Hi! We’ve been trying to extract channel estimates inside cuBB. Are there any logs or functions for this task?
Hi @snkim94 ,
As you can see in the public document of the diagram of the PUSCH pipeline, channel estimation is implemented as a CUDA kernel of the CUDA Graph PUSCH pipeline.
By default, the CUDA codes of Aerial do not output logs. If you want to check the estimated value, you can add some codes to show it and recompile the whole package. You can find many commented printf statements for debugging in the CUDA source code files. It can help you with how you can add. Please note that this may impact performance.
Also, you can use the 5GModel MATLAB simulator to check the algorithm. It has the same algorithms implemented in cuPHY, and it is much more user-friendly to see estimated values.
Thank you.
Nori
Thank you very much for the reply!
Can you also specify which file should we look for the commented lines?
Hi @snkim94 ,
You can run a cuPHY stand-alone test as explained in our release documentation in Section “Running the cuPHY Examples”.
You can run the PUSCH pipeline with the command cuPHY/build/examples/pusch_rx_multi_pipe/cuphy_ex_pusch_rx_multi_pipe -i ~/<tv_name>.h5. If you define an additional argument for an output file such as -o debug_sample_dump.h5, the test will write some important PHY layer configuration and data samples at different stages into this file.
The channel estimation output will be saved in the data structure Hest0. The following shows only part of the data in sample output:
h5ls debug_sample_dump.h5
CfoEst0 Dataset {1, 14}
DataEq0 Dataset {13, 3276, 2}
DataRx0 Dataset {4, 14, 3276}
EqCoeff0 Dataset {1, 273, 2, 12, 4}
G_csi1 Dataset {1}
G_csi2 Dataset {1}
G_harq Dataset {1}
G_harq_rvd Dataset {1}
G_schAndCsi2 Dataset {1}
HEst0 Dataset {1, 3276, 2, 4}
LLR0 Dataset {13, 3276, 2, 8}
LwInv0 Dataset {273, 4, 4}
N_slot_frame Dataset {1}
N_symb_slot Dataset {1}
NoiseVarPostEq Dataset {1}
NoiseVarPreEqPerUe Dataset {1}
Ree0 Dataset {1, 273, 2, 12}
Let us know if any additional help is needed.
Thank you,
Balkan
This is great! Thank you so much for the detailed explanations. We will definitely try these code. One last question is that similar to the channel estimates, if we want to extract the I/Q samples of UL SRS measurements, which files should we look for?
Hi @snkim94,
Yes, you can use the same command to trigger debug output for the srs pipeline example.
The following command will output debug data into debug_sample_dump_srs.h5.
./cuPHY/examples/srs_rx_pipeline/cuphy_ex_srs_rx_pipeline.cpp -i Rel_24_1_tvs_srs/TVnr_8002_SRS_gNB_CUPHY_s0p0.h5 -o debug_sample_dump_srs.h5
The data file has the following information:
h5ls debug_sample_dump_srs.h5
Toffset Dataset {1}
Tsrs Dataset {1}
bandwidthIdx Dataset {1}
cellIdx Dataset {1}
chEst0 Dataset {1, 4, 136}
chEstBuffIdx Dataset {1}
chEstToL20 Dataset {1, 4, 2}
combOffset Dataset {1}
combSize Dataset {1}
configIdx Dataset {1}
cyclicShift Dataset {1}
dataRx0 Dataset {4, 1, 3276}
frameNum Dataset {1}
frequencyHopping Dataset {1}
frequencyPosition Dataset {1}
frequencyShift Dataset {1}
groupOrSequenceHopping Dataset {1}
mu Dataset {1}
nAntPorts Dataset {1}
nRepetitions Dataset {1}
nRxAntSrs Dataset {1}
nSrsSym Dataset {1}
nSyms Dataset {1}
prgSize Dataset {1}
resourceType Dataset {1}
sequenceId Dataset {1}
slotNum Dataset {1}
srsAntPortToUeAntMap0 Dataset {1}
srsAntPortToUeAntMap1 Dataset {1}
srsAntPortToUeAntMap2 Dataset {1}
srsAntPortToUeAntMap3 Dataset {1}
srsStartSym Dataset {1}
startSym Dataset {1}
toEstMicroSec Dataset {1}
usage Dataset {1}
widebandCsCorrNotUse Dataset {1}
widebandCsCorrRatioDb Dataset {1}
widebandCsCorrUse Dataset {1}
widebandNoiseEnergy Dataset {1}
widebandSignalEnergy Dataset {1}
widebandSnr Dataset {1}
Please note that these tests (cuphy_ex_*_rx_pipeline) run only specific channels. Therefore, they are not cuBB tests.
