Experience

HARMAN, HALOsonic Innovation Group, Novi, MI

Invented adaptive algorithms for musical interference cancellation and online secondary path IR estimation to prevent MFxLMS ANC systems from misadjusting due to the presence of music signals or changes in cabin acoustics
Developed LPC based noise shaping and power scheduling techniques to mask the presence of auxiliary noise used in online secondary path estimation algorithms

Developed a narrowband active road noise cancellation algorithm. Implemented a real-time frequency estimator by applying FFT Zoom and Spectral Centroids on accelerometer signals to provide an accurate reference to the ANC algorithm. Developed mathematical models to theoretically describe the tradeoffs between stepsize, leakage, notch filter response, and out-of- band noise boosting
Invented a virtual mic ANC algorithm to reduce engine noise at locations far away from error microphones using an adaptive array processing algorithm. This technology helps in achieving better engine noise control performance in underdetermined MIMO ANC systems with reduced or non-optimum error mic placement
Led the R&D efforts on auto-tuning algorithms for ANC systems using eigen-analysis of impulse response matrices.
Demonstrated improvements in noise cancellation and reduced total production tuning time by 70%
Developed simulation tools with GUIs in MATLAB for predicting & analyzing ANC performance
Developed audio & vibration analysis tools using CQT & Bark filterbanks, Variable Size DFT & multirate DSP algorithms

Implemented source separation algorithms using Non-negative Tensor Factorization (NTF), 8-9 dB SDR on audio mixtures
Developed SVM and CNN based audio classifiers using MFCCs & log-mel spectrograms on the UrbanSounds8K dataset
Prototyped an auditory augmented reality system pipeline to separate audio mixtures using NTF and identify individual sources using Deep Learning