Active Noise Control for Motorcycle Helmet

Transcription

1 Active Noise Control for Motorcycle Helmet Lichuan Liu, Sen M. Kuo, and Kishan P. Raghuathan Abstract This paper presents an active noise control (ANC) system for motorcycle helmets. The characteristics of the motorcycle noise were studied. Experimental setup with real ambience of a person riding a motorcycle, wearing a helmet under noisy conditions, was built to measure transfer functions for computer simulations and to conduct real-time experiments. Simulation and experiment results show the feasibility of using ANC system for motorcycle helmets to reduce noise. Keywords Active noise control (ANC), FXLMS algorithm, motorcycle helmet, real time experiments. I. INTRODUCTION OTORCYCLES are the one of most affordable and M common type of transportation vehicles in the world [1]- [3]. There are around 200 million motorcycles (including mopeds, motor scooters and other powered two- and three-wheelers) in use worldwide [4]. With increasing gas price, the motorcycle is an attractive alternative to a car because it provides mileage as high as 107 miles per gallon [5]. However, the motorcycle noise not only distracts the concentration of the rider, but also results in permanent hearing loss. Motorcycle noise mainly consists of engine noise, wind noise, tire noise, road noise, etc. [6]. Motorcycle noise levels on an open road will be in the range of dba at 30 mph (miles per hour) to 116 dba at 120 mph, depending on the type of motorcycles and the road conditions [7]. The average rate of increase in noise levels is 15.5 db per doubling the speed. Some of the common ailments of noise while riding a motorcycle are summarized as follows: (1) Noise induces hearing loss [8], (2) cardiovascular effects [9], (3) tinnitus [8], (4) stress and sleep disturbances [9], (5) temporary threshold shift [10], and (6) increased risk of accidents. Exposure to high noise levels for long duration may distract the concentration of the rider and make it difficult to hear or communicate to any emergency signal on the road, which may even result in fatal accidents. The first energy-absorbing helmet was designed for the United States Air Force for protecting aviators from the effects of impacts [11], and then applied to others such as motorcyclists. The conventional motorcycle helmet has two principal protective components: a hard outer shell prevents penetration of the helmet by a pointed object, and an inner foam L. Liu, S. Kuo and K. Raghuathan are with the Department of Electrical Engineering, Northern Illinois University, Dekalb, IL USA. ( {lichuan, kuo}@niu.edu and kishanprakash@gmail.com). liner increases the distance and period of time over which the head stops and reducing its deceleration [12]. Motorcycle helmets are estimated to be 37% effective in preventing fatal injuries [13]. According National Highway Traffic Safety Administration (NHTSA), it is estimated that over $13.3 billion was saved between 1984 and 1999 from helmet use [14]. A motorcycle helmet not only protects the rider s head from crash but also provides protection from the external harsh environment like rain, wind, sun glare, and noise, etc. [15]. The liner inside the motorcycle helmet also acts as sound seal to attenuate the high-frequency acoustic noise above 500 Hz, which can be considered as the passive noise control. However, the liner is ineffective to reduce the low-frequency noise that needs very thick absorbers and barriers. There is a limitation on thickness of the helmet so it quickly becomes impractical if the liner is thicker than 1 2 inches [12]. Therefore, the best solution is an active noise control (ANC) system that can effectively cancel low-frequency noise. This paper presents an ANC system using broadband feedforward ANC algorithm [16] to attenuate the harmful motorcycle noise at both ears of the rider. This paper is organized as follows: Section II analyzes the motorcycle noise. Section III introduces the ANC system for motorcycle helmet. The simulation results are presented in Section IV. In Section V, the real-time experimental setup and the experiment results are presented and discussed. Section VI offers the conclusion. II. ANALYSIS OF MOTORCYCLE NOISE This section analyzes the motorcycle noise at different speeds and situations. The motorcycle used for acquiring real motorcycle noise is Harley Davidson s Heritage Softtail with a two-cylinder, 4-stroke engine. The noise was recorded using RadioShack s omni-directional microphones and 722 high-resolution digital audio recorder with 16-bit resolution. Motorcycle noise consists of periodic engine noise and environmental noise such as wind, tire, and road noises. The motorcycle noise can be classified into low-frequency, medium-frequency, and high-frequency noises [6]. Low-frequency noise is often due to the motorcycle engine and can be very annoying in long journeys. The primary source of medium-frequency noise is the mechanical elements such as the transmission and some engine components. High-frequency noise is usually due to aerodynamic phenomena like wind noise, whistling, etc. Both medium- and high-frequency noises can be reduced by the inner liner of helmet. To record real motorcycle noise, two microphones are 799

