Amplitude demodulation of arbitrary-bandwidth signals
* Presenting author
Abstract:
The factorization of signals into slow-varying modulator-envelopes and fast-varying carriers, commonly known as amplitude demodulation (AD), is widely used in different domains of acoustic signal processing. For example, AD is a part of signal processing strategies behind ultrasound imaging, source separation, speech synthesis and recognition, and cochlear prosthesis. Originally, AD was devised for locally narrowband, i.e., sinusoidal signals. However, many relevant problems require demodulating wideband signals. This poses a still-unresolved challenge.Here, we introduce a new convex-programming-based approach to the AD of arbitrary-bandwidth signals where demodulation is framed as a problem of signal recovery from an unlabeled mix of its true and corrupted sample points. While featuring recovery guarantees for many practically relevant signal classes, the new method surpasses all competing techniques in computational efficiency by many orders of magnitude. This opens the door for AD applications in previously inaccessible settings. In particular, the new method makes online and large-scale offline data processing feasible, including the calculation of modulator-carrier pairs in higher dimensions and poor sampling conditions independent of the signal bandwidth. We demonstrate the power and versatility of the new method by applying it to process synthetic and natural speech signals.More information is available at https://arxiv.org/abs/2102.04832.