The paper explores the Channel Sounding (CS) method introduced in the Bluetooth Low Energy 6 (BLE6) specification with a focus on its evaluation under dificult adverse radio conditions in an utility basement employing BLE6 hardware. Two measurement series were conducted, one for reference and calibration in open ground, and one in the utility basement of an university building. Multiple distance estimation algorithms are assessed. In addition to the standard algorithms implemented in the evaluation board, namely Inverse Fast Fourier Transform (IFFT), Slope, and Round Trip Time (RTT), some enhancements for the IFFT-based method have been implememted and tailored for improved robustness and precision. Experimental results indicate that Channel Sounding remains efective even in complex and obstructed environments, with the proposed algorithm outperforming the default implementation.
phone application scenario, but as BLE6 has not yet found its way into real smartphones, it is based
on simulation. A car key fob scenario is investigated by [
To support the discussion of results, the employed algorithms are reviewed in greater detail: The RTT algorithm is already present in the vendor-supplied firmware and directly interfaces with the hardware to determine the signal round-trip time for each individual transmission. In contrast to CS algorithms, it would work already with one single channel.
The Slope– and IFFT algorithms are also part of the default firmware. Both operate on IQ values representing the measured phase diferences for the channels, which are provided by the hardware/ifrmware. The Slope algorithm estimates the distance by evaluating the slope of the unwrapped phase diferences over the frequency, which is ideally a line. The IFFT algorithm performs an inverse fourier transformation (IFFT) on the phase function. The index of the IFFTs peak value is proportional to the signal’s propagation distance. As the index is an integer, the accuracy is limited. Therefore a more accurate value is obtained by parabolic intepolation of values around the index of the maximum value.
In addition to these 3 default algorithms implemented in the vendor firmware, the authors proposeand implemented a modified IFFT algorithm “IFFT2”. This algorithm currently works ofline. For this purpose, the firmware was modified to transmit raw IQ values alongside the distance estimates generated by the firmware via a serial interface to the connected PC in real time. The proposed ofline algorithm “IFFT2”, which is analyzed in this paper, also relies on an inverse FFT. Distinctive features of this approach include: • Neglecting of the signal amplitude by normalization of the IQ values. • Interpolation of missing data points on the unit circle, taking into account the direction of the phase rotation.
• Zeroing of missing data points which are at the beginning- or end of the frequency range The final peak is determined by applying a parabolic approximation around the maximum of the transformed signal, as also performed by the original IFFT algorithm.
The Experiments are employing Nordic nRF54L15 Development Kits (DK), which are equipped with the corresponding BLE6 Bluetooth System-on-Chip (SoC). This chip supports BLE6 Channel Sounding. An SDK provided by the manufacturer is available for the development kits, including example code for CS devices. These examples already provide distance estimates for both CS and Round Trip Time (RTT). Two CS algorithms are implemented by default: Inverse Fast Fourier Transform (IFFT) and Slope. For our experiments, we modified the initiator firmware to extract also the raw data which constists of the quadrature (IQ) values from the 75 acitve measurement channels.
The DK modules were mounted on tripods, with the feedpoint of the vertically oriented antenna positioned at a height of 1.20 meters above ground level. The RF modules use a PCB antenna ( /4 monopole), connected to the chip via a 5 mm short transmission line. The reflector does not require any I/O connections and starts operation immediately upon battery connection. The initiator outputs measurement data via a serial interface using a USB converter, and a connected laptop is used for data acquisition.
A complete channel scan is performed approximately every 0.5 seconds. The IQ values from each scan are averaged already in the firmware. Every 5 seconds, an aggregated dataset (comprising typically 10 averaged scans) is sent to the PC via the serial interface, comprising averaged IQ data, along with the firmware-computed distances.
The actual measurement procedure for each position was as follows: • Adjusting of the devices to the forseen positions • Performing of 50 seconds of measurement which gives typically 10 measurement results. • Recording of measurement label and the set of results to allow for ofline evaluation of mean value and standard derivate for each position.
To validate and calibrate the setup and to evaluate the algorithms under more or less ideal conditions, initial measurements were conducted in open field conditions (see also Fig. 2). This was carried out on a grassy area in a public park (see photograph). Ground truth distances were established using a surveying tape. The ground truth accuracy, estimated by repeated bidirectional tape measurements, was determined to be well below 5 cm. Measurements were taken over a distance range of 1..40 meters, with the initiator remaining stationary while the responder was repositioned for each measurement.