2 installed inside a commercial-available helmet, one at the left ear and the other near the mouth of the rider inside the helmet. A passenger at the back seat performs digital recording using sampling rate of 32,000 Hz. The motorcycle noise is recorded at different fixed and changing speeds, with and without the rider s speech. The recorded data files are decimated to 4 khz using MATLAB for further processing and experiments. Fig. 1 depicts the spectral characteristics of the recorded motorcycle noise at the speed of 40 mph, which shows that the motorcycle noise is a broadband noise and most of the energy is concentrated at low-frequency region from 100 to 400 Hz. When listening to the recorded sound files, it can be observed that most of noise components are due to the motorcycle engine, which also shows that the inner layer of helmet acts as a passive attenuator for reducing high-frequency noise. Unfortunately, it is inefficient to block the low-frequency motorcycle engine noise. This fact validates the need of ANC system for motorcycle helmets.. Fig. 1 and Fig. 2 show the energy of motorcycle noise is concentrated below 400 Hz and is spread over a wide frequency range, thus it can be classified as a broadband noise. As shown in Fig. 2, the spectral contents at each time frame are changed rapidly when the speed changed from 0 to 60 mph. Therefore, the ANC system should converge fast in order to track the time-varying frequency components of the motorcycle noise during acceleration and deceleration. III. ACTIVE NOISE CONTROL FOR MOTORCYCLE HELMET ANC system cancels the primary noise based on the acoustical superposition at the locations of the error sensors. The feedforward ANC system uses a reference sensor to pick up the primary noise and use it as a reference signal. This paper considers the feedforward ANC system with the filtered-x least mean square (FXLMS) algorithm [17] because the motorcycle noise consists of unpredictable broadband components as discussed in Section II. A. Filtered-x LMS Algorithm Quiet zone x(n) Primary path d(n) + P(z) - e(n) Adaptive filter W(z) y(n) S(z) y'(n) Ŝ(z) x'(n) LMS Fig. 1 Spectrum of motorcycle noise at the speed of 40 mph To analyze the transient behaviors of noise during speed changes, the recorded motorcycle noise data sets are divided into five frames of equal intervals (5 seconds each) and the spectrum of each frame is computed and plotted using MATLAB. Fig. 2 shows the time-varying spectrum when the motorcycle speed is changed from 0 to 60 mph. Fig. 2 Spectrum of the motorcycle noise as a function of time for the speed changed from 0 to 60 mph Fig. 3 Block diagram of the FXLMS algorithm The block diagram of broadband feedforward ANC using the FXLMS algorithm is shown in Fig. 3 [18]. The transfer function P(z) is the primary path between the noise source and the error microphone. The adaptive filter W(z) uses the reference signal x(n) and the least-mean-square (LMS) algorithm to update its coefficients in order to estimate and track the changes of primary path. The filter output y(n) is send to the secondary loudspeaker to produce anti-noise y'(n), which acoustically cancels the primary noise d(n) to minimize the error signal (residual noise) e(n). The transfer function S(z) represents the secondary path from the secondary loudspeaker to the error microphone. The FXLMS algorithm compensates the effect of secondary path by placing the secondary-path model S ˆ( z ) in the signal path before the weight update of the LMS algorithm [18]. In practical applications, the secondary path S(z) is unknown and it must be estimated by the adaptive filter S ˆ( z ) using the off-line modeling (or training) technique [17]. This preliminary identification may be repeated as the on-line secondary-path modeling if the environment changes. As shown in Fig. 3, the residual noise can be described as en ( ) = dn ( ) sn ( ) yn ( ) (1) 800

