Synchronized Four-Channel IQ Waveform Dataset Acquired with ULA, Nested, and Coprime Antenna Arrays

Description

This dataset contains real synchronized four-channel in-phase and quadrature (IQ) waveforms acquired using three four-element antenna-array geometries: coprime, nested, and uniform linear array (ULA). The raw multichannel IQ measurements preserve spatial, amplitude, phase, and inter-channel information that can support a broad range of telecommunications, array-signal-processing, and data-driven applications. The measurements were collected using a software-defined radio experimental platform with a USRP-2900 transmitter and a four-channel USRP X410 receiver. The transmitted signal was received simultaneously by four antennas under controlled spatial configurations and known transmitter positions. The waveforms are stored as NumPy files using the complex64 data type, with 256,000 complex IQ samples per receiver channel. The released dataset contains 950 multichannel acquisitions represented by 3,800 IQ files. The coprime subset contains 364 acquisitions and 1,456 files covering 17 physical angles; the nested subset contains 339 acquisitions and 1,356 files covering 16 physical angles; and the ULA subset contains 247 acquisitions and 988 files covering 27 physical angles. Each acquisition consists of four synchronized files corresponding to receiver channels ch0–ch3. Experimental position labels are linked to their corresponding physical angles through the accompanying metadata/angle_mapping.csv file. Before release, the waveform collection was subjected to signal-quality and integrity checks, while the retained files remain available as raw synchronized IQ waveforms without imposing a predefined feature representation or estimation method. The dataset can therefore be reused for multiple research purposes, including array signal processing, channel characterization, spatial correlation analysis, covariance analysis, calibration studies, spectral analysis, feature extraction, machine-learning and deep-learning methods, transfer learning, signal classification, localization, direction finding, and direction-of-arrival estimation. The associated technical validation provides an example of how the multichannel measurements can be processed and analyzed, while the availability of the raw IQ waveforms allows users to develop and evaluate alternative processing pipelines.

Steps to reproduce

The dataset can be accessed using standard Python tools. NumPy is required to load the IQ waveform files (.npy), while a spreadsheet application or a data-analysis library such as pandas can be used to read the accompanying angle_mapping.csv metadata file. Each acquisition is represented by four synchronized files corresponding to receiver channels ch0, ch1, ch2, and ch3. Within each array-geometry subset, files belonging to the same acquisition share the same acquisition date, timestamp, and experimental position label. These four files should therefore be grouped using these common identifiers. Example loading procedure: import numpy as np iq = np.load("IQ_ch0_<date>_<time>_<label>.npy") Each loaded array contains 256,000 complex64 IQ samples from one receiver channel. The experimental position label contained in the filename can be associated with its corresponding physical Direction-of-Arrival (DoA) angle using metadata/angle_mapping.csv. The accompanying README.md describes the repository folder structure, filename convention, dataset composition, and basic usage. More detailed information about the experimental acquisition setup, signal-quality assessment, preprocessing procedure, and technical-validation workflow is provided in the associated article.

Institutions

• Instituto Tecnológico Metropolitano

  Antioquia, Medellín

Categories

Licence