Application of active noise control based on neural network to vehi- cle's engine sound Donghyeon Lee 1 Mechnical Engineering, Hanyang University, Korea 222, Wangsimni-ro, Seongdong-gu, Seoul Narae Kim 2 Mechnical Engineering, Hanyang University, Korea 222, Wangsimni-ro, Seongdong-gu, Seoul Junhong Park 3 Mechnical Engineering, Hanyang University, Korea 222, Wangsimni-ro, Seongdong-gu, Seoul

ABSTRACT Active noise control(ANC) is a particularly effective system for reducing low-frequency noise to com- pensate passive noise control. With the recent development of digital signal processor(DSP) perfor- mance, ANC has the potential to be developed with various algorithms. Accordingly, several ANC algorithms using various controlloer such as artificial neural network(ANN) are being proposed. In nonlinear system; at many practical applications, the ANC algorithm using a neural network gets more reduction performance compared to the linear ANC. In this study, the methology proposed neural network based FxLMS algorithm to reduce noise for non-linear system by predicting time series data for near future. This proposed algorithm is applied to reduce the engine noise of vehicle to construct silent inner environment and verify the performance by below.

1. INTRODUCTION

ANC (Active Noise Control) is a method of removing noise by generating a control sound[1]. Since ANC was first proposed, it has continuously received a lot of attention and many related studies have been conducted. However, the performance of the computational processor did not keep up until the real-time ANC system at that time came to the actual experiment. Therefore, in recent years, when considerable scientific and technological progress has been made, as the computing power of the DSP board has improved, an attempt has been made to apply it to a place where various noises are severe. ANC developed a feed-forward method that acquires the signal of a noise source in advance before noise removal and an adaptive filter that responds to environmental changes in order to apply it to changing environmental conditions and noise characteristics. The most frequently utilized adaptive algorithm when performing ANC is filtered-x least mean square (FxLMS). The reference microphone

1 alajju3@hanyang.ac.kr

2 dogmecome@hanyang.ac.kr

3 parkj@hanyang.ac.kr

worm 2022

measures a reference noise x(n), and the control speaker generates an interfering wave signal y(n) through an adaptive filter W(z). The noise and control signals are summed for each primary path P(z) and secondary path S(z) and the noise residue is measured by an error microphone. The FxLMS algorithm performs an adaptation process on the filter W(z) that produces y(n) to minimize the noise residue e(n).

2. STRUCTURE OF RECURRENT NEURAL NETWORKS AND LONG SHORT-TERM MEMORY

RNN is one of the frequently used algorithms for processing sequence data. The RNN layer accepts time-varying input data x and outputs a value y. Because the nodes in the RNN layer are directly connected to each other, the RNN can take into account the sequence input data and generate an output. Long short-term memory (LSTM) is sometimes used as a special kind of RNN. RNN has the disadvantage that the importance of the initial sequence decreases as the ordinal sequence of the se- quence data increases. However, LSTM retain the importance of previous sequences better than RNN by input, output, and forgetting gates. These features allow LSTM to handle long-term data predic- tion. 3. ACTIVE NOISE CONTROL SYSTEM USING NEURAL NETWORK ALGORITHM

In this study, we propose an applied FxLMS algorithm using a neural network to predict a sound signal in the very close future. Unlike the existing FxLMS, which uses the estimated secondary path and the update function using an adaptive filter, the proposed algorithm is replaced with one neural network controller block. The input data of the block is a reference signal close to noise and signal y that operates the control speaker is output.

worm 2022

Figure 1 : Block diagram of ANC system consist with neural network controller 4. CONCLUSIONS

In this paper, an advanced FxLMS ANC system using a neural network is proposed and simulated. As a result, it was confirmed that the proposed algorithm has better cancellation performance in var- ious situations than the conventional FxLMS ANC.

x(n) dq) *L Pe) POT aw Neural Network |_*) y(n) Controller Siz)

5. REFERENCES

1. Kuo, S. M., & Morgan, D. R. Active noise control: a tutorial review. Proceedings of the

IEEE , 87(6) , 943-973 (1999). 2. Zhang, H., & Wang, D. Deep ANC: A deep learning approach to active noise control. Neural

Networks , 141 , 1-10 (2021).

worm 2022