We introduce CRADLE, or combined RF/acoustic detection and localization, for the remote parametric ranging of passive tags. In this approach, a tag consists of an RF antenna connected to an ultrasonic transducer via a passive electrical network. A reader looks for tags by transmitting both continuous-wave RF and pulsed ultrasound into a region of interest; when a tag is within that region, the ultrasound incident on the tag transducer varies the transducer capacitance, creating a time-varying load for the tag antenna and thus parametrically modulating the RF waves reradiated from the tag. The reader localizes the tag by demodulating the received signal to recover the acoustic time of flight. In this paper, we present a proof-of-concept CRADLE system, describing the design of a custom tag and reader. The results of our localization experiments show that we can achieve sub-decimeter accuracy at reader-tag distances of up to 6 m using the proposed system; future improvements in the design of the tag and reader can result in localization of centimeter-sized tags with sub-centimeter accuracy out to distances of several meters.

Analog/mixed-signal (AMS) computation can be more energy efficient than digital approaches for deep learning inference, but incurs an accuracy penalty from precision loss. Prior AMS approaches focus on small networks/datasets, which can maintain accuracy even with 2b precision. We analyze applicability of AMS approaches to larger networks by proposing a generic AMS error model, implementing it in an existing training framework, and investigating its effect on ImageNet classification with ResNet-50. We demonstrate significant accuracy recovery by exposing the network to AMS error during retraining, and we show that batch normalization layers are responsible for this accuracy recovery. We also introduce an energy model to predict the requirements of high-accuracy AMS hardware running large networks and use it to show that for ADC-dominated designs, there is a direct tradeoff between energy efficiency and network accuracy. Our model predicts that achieving < 0.4% accuracy loss on ResNet-50 with AMS hardware requires a computation energy of at least ~300 fJ/MAC. Finally, we propose methods for improving the energy-accuracy tradeoff.