PANDA: Performance Acceleration through Nonuniform Data Acquisition

Panda is a technique to reduce the energy required to sample analog signals on battery-powered IoT devices. We sample the analog signal nonuniformly in time which reduces the energy overhead of repeatedly waking and sleeping the microcontroller. Using algorithms we developed, we can reconstruct frequency domain information from the nonuniformly spaced samples.

Papers

PANDA: Performance Acceleration through Nonuniform Data Acquisition
Neil Klingensmith, Suman Banerjee
ACM eEnergy 2018


A Fast Algorithm for Energy-Efficient Sampling of Analog to Digital Converters
Neil Klingensmith, Suman Banerjee
ACM TIOT 2019 (To Appear)

Hermes Real-Time Hypervisor

Watch video

Hermes is a hypervisor for MCU-based systems with real-time requirements. The goal of Hermes is to manage tradeoffs between performance and flexibility in software that runs on embedded systems. Bare-metal software often has the best performance—low I/O latency, less CPU and memory overhead, etc. RTOS-based software has comparatively higher flexibility in terms of its ability to share resources and manage multiple jobs, but often sacrifices latency. Hermes aims to provide software developers with the option to run some portions of code in a bare-metal like environment with minimal latency while allowing other jobs to run within a more flexible RTOS.

Papers

Hermes: A Real Time Hypervisor for Mobile and IoT Systems
Neil Klingensmith, Suman Banerjee
ACM HotMobile, Tempe, AZ, February 2018

A Hypervisor-Based Privacy Agent for Mobile and IoT Systems
Neil Klingensmith, Suman Banerjee
ACM HotMobile, Santa Cruz, CA, February 2019

Emonix Water Treatment Controller

Existing water softeners use open-loop controllers that do not include water quality sensors to trigger regeneration. As a result, they generally regenerate too frequently, wasting salt and water and polluting the environment. Emonix aims to reduce the frequency of water softener regenerations by sensing water quality and applying domain knowledge to trigger regenerations at optimal times. The sensor readings, which may be noisy or inaccurate, are processed by backend algorithms to accurately predict when regeneration is required.

Papers


SPOCK: A Sensor Value Prediction and Online Control Algorithm for Building Resource Management
Neil Klingensmith, Ananthraghavan Sridhar, Zachary LaVallee, Suman Banerjee
ACM BuildSys, Stanford, CA, November 2016


Water or Slime? A platform for automating water treatment systems
Neil Klingensmith, Ananthraghavan Sridhar, Zachary LaVallee, Suman Banerjee
ACM BuildSys, Seoul, KR, November 2015

Hot, Cold, and In Between

Papers

Hot, Cold and In Between: Enabling Fine-Grained Environmental Control in Homes for Efficiency and Comfort
Neil Klingensmith, Joseph Bomber, Suman Banerjee
ACM eEnergy, Cambridge, UK, June 2014