3 where s(n) is the impulse response of the secondary path S(z), and * denotes linear convolution. The adaptive filter output is computed as T yn ( ) = w ( n) x ( n), (2) where n = [ w n w n w n ] w ( ) ( ) ( )... ( ) T is the adaptive filter 0 1 L 1 coefficient vector of length L, T denotes transpose operation, x ( n) = x( n) x( n 1)... x( n L+ 1) T is the signal vector and [ ] at time n. The weight vector the adaptive filter is updated by the FXLMS algorithm w( n+ 1) = w( n) + [ μ x ( n) e( n)], (3) where μ is the step size that determines the convergence speed of the algorithm, and the filtered signal vector x ( n) = [ x ( n) x ( n 1)... x ( n L+ 1) ] T has elements x ( n) = sˆ ( n)* x( n), (4) where sˆ( n ) is the impulse response of Sz ˆ( ) that is the estimate of the secondary path S(z). IV. COMPUTER SIMULATIONS A. Experimental Setup We test the performance of ANC system only at the left ear due to hardware constraints, an identical setting can be applied at the right ear. Fig. 4(a) depicts the overall experimental setup of the ANC system, which uses a Bruel and Kjaer s torso and head simulator (KEMAR) to emulate a rider and resemble a person on a motorcycle wearing a helmet. The primary loudspeaker is placed 65 cm in front of the rider (KEMAR) to emulate the noise source. Fig. 4(b) shows detailed setup of helmet, where a headphone is installed close to the left ear of KEMAR inside the helmet to serve as a secondary speaker, and the error microphone is placed near the left ear of KEMAR [20]. The primary microphone is placed inside the helmet near the chin of the rider, which can also pick up rider s speech for communication purpose. B. Locations of Error Microphones As shown in Fig. 3, the objective of the ANC system is to minimize the residual noise e(n) picked up by the error microphone, which in turn creates quiet zones near the ears of the rider. The location of the error microphone that monitors the performance of ANC system is a critical design consideration [19]. Eight symmetrical locations shown in Fig. 5 are evaluated for the error microphone location. Magnitude responses of the secondary-path transfer functions obtained at eight error microphone locations around the left ear of KEMAR are shown in Fig. 5. These transfer functions are obtained by sending white noise to the secondary speaker and measuring the signal sensed by the error microphone at the corresponding locations using a spectrum analyzer. Fig. 5 Magnitude responses of secondary path S(z) at eight different microphone locations (a) Secondary loudspeaker Error microphone Primary microphone (b) Fig. 4 Experimental setup for motorcycle helmet ANC system, (a) overall setup, (b) detailed helmet setup Because the function of adaptive filter W(z) is to model the primary path P(z) and inverse model the secondary path S(z), a flat magnitude response of S(z) in the frequency range of interest is desirable [19]. Fig. 5 shows that the location #8 has the flattest magnitude response as compared with all other locations, thus the error microphone is placed at that location for experiment. C. Simulation Results Recorded motorcycle noise data sets are used in computer simulations to evaluate the performance of the ANC system. The step size is experimentally selected from.003 to.006, and the adaptive filter length is 128 [20]. 801

4 The performance of the ANC system for the motorcycle noise recorded at the speed of 40 mph is shown in Fig. 6, which shows the spectra of signal measured at the error microphone before () and after () noise cancellation. The average cancellation of 40 db is achieved in the frequency range up to 200 Hz, and 15 db in the frequency range from 200 to 600 Hz. Fig. 7 shows the spectra of signal measured at the error microphone before () and after () noise cancellation at the speed of 60 mph. Average cancellations of 40 db in the frequency range up to 200 Hz, and 20 db in the frequency range from 200 to 600 Hz are achieved. These results show that the ANC algorithm has achieved satisfactory noise reduction at low frequencies for motorcycle noise at speeds of 40 mph and 60 mph. Fig. 8 Experimental setup for real-time experiment of motorcycle helmet ANC system Fig. 6 Spectra of motorcycle noise at the speed of 40 mph before (ANC OFF) and after () noise cancellation A. Multiple Sinusoids In this case, multiple sinusoids at 210, 250, and 300 Hz are used as primary noise. These tones are individually generated from different signal generators, mixed, and played through the primary loudspeaker. Fig. 9(a) depicts the waveforms of the transition from to, 9(b) zooms in transition region from OFF to ON, and Fig. 9(c) shows the spectra of signals measured at the error microphone for both and. Fig. 9(c) shows that noise reduction of 70 db, 55 db, and 60 db are achieved at 210, 250, and 300 Hz, respectively. Figs. 9(a) and 9(b) show that the convergence time for ANC algorithm to reach steady state after the system is turned on is about 400 ms. Fig. 7 Spectra of motorcycle noise at the speed of 60 mph before (ANC OFF) and after () noise cancellation V. REAL TIME EXPERIMENTS The performance of ANC system is further evaluated by real-time experiments using sinusoidal noise and recorded motorcycle noise at different speeds. To get the ambience of a person riding a motorcycle with a helmet under real noisy conditions, the setup depicted in Fig. 4 was modified to a more realistic setup as shown in Fig. 8. The difference is the placement of the primary loudspeaker that acts as the source of motorcycle noise and the full-face helmet is replaced by an open-face helmet. The primary speaker is placed 100 cm in front of KEMAR that mimics the engine location in response to the rider [20]. Fig. 9 Waveforms of the transition from to, (b) zoom in transition region, and (c) spectra of and B. Motorcycle Noise at Speed 40 mph The performance of the ANC system is further evaluated using recorded motorcycle noise at the speed of 40 mph. As discussed in Section II, motorcycle noise is a nonstationary 802

