Detailed Description
This artifact provides datasets containing raw Channel Frequency Response (CFR) measurements collected from uplink Sounding Reference Signals (SRS), the corresponding FAPI metadata, and the slot of fronthaul (FH) IQ samples where SRS was transmitted by a 5G User Equipment (UE).
Both indoor and outdoor measurements are made available for two scenarios. First, in the case where data was captured while the channel state was kept static. Second, in the case, where data is captured while introducing dynamics to the channel via a moving scatterer. The scatter is a moving person in the indoor scenario and a UAV in the outdoor scenario.
The platform used for measurements is the NVIDIA Aerial Testbed installed at the Eurecom campus. The respective RUs for indoor and outdoor deployments are Benetel RAN550 and RAN650 COTS units. A Quectel Unit is acting as the UE in the measurements transmitting the signal. The MIMO configuration used is 2TX x 4RX.
The dataset is designed to support research in wireless localization, detection, and AI/ML-based radio analytics. It provides MIMO CFR estimates together with related metadata. Additionally, one slot of FH IQ samples used for estimation is provided.
Deployment Description
In all settings a band n77 deployment with center frequency 3.9504GHz was used. The system bandwidth is 100MHz and there are N_ACTIVE_RB = 272 active resource blocks. In the fronthaul data there are N_SC = 3,276 subcarriers. The TDD pattern used is DDDDDDSUUU. The MIMO configuration used is 2TX x 4RX.
Outdoor
The Benetel RAN650 O-RU is place on the roof of the Eurecom building and connected to 4 ports of the Amphenol 8 port antenna. The Quectel UE is placed on the terrace of the Eurecom forum.
Indoor
The Benetel RAN550 O-RU is mounted on the ceiling of the -3rd floor in the Eurecom building and transmits using integrated antennas. The Quectel UE is mounted on the ceiling of the same hallway with direct line of sight to the RU.
Dataset Structure
The dataset contains 4 Parquet files for both static and dynamic scenarios. Each file is generated from a synchronized join of the srs, srs_hest, and fh database tables. The resulting tables store the raw fronthaul IQ grids, channel estimates, and over-the-air telemetry parameters over time.
T is the number of recorded time-steps (rows) in the Parquet file.
K is the length of the flattened and interleaved CFR matrix. It has size N_TX x N_RX x N_ACTIVE_RB x 2.
M is the length of the flattened raw FH data. It has size N_RX x N_SYMB x N_SC x 2
{
CellId: [T×1 uint16] # Physical cell identifier
TsTaiNs: [T×1 datetime64] # TAI timestamp of the reception (nanosecond precision)
SFN: [T×1 uint16] # System Frame Number (0-1023)
Slot: [T×1 uint16] # Slot number within the frame
rnti: [T×1 uint16] # Radio Network Temporary Identifier (UE ID)
nRxAnt: [T×1 uint16] # Number of receive antennas at the gNB
widebandSnr: [T×1 float32] # Estimated wideband Signal-to-Noise Ratio (dB)
toaUs: [T×1 float32] # Time of Arrival (ToA) in microseconds
startSym: [T×1 uint8] # Starting OFDM symbol index for the SRS allocation
nSyms: [T×1 uint8] # Number of allocated OFDM symbols
combSize: [T×1 uint8] # SRS comb size (subcarrier spacing factor)
combOffset: [T×1 uint8] # Subcarrier offset for the comb grid
hestSize: [T×1 uint32] # Length of the channel estimate array
hestData: [T×K int16] # Flattened I/Q array of the Channel Frequency Response (CFR)
fhData: [T×M int16] # Flattened I/Q array of the raw Fronthaul SRS IQ grid
rbSnrData: [T×N_ACTIVE_RB float32]# Array of per-PRB Signal-to-Noise Ratio values
... [20+ additional configuration fields including cyclicShift, bandwidthIdx, etc.]
}
Loading Data:
To load the data we recommend to first look at the following repository containing analysis scripts: https://gitlab.eurecom.fr/mayersho/aerial-srs-campaign-data-lake
