Proceedings, APSIPA Annual Summit and Conference 2018 12-15 November 2018, Hawaii Effectiveness of Active Noise Control System for Nonstationary Noise in Consideration of Psychoacoustic Properties Rina Hasegaway, Hiroto Yamashita and Yoshinobu Kajikawaz Kansai University, Osaka, Japan y E-mail: [email protected] z E-mail: [email protected] Abstract—The conventional Active noise control (ANC) system and examine its effectiveness for nonstationary noise through obtains the greatest noise reduction around the error microphone the results of a simulation and a subjective evaluation. and minimizes the mean square error signal. However, the This paper is organized as follows. In Section II, the ANC impression of auditory sensation may sometimes be less than the numerical value when a person actually listens to noise system considering psychoacoustic properties is explained. with ANC. This is because of the complicated characteristic The effectiveness of an A-weighting filter and ITU-R 468 of psychoacoustic properties. The noise control filter optimized noise weighting filter is examined for nonstationary noise via based on the mean square error (conventional ANC) does not simulation results in Section III. Furthermore, the results of a necessarily give optimal results for the human auditory system. subjectivity evaluation are given to demonstrate the significant To mitigate this problem, ANC systems considering psychoacous- tic properties have been proposed and their effectiveness has difference between conventional ANC and ANC based on also been demonstrated for magnetic resonance imaging (MRI) noise-weighting filters in Section IV. Conclusions are given noise and synthesized random noise. In this ANC system, noise in Section V. weighting is incorporated into the conventional ANC structure. In this paper, we examine the effectiveness of the ANC system II. ANC SYSTEM CONSIDERING PSYCHOACOUSTIC considering psychoacoustic properties based on A weighting and PROPERTIES ITU-R 468 noise weighting for nonstationary noise through some experimental results and a subjective evaluation. A. Psychoacoustic Properties The hearing sense has various characteristics such as mask- I. INTRODUCTION ing, loudness and pitch. We focus on loudness in this paper. Loudness is a subjective perception, which is defined as 1 sone Active noise control (ANC) is one of the techniques based when the sound pressure level is 40 dB at the frequency 1 kHz. on the superposition used to reduce unwanted acoustic noise[1- Then, the sound with n times the percieved loudness is defined 4]. Conventionally, an ANC system attempts to minimize the as having loudness n sone. According to the definition, two mean squared error of the error signal picked up at the error sine waves with different frequencies are regarded as having microphone. Although the sound pressure level around the equal loudness level if they are perceived as equally loud by an error microphone becomes small, humans may not perceive average young person without significant hearing impairment. sufficient noise reduction through the auditory system. This The curve is called the equal-loudness contours, which are is because of complicated psychoacoustic properties. Hence, standardized in ISO 226:2003. However, acoustic noise, which the ANC system should be adjusted to consider psychoa- is mainly a broadband signal and nonstationary, has different coustic properties. Bao and Panahi have already proposed characteristics in terms of the hearing sense. an ANC system considering psychoacoustic properties and demonstrated its effectiveness for MRI noise and a synthesized B. Noise-Weighting Filter random signal [5]. In [5], noise-weighting filters based on A- Noise weighting has been suggested as a means of quan- weighting defined in IEC 61672:2003 and ITU-R 468 noise tifying the hearing sensitivity of humans to the frequency. In weighting were incorporated into conventional ANC systems. this paper, we use two different noise-weighting filters, i.e., These noise-weighting filters modify the error and reference the A-weighting filter and the ITU-R 468 noise-weighting signals, and these modified signals are used to adjust the filter. A-weighting, defined in IEC 61672:2003, is based on noise control filter in the ANC structure. However, acoustic the equal-loudness contour for the 40 dB (phon) pure tone. noise often has nonstationary noise. Hence, the ANC system The frequency response of A-weighting is shown in Fig. 1 considering psychoacoustic properties should be examined for [6]. In Fig. 1, the original A-weighting filter is displayed nonstationary noise. Moreover, such ANC systems should be by a red line. However, we use a different characteristic in examined through a subjective evaluation. In this paper, we part of the frequency range because the A-weighting filter is develop an ANC system integrated with noise-weighting filters approximated by an FIR filter with short taps in this paper, 978-988-14768-5-2 ©2018 APSIPA 1256 APSIPA-ASC 2018 Proceedings, APSIPA Annual Summit and Conference 2018 12-15 November 2018, Hawaii TABLE I 20 SIMULATION CONDITIONS. 