Thank you,
Balkan
Thank you! These codes work also work for cuBB 23-4?
@snkim94 Yes, they also work for 23-4. Please let us know if you have any issues.
Thank you.
Hi, we are trying the codes from “Using Aerial Python mcore Module” in 24-1 documents to generate the test vector.
cd ${cuBB_SDK}/5GModel/aerial_mcore
source scripts/setup.sh
./scripts/gen_e2e_ota_tvs.sh
ls -lh GPU_test_input/
cp GPU_test_input/* ${cuBB_SDK}/testVectors/
But we faced error saying “X does not support locale C.UTF-8” and the docker stoped.
aerial@winslab-arc-gnb2:/opt/nvidia/cuBB/testVectors$ cd ${cuBB_SDK}/5GModel/aerial_mcore
aerial@winslab-arc-gnb2:/opt/nvidia/cuBB/5GModel/aerial_mcore$ source scripts/setup.sh
[Aerial Python]aerial@winslab-arc-gnb2:/opt/nvidia/cuBB/5GModel/aerial_mcore$ ./scripts/gen_e2e_ota_tvs.sh
/opt/nvidia/cuBB/5GModel/aerial_mcore/scripts/gen_e2e_ota_tvs.sh starting…
/opt/nvidia/cuBB/5GModel/aerial_mcore/examples /opt/nvidia/cuBB/5GModel/aerial_mcore
Warning: X does not support locale C.UTF-8
Also, when we try to use the test vector in /opt/nvidia/cuBB/testVectors/cuPhyChEstCoeffs.h5, it doesn’t work
Hi @snkim94,
cuPhyChEstCoeffs.h5
is not a test vector. It is a data structure that contains filter coefficients.
What is the OS you are running the aerial_mcore to generate the TVs?
Thanks.
We are running the codes inside the cuBB 23-4 container.
I see the same message “Warning: X does not support locale C.UTF-8” but this is just a warning message. It should not stop the container.
Can you try the following steps to generate the test vectors?
cd ${cuBB_SDK}/5GModel/aerial_mcore/examples
source ../scripts/setup.sh
export REGRESSION_MODE=1
time python3 ./example_5GModel_regression.py allChannels
echo $?
ls -alF GPU_test_input/
du -h GPU_test_input/
Thank you.
We have tried these codes and waited 4~5 hours but there were no results after receiving the following outputs:
Run genCfgTemplate …
Test runSim for DL …
Read config from cfg_template_DL.yaml
Thank you
You should see the following prints as soon as the script starts.
Run genCfgTemplate ...
Test runSim for DL ...
Read config from cfg_template_DL.yaml
Channel: ssb pdcch pdsch csirs prach pucch pusch srs
----------------------------------------------------
Alloc: 1 1 1 1 0 0 0 0
==> PASS
Test runSim for UL ...
Read config from cfg_template_UL.yaml
Channel: ssb pdcch pdsch csirs prach pucch pusch srs
----------------------------------------------------
Alloc: 0 0 0 0 1 1 1 1
UE # 0: PUSCH TB detected
RO # 0: 1 Prmb detected
UE # 0: PUCCH UCI detected
==> PASS
.....
Generating full set of test vectors can take a few hours.
Can you please check your system configuration to see if you have at least 64 GB memory and 300 GB of disk space?
Thank you.
