Ali A. Milani

On maximum achievable noise reduction in ANC systems

Active noise control (ANC) is primarily a signal estimation problem. The nature of noise to be canceled, the acoustic paths, and the transfer functions of microphones/loudspeakers place fundamental limits on the performance of ANC systems. In this paper we analyze the performance of ANC systems from the maximum achievable noise attenuation level (NALmax) prospective for different ANC system structures i.e feed-forward, feedback and hybrid. We first derive NALmax for stochastic noise and then generalize the result to accommodate tonal (sinusoidal) noises. These results are invaluable in choosing an ANC design scheme.

A New Delayless Subband Adaptive Filtering Algorithm for Active Noise Control Systems

Subband adaptive filtering (SAF) techniques play a prominent role in designing active noise control (ANC) systems. They reduce the computational complexity of ANC algorithms, particularly, when the acoustic noise is a broadband signal and the system models have long impulse responses. In the commonly used uniform-discrete Fourier transform (DFT)-modulated (UDFTM) filter banks, increasing the number of subbands decreases the computational burden but can introduce excessive distortion, degrading performance of the ANC system. In this paper, we propose a new UDFTM-based adaptive subband filtering method that alleviates the degrading effects of the delay and side-lobe distortion introduced by the prototype filter on the system performance. The delay in filter bank is reduced by prototype filter design and the side-lobe distortion is compensated for by oversampling and appropriate stacking of subband weights. Experimental results show the improvement of performance and computational complexity of the proposed method in comparison to two commonly used subband and block adaptive filtering algorithms.

An Efficient Feedback Active Noise Control Algorithm Based on Reduced-Order Linear Predictive Modeling of fMRI Acoustic Noise

Functional magnetic resonance imaging (fMRI) acoustic noise exhibits an almost periodic nature (quasi-periodicity) due to the repetitive nature of currents in the gradient coils. Small changes occur in the waveform in consecutive periods due to the background noise and slow drifts in the electroacoustic transfer functions that map the gradient coil waveforms to the measured acoustic waveforms. The period depends on the number of slices per second, when echo planar imaging (EPI) sequencing is used. Linear predictability of fMRI acoustic noise has a direct effect on the performance of active noise control (ANC) systems targeted to cancel the acoustic noise. It is shown that by incorporating some samples from the previous period, very high linear prediction accuracy can be reached with a very low order predictor. This has direct implications on feedback ANC systems since their performance is governed by the predictability of the acoustic noise to be cancelled. The low complexity linear prediction of fMRI acoustic noise developed in this paper is used to derive an effective and low-cost feedback ANC system.

Active noise control of noisy periodic signals using signal separation

Periodic signals (since they can be easily predicted) can be canceled much more effectively when compared to non- periodic/stochastic signals. A large class of acoustic noise sources have an underlying periodic process that generates a periodic noise component, and thus the acoustic noise can in general be modeled as the sum of a periodic signal and a random signal (usually a background noise). In this paper we present the idea that separating the acoustic noise into periodic and random noise components and doing separate active noise control(ANC) for each tends to increase the over-all noise attenuation level (NAL). Formulae for the exact improvement in noise attenuation levels are derived. A novel signal separation and noise cancelation scheme based on adaptive filtering is developed and its effectiveness is shown for several periodic signal in white noise cases.

Speech Enhancement in Functional MRI Environment using Adaptive Sub-Band Algorithms

Very high level of acoustic noise in fMRI scanner rooms disrupts speech communication between the subject and the physician/researcher. Enhancing speech in such an environment is challenging due to the broadband and dynamic nature of the noise. Sub-band adaptive methods prove to be very effective in cancelling such noise. In this paper we present the results of using sub-band adaptive methods for enhancing speech corrupted by noise from a 3-Tesla fMRI scanner. We also observe that the performance depends on the synthesis filter bank structure.

Weight Stacking Analysis of Delayless Subband Adaptive Algorithms for fMRI Active Noise Cancellation

High level acoustic noise in fMRI scanners is a source of concern to patients and health care providers. Active noise control systems employing delayless subband adaptive filters have been shown effective in fMRI acoustic noise reduction [3] [4]. In this method [5], adaptive filtering is done in subbands and the subband weights are stacked together to construct the fullband filter weights. There are two types of stacking methods called FFT and FFT-2. These stacking methods introduce distortion which limit the noise reduction level. In this paper, we model the distortion and analyze the effect of distortion when different adaptive schemes (nLMS, APA, RLS) are used. This analysis helps in selecting the appropriate adaptive scheme and determining the optimum number of subbands.

Objective and subjective evaluation of adaptive speech enhancement methods for functional MRI

To recover speech corrupted by functional magnetic resonance imaging (fMRI) acoustic noise using two‐channel adaptive speech enhancement techniques.

Performance Analysis of Sub-Band nLMS, APA and RLS in fMRI ANC with a Non-Minimum Phase Secondary Path

We present the performance comparison of sub-band FXLMS algorithm for fMRI acoustic noise cancellation when the secondary path is non-minimum phase. Three types of least square adaptive filtering methods (nLMS, APA, RLS) are used in sub-bands. A series of simulations have been done using recorded fMRI acoustic noise and the results are given and compared based on the noise attenuation level, convergence rate and the quality of primary path estimation. It is verified that the spectrum structure of the fMRI acoustic noise has the main role in the performance of the active noise control especially when sub-band filtering is used.

Analysis and optimal design of delayless subband active noise control systems for broadband noise

Active control of wide-band noise presents certain unique challenges many of which can be addressed using delayless subband adaptive filtering techniques. The performance of a delayless subband active noise control (DSANC) system depends on a complex interplay between the (1) choice of adaptation algorithm, (2) number of subbands, (3) weight stacking scheme, (4) input noise spectrum, and (5) primary, secondary paths. This interplay is studied in this paper for two different kinds of broadband noise. Distortion introduced by the weight stacking methods is investigated and quantified. It is shown that the computational complexity decreases and the stacking distortion increases with the number of subbands. The performance limiting effect of the non-minimum phase property of secondary path on the system performance is evaluated and analytically formulated. An upper bound for the obtainable noise attenuation level (NAL) is derived. A step by step optimal design procedure for the best performance is developed taking computational complexity into consideration. Simulation results support the analytical development and the proposed approach for optimal design of DSANC systems.

Integrated speech enhancement for functional MRI environment

This paper presents an integrated speech enhancement (SE) method for the noisy MRI environment. We show that the performance of SE system improves considerably when the speech signal dominated by MRI acoustic noise at very low SNR is enhanced in two successive stages using two-channel SE methods followed by a single-channel post processing SE algorithm. Actual MRI noisy speech data are used in our experiments showing the improved performance of the proposed SE method.