FR3 coherent multiband monostatic channel measurements

The dataset includes 475 over-the-air Channel Frequency Response (CFR) measurements obtained with a multiband monostatic transceiver spanning the frequency range 6-22 GHz. The transceiver sweeps the full frequency range, transmitting OFDM pilot signals with a bandwidth of 500 MHz or 1 GHz with different carrier frequencies. The dataset consists of 200 calibration measurements using a loopback cable (non over-the-air), 150 single-target over-the-air measurements, and 125 over-the-air multitarget measurements. We consider three target types: a calibrated corner reflector, a metal plate, and humans. We collect our measurements in a laboratory environment where we cover walls, furniture, and the support of the targets with panels made of RF absorbing material. Measurement system We utilize a Software Defined Radio (SDR) platform in monostatic configuration, using 1 transmit and 1 receive chains. The SDR includes the following components: Digital baseband RFSoC: A Xilinx Radio Frequency System on a Chip (RFSoC) 4x2 Kit equipped with a ZU48DR processor is used for baseband signal processing, including digital-to-analog (and vice-versa) conversion. The board is capable of generating Radio Frequency (RF) signals up to 6 GHz, and therefore cannot directly generate FR3 signals. It is instead utilized to generate an Intermediate Frequency (IF) signal centered at 1 GHz that is up/down converted in a later stage. RF transceiver Pi-Radio board: The IF signal generated by the RFSoC 4 × 2 Kit is up/down converted by a Pi-Radio TRX board, which translates the fixed-frequency IF to a configurable frequency in the 6-24 GHz range, enabling the frequency sweep. The up and down conversions are performed coherently, ensuring consistent phase measurements. Vivaldi wideband antenna: The measurements are performed over a bandwidth of over 15 GHz centered around 14 GHz, which constitutes a fractional bandwidth greater than 100%. Such a fractional bandwidth is difficult to achieve with standard antennas such as patches or dipoles, which usually provide a percentage fractional bandwidth in the order of units to a few tens. To address this, we use a wideband Vivaldi antenna which, thanks to its exponentially tapered structure, can operate across the whole 6-24 GHz band. Frequency sweep To configure the RFSoC and Pi-Radio board, we use he code available at https://github.com/ali-rasteh/RFSoC_SDR. To obtain coherent CFR estimates over the bandwidth 6-22 GHz, we use the Pi-Radio FR3 board to implement a frequency sweep across the full bandwidth. This is done by transmitting random OFDM pilot signals from a 4-Quadrature Amplitude Modulation (QAM) modulation over a configurable bandwidth centered around a configurable carrier frequency The carrier frequency is changed in subsequent OFDM symbols. The switching between different carrier frequencies requires tswitch = 10 ms in the Pi-Radio board, so this is the minimum timing between OFDM symbols that can be configured in our measurements.In our experiments, we use two different bandwidth values 0.5 or 1 GHz, depending on the specific experiment. We select the carrier frequencies to obtain a contiguous setof CFR estimates. When using 0.5 GHz, we select the set {6.5, 7, . . . , 22} GHz, with 32 subbands, while using 1 GHz we select {6.5, 7.5, . . . , 21.5} GHz, with 15 subbands.The different cardinality of the two sets has the important consequence that the total duration of the frequency sweep using 0.5 GHz is 32tswitch = 320 ms, while using1 GHz it is 15tswitch = 150 ms. For this reason, we use 1 GHz in the measurements involving human targets since a shorter total measurement time is preferable to avoidincoherence due to small involuntary movements of the person. Considered targets Corner reflector: We use a corner-cube reflector with 15 cm side length. The corner reflector is calibrated for radar applications. Metal plate: We use a flat metal