ANCシステムにおけるオンライン2次経路とフィードバック経路モデリングのための補助ノイズ電力スケジューリングに関する研究

摘要

The idea of cancelling the acoustic noise by generating an anti-noise signal is very fascinating, and was first proposed by P. Lueg in 1936. In feedforward active noise control (ANC) systems, the anti-noise signal is generated with the help of reference and error microphones, an adaptive filtered-x-LMS (FxLMS) algorithm based ANC filter, and an electro-acoustic path named as the secondary path. For stable operation of ANC systems, the FxLMS algorithm needs an estimate of the secondary path. The anti-noise signal generated by the loudspeaker (part of secondary path) causes interference with the reference microphone signal. This interference is due to the presence of electro-acoustic path, named as feedback path, between the loudspeaker and the reference microphone. It is required to neutralize the effect of this feedback path, and hence an estimate of the feedback path is required. For online modeling of the secondary and feedback paths, an additional auxiliary noise is injected. This auxiliary noise contributes to the residual error, and thus degrades the noise-reduction-performance (NRP) of ANC system. In order to improve the NRP, a gain scheduling strategy is used to vary the variance of the injected auxiliary noise. The purpose of the gain scheduling is that when the model estimates of the secondary and the feedback paths are far from the actual unknown paths, auxiliary noise with large variance is injected. Once the model estimates are closer to the actual unknown paths, the variance of auxiliary noise is reduced to a small value. In this way, on one hand the gain scheduling can help us to achieve the required model estimates of secondary and feedback paths, and on the other hand to improve the NRP at the steady-state. In this thesis, we discuss the two most important issues, i.e., 1) online secondary path modeling (OSPM), and 2) online feedback path modeling and neutralization (FBPMN) with gain scheduling. In chapter 1, the basic underlying physical principle and configurations of active noise control (ANC) systems are explained. The application of the basic building block of an ANC system i.e. An adaptive filter, in different system identification scenarios is discussed. The most popular adaptive algorithm for ANC system, i.e., FxLMS algorithm is derived for the general secondary path. A brief overview is given for the two fundamental issues in ANC systems, i.e., 1) OSPM and 2) online FBPMN. The use of optimal excitation signal, i.e., Perfect sweep signals for system identification is described. In chapter 2, the existing methods for OSPM without gain scheduling, where the auxiliary noise with fixed variance is used in all operating conditions, are discussed. In this chapter a simplified structure for OSPM with the modified FxLMS (MFxLMS) adaptive algorithm is proposed. The advantage of the simplified structure is that it reduces the computational complexity of the MFxLMS algorithm based OSPM without having any compromise on the performance of ANC system. In chapter 3, the existing methods for OSPM with gain scheduling are discussed. The drawbacks with the existing gain scheduling strategies are highlighted, and some new gain scheduling strategies are proposed to improve the modeling accuracy of SPM filter and the NRP of an ANC system. In existing methods, the gain is varied based on the power of residual error signal which carries information only about the convergence status of ANC system. In the Proposed methods the gain is varied based on the power of error signal of SPM filter. This is more desirable way of controlling the gain because the power of error signal of SPM filter carries information about the convergence status of both the ANC system and the SPM filter. The performance comparison is carried out through the simulation results. In chapter 4, the second most important issue associated with the feedforward configuration of ANC system, i.e., the issue of online FBPMN is deal with. In the first part, the existing methods for online FBPMN without gain scheduling are discussed. A new structure is proposed for online FBPMN without gain scheduling. The performance of the existing methods is compare with the proposed method through the simulation results. In the new structure the good features from the existing structures are combined together. The predictor is used in the new structure to remove the predictable interference term from the error signal of adaptive FBPMN filter. In addition to this, the action of FBPM filter and the FBPN filter is combined into a single FBPMN filter. The advantage of the new structure over the existing structures is that it can better neutralize the effect of feedback coupling on the input signal of ANC filter, thus improves the convergence of ANC system. In the second part, a gain scheduling strategy is proposed to improve the NRP of ANC system. In addition to this, a self-tuned ANP scheduling strategy with matching step-size for FBPMN filter is also proposed that requires no tuning parameters and further improves the NRP of ANC systems. In chapter 5, the concluding remarks and some future research directions are given.