Enabling ultra-low power operation in high-end wireless sensor networks nodes

Abstract

This paper presents a prototype hardware/software architecture for minimizing energy consumption on high-end microcontrollers, while simplifying the development of applications providing a general-purpose-like programming environment. The key features enabling this twofold goal are operating system support to processes, optimized sensing and hibernation of the system state. To balance performance offered by high-end microcontrollers with the need for ultra-low power operation-especially required by WSNsit is essential to minimize the duty cycle by keeping the microcontroller in a low-power state as long as possible, without affecting the performance required by the application. To cope with the increasing leakage current observed in today's microcontrollers a software-based hibernation mechanism has been implemented to transparently save the memory of processes in a non-volatile memory, allowing to completely switch off the microcontroller. To further avoid costly process loading overheads during the sleeping periods, we devised a smart sensing mechanism capable of gathering data from peripherals without restoring the state of processes. Preliminary results show that the lifetime of the proposed node architecture is of the same order of magnitude of that of ultra low-power nodes and significantly better than that of high-end ones.

Optimal hibernation policies for energy efficient stateful operation in high-end wireless sensor nodes

“…The hibernation mechanism has been implemented in MiosixOS, on the PoliNode prototype infrastructure [1]. This platform is based on a STM32F2 microcontroller, an external MR25H10 magnetoresistive RAM and an ultra low-power transceiver.…”

“…As a consequence, both private and public institutions are often resorting to universities and research centers asking for solutions which can mask this complexity and make WSNs more practical for commercial applications. In this direction moved the work we presented in [1], which proposed a joint development of hardware and software, especially at Operating System level, providing to the end user features similar to those of a general-purpose programmable system: a truly standard programming language, standard libraries, standard processes and threads, standard synchronization methods and programming paradigms, memory isolation and protection, application linking and loading. The operating model is inspired to the principle of leveraging data processing inside the network, pointed out since the very beginning of research on WSNs [2] and relying on the consideration-already true-that thousands or millions of operations per second can be done using the energy of sending a bit over a few tens of meters [3].…”

Tiny Neural Networks for Environmental Predictions: An Integrated Approach with Miosix

“…With our experiment we adopted a distributed architecture (e.g., Edge Computing) embedding Tiny ML, trying to take advantages from both of them. In order to ease and better integrate the Tiny ML algorithm at the edge of the system we selected Miosix, a configurable, tiny, open and real time operating system described in [4]: this choice enabled the design of a solution that exploits multithreading and process multi-sensors data patterns. The result of this implementation is a scalable and power-efficient system.…”

A Formal Model for Optimal Autonomous Task Hibernation in Constrained Embedded Systems

“…New-breed magnetoresisteve (MRAM), ferroelectric (FeRAM) or phase-changing RAM offer a good solution to both problems thanks to their unlimited number of write cycles and their limited energy consumption. Based on such considerations, we proposed [1] a powerful and energyefficient hardware platform and operating system for Wireless Sensor Networks (WSN), capable of overtaking some of the highlighted limitations.…”

“…Due to the timing and energy overheads required to swapout the system status, shut-down the operating system and resume it, the optimal choice whether to hibernate the system or put it into sleeping mode-with memory retention-is a very harsh problem, though simpler suboptimal policies exist. The decision, in fact, depends on a broad range of factors, such as the number of active processes, their periods and their sizes in RAM 1 , the specific energy levels of the microcontroller and the electrical characteristics of the external memory. The hibernation policy itself, moreover, should be autonomous, effective, dynamic, robust, computationally efficient and reliable.…”