Analog Sparse Approximation For Compressed Sensing Recovery

Non-smooth convex optimization programs such as L1 minimization produce state-of-the-art results in many signal and image processing applications. Despite the progress in algorithms to solve these programs, they are still too computationally expensive for many real-time applications. Using recent results describing dynamical systems that efficiently solve these types of programs, we demonstrate through simulation that custom analog ICs implementations of this system could potentially perform compressed sensing recovery for real time applications approaching 500 kHz. Furthermore, we show that this architecture can implement several other optimization programs of recent interest, including Smoothly Clipped Absolute Deviations and group L1 minimization.