A multi-source fusion algorithm for high-accuracy signal reconstruction of vehicle interior noise on passenger ear-sides. (May 2019)

• Record Type: Journal Article
• Authors: Yang, Dongpo; Wang, Xiaolan; Wang, Yansong; Guo, Hui; Liu, Ningning; Li, Wenwu
• Is Part Of: Applied acoustics. Volume 148(2019)
• Page Start: 75
• Page End: 85
• Publication Date: 2019-05
• Subjects: Signal decomposition optimisation -- Component fitness calculation -- Ear-side noise reconstruction -- Energy-extreme division; Acoustical engineering -- Periodicals; Periodicals; 620.2
• DOI: 10.1016/j.apacoust.2018.12.017
• Languages: English
• ISSNs: 0003-682X

Abstract

Highlights: A new reconstruction method to facilitate active noise control. The optimal fitness value optimizes the network model, improve the reconstruction accuracy. A new method is proposed for reducing the non-stationary noise signals. The proposed reconstruction method can provide high reconstruction accuracy. Abstract: The use of reconstructed noise signal as a primary reference signal is critical to active noise control in passenger ear-sides under high-speed conditions. A signal decomposition optimisation-based BP neural network for ear-side noise reconstruction (DBENR) algorithm is proposed. This algorithm contains the processes of signal decomposition optimisation (SDO), component fitness calculation (CFC) and ear-side noise reconstruction (ENR). The SDO method is divided into two steps. Firstly, multi-source noise signals are decomposed into a finite number of intrinsic mode function (IMF) components by empirical mode decomposition. Secondly, according to a proposed energy-extreme division method, the IMFs are reconstructed into three signal components, namely, high-, intermediate- and low-frequency components. CFC calculates the fitness of a component in each forward training process of a signal reconstruction BP network to obtain the optimal fitness value. The ENR model is obtained by regarding the optimal fitness values as the initial weights and the thresholds of the signal reconstruction BP network and training. The effectiveness of the proposed DBENR algorithm is validated using five noise signal sources collected from a vehicle. Compared with the signal reconstruction BP algorithm, the proposed algorithm is superior in reconstruction accuracy.