plate of dimensions 25 × 10 cm. The thickness of the plate is a few millimeters. Static human: Human subjects are instructed to stand as still as possible in front of the measurement device. To mitigate the impact of respiration on the phasemeasurements, they are asked to hold their breath for 2-3 seconds during the data collection, which lasts 150 ms for the full frequency sweep. Dataset information Information about each test, the employed bandwidth, the target types and locations can be found in file measurements_info.xlsx. CFR measurements are provided in .npz format according to the following naming convention test_09_07_2025_TEST_XXX.npz where XXX is the test number referenced in the first column in measurements_info.xlsx. Different test numbers correspond to different targets, and experiments. Each .npz file contains a field "txtd" with the transmitted signal in the time domain and 32 or 15 files containing the received signals (depending on whether 0.5 or 1 GHz are used). In the script data_loader_zenodo.py we provide a minimal example of how to load the CFR for each subband and estimate the CIR. Reference If you use this dataset in your work, please cite our associated paper @article{pegoraro2025toward, title={Toward Multiband Sensing in FR3: Frequency Anisotropy Characterization and Non-Contiguous Bands Aggregation Algorithms}, author={Pegoraro, Jacopo and Ventura, Gianmaria and Tagliaferri, Dario and Mezzavilla, Marco and Bedin, Andrea and Rossi, Michele and Widmer, Joerg}, journal={arXiv preprint arXiv:2510.03787}, year={2025} }本数据集包含475组空口信道频率响应（Channel Frequency Response, CFR）测量数据，由覆盖6~22 GHz频段的多频段单静态收发机采集得到。该收发机对全频段进行扫频操作，以不同载波频率发射带宽为500 MHz或1 GHz的正交频分复用（Orthogonal Frequency Division Multiplexing, OFDM）导频信号。 数据集包含200组采用环回电缆的校准测量数据（非空口场景）、150组单目标空口测量数据以及125组多目标空口测量数据。本次实验涵盖三类目标：经过校准的角反射器、金属平板以及人体目标。所有测量均在实验室环境中完成，我们使用射频吸波材料制成的吸波面板覆盖了实验室墙面、家具以及目标支架。 ## 测量系统 本系统采用单静态配置的软件定义无线电（Software Defined Radio, SDR）平台，配备1条发射链路与1条接收链路。该SDR平台包含以下组件： 1. 数字基带射频片上系统（Radio Frequency System on a Chip, RFSoC）：采用搭载ZU48DR处理器的Xilinx RFSoC 4×2套件，用于基带信号处理，包括数模/模数转换。该板卡仅能生成最高6 GHz的射频（Radio Frequency, RF）信号，无法直接生成FR3频段信号，因此仅用于生成中心频率为1 GHz的中频（Intermediate Frequency, IF）信号，后续再通过上下变频处理至目标频段。 2. 射频收发机Pi-Radio板卡：由RFSoC 4×2套件生成的中频信号通过Pi-Radio TRX板卡进行上下变频，将固定中频转换为6~24 GHz范围内的可配置频率，从而支持频段扫频。该板卡的上下变频过程采用相干方式，确保相位测量的一致性。 3. 维瓦尔第宽带天线：本次测量的有效带宽超过15 GHz，中心频率约为14 GHz，相对带宽大于100%。常规天线如贴片天线、偶极子天线的相对带宽通常仅为数个百分点至数十个百分点，难以实现如此高的相对带宽。为此，我们采用了维瓦尔第宽带天线，其采用指数渐变结构，可覆盖6~24 GHz全频段。 ## 频段扫频机制 针对RFSoC与Pi-Radio板卡的配置，我们使用了开源代码库https://github.com/ali-rasteh/RFSoC_SDR中的代码。 为在6~22 GHz频段内获取相干的CFR估计值，我们通过Pi-Radio FR3板卡实现全频段扫频。具体方式为：以4正交幅度调制（4-Quadrature Amplitude Modulation, 4-QAM）方式发射随机OFDM导频信号，该信号的带宽与载波频率均可配置，且在后续OFDM符号中切换载波频率。Pi-Radio板卡在不同载波频率间切换的耗时为tswitch=10 ms，这也是本次测量中可配置的OFDM符号间最小间隔。在实验中，我们根据具体场景选用500 MHz或1 GHz的信号带宽。 我们通过选择载波频率以获取连续的CFR估计值：当使用500 MHz带宽时，选取{6.5,7,…,22} GHz的载波频率，共包含32个子带；当使用1 GHz带宽时，选取{6.5,7.5,…,21.5} GHz的载波频率，共包含15个子带。由于两组子带数量不同，500 MHz带宽下的总扫频时长为32×tswitch=320 ms，而1 GHz带宽下的总扫频时长为15×tswitch=150 ms。基于此，针对人体目标的测量我们选用1 GHz带宽，因为更短的总测量时长可有效避免因人体轻微无意识移动导致的信号相干性损失。 ## 待测目标类型 1. 角反射器：采用边长为15 cm的直角角反射器，该反射器经过雷达应用校准。 2. 金属平板：采用尺寸为25×10 cm的平面金属板，板厚仅为数毫米。 3. 人体目标：要求受试人员在测量装置前尽可能保持静止。为缓解呼吸对相位测量的影响，在总时长150 ms的全频段扫频数据采集过程中，受试人员需屏息2~3秒。 ## 数据集说明 每一次测试的相关信息、所用带宽、目标类型与位置均可在`measurements_info.xlsx`文件中查阅。CFR测量数据以.npz格式存储，命名规则为`test_09_07_2025_TEST_XXX.npz`，其中`XXX`对应`measurements_info.xlsx`第一列中的测试编号，不同测试编号对应不同的目标与实验场景。 每个.npz文件包含一个名为`txtd`的字段，存储时域发射信号，同时包含32或15个接收信号文件（取决于所用带宽为500 MHz还是1 GHz）。我们在`data_loader_zenodo.py`脚本中提供了加载各子带CFR并估计信道冲激响应（Channel Impulse Response, CIR）的极简示例。 ## 引用说明 若您在研究中使用本数据集，请引用以下论文： @article{pegoraro2025toward, title={Toward Multiband Sensing in FR3: Frequency Anisotropy Characterization and Non-Contiguous Bands Aggregation Algorithms}, author={Pegoraro, Jacopo and Ventura, Gianmaria and Tagliaferri, Dario and Mezzavilla, Marco and Bedin, Andrea and Rossi, Michele and Widmer, Joerg}, journal={arXiv preprint arXiv:2510.03787}, year={2025} }