TL;DR
This paper introduces and compares two Bluetooth Low Energy (BLE) based indoor localization methods: Received Signal Strength Indicator (RSSI) and Multi-Carrier Phase Difference (MCPD). The authors demonstrate that MCPD offers significantly higher accuracy (0.52m average error) compared to RSSI (1.39m average error) for indoor positioning, making it a more promising technology for applications like hospital asset tracking.
What Problem Does This Paper Address?
Indoor Positioning Systems (IPS) are becoming increasingly critical in complex environments like hospitals, where accurately locating people and assets can save lives and improve efficiency. While GPS is effective outdoors, it fails indoors. Various technologies like Wi-Fi, UWB, and RFID have been explored, but each has drawbacks such as high power consumption, limited range, or cost. Bluetooth Low Energy (BLE) presents an attractive alternative due to its low power consumption and widespread adoption in modern devices. However, traditional BLE localization methods, primarily relying on RSSI, suffer from significant inaccuracies due to environmental factors like signal reflections and multipath fading. This paper addresses the need for a more accurate, low-cost, and energy-efficient BLE-based IPS by investigating and comparing the performance of RSSI with a newer technique, MCPD.
Methods
The core of this research lies in the implementation and comparison of two BLE ranging technologies:
- RSSI (Received Signal Strength Indicator): This is a well-established method that measures the power of the incoming signal. The paper uses a logarithmic distance loss model to estimate distance based on RSSI values.
- MCPD (Multi-Carrier Phase Difference): This newer technique leverages the direction-finding capabilities introduced in BLE 5.2. By measuring the phase difference of incoming signals across multiple antennas and at different frequencies, MCPD can determine distance with higher precision. The paper utilizes the Nordic Distance Toolbox and its IFFT implementation for MCPD.
Both methods are integrated into a custom-designed hardware and software system built around the Nordic nRF5340 SoC, chosen for its low power consumption and dual-core architecture. The localization algorithm employed is Weighted k-Nearest Neighbors (wkNN), a supervised learning algorithm that uses a "fingerprint" of known beacon signal strengths/phases at various locations to estimate the position of a mobile device. The wkNN algorithm is chosen for its ability to run on resource-constrained microcontrollers and its effectiveness in improving localization accuracy over basic kNN. The paper evaluates wkNN using both Euclidean and Chebyshev norms and explores the impact of varying the number of nearest neighbors (k).
Datasets
The paper does not introduce a new benchmark dataset. Instead, it describes an experimental setup within a 7.3m x 8.9m room. Training data was collected by placing six BLE beacons around the room and recording their RSSI and MCPD values at nine predefined training positions. Validation data was collected at five randomly chosen positions, distinct from the training positions, to provide a realistic evaluation of the algorithm's performance.
Key Insights and Ideas
- MCPD's Superiority: The most significant insight is the clear superiority of MCPD over RSSI for indoor localization accuracy. This is attributed to MCPD's ability to overcome the inherent environmental noise and multipath effects that plague RSSI measurements.
- Low-Power Edge Computing: The use of the nRF5340 SoC highlights the feasibility of implementing sophisticated localization algorithms like wkNN directly on low-power edge devices, enabling real-time processing without the need for constant cloud communication.
- Impact of Averaging: While averaging RSSI measurements can improve accuracy, it comes at the cost of increased power consumption and latency. MCPD, with its inherently lower variance, achieves high accuracy even with single measurements, allowing for drastic reductions in power consumption.
- Beacon Subset Influence: The selection of beacons significantly impacts localization accuracy. Even with a small number of beacons (three), MCPD can achieve sub-meter accuracy, and careful selection can yield better results than using more beacons suboptimally.
Key Results
- Accuracy: The MCPD-based approach achieved an average error of 0.50 m in a 7.3m x 8.9m room, while the RSSI-based approach had an average error of 1.39 m.
- wkNN Performance: Using k=3 and the Chebyshev norm for MCPD, an average error of 0.501 m was achieved with 6 beacons after averaging 15 measurements. For RSSI, using k=3 and the Euclidean norm, the average error was 1.369 m under the same conditions.
- Single Measurement Accuracy: Even with single measurements, MCPD achieved an average error of 0.52 m (k=3, Chebyshev norm), demonstrating its robustness.
- Power Consumption: The nodes consume approximately 0.185 mW for RSSI and 1.850 mW for MCPD when measuring every 100 ms. By increasing the MCPD measurement interval to 1.5 s, power consumption drops to 0.122 mW, leading to significantly longer battery life (139 days vs. 92 days for RSSI on a 110 mAh battery).
Limitations and Future Work
- Experimental Environment: The experiments were conducted in a single, relatively small room. The performance of both methods in larger, more complex hospital environments with varied layouts and materials needs further investigation.
- Training Data Dependency: The wkNN algorithm relies on pre-collected training data. Any significant changes in the environment or beacon placement would require re-training the system.
- Comparison to Prior Work: The authors note that their RSSI results are slightly worse than a previous study [17], attributing this to differences in validation methodology (same position vs. different positions), the number of measurements used for averaging (15 vs. 1000), and the hardware used.
- Future Work: The paper suggests that the increased availability of BLE Direction Finding hardware and the energy-efficient nature of MCPD make it a strong candidate for future low-cost, accurate indoor localization systems, particularly in medical applications. Further research could explore optimizing beacon placement, developing adaptive algorithms, and testing in more diverse real-world scenarios.
Summary updated