Design Principles for Efficient Smart Sensor Systems

1. Modular Platform for High Density Wireless Sensing Problem Overview Sample Problems Possible Solutions • The goal of this project is to create a set of design principles to aid in the creation of efficient smart sensor systems. We define such systems as the general class of sensors systems that • have the ability to analyze their state and surroundings • and • respond to them by changing their mode of operation in some beneficial fashion. • Given this definition of smart sensors, we need to consider the properties which comprise efficiency. Of the possible criteria: • Power usage • Sensor area/volume • Processing power • RF bandwidth / Storage volume • we note that power cell density is advancing at a particularly slow rate (doubling every 15 years). Further, the use of rest of the above draws substantial power, again making it a key limitation in smart sensor systems. • Therefore, our goal is to design smart sensors systems to accomplish their given task using as little power as possible (on average). The following procedure will be followed for any given problem: • Identify world states of interest for the system. • Determine the parameters to be calculated in each state. • Build recognition system to determine current state. • In real-time, recognize state and alter operation dynamically to be optimal in each. • We consider two scenarios: • a system which must decide solely on its owninformation what data to collect and how to do so • and • 2. a system that must collate data from a multitude of other sources to determine how it can best contribute to an analysis. • 1. Medical logging device • Stand-alone system (ie no external data sources) • Collects and logs data over extended time • Sample states: • Still: Sample tilt switch to make sure state doesn’t change. • Twitching: Sample accelerometer at lower rate to ensure no forward progress is being made. • Walking: Sample accelerometer at full rate and calculate gait parameters • 2. Node in ad-hoc sensing network • One of many similar sensor systems • Combines its own and others’ data to make decisions • Sample states: • Interesting data? • None: Sample at very low rate looking for state change. • Own: Sample at highest rate, calculate parameters, send data to others. • Others’: Use others’ data to determine rate at which to sample, how much data to send. We are currently considering three main approaches to reducing the power consumption of the smart sensor systems in their various states. We will concentrate on sensor and data related issues. A schematic of the proposed system is shown below. As a first platform for testing these ideas, we have designed and built a modular architecture based on 1.4” sq. boards. Each board has a pair of 26 (14 + 12) pin headers to connect signals and power to other panes in the stack. The simplicity will which sensors can be added and removed with greatly aid in prototyping new designs as described at left. Microcontroller Data State Analysis Sensor 1 Sensor Controller Sensor 2 Transceiver State Algorithm Selector Enable Sensor n Data Compression Storage Processed Data 1. Real-time sensor selection System will choose and activate the sensor which provides just enough information to confirm world state and calculate any parameters of interest. Sampling accuracy and rate are also variable. 2. Variable accuracy processing Alter order of processing to calculate parameters based on most relevant data first, thereby achieving a smaller percentage error much more quickly than conventional algorithms. Error monotonicity allows for power/accuracy trade-offs. 3. Selective transmission/storage of data Save bandwidth / power by using current and previous data (from all sources) to compute importance of most recent data through information theoretic techniques (ie find the innovation or differential entropy). • Counter clockwise from bottom left: • Main board • 22 MIPS processor with 12 bit analog to digital converter • 115.2 kBps wireless transceiver with TDMA channel sharing • Inertial Measurement board • Full 6 DOFs in flat package • 3 axes of accelerometers • 3 axes of gyros • Tactile board • Inputs & signal processing for • FSRs • PVDFs • Bend Sensors • Power regulation board • Provides for battery input • Outputs of 3.3V and 5V Design Principles for Efficient Smart Sensor Systems