we are happy to inform you about the next date of our Communication Technology Colloquium.
Wednesday, 18. October 2023
Speaker: Johannes Fabry
Time: 11:00 a.m.
Location: Lecture room 4G
Doctoral-Lecture: Least-Squares Methods for Individualization of Hearables with Active Noise Cancellation
Motivation, Goal and Task of the Dissertation
According to the World Health Organization (WHO), noise exposure is a major contributor to public health problems. Long-term exposure to occupational or recreational noise leads to noise-induced hearing loss or tinnitus, whereas environmental noise causes stress, sleep disturbances, chronic high annoyance, and cardiovascular diseases.Active noise cancellation (ANC) in headphones effectively supplements passive hearing protection to drastically reduce the perceived loudness of ambient noise, especially at lower frequencies. An ANC headphone plays back a cancellation signal via its loudspeaker, which destructively interferes with passively attenuated ambient noise. The cancellation signal is based on information from reference sensors integrated into the headphone.Due to strict processing power and battery constraints, the state of the art of ANC in headphones consists of either time-invariant or simple adaptive filters. The time-invariant filter approach implements a one-size-fits-all ANC solution. Consequently, the active attenuation of ambient noise varies strongly between users depending on the individual fit of the headphone. On the other hand, simple adaptive filters, such as the least mean squares algorithm, exhibit a slow convergence and tracking speed. Thus, the system cannot adjust to rapid changes in the ambient noise.The first goal of this dissertation is to develop methods that yield a more reliable ANC performance with respect to the fit of a headphone for individual users. The methods should allow a calibration of ANC headphones in the field as well as a configuration of the ANC transfer characteristic. The second goal is to develop efficient adaptive algorithms that yield a close-to-optimal ANC performance, are suitable for relevant audio codecs, and require almost no prior knowledge about the acoustic system.
Major Scientific Contributions
The first major contribution is a method to calibrate an ANC headphone in the field. It considers the individual fit of a user by estimating the acoustic primary and secondary path, which characterize the acoustic properties of a headphone and fit and from which an individualized feed-forward filter can be designed. A novel secondary path estimator explicitly considers the interference of a measurement signal with ambient sound. Compared to other methods it considerably increases the accuracy of the secondary path estimate. Since a direct measurement of the primary path in the field is not suitable due to lacking control over the ambient sound field, we developed an estimator for the primary path based on the coherency between the primary and secondary path. Evaluations on a vast dataset of acoustic paths for different users, which was created in the course of this work, show significant improvements of the active attenuation for an individualized filter compared to a one-size-fits-all feed-forward filter. The second major contribution introduces active acoustic equalization (AAE) as a framework for designing feed-forward filters that result in an arbitrary target transfer function for a given ANC headphone. Thereby, users can adjust how they hear ambient sound to their personal liking. We analyze the influence of the secondary path delay as well as the desired attenuation on the actual transfer function and propose effective measures to improve the accuracy of the filter design with respect to the desired transfer function. Since only the magnitude spectrum of the transfer function is relevant for the perception of AAE, we furthermore propose a frequency domain filter optimization that considers the energy spectrum of a desired transfer function. A novel cost function explicitly considers the nonlinear perception of magnitude and frequency of the human auditory system. By the example of a hear-through, we show how the proposed optimization results in a better median performance compared to the state of the art. The third major contribution is the development of a system-theoretic model and of adaptive algorithms for feed-forward ANC. The model explicitly considers self-induced disturbances that couple into the reference sensors, such as footfall or chewing sounds. We then introduce the Kalman filter as the optimal, unbiased estimator for the model, highlight the importance of accurate process parameter estimates, and subsequently introduce respective estimators for the measurement noise, process noise, and fading factor. Furthermore, we derive the novel composed Kalman filter (CKF) as an efficient implementation of the time domain Kalman filter. The CKF assumes that the Kalman filter’s state error covariance matrix is subject to a band-matrix-like structure. We validate this assumption by deriving a stationary solution for this matrix. For reasonable settings, the CKF's computational complexity is almost an order of magnitude lower whilst its performance is comparable to that of the original Kalman filter. We also propose a numerically stable implementation of the CKF based on a UD factorization and experimentally validate its robustness using a quasi fixed point arithmetic. Lastly, we propose an online secondary path estimator, which, compared to the state of the art, improves the frequency-dependent signal-to-noise ratio of the measurement signal whilst rendering it inaudible to users of an ANC headphone. The process parameter estimators, the CKF, as well as the online secondary path estimator are evaluated and compared under laboratory as well as realistic conditions, whereby various measured acoustic paths and disturbances, ambient sounds, and varying directions of arrival are considered.
All interested parties are cordially invited, registration is not required.
General information on the
colloquium, as well as a current list of dates of the
Communication Technology Colloquium can be fount at:
https://www.iks.rwth-aachen.de/aktuelles/kolloquium
Simone Sedgwick Secretariat Institute of Communication Systems(IKS) Prof. Dr.-Ing. Peter Jax RWTH Aachen University Muffeter Weg 3a, 52074 Aachen, Germany +49 241 80 26956(phone) +49 241 80 22254(fax) sedgwick@iks.rwth-aachen.de https://www.iks.rwth-aachen.de/