I have been getting this version mismatch issue while running ch_est example with cuPhyEstCoeffs.h5 configuration:
[Aerial Python]aerial@winslab-arc-gnb1:/opt/nvidia/cuBB/build/cuPHY/examples/ch_est$ ./cuphy_ex_ch_est -i ../../../../testVectors/cuPhyChEstCoeffs.h5
./cuphy_ex_ch_est: /usr/local/MATLAB/MATLAB_Runtime/R2023a/sys/os/glnxa64/libstdc++.so.6: version `GLIBCXX_3.4.29' not found (required by ./cuphy_ex_ch_est)
./cuphy_ex_ch_est: /usr/local/MATLAB/MATLAB_Runtime/R2023a/sys/os/glnxa64/libstdc++.so.6: version `GLIBCXX_3.4.29' not found (required by /opt/nvidia/cuBB/build/cuPHY/src/cuphy/libcuphy.so)
./cuphy_ex_ch_est: /usr/local/MATLAB/MATLAB_Runtime/R2023a/sys/os/glnxa64/libstdc++.so.6: version `GLIBCXX_3.4.29' not found (required by /opt/nvidia/cuBB/build/cuPHY/nvlog/libnvlog.so)
./cuphy_ex_ch_est: /usr/local/MATLAB/MATLAB_Runtime/R2023a/sys/os/glnxa64/libstdc++.so.6: version `GLIBCXX_3.4.29' not found (required by /opt/nvidia/cuBB/build/cuPHY/external/fmtlog_flat/libfmtlog-shared.so)
./cuphy_ex_ch_est: /usr/local/MATLAB/MATLAB_Runtime/R2023a/bin/glnxa64/libcurl.so.4: no version information available (required by /usr/lib/x86_64-linux-gnu/libhdf5_serial.so.103)
[Aerial Python]aerial@winslab-arc-gnb1:/opt/nvidia/cuBB/build/cuPHY/examples/ch_est$ strings /usr/local/MATLAB/MATLAB_Runtime/R2023a/sys/os/glnxa64/libstdc++.so.6 | grep GLIB
GLIBCXX_3.4
GLIBCXX_3.4.1
GLIBCXX_3.4.2
GLIBCXX_3.4.3
GLIBCXX_3.4.4
GLIBCXX_3.4.5
GLIBCXX_3.4.6
GLIBCXX_3.4.7
GLIBCXX_3.4.8
GLIBCXX_3.4.9
GLIBCXX_3.4.10
GLIBCXX_3.4.11
GLIBCXX_3.4.12
GLIBCXX_3.4.13
GLIBCXX_3.4.14
GLIBCXX_3.4.15
GLIBCXX_3.4.16
GLIBCXX_3.4.17
GLIBCXX_3.4.18
GLIBCXX_3.4.19
GLIBCXX_3.4.20
GLIBCXX_3.4.21
GLIBCXX_3.4.22
GLIBCXX_3.4.23
GLIBCXX_3.4.24
GLIBCXX_3.4.25
GLIBCXX_3.4.26
GLIBCXX_3.4.27
GLIBCXX_3.4.28
What could be the possible way forward?
Hi @subhams
cuPhyChEstCoeffs.h5 is not a test vector but is a data structure that contains filter coefficients. It is used for cuBB level test.
The command “cuphy_ex_ch_est ~/<tv_name>.h5” is used for channel-est-kernel unit test based on a pre-generated test vector, e.g., TVnr_7201_PUSCH_gNB_CUPHY_s0p0.h5. Please refer to our release documentation for the steps of generating test Vectors.
Instead of running channel-est kernel unit test by using “cuphy_ex_ch_est ~/<tv_name>.h5”, running cuPHY channel pipeline test is recommended. As recommended above by @bkecicioglu, "you can run a cuPHY stand-alone test as explained in our release documentation in Section ‘Running the cuPHY Examples’. you can run the PUSCH pipeline with the command cuPHY/build/examples/pusch_rx_multi_pipe/cuphy_ex_pusch_rx_multi_pipe -i ~/<tv_name>.h5. By adding an additional argument for an output file such as -o debug_sample_dump.h5, the test will write some important PHY layer configuration and data samples at different stages into this file. The channel estimation output will be saved in the data structure Hest0. "
Thanks!