WorksEngineering
RadarML: A Multi-Channel Over-the-Air Dataset of Common Radar Waveforms for ES Evaluation
Abstract
This paper introduces "RadarML", a publicly available dataset of modulated radar pulses recorded over-the-air across multiple semi-independent receiver channels. The dataset was generated using the UCL ARESTOR platform which is a Radio Frequency System on a Chip (RFSoC) based Software Defined Radio (SDR). The dataset contains a library of around two million waveforms with varying centre frequencies, bandwidths, pulse widths and modulations, with 1,000 examples of each signal variation per receiver channel. The goal of this dataset is to enable comparative analysis of radar modulation classification techniques on real data and stimulate research into multi-channel signal detection and modulation classification methods.
Introduction
As utilisation of the Radio Frequency (RF) spectrum continues to grow, the density of emissions within any given frequency band increases. For Electronic Surveillance (ES) systems, the ability to reliably detect and classify radar signals has therefore become increasingly important. Modern radar systems, particularly software defined radars are waveform agile and capable of adapting parameters such as centre frequency, bandwidth, pulse duration, power and modulation on a pulse-to-pulse basis. This agility, combined with the proliferation of complex modulation schemes, creates a significantly more challenging electromagnetic environment for ES systems.
There are many publications that deal with the problem of signal detection and modulation classification of radar signals. However, many of these papers use different definitions and implementations of pulse modulation types making comparative analysis of techniques difficult, and very few utilise data that was captured experimentally. As far as the authors of this paper are aware, none use multiple co-located receiver channels simultaneously capturing the same radar pulses.
To address this gap in available data, we present "RadarML", a publicly available dataset of both frequency and phase modulated radar pulses that were captured experimentally over-the-air across multiple Analogue-To-Digital (ADC) channels using an RFSoC based system.
Dataset Description
The dataset contains a broad range of representative radar waveforms with parameters as described below in Table 1. These waveforms vary in centre frequency, time duration or bandwidth (if applicable). Many of the waveforms are derived from open literature [1], however some more exotic examples such as non-linear chirps are included, which are more challenging than traditional radar waveforms for ES techniques. For each permutation of waveform there were 1,000 pulses transmitted.
| Modulation | Baseband Centre Frequencies (MHz) | Bandwidths (MHz) | Pulse Widths (s) |
| CW | -30, -10, 10, 30 | N/A | 10, 50, 100 |
| Linear FM | -30, -10, 0, 10, 30 | +/- 10, 20, 40, 80 | 10, 50, 100 |
| Exponential FM | -30, -10, 0, 10, 30 | +/- 10, 20, 40, 80 | 10, 50, 100 |
| Quadratic FM | -30, -10, 0, 10, 30 | +/- 10, 20, 40, 80 | 10, 50, 100 |
| Tangential FM | -30, -10, 0, 10, 30 | +/- 10, 20, 40, 80 | 10, 50, 100 |
| 4 CP-FSK | -30, -10, 0, 10, 30 | 10, 20, 40, 80 | 10, 50, 100 |
| Barker 5 | -30, -10, 10, 30 | N/A | 10, 50, 100 |
| Barker 7 | -30, -10, 10, 30 | N/A | 10, 50, 100 |
| Barker 11 | -30, -10, 10, 30 | N/A | 10, 50, 100 |
| Barker 13 | -30, -10, 10, 30 | N/A | 10, 50, 100 |
| Barker 5 11 | -30, -10, 10, 30 | N/A | 10, 50, 100 |
| Barker 5 13 | -30, -10, 10, 30 | N/A | 10, 50, 100 |
Table 1. Synthetic Waveform Library
Some example spectrograms generated from the dataset are shown in Figure 1.

Figure 1. Spectrograms generated from the raw data, showing examples of a linear chirp, Barker5x13, and non-linear chirp pulse modulations respectively.
For every waveform combination, the dataset contains a NumPy file with dimensions [3, 1000, 33600], the dimension definitions are summarised in Table 2.
| Dimension | Size | Description |
| 1 | 3 | ADC channels |
| 2 | 1000 | Number of pulses acquired per waveform configuration |
| 3 | 33600 | Interleaved I and Q samples per acquisition |
Table 2. Structure of each waveform dataset file
The samples for each pulse are provided unprocessed, as digitised by the ADCs, in the np.int16 format with no synthetic noise added. To evaluate ES algorithms across a range of signal to noise ratios (SNRs), additive white Gaussian noise (AWGN) can be added to the baseline samples at load time.
Along with the data, a code repository is provided (see Section IV) which includes a PyTorch dataloader for viewing the raw data, and a wrapper dataloader that supports augmentations in SNR, frequency, phase and time to improve generalisation (as shown in Figure 5). A spectrogram transform and a baseline VGG13-style classifier are also included. Two jupyter notebooks demonstrate their use:

