiNs,fhData from fh where
            TsTaiNs == {pusch_record.TsTaiNs.timestamp()}
            """

    fh = client.query_df(query)

    # Make sure that the fronthaul database is complete for the SFN.Slot we've chosen
    if fh.index.size < 1:
        pusch_records = pusch_records.drop(index)
        continue;

    fh_samp = np.array(fh['fhData'][0], dtype=np.float32)
    rx_slot = np.swapaxes(fh_samp.view(np.complex64).reshape(4, 14, 273*12),2,0)

  # Extract all the needed parameters from the PUSCH record.
    slot = int(pusch_record.Slot)
    rntis = [pusch_record.rnti]
    layers = [pusch_record.nrOfLayers]
    start_prb = pusch_record.rbStart
    num_prbs = pusch_record.rbSize
    start_sym = pusch_record.StartSymbolIndex
    num_symbols = pusch_record.NrOfSymbols
    scids = [int(pusch_record.SCID)]
    data_scids = [pusch_record.dataScramblingId]
    dmrs_scrm_id = pusch_record.ulDmrsScramblingId
    num_dmrs_cdm_grps_no_data = pusch_record.numDmrsCdmGrpsNoData
    dmrs_syms = dmrs_fapi_to_bit_array(int(pusch_record.ulDmrsSymbPos))
    dmrs_ports = [pusch_record.dmrsPorts]
    dmrs_max_len = 1
    dmrs_add_ln_pos = 2
    num_subcarriers = num_prbs * 12
    mcs_tables = [pusch_record.mcsTable]
    mcs_indices = [pusch_record.mcsIndex]
    coderates = [pusch_record.targetCodeRate/10.]
    tb_sizes = [pusch_record.TBSize]
    mod_orders = [pusch_record.qamModOrder]
    tb_input = np.array(pusch_record.pduData)


    # Run PyAerial (cuPHY) channel estimation.
    ch_est = channel_estimator.estimate(
        rx_slot=rx_slot,
        num_ues=num_ues,
        layers=layers,
        scids=scids,
        slot=slot,
        dmrs_ports=dmrs_ports,
        dmrs_syms=dmrs_syms,
        dmrs_scrm_id=dmrs_scrm_id,
        dmrs_max_len=dmrs_max_len,
        dmrs_add_ln_pos=dmrs_add_ln_pos,
        num_dmrs_cdm_grps_no_data=num_dmrs_cdm_grps_no_data,
        start_prb=start_prb,
        num_prbs=num_prbs,
        prg_size=1,
        num_ul_streams=1,
        start_sym=start_sym,
        num_symbols=num_symbols
    )

    fig, axs = plt.subplots(1)
    fig.suptitle("Channel estimates for SFN.Slot: "+str(pusch_record.SFN)+"."+str(pusch_record.Slot))
    axs.set_title(pusch_record.TsTaiNs)
    for ant in range(4):
         axs.plot(np.abs(ch_est[0][ant, 0, :, 0]))
    axs.grid(True)
    plt.show()