RFFI-Temporal: A Long-Term RF Fingerprinting Dataset for Temporal Drift Analysis

Dataset Overview

RF-Predict is a longitudinal dataset designed to study the long-term stability of Radio Frequency Fingerprint Identification (RFFI). RFFI systems typically degrade over time due to environmental changes, hardware aging, and battery decay. This dataset enables systematic investigation of these effects.

The dataset contains packet-aligned complex baseband IQ recordings collected from 30 TI CC13xx IoT devices transmitting 2-GFSK signals at 866 MHz. Data was captured using three software-defined radio receivers. Each packet is stored with pre- and post-packet margins, preserving transmitter startup and shutdown transients for transition-based analysis.

Each packet includes metadata such as transmitter ID, sequence number, internal temperature, battery level, capture timestamp, and internal clock (RTC). The 30 devices are split into 15 battery-powered and 15 DC-powered units, with transmission intervals ranging from 15 seconds to 24 hours. Data collection spans two phases: Phase 1 provides baseline data with uniform 6-second intervals over approximately 45 hours, while Phase 2 covers 67 days of long-term evaluation with device-specific transmission intervals.

The released dataset consists of approximately 6,360,482 valid packets totaling 130.75 GB, organized into 6 HDF5 files, with accompanying Phase 2 weather files. All dataset timestamps are reported in UTC; during data collection, local time corresponded to UTC+3.

This dataset is intended for long-term RFFI stability analysis, temporal drift studies, hardware aging analysis, and device identification research under realistic long-duration conditions.

Experimental Setup

The experimental setup consists of 30 transmitters and 3 receivers arranged in an indoor environment. The transmitters are placed in a 6×5 grid with 13 cm spacing between units. The receivers are positioned 128 cm from the nearest transmitter group, with 4 cm spacing between receiver antennas.

The transmitters are TI CC13xx-based IoT sensor nodes operating at 866 MHz with 2-GFSK modulation. Of the 30 devices, 15 are battery-powered (odd-numbered: T01, T03, ..., T29) and 15 are DC-powered (even-numbered: T02, T04, ..., T30). Each battery/DC pair shares the same transmission interval. Transmission intervals range from 15 seconds to 24 hours.

Three software-defined radio receivers capture transmitted packets simultaneously: two Fairwaves XTRX receivers (R01, R02) and one Ettus USRP B200 Mini receiver (R03). All receivers are tuned to 866 MHz with a sample rate of 400 kHz and operate continuously throughout the collection period. Receivers operate independently; minor timestamp offsets of up to 5 seconds may occur between receivers.

Packet Structure

Each packet is 23 bytes long and contains a preamble, sync word, payload length, sequence number, device ID, RTC timestamp, temperature, battery level, and CRC. At a 400 kHz sample rate with 8 samples per symbol, each packet corresponds to 1472 complex IQ samples. Each packet is stored with an additional 400 pre-packet samples and 800 post-packet samples to preserve startup and shutdown transients.

Collection Phases

Phase 1 (Initial Phase - Baseline): all transmitters sent packets at a fixed 6-second interval.  
Note: A temporary reception outage was observed on receiver R01 during Phase 1 (approximately 2025-10-08 21:25 UTC to 2025-10-09 06:02 UTC).

Phase 2 (Interval-Based Long-Term Transmission): after Phase 1, all devices were restarted and configured with transmitter-specific intervals. Phase 2 data collection started around 2025-10-09 19:20 UTC. The earliest valid packet was observed at 2025-10-09 19:20:29 UTC. The battery-powered transmitter T01 stopped transmitting on 2025-12-13 02:00:22 UTC. The dataset is bounded at 2025-12-15 06:00:00 UTC.

Weather Data (Temperature and Humidity): Weather data was recorded in two separate files containing ambient temperature and relative humidity measurements of the office environment. Data collection was introduced during Phase 2, approximately one week after its start. A new record is added whenever a change in measurement is observed.

Dataset Statistics

Across both phases, the dataset contains approximately 6,360,482 valid packets. Phase 1 includes 2,159,475 received packets out of 2,423,607 expected packets. Phase 2 includes 4,201,007 received packets out of 4,231,899 expected packets.

Receiver overlap statistics are as follows.  
In Phase 1, 80.5% of packets were received by all three receivers, 19.4% by two receivers, and 0.1% by one receiver. These statistics include a short receiver-specific outage affecting R01.  
In Phase 2, 98.5% of packets were received by all three receivers, 1.4% by two receivers, and approximately 0.1% by one receiver.

File Format and Usage

The dataset is distributed as HDF5 files. A README file should accompany the record and describe file organization, metadata fields, loading instructions, and citation guidance.

Availability

This Zenodo record is currently restricted. It will become publicly accessible once the relevant paper has been published.

Note: This dataset was created as part of ongoing research at GOHM Electronics, and this record will be updated with the related publication when it becomes available.