Region-Based Distributed Estimation Using Quantized Data

Abstract

SUMMARYIn this paper, we consider distributed estimation where the measurement at each of the distributed sensor nodes is quantized before being transmitted to a fusion node which produces an estimate of the parameter of interest. Since each quantized measurement can be linked to a region where the parameter is found, aggregating the information obtained from multiple nodes corresponds to generating intersections between the regions. Thus, we develop estimation algorithms that seek to find the intersection region with the maximum likelihood rather than the parameter itself. Specifically, we propose two practical techniques that facilitate fast search with significantly reduced complexity and apply the proposed techniques to a system where an acoustic amplitude sensor model is employed at each node for source localization. Our simulation results show that our proposed algorithms achieve good performance with reasonable complexity as compared with the minimum mean squared error (MMSE) and the maximum likelihood (ML) estimators. key words: distributed estimation, maximum a posteriori (MAP), quantization, source localization, sensor networks

Efficient design of quantisers for distributed estimation

“…The description of the estimation based on non‐regular quantised data is beyond this work. See [7] for the details.…”

“…This compression is made possible by constructing for each of the samples in the partitions the interval in which the possible codeword for the sample can be located for perfect encoding and by finding the codeword that resides in as many intervals as possible. Since the algorithm in [6] sends a single codeword (not a codeword set) at each node due to the impracticality of using a large number of codewords for estimation, compression of partitions enables us to transmit a codeword set of multiple codewords at each node for estimation, which was shown to further improve the estimation accuracy [7]. The proposed algorithm is evaluated for source localisation in acoustic amplitude sensor networks to show through simulations that it achieves a significantly improved the performance over traditional designs and also performs well with about reduced codeword sets as compared with the novel design techniques [3, 6].…”

Improving data quality using a deep learning network

“…IoT data has time-series characteristics because it periodically collects data from sensors. Using this characteristic, Recurrent Neural Network such as LSTM is used [16][17][18][19][20][21][22][23][24][25][26][27][28].…”