Fully Integrated Analog Machine Learning Classifier Using Custom Activation Function for Low Resolution Image Classification

摘要

This paper analyses the noise performances of coupled phase-locked loops, providing closed-form expressions for the transfer functions of the various noise sources, and presents a novel coupling topology, whose goal is the reduction of both in-band and out-of-band phase-noise. The proposed circuit consists of two, or more, standard loops coupled via an additional phase detector. It will be demonstrated that in this architecture the impact of the main PLL noise sources, the oscillator and the reference buffer, is efficiently traded with the power dissipation without resorting to lossy and area-consuming passive coupling networks. The work then shows how to derive the set of design parameters that grant the desired performance for a given case study. The entire procedure is verified by simulations.