5 signal. The gains of the pre-amplifiers and the power amplifiers can be adjusted such that the volumes of the primary speaker and the secondary speakers are comparable and match the peak-to-peak voltage of the A/D converter. The step size of 0.1 is used to update the secondary-path modeling filter Sz ˆ( ) with the filter length of 64. Based on the trial and error approach, step size of and the filter length of 128 are used to update the adaptive filter W(z) for best performance [20]. Fig. 10 shows the spectrogram (a), the time waveform (b), and the power estimate (c) that representing the transition of signal measured at the left ear of the KEAMER before (ANC OFF) and after () noise cancellation for motorcycle noise at the speed of 40 mph. Fig. 10(a) shows that noise reduction of 30 db is achieved at the dominant noise component at 100 Hz, and noise reduction of 20 to 30 db is achieved in the frequency range of 100 to 200 Hz. Fig. 10(b) and Fig. 10(c) show that the convergence time of ANC is about 600 ms. In spite of nonstationary and broadband characteristics of motorcycle noise, the ANC algorithm converged, maintained stability, and showed satisfactory noise reduction. W(z). Fig. 11 depicts the spectrogram (a), time waveform (b), and power estimates (c) that representing the transition of signal measured at the left ear of the KEAMER before () and after () noise cancellation for motorcycle noise at the speed of 60 mph. An average noise reduction of 25 to 30 db is achieved at the dominant frequency range of 120 to 160 Hz, and about 20 db of noise reduction is achieved at 100 Hz. The convergence time is about 3 seconds. Fig. 11 (a) Spectrogram, (b) time waveform, and (c) power estimate of the transition from to Fig. 10 (a) Spectrogram, (b) time waveform, and (c) power estimate of the transition from to C. Motorcycle Noise at Speed of 60 mph The validity of the ANC system is also evaluated with the motorcycle noise at highway speed of 60 mph. As discussed in Section I, higher speed produces higher levels of noise. In general, motorcycle at the speed of 60 mph produces sound levels of about 90 db. The gain of the power amplifiers that drives the secondary speaker and the noise generator is set such that the volumes are comparable. For better results, step size of and the filter length of 128 are used for the adaptive filter D. Motorcycle Noise at Speed of 0-60 mph In reality, a motorcycle might have sudden changes in speed during acceleration and deceleration. For example, after the traffic stop (0 mph), the motorcycle rider increases the speed to 60 mph, or the rider may stop from high speed. This section evaluates the performance of the ANC system during transition of motorcycle speed from 0 to 60 mph as shown in Fig. 12. It is a challenge for the ANC system to adapt to such fast-changing nature of the primary noise. The gains of the pre-amplifiers are set to match the voltage of the A/D converter (+/- 3V), and comparable gains are set for power amplifiers to drive the primary and secondary speakers. Step size of and filter length of 128 are used to achieve best results [20]. In this case, the motorcycle takes about 10 seconds to rise from 0 to 60 mph. The motorcycle noise with the speed transition is played for the first 10 seconds with and the same motorcycle noise is played for another 10 seconds with. The spectrogram, the time waveform, and the power estimate representing the transition of signal measured at the left ear of the KEAMER before and after noise cancellation 803

6 with speed transition from 0 to 60 mph are shown in Fig. 12 (a), (b) and (c), respectively. Inspire of speed changes from 0 to 60 mph, the ANC system remains stable, which proves the feasibility of the ANC system in practical situations. Cancellation of 20 to 25 db in the frequency range of 100 to 300 Hz is achieved at the speed transition of motorcycle noise. [7] M. C. Lower, D. W. Hurst, A. R. Claughton, and A. Thomas, Sources and Levels of Noise Under Motorcyclists Helmets, Proc. of Institute of Acoustics, vol. 16, part 2, pp , [8] Noise-Induced Hearing Loss, National Institute on Deafness and Other Communication Disorders, NIH Pub. No , October, [9] What Problems Can Noise Cause, European Agency for Safety and Health at Work, [10] D. Miller, Will You Ever Recover from Temporary Threshold Shift, Audio Masterclass.com, [11] Roth, Lombard, Protected Helmet, United States Patent 2,923,941, January 20, 1953 [12] Motocycle helmet, [13] Motorcycle Helmet safety