Compressed Level Crossing Sampling for Ultra-Low Power IoT Devices

Level crossing sampling (LCS) is a power-efficient analog-to-digital conversion scheme for spikelike signals that arise in many Internet of Things-enabled automotive and environmental monitoring applications. However, LCS scheme requires a dedicated time-to-digital converter with large dynamic range specifications. In this paper, we present a compressed LCS that exploits the signal sparsity in the time domain. At the compressed sampling stage, a continuous-time ternary encoding scheme converts the amplitude variations into a ternary timing signal that is captured in a digital random sampler. At the reconstruction stage, a low-complexity split-projection least squares (SPLSs) signal reconstruction algorithm is presented. The SPLS splits random projections and utilizes a standard least squares approach that exploits the ternary-valued amplitude distribution. The SPLS algorithm is hardware friendly, can be run in parallel, and incorporates a low-cost k-term approximation scheme for matrix inversion. The SPLS hardware is analyzed, designed, and implemented in FPGA, achieving the highest data throughput and the power efficiency compared with the prior arts. Simulations of the proposed sampler in an automotive collision warning system demonstrate that the proposed compressed LCS can be very power efficient and robust to wireless interference, while achieving an approximately eightfold data volume compression when compared with Nyquist sampling approaches.