T-Arshad/POWDER_CoChannel_Protocol_Dataset

POWDER Co-Channel Protocol (PCP) Dataset

768 real-world over-the-air (OTA) IQ captures for multi-label RF fingerprinting under co-channel interference, with heterogeneous waveforms (802.11a Wi‑Fi, 4G LTE, and 5G NR) collected on the POWDER PAWR testbed at the University of Utah by the CREDIT Center, Prairie View A&M University.

Each capture records the superposition of up to six simultaneously transmitting USRP radios on a shared 20 MHz channel at 2.425 GHz. Every .bin IQ file includes a matching .json with receiver settings, per-transmitter labels, gains, and waveform paths.

Dataset Summary

Captures	768 (3 rounds × 4 gain levels × 64 TX on/off permutations)
Files	768 × .bin + 768 × .json (~123 GB total)
IQ format	complex64, 20,000,000 samples per capture (~160 MB/file)
Sample rate	33.33 MS/s
Receiver	USRP X310 , RX gain 25 dB
Transmitters	2 X310 USRP's and 4 B210 USRP's
Protocols	802.11a, 4G LTE, 5G NR

Transmitter gain was swept at four percentages of each radio's maximum gain range (X310: 0–31.5; B210: 0–89). Folder names use the percentage (e.g. Round1_Gain90/)

Gain %	Folder suffix	X310 TX gain	B210 TX gain
60%	Gain60	18.9	53.9
70%	Gain70	22.1	62.9
80%	Gain80	25.2	71.8
90%	Gain90	28.4	80.8

Transmission protocol assignments

Data were collected in 3 rounds. In each round, the six transmitters use a fixed protocol assignment, but the mapping rotates across rounds so each waveform (802.11a, 4G LTE, 5G NR) is broadcast from different radio pairs.

Round	X310_3 & X310_4	B210_1 & B210_2	B210_3 & B210_4
1	5G NR	4G LTE	802.11a (Wi‑Fi)
2	4G LTE	802.11a (Wi‑Fi)	5G NR
3	802.11a (Wi‑Fi)	5G NR	4G LTE

Repository Structure

Each folder Round{R}_Gain{PCT}/ contains 64 paired files:

Round1_Gain90/otax3102node_tx011010_id26.bin
Round1_Gain90/otax3102node_tx011010_id26.json

Filename part	Meaning
otax3102node	Receiver node
tx011010	6-bit TX on/off mask (1 = active)
id26	Capture index in folder (0–63)

Metadata(.json)

Field	Description
rx.usrp	Receiver hardware (X310)
rx.sample_rate	Hz (e.g. 33333000.0)
rx.center_freq	Hz (e.g. 2425000000.0)
rx.gain	Receiver gain (dB)
rx.dtype	complex64
tx[].id	POWDER node ID
tx[].usrp	X310 or B210
tx[].index	Label index 0–5
tx[].state	"0" or "1" (inactive / active)
tx[].gain	Transmitter gain (dB)
tx[].signal_path	Source waveform used in collection
Datetime	Capture timestamp
samples	Number of IQ samples

Example (Round1_Gain90/otax3102node_tx011010_id26.json):

{
  "rx": {"usrp": "X310", "sample_rate": "33333000.0", "center_freq": "2425000000.0", "gain": "25.0", "dtype": "complex64"},
  "tx": [
    {"id": "otax3103node", "usrp": "X310", "index": 0, "state": "0", "gain": 28.4, "signal_path": "Samples/5G.bin"},
    {"id": "otax3104node", "usrp": "X310", "index": 1, "state": "1", "gain": 28.4, "signal_path": "Samples/5G.bin"},
    {"id": "otanuc1", "usrp": "B210", "index": 2, "state": "0", "gain": 80.8, "signal_path": "Samples/4G.bin"},
    {"id": "otanuc2", "usrp": "B210", "index": 3, "state": "1", "gain": 80.8, "signal_path": "Samples/4G.bin"},
    {"id": "otanuc3", "usrp": "B210", "index": 4, "state": "1", "gain": 80.8, "signal_path": "Samples/wifi.bin"},
    {"id": "otanuc4", "usrp": "B210", "index": 5, "state": "0", "gain": 80.8, "signal_path": "Samples/wifi.bin"}
  ],
  "Datetime": "2026-03-12 19:34:35.916275",
  "samples": "20000000.0"
}

Usage

Download

from huggingface_hub import snapshot_download

data_dir = snapshot_download(
    repo_id="T-Arshad/POWDER_CoChannel_Protocol_Dataset",
    repo_type="dataset",
    allow_patterns=["Round1_Gain90/*"], local_dir = "path/to/local_dir"
)

Load IQ Samples

import numpy as np

def load_signal(file_path ="POWDER_Cochannel_Protocol/Round1_Gain90/otax3102node_tx011010_id26.bin)
    """
    - file_path: path the signal file
    """
    samples = np.fromfile(filepath, dtype = np.complex64)
    return samples

def sample_signal(file_path = "POWDER_Cochannel_Protocol/Round1_Gain90/otax3102node_tx011010_id26.bin", sample_len = 1024, sample_offset = 10000):
    """
    - file_path: path the signal file
    - sample_len: number of timesteps to sample
    - sample_offset: timestep to start sampling from
    """
    samples = np.memmap(file_path, dtype = np.complex64, mode = "r", offset=samples_offset*np.dtype(np.complex64).itemsize, shape = (sample_len,))
    samples = np.array(samples)
    return IQ_samples

Citation

@inproceedings{abdulquddoos2026calibrated,
  title={Calibrated {RF}-Fingerprinting Under Interference With Heterogeneous Transmission Protocols},
  author={Abdul-Quddoos, Tariq and Li, Xiangfang and Qian, Lijun},
  booktitle={IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC)},
  year={2026},
  organization={IEEE}
}

License

This dataset is released under the MIT License.