fact sheet, National Highway traffic Administration Publication, Dec., 2008 [14] G. Marples, The History of Motorcycle Helmets Dead or Alive? Sep [1] Cycle safety information: what you should know about motorcycle helmet, Motorcycle Safety Foundation, Sept [15] S. M. Kuo and D. R. Morgan, Active Noise Control Systems Algorithms and DSP Implementations, New York: Wiley, [16] Morgan, D. R., An analysis of multiple correlation cancellation loops with a filter in the auxiliary path, IEEE Trans. Acoust., Speech, Signal Processing, vol. ASSP-28, pp , [17] S. M. Kuo and D. R. Morgan, Active Noise Control: A Tutorial Review, Proceedings of the IEEE, vol. 87, pp , June [18] S. Mitra, Single-Channel Adaptive Feedback Active Noise Control for Headsets, M. S. Thesis, Northern Illinois University, DeKalb, IL, [19] K. Raghuathan, Audio Integrated Active Noise Control for Motorcycle Helmet, M. S. Tthesis, Northern Illinois University, DeKalb, IL Fig. 12 (a) Spectrogram, (b) time waveform, and (c) power estimate of the transition from to VI. CONCLUSION This paper developed and implemented a motorcycle helmet ANC system to cancel the harmful motorcycle noise. The characteristics of motorcycle noise at different fixed and time-varying speeds were analyzed. Optimum locations of the error microphone were studied and the location that has the flattest magnitude response of secondary path was used. The performance and feasibility of the ANC algorithm was evaluated with computer simulations and real-time experiments using sinusoidal noise and recorded motorcycle noise. Experiment results successfully showed that the motorcycle helmet ANC system can dramatically reduce the noise levels. REFERENCES [1] H. Nakata, "Motorcycle makers battle it out in Vietnam". Japan Times, Oct [2] P. McCracken, "Vietnam eats, sleeps, and dreams on motorbikes". The Christian Science Monitor, Oct [3] Shuhei, Adachi (2006). "Fuel Cell Powered Motorcycles". Journal of the Society of Automotive Engineers of Japan, Vol. 60 (1): pp [4] Gekgo Worldwide, "Verucci Gas Scooters"., Retrieved 15 August 2006 [5] A. Barrionuevo, "That Roar in the Jungle Is 15,000 Motorbikes". The New York Times. Nov [6] Technical Research Support for DEFRA Noise Programme, Low Frequency Noise, Department of Environment of North Ireland, Lichuan Liu (M 06) received her B.S. and M.S. degree in Electrical Engineering in 1995 and 1998 respectively from University of Electronic Science and Technology of China, and Ph. D. degree in Electrical Engineering from New Jersey Institute of Technology, Newark, NJ in She joined Northern Illinois University in 2007 and is currently an Assistant Professor of Electrical Engineering. Her current research includes wireless communication, wireless sensor networking and ad-hoc networking, digital signal processing. She has over 30 publications including 14 journal papers. She served as a program committee member for the 2010 IEEE Int. Conf. on Networking Sensing and Control (ICNSC), and the 2009 IEEE Int. Conf. on Service Operations and Logistics, and Informatics (SOLIS). Dr. Liu is the receipt of the Best Conference Paper Finalist of IEEE CASE 2008, the Student Poster Award of Sarnoff, ICNSC Best Student Paper Award, WOCC Student Poster Award and the student presentation Award of CIE. Sen M. Kuo (M 84 SM 04) received the B.S. degree from National Taiwan Normal University, Taipei, Taiwan, in 1976 and the M.S. and Ph.D. degrees from the University of New Mexico, Albuquerque, in 1983 and 1985, respectively. He is currently a Professor in the Department of Electrical Engineering, Northern Illinois University, DeKalb. In 1993, he was with Texas Instruments, Houston, TX, and with Chung-Ang University in He is the leading author of three books: Active Noise Control Systems (New York: Wiley, 1996), Real-Time Digital Signal Processing (Wiley, 2001, 2007), and Digital Signal Processors (Englewood Cliffs, NJ: Prentice-Hall, 2005), and co-author of two books: Embedded Signal Processing with the Micro Signal Architecture (Wiley, 2007) and Subband Adaptive Filtering (Wiley, 2009). His research focuses on active noise control, real-time DSP applications, adaptive echo and noise cancellation, digital audio applications, and digital communications. Dr. Kuo received the IEEE 1993 Chester Sall Award for the first-place transactions (Consumer Electronics) paper award, and the faculty-of-year award in 2001 for accomplishments in research and scholarly areas. Kishan P. Raghunathan received his B.Tech. degree in Electronic and Communication Engineering in 2006 from Vignan Institute of Technology and Science, Deshmukhi-Nalgonda, India, and the his M.S. degree in Electrical Engineering in Northern Illinois University, Dekalb, IL, USA in He is currently a Hardware/software applications engineer at Jet Propulsion Laboratory, NASA, Pasadena, CA, USA. His current research includes digital signal processing and active noise control.. 804