Figure 5. A plot showing how frequency, phase and time augmentations affect the data by plotting the FFT and in-phase and magnitude component against time.
dataloader_demo.ipynb - Provides a step-by-step demo of the dataloaders. It covers loading and viewing the raw pulses, comparing the same physical pulse across ADCs, balancing the number of pulses per class, applying each augmentation, and seeding for repeatable results.
vgg13_training_demo.ipynb - Provides an end-to-end example that trains and evaluates the VGG13 classifier on spectrograms from a single ADC. Training uses pulses degraded from 30 dB to 0 dB SNR, with random frequency, phase and time shifts. The trained model is evaluated from 15 dB to −15 dB SNR, with per-SNR confusion matrices and macro F1 plotted against SNR. An example confusion matrix at 0 dB SNR generated from this script is shown in Figure 2.

Figure 2. Confusion matrix at 0 dB SNR for a VGG13 Model trained on ADC0.
Waveform Definitions
The equations that describe the instantaneous frequency of the frequency modulated chirps in the dataset are given below:
where
The derivation of the instantaneous frequency of the tangential chirp is given in [2] by Widyantara et al.
where ,
In half of the frequency-modulated examples, the start () and stop () frequencies were swapped to generate the corresponding "down" chirps.
For the continuous-phase frequency-shift-keyed (CP-FSK) waveforms, four frequencies are chosen at random that satisfy the bandwidth and centre-frequency constraints provided. The pulse is split equally into sub-pulses which are each assigned one of the frequency values. To ensure spectral purity and minimise abrupt phase changes, the phase of each sub-pulse transitions continuously from the previous sub-pulse, maintaining phase continuity across frequency shifts.
In the binary Barker code implementation, the pulse is again split into sub-pulses, each of which is given a code value of "1" or "-1". Whenever the code changes between sub-pulses, the phase is shifted by 180 degrees.
Some compound Barker codes are also present in the dataset. These are generated by taking the Kronecker product of two different Barker sequences to create a longer code, which is split into sub-pulses and phase shifted using the same method described in the previous paragraph.
Hardware Description
The dataset introduced in this paper was transmitted and received over-the-air in the 2.4 GHz ISM (Industrial, Scientific and Medical) band in an anechoic chamber using the UCL "ARESTOR" system shown in Figure 3, which is an SDR based on an RFSoC architecture. ARESTOR has enabled experiments in active and passive radar [3], Integrated Sensing and Communications (ISAC) [4] [5] [6] as well as ES [7] [8] [9]. More details about the ARESTOR system and its hardware are provided in previous papers authored by the UCL Radar Group [10] [11].

Figure 3. The UCL ARESTOR System.
To generate the RadarML dataset, a single ARESTOR node acting as both the transmitter and receiver(s) was used to avoid the requirement of synchronising multiple nodes. A single DAC (Digital-to-Analogue) channel was used as the transmitter and 3 semi-independent ADC channels on different RF-Tiles were used for receive. Whilst the PLLs (Phase-Locked Loops) on each RF-Tile are locked to the same reference clock, no phase synchronisation (Multi-Tile Synchronisation) is performed between them, therefore the ADC channels are considered mesochronous.
The dataset waveforms were generated at an intermediate centre frequency of 240 MHz at a sample rate of 720 MS/s before interpolation by a factor of 8 by the DAC tile on the RFSoC to achieve a final sample rate of 5.76 GS/s. The waveforms are then digitally up-converted by mixing with a complex NCO (Numerically Controlled Oscillator) operating at a frequency of +2.2 GHz to enable transmission within the ISM band at a centre-frequency of 2.44 GHz.
The receivers on the RFSoC operate at a sample rate of 3.84 GS/s, which is then decimated by a factor of 32 in two stages (8× by the RF Tile and 4× by further FPGA processing) to achieve a final sample rate of 120 MS/s. As the desired
signal is sampled in the 2nd Nyquist zone, spectral inversion occurs. The actual intermediate frequency after sampling is given in (5). The digitised signals are then digitally down converted using an NCO operating at -1.4 GHz, which mixes the centre-frequency of our target signals to baseband.
The receiver(s) are configured to record for 140 μs per acquisition, providing sufficient temporal coverage for all pulse durations in the dataset. Acquisitions are scheduled at 1 kHz to align with the transmitter’s PRF of 1 kHz.
To ensure each transmitted pulse is centred within its corresponding receiver acquisition window, a variable pre-transmission delay is applied based on the pulse duration:
Each receive channel uses the same model of antenna (RFSPACE UWB3) and filter (Mini Circuits VBF-2435+), however, a different amplifier is used with each to introduce further variation into the digitised data. Attenuators are used to ensure each channel has roughly the same SNR regardless of the gain and noise factor of the amplifier chosen. All the information on the physical hardware attached to each DAC/ADC channel used in this setup are described below in Table 3.
| Port | Antenna | Amplifier | Filter | Attenuation |
| DAC0 | L-COM HG2418DPP | Mini-Circuits ZX60-6013E-S+ | Mini-Circuits VBF-2435+ | N/A |
| ADC0 | RFSPACE UWB3 | Mini-Circuits ZX60-6013E-S+ | Mini-Circuits VBF-2435+ | N/A |
| ADC2 | RFSPACE UWB3 | Mini-Circuits ZJL-4GH+ | Mini-Circuits VBF-2435+ | 1 dB |
| ADC4 | RFSPACE UWB3 | Mini-Circuits ZX60-83LN12+ | Mini-Circuits VBF-2435+ | 13 dB |
Table 3. DAC/ADC Channel Configurations
The physical setup of the chamber is shown in Figure 4.

