Combining Locally Weighted Scatterplot Smooth and Least Squares Predication to Reduce TOA Ranging Error
Shihong Duan, Yanzhong Liu, Jie He, University of Science and Technology Beijing, China
Location: Big Sur
Alternate Number 4
Common filter algorithm for improving TOA range accuracy originally are designed to smooth White Gaussian noise, and not effective for correcting range error caused by serious multi-path scenario. In this paper, an enhancement algorithm combing locally weighted scatterplot smooth (RLOWESS) and least squares predication is proposed to reduce TOA ranging error in real-time. Mobile distance sequences affected by multipath are measured in diverse test environment with IEEE 802-15-4A CSS standard RF chip. Rlowess algorithm is verified to be more effective to achieve higher range accuracy firstly than Kalman Filter scheme, and the enhancement algorithm base on Rlowess is described to improve the real-time performance of correcting streaming range data . The algorithm is validated based on datasets obtained in 6 different test scenarios and the theoretical derivation of time complexity is detailed. The statistical data results and the theoretical analysis show that the enhanced Rlowess algorithm has a better performance in real time accuracy comparing with Rlowess algorithm and Kalman Filter algorithm.
The measurement system employs the IEEE 802-15-4A CSS standard radio chip NanoLOC to perform the TOA ranging function. According to the receiving strength of Direct Path (DP) signal, the channel is divided into DDP (Dominant Direct Path), NDDP (Non-dominant Direct Path), UDP (Undetected Direct Path). So, the measurement was conducted in different scenario of DDP, NDDP and UDP. During the test, the base station node and the target node are fixed on the tripod to avoid the occlusion and movement of the human body. A large number of TOA ranging samples was obtained and statistically analyzed.
The direct path can be blocked by large metallic objects and large concrete walls or its first peak used for time of flight measurements may shift due to multipath components arriving close to the direct path. Multipath and UDP condition would cause huge ranging error and significantly affects the localization accuracy. For matching problem, the accuracy is mainly depended on the signal noise ratio. Here the ranging error plays the role of noise. Thus, the huge ranging error becomes the outlier of the signal and decreases the matching accuracy. Therefore, specific approaches of reducing the effect of ranging error should also be taken into account.
Our research work mainly focused on:
(1) In order to avoid assumption of channel signal noise is white gaussian noise. Our former has proposed optimized Kalman filter algorithm, Combination of Channel Classi?cation and Kalman Filter (CC-KF), which effectively improve the accuracy of range correction. Firstly, this paper verified the Robust locally weighted regression and smoothing scatterplots (Rlowess) algorithm. The result showed that Rlowess is better than CC-KF in range accuracy, but the use of stream data window caused the poor real-time, which is formally analyzed the time complexity of Rlowess.
(2) The paper proposed enhanced Rlowess scheme, which introduced the least squares method to predict the subsequent distance data, can significantly improve the correction on time and ensure the distance correction accuracy. The enhanced Rlowess scheme framework including TOA range error model, measurement data smoothing and weighted matching.
Enhanced Rlowess scheme is validated against measured range dataset stream. The good performance with high positioning accuracy with average error of 0.467m and real-time with ms-level delay.
The paper proposed the enhanced algorithm to correct streaming range data for multi-path scenario, where channel signal noise is not White Gaussian noise. Enhanced Rlowess algorithm introduced least squares predication to reduce operation time and weighted matching to ensure the correction accuracy.