TL;DR
This paper introduces an efficient method for precisely locating devices indoors using 5G signals by jointly estimating the direction and time of arrival of signals. It tackles the challenges of array modeling errors in 5G base stations and the computational complexity of traditional methods, achieving real-time performance with high accuracy.
What Problem Does This Paper Address?
Accurate indoor positioning is crucial for many applications, but GPS signals are unreliable indoors. While 5G technology offers a promising solution due to its wide bandwidth and dense deployment, existing methods for estimating signal direction-of-arrival (DOA) and time-of-arrival (TOA) jointly (JADE) face two major hurdles:
- Array Modeling Errors: Small-scale antenna arrays in 5G picocell base stations (gNBs) are prone to errors in how they model incoming signals, especially direction-dependent ones. These errors significantly degrade positioning accuracy.
- Computational Complexity: Traditional JADE algorithms are computationally intensive, making real-time indoor positioning challenging. This is due to the high dimensionality of the data and the need for complex joint processing.
Methods
The proposed solution, a Joint Angle and Delay Estimation (JADE) scheme, addresses these problems through several key innovations:
- Direction-Dependent Array Error Calibration: The authors first characterize and model array modeling errors as a vector-valued function that captures direction-dependent characteristics. This model is then used to pre-calibrate the ideal steering vector, effectively counteracting these errors during online estimation.
- Decoupled DOA and TOA Estimation: To reduce complexity, the scheme decouples the estimation of TOA and DOA. It first uses the Iterative Adaptive Approach (IAA) to estimate TOA and resolve multipath signals in the time domain. Subsequently, a Conventional Beamformer (CBF) is used to retrieve the DOA of the line-of-sight (LOS) signal.
- Efficient Implementation:
- CFR Denoising and Dimension Reduction: A pre-processing module is introduced that truncates the channel impulse response (CIR) and re-transforms it to the frequency domain. This reduces noise and the dimensionality of the Channel Frequency Response (CFR) matrix.
- FFT-Accelerated TOA Estimation: The IAA-based TOA spectral estimator is accelerated using Fast Fourier Transforms (FFTs), leveraging the structured nature of 5G reference signals.
Datasets
The paper utilizes the following for evaluation:
- Real-measured data: Collected from commercial 5G equipment to analyze array modeling errors.
- Simulated multipath data: Generated based on the proposed signal model (equation 6) to evaluate performance under varying multipath conditions and SNRs.
- 5G indoor channel models: Specifically, 3GPP_38.901_InF_LOS and 3GPP_38.901_Indoor_LOS models are used to simulate realistic indoor environments.
- Field test data: Collected from a real indoor environment (an underground parking lot) using a commercial 5G picocell gNB.
Key Insights and Ideas
- Direction-Dependent Errors are Key: The paper highlights that array modeling errors in 5G picocell gNBs are not uniform but depend on the signal's direction, significantly impacting DOA estimation.
- Calibration is Crucial: Pre-calibrating the steering vector based on measured array errors is essential for accurate DOA estimation in realistic indoor scenarios.
- Decoupling for Efficiency: Separating TOA and DOA estimation, and then using efficient algorithms like IAA for TOA and CBF for DOA, drastically reduces computational load.
- FFT is a Game-Changer: Accelerating the TOA estimation with FFTs makes real-time implementation feasible, achieving near real-time performance.
- 5G Signals are Rich for Localization: The paper demonstrates how the inherent properties of 5G signals (wide bandwidth, reference signals) can be effectively leveraged for precise indoor positioning.
Key Results
- DOA Accuracy Improvement: The proposed method, using calibrated steering vectors, significantly reduces DOA estimation bias, especially at large incident angles (e.g., reducing error from 8.11° to 1.28° at ±60°).
- Real-time Performance: The efficient implementation achieves an average running time of 10.1 ms on a PC for typical configurations, which is orders of magnitude faster than traditional methods like MUSIC-based JADE.
- High Positioning Accuracy: Field tests demonstrated a triangulation positioning accuracy of 0.44 m for 90% of cases using DOAs from just two base stations, meeting 3GPP R17 requirements.
- Robustness: The method shows superior performance compared to existing JADE methods in simulations, especially under challenging multipath conditions and varying SNRs.
Limitations and Future Work
- Limitations:
- The paper primarily focuses on 2D positioning and assumes a single user terminal.
- While the method is adaptable, extensions to 3D positioning and multi-user scenarios would require further development.
- The field test used a specific picocell gNB configuration; performance might vary with different hardware.
- Future Work:
- The authors suggest extending the framework to 3D localization by incorporating elevation angle estimation and using planar arrays.
- Hybrid positioning methods that combine DOA/TOA estimates with tracking frameworks (like EKF, UKF, PF) are mentioned for further performance enhancement.
- The adaptability of the method to other scenarios, such as non-uniformly distributed subcarriers or arbitrary array structures, is discussed as a potential avenue for future research.
Summary updated