0 Input signal Factory noise Sampling frequency 44100 Hz Tap length of unknown system P 500 -20 Tap length of noise control filter filter W 500 Tap length of secondary path model C^ 500 Tap length of noise weighting filter H 500 Amplitude [dB] Amplitude -40 Standard Update algorithm NLMS FIR filter Step size parameter α 0.01 -60 Regularization parameter β 1:0 × 10−6 10-2 10-1 100 101 Frequency [kHz] Fig. 1. Frequency response of A-weighting filter. C. Feedforward ANC System Considering of Psychoacoustic Properties 20 Feedforward ANC with a noise-weighting filter can ef- fectively reduce noise over a specific frequency band by incorporating a noise-weighting filter in a conventional ANC 0 system. A block diagram of a feedforward ANC system that considers psychoacoustic properties is shown in Fig. 3. In Fig. -20 3, P (z) indicates the primary path from the noise source to the error microphone, C(z) is the secondary path from the secondary source to the error microphone, C^(z) is the model Amplitude [dB] Amplitude -40 of C(z), W (z) is the noise control filter, and H(z) is the Standard FIR filter noise-weighting filter. In this system, noise reduction in terms -60 of acoustic impression can be expected by using the filtered -2 -1 0 1 10 10 10 10 reference signal rH (n) and the error signal eH (n) through the Frequency [kHz] use of the noise-weighting filter to update the noise control filter W (z). The update equation of the filter coefficients in Fig. 2. Frequency response of ITU-R 468. the ANC system in consideration of the hearing characteristic is given by x(n) d(n) + e(n) P(z) ∑ α w(n + 1) = w(n) + 2 rH (n)eH (n); (1) – β + krH (n)k where α is the step size parameter and β is the regularization y(n) y' (n) W(z) C(z) parameter. w(n) is the noise control filter vector, rH (n) is obtained by convolution of the filtered reference signal r(n) Ĉ(z) with the noise weighting filter H(z), and eh(n) is obtained r(n) by convolution of the error signal e(n) with H(z). H(z) III. SIMULATION RESULTS A. Simulation Conditions and Evaluation Criteria NLMS H(z) rH(n) eH(n) In this section, we examine the effectiveness of the ANC system in consideration of psychoacoustic properties by sim- Fig. 3. Block diagram of the ANC system in consideration of psychoacoustic ulation. Table 1 lists the simulation conditions. In the simula- properties. tions, the noise reduction performance is evaluated by P 2 P d (n) Reduction = 10 log10 2 : (2) which is indicated by the green line. On the other hand, the e (n) ITU-R 468 noise-weighting filter, recommended by CCIR is The primary and secondary paths are actual acoustic paths, used to evaluate random noise. The frequency response of the which are estimated in advance using white noise. The impulse ITU-R 468 noise-weighting filter is shown in Fig. 2 [7]. In Fig. responses of the primary and secondary paths are shown in 2, the red line indicates the frequency response of the original Figs. 4 and 5, respectively. In addition, the A-weighting filter ITU-R 468 noise-weighting filter and the green line indicates and ITU-R 468 noise weighting filter are realized by a 500- the approximated response by an FIR filter with short taps. tap FIR filter, which is designed by the window-function 1257 Proceedings, APSIPA Annual Summit and Conference 2018 12-15 November 2018, Hawaii 10 1.0 0 -10 0.5 -20 -30 0 -40 Amplitude -50 Amplitude [dB] Amplitude -0.5 -60 -70 -1.0 0 5 10 15 20 0 10 20 30 Frequency [kHz] Time [s] (a) Time waveform. Fig. 4. Frequency spectrum of primary path. 10 -30 0 -40 -10 -50 -20 -60 -70 -30 -80 -40 -90 -50 Amplitude [dB] Amplitude [dB] Amplitude -100 -60 -110 -70 -120 0 5 10 15 20 0 5 10 15 20 Frequency [kHz] Frequency [kHz] (b) Frequency spectrum. Fig. 5. Frequency spectrum of secondary path. Fig. 6. Factory noise. method, where the sampling frequency is 44,100 Hz and the consequently becomes strong. Moreover, the noise reduction number of points in the FFT is 8,192. Furthermore, the time performance becomes larger in the frequency band where the waveform and frequency spectrum of the factory noise used error and reference signals are largely weighted by the noise as nonstationary noise are shown in Fig. 6. As can be seen weighting filter, whereas the noise reduction performance from Fig. 6, the factory noise is broadband noise. In the deteriorates in the other frequency bands as shown in Figs. simulations, we compare the noise reduction between the ANC 7 and 10.