An effective way to address non-Gaussian estimation problems with measurement fading
The advancement of wireless communication technologies involves extensive research on the networked systems comprising sensors and processing units. Typically, bandwidth limitation and measurement fading are the most encountered phenomena in wireless transmission. The two, along with other network-induced phenomena like transmission delays, packet dropouts, and signal quantization, must be taken into consideration to ensure accurate state estimation. Among the available tools for solving various state estimation problems, the Kalman filter and its variants, analytical or approximate solutions to the sequential Bayesian filtering problem, are commonly used. Unfortunately, the analytical solutions of the sequential Bayesian filtering do not exist for general non-Gaussian models. This has resulted in numerical solutions such as the particle filter for tracking the non-Gaussian probability density function (PDF) problems.
On the other hand, measurement fading, which describes the distortion of transmitted raw signals, is a major challenge in wireless communication. This can be attributed to the complexity caused by the non-Gaussianity of the fading models, which has remained underexplored despite its significance in networked systems. Currently, hard-decoding and soft-decoding are two schemes used to describe channel fading. The latter does not require a detector and is thus easy to realize. Nonetheless, the state estimation under fading measurements remains a challenge owing to numerous drawbacks that hinder obtaining accurate measurements.
Recently, Dr. Wenshuo Li and Professor Lei Guo from Beihang University, Professor Zidong Wang from Shandong University of Science and Technology, and Professor Yuan Yuan from Northwestern Polytechnic University developed a two-stage particle filtering to address the non-Gaussian filtering problems in systems with fading measurements. Taking into consideration the practical advantages of the cascade structure of the system, the two-stage algorithm was expected to achieve state estimation and measurement recovery simultaneously. Their work is currently published in the research journal, Automatica.
In their approach, the original problem was decomposed into two cascaded subproblems: measurements recovery from the faded ones and state estimation based on the recovered measurements. The decomposition was made possible by decentralization of the particle filter that provided an alternative means of estimating the augmented state of the model. The two subproblems were solved by first- and second-stage particle filters, respectively, using the proposed two-step particle filter algorithm. The feasibility of the proposed algorithm was validated using two examples in which the first case involved tackling the nonlinear filtering, while the second case involved tracking object problem in which the measured signals were distorted using the communication channels. The authors also investigated the relationship between the proposed algorithmand existing schemessuch as brute-force particle filter and decentralized particle filter.
Results showed that the second-stage resampling procedure could be implemented simultaneously and in parallel resulting in a significant reduction in the algorithm execution time. The parallelized architecture further proved suitable for implementing distributed systems resulting in a further improvement in the execution efficiency. Moreover, the two-stage particle filter outperformed the brute force particle filter as it was capable of striking a better balance between the execution time and tracking accuracy, owing to the advantage of the cascaded model structure.
In summary, the research team proposed a two-stage particle filter to solve the non-Gaussian state estimation problems with fading measurements. The algorithm was reportedly suitable for distributed implementations with reduced execution time and improved accuracy and efficiency. Moreover, it outperformed a majority of the existing schemes indicating its potential practical applications. In a statement to Advances in Engineering, Professor Lei Guo admitted that the proposed algorithm could be extended to solve more complicated problems, thus cementing its significance in the research field.
Li, W., Wang, Z., Yuan, Y., & Guo, L. (2020). Two-stage particle filtering for non-Gaussian state estimation with fading measurements. Automatica, 115, 108882.