Abstract
In the field of indoor localization, ultra-wideband (UWB) technology is no longer dispensable. The market demands that the UWB hardware has to be cheap, precise, and accurate. These requirements lead to the popularity of the DecaWave UWB system. The great majority of the publications about this system deal with the correction of the signal power, hardware delay, or clock drift. It has traditionally been assumed that this error only appears at the beginning of the operation and is caused by the warm-up process of the crystal. In this article, we show that the warm-up error is influenced by the same error source as the signal power. To our knowledge, no scientific publication has explicitly examined the warm-up error before. This work aims to close this gap and, moreover, to present a solution, which does not require any external measuring equipment and only has to be carried out once. It is shown that the empirically obtained warm-up correction curve increases the accuracy for the two-way ranging significantly.
Introduction
Ultra-Wideband (UWB) technology has become a cornerstone for high-precision indoor localization, powering applications from industrial robotics to consumer electronics. Despite its widespread adoption, a persistent challenge remains: the "warm-up error." When UWB chips, such as the popular DecaWave DW1000, are initially powered on, their distance measurements exhibit a significant drift that typically takes about 15 to 30 minutes to stabilize.
Historically, this drift was attributed solely to the physical heating of the crystal oscillator, leading to the common practice of allowing devices a "warm-up" period before use. This research investigates the underlying mechanisms of this error, demonstrating that it is not merely a thermal issue but is deeply linked to signal power and frequency offsets.
Methodology
The study utilized DecaWave EVK1000 hardware in a controlled indoor environment. To minimize errors from clock drift between unsynchronized devices, the researchers employed the Alternative Double-Sided Two-Way Ranging (AltDS-TWR) protocol.
The experimental design focused on isolating the warm-up error by systematically varying three parameters:
- Distance: Measurements were taken at fixed intervals (1m, 2m, 3m) to check for distance-dependent scaling.
- Transmitter Gain: Signal strength was adjusted (0 dB to 12 dB) to observe its impact on timestamping.
- Frequency Trimming: The crystal oscillator's "trim" was manipulated to simulate various frequency offsets between the tag and the reference station.
By logging these variables alongside ranging data over 30-minute sessions, the authors developed an empirical correction curve—a multi-dimensional mapping of received signal power and frequency difference to observed distance error.
Key Insights
The research revealed that the warm-up error is primarily driven by the same factors as signal power error. As the device warms up, the crystal's frequency shifts. This shift affects the chip's internal Phase-Locked Loop (PLL) and the Leading Edge Detection (LDE) algorithm.
The DW1000's LDE algorithm, which identifies the "first path" of a radio pulse, is sensitive to the pulse's shape. Changes in frequency offset and signal amplitude alter this shape, leading to deterministic errors in timestamping. Crucially, because these errors are predictable based on measurable internal states (power and frequency), self-calibration is possible. A device can sweep its own gain and trim settings to generate a correction curve without external ground-truth equipment.
Results and Impact
Implementing the proposed correction curve yielded a dramatic improvement in ranging performance:
- Sub-Centimeter Accuracy: While standard DecaWave systems typically achieve ~10 cm accuracy, the corrected system achieved accuracy better than 1 cm.
- Error Reduction: In validation scenarios with small frequency differences, the mean absolute error dropped from 1.08 cm to just 0.39 cm.
- Instant Readiness: The correction eliminates the need for a 15-minute thermal stabilization period, allowing for high-precision localization immediately upon power-up.
Limitations and Future Work
Despite the success of the correction model, some implementation challenges remain. The DW1000's crystal trim has a relatively coarse step size (~1.5 ppm), making it difficult to sample the error curve near zero-offset points. Additionally, the non-linear relationship between frequency, power, and error requires a sophisticated calibration routine for each individual device. Future research will explore the impact of extreme ambient temperature fluctuations and voltage variations on these correction curves.
Summary updated