Figure 4. Receiver setup in the anechoic chamber, showing the 3x co-located receive antennas and the transmit antenna.
Data and Code Availability
The dataset and some source code to get started are publicly available, the links for which are provided below:
Dataset: The full dataset as described in this paper is available at https://doi.org/10.5522/04/30752767.
Example Processing Scripts: The pre-processing and plotting scripts plus the example neural network as described in section II are available at https://github.com/UCLRadarGroup/Radar_ML.
The dataset is released under the CC BY 4.0 license and the code is released under the GPL-3.0 license.
Conclusions
This paper has presented RadarML, a multi-channel dataset of representative radar waveforms captured over-the-air to support the evaluation of electronic surveillance and modulation classification techniques. The data was generated using the UCL ARESTOR platform in order to systematically produce large quantities of pulses over a broad range of modulations and pulse parameters. ARESTOR operated as both the radar as well as the ES receiver(s) in this case.
It is hoped that RadarML will enable reproducible comparison of algorithms and stimulate further research into multi-channel signal processing and fusion approaches.
Acknowledgments
This dataset was generated under an iCASE PhD Studentship funded by the Engineering and Physical Sciences Research Council (EPSRC) and the Defence Science and Technology Laboratory (Dstl). Prof. Ritchie’s time is supported by The Royal Academy of Engineering and Leonardo UK under the Research Chairs and Senior Research Fellowships programme.
References
- [1]P. E. Pace, Detecting and classifying low probability of intercept radar. Artech House, 2004.
- [2]M. R. Widyantara, Sugihartono, F. Y. Suratman, S. Widodo, and P. Daud, “Analysis of non linear frequency modulation (nlfm) waveforms for pulse compression radar,” Jurnal Elektronika dan Telekomunikasi, 2018.DOI
- [3]M. Ritchie, N. Peters, and C. Horne, “Joint active passive sensing using a radio frequency system-on-a-chip based sensor,” in 2022 23rd International Radar Symposium (IRS), pp. 130–135, 2022.DOI
- [4]M. Temiz, C. Horne, N. J. Peters, M. A. Ritchie, and C. Masouros, “An experimental study of radar-centric transmission for integrated sensing and communications,” IEEE Transactions on Microwave Theory and Techniques, vol. 71, no. 7, pp. 3203–3216, 2023.DOI
- [5]M. Temiz, N. J. Peters, C. Horne, M. A. Ritchie, and C. Masouros, “Radar-centric isac through index modulation: Over-the-air experimentation and trade-offs,” in 2023 IEEE Radar Conference (RadarConf23), pp. 1–6, 2023.DOI
- [6]R. S. C. Winter, A. Catherall, C. Masouros, and M. Ritchie, “Impact and mitigation of index modulation on radar quality for a fmcw-based communication system,” in 2024 IEEE International Workshop on Technologies for Defense and Security (TechDefense), pp. 485–489, 2024.
- [7]C. Horne, N. J. Peters, and M. A. Ritchie, “Classification of lora signals with real-time validation using the xilinx radio frequency system-on-chip,” IEEE Access, vol. 11, pp. 26211–26223, 2023.DOI
- [8]R. White, C. Horne, and M. A. Ritchie, “An empirical evaluation of pdw extraction techniques against weak lpi pulsed radar,” in 2024 21st European Radar Conference (EuRAD), pp. 260–263, 2024.
- [9]R. White, R. S. C. Winter, C. Horne, and M. A. Ritchie, “Evaluation of yolo for automatic radar waveform detection and classification in congested eme,” in 2025 IEEE International Radar Conference (RADAR), pp. 1–6, 2025.
- [10]N. Peters, C. Horne, and M. A. Ritchie, “ARESTOR: A multi-role RF sensor based on the Xilinx RFSoC,” in 2021 18th European Radar Conference (EuRAD), pp. 102–105, 2022.DOI
- [11]N. J. Peters, C. P. Horne, A. D. Amiri, P. Beasley, and M. A. Ritchie, “Modular multi-channel rfsoc system expansion and array design,” in 2023 IEEE Radar Conference (RadarConf23), pp. 1–4, 2023.DOI