The measurements in the basement were conducted in the same manner (see photo). The building dates back to around 1910. The walls are partly made of brick and partly of concrete, while the ceilings are entirely constructed from concrete. The measurement points and corresponding distances are marked on the building plan. The metal doors in the corridors were left open during the measurements, as performance through closed fire doors would have been very poor.
Table 1 presents a summary of the results of the outdoor measurements. For each of the four methods ITTF2, IFFT, Slope and RTT the mean distance error (Delta) and the standard derivative (SD) for 10 successive measurements is given. Table 3 (at the end of the paper) presents a detailled view on the measurements for each position. The Fourier-based methods demonstrate high accuracy in outdoor environments. Notably, for the proposed IFFT2 method, the standard derivative reveals exceptionally low variation within individual distance measurements typically within a few centimeters.
The results were calibrated with an ofset: each method exhibited a constant distance ofset, sometimes on the order of meters, which was subtracted for the phase-based methods. These ofsets (calculated from the mean deviation from ground truth) were 0.93 m for IFFT2, 0.46 m for IFFT, 1.91 m for Slope and 3.78 m for RTT. Such a constant ofset could, for example, be caused by antenna cable length; however, in our setup, the antennas were mounted directly on the chip. The antennas themselves, especially near their resonance frequency, might exhibit a frequency-dependent phase response that is influenced by their quality factor. There is already a compensation factor considerered in the firmware, but it did not perfectly match the given antenna configuration. In contrast to the very reliable and accurate results outdoors, the situation in the basement is significantly diferent. Table 2 lists the corresponding indoor measurement results. The pairs of numbers in the first column indicate the measurement points between which the distance was recorded. Two types of ground truth distances are provided: Manhattan and Euclidean. The Delta values were computed relative to the Manhattan distance. The mean standard deviation is 1.61 m for IFFT2, 2.05 m for IFFT, 1.21 m for Slope and 1.25 m for RTT.
The performance of CS is noticeably degraded in the utility basement corridors, likely due to multipath propagation and signal shadowing. Positions 5a, 6, 8a, and 12 could not be reached from the initiator positions 4 and 1 used in the measurements.
CDF outdoor
CDF indoor IFFT Slope RTT IFFT2
IFFT Slope RTT IFFT2
The CDF plots (Fig. 3) confirm the discussion: Outdoors, reasonable accuracies have been reached: Looking for example at the 80. percentile, CS with the FFT methods show good performance. IFFT2 is below 0.2 meters, IFFT about 0.3 m. Slope can not hold up and is above one meter. Even RTT, which has relaxed requirements on channel availability, reaches 0.6 m.
Indoors, trhe situation is worse. The IFFT variants reach 80% only close to 5 meters of error. IFFT2 error is below 9 m for 95% of the measurents, others at or above 11 m.
This work investigated the performance of Bluetooth Channel Sounding (CS) and Round Trip Time (RTT) based distance estimation methods under both ideal and challenging radio conditions, employing stock BLE6 hardware. Particular attention was given to an underground utility basement environment, which introduces propagation efects like multipath and signal shadowing.
Outdoor experiments demonstrate that Fourier-based approaches achieve high precision, with intrameasurement deviations (StdDev) in the range of a few centimeters. Slight systematic errors were observed as function of the distance, which might be caused by constant phase ofsets i.e. originating from antenna characteristics. These ofsets were compensated to some extent during post-processing.
In contrast, the indoor measurements in the utility basement revealed significant performance degradation. The complex structure of the building with concrete ceilings and metal doors seemed to severely impact the signal propagation. Certain locations were unreachable, and the standard deviations of the distance errors increased substantially. However, the performance is still comparable to WiFi. Employing more reflectors at suitable places might help to overcome mentioned indoor issues.
Optimization of the IFFT algorithm lead to the proposed IFFT2 algorithm: The signal amplitude was normalized such that only the phase was considered. Missing data points were interpolated on the unit circle considerung also data points at the beginning- and end of the frequency range. IFFT2 outperformed IFFT under most conditions.
While reporting issues in complex environments, BLE6 CS showed a robust and, depending on the setting, very high ranging accuracy. When available at smartphones, we will propably see very interesting and helpful applications.
During the preparation of this work, the authors used GPT-4 in order to perform Grammar and spelling check as well as for generating python code for data processing e.g. file parsing, averaging, conversions etc. After using these tools, the authors reviewed and edited the content as needed and take full responsibility for the publication’s content.