The objective of this effort is to develop a flexible platform for experimentation with densely populated, physically situated wireless networks.  This is a follow-on to an initial development effort to develop a simple "tag" which beaconed over a radio transmitter at periodic intervals.  This talk summarizes some of the things we learned from the initial implementation.  The project has inherited the name "tags", although the envisioned scope of applications is much more broad.

One critical aspect of this project is the development of localization systems. Without localization it is very difficult to create interesting applications.

Design Overview

The design we have selected is a modular design in which a device is composed of a collection of smaller modules, each of which has its own processing unit.  The top-level controller is usually powered, and is designed to be very efficient.  It coordinates the activities of the other modules and can selectively power them down to conserve energy.  Rather than having a single CPU that globally schedules a number of real-time tasks, real-time scheduling is delegated to separate controllers that are dedicated to real-time tasks.  The use of local controllers may save energy by enabling resource allocation at a finer granularity, and by reducing overhead associated with task switching.

Before settling on this design, we decided against several other options.

• Off-the-shelf PC hardware.  This approach is too expensive, both in terms of money and power.  Although software development using existing tools and operating systems is often easier in the short term, the operating system environments that are designed for PCs often get in the way of building lightweight, power-aware systems.
• Integrated devices built around a single large processor.  There are two aspects that made this suboptimal for our needs.  First, having a single CPU that does everything often requires a significant amount of RAM, increasing the power consumption.  Second, because we don't know exactly what hardware features we want, a more modular approach leads to less dependency on choices we make today.

• The Radio Module.  This module is based on the Atmel MegaAVR 103L, a RISC microcontroller with 4K of on-board SRAM, and 128K of flash program store (64K AVR instructions). This processor is a larger version of the processor used in the Berkeley MacroMote, developed as part of the "Smart Dust" project. The Radio Module has a 32K battery-backed SRAM that enables it to store more state and perform more complex tasks. The backup battery allows it to maintain its state even while it is in a low-power sleep mode. It can communicate using an on-board RFM TR-1000 radio module. The board can also measure approximate RSSI and control the transmit power. A port of the TinyOS system, which was designed on very similar hardware, is currently underway. Schematics for the Radio module can be found here.
• The Sensor Module.  This module has a suite of sensors and actuators, currently driven by a PIC 16F877 microcontroller. A follow-on version of this board will instead incorporate an Atmel Mega 103. The current sensor module includes:
  • Light Sensor
  • Accelerometer
  • Temperature Sensor
  • Microphone
  • D2A converter, with amplifier for driving a speaker
  • IR LED
  The Sensor board also includes a low power acoustic wakeup circuit that can wake the system when a particular frequency acoustic signal is detected. Schematics for the Sensor module can be found here.

The system bus is composed of

• a power bus,
• an I²C synchronous serial bus,
• an SPI serial bus,
• room for several other serial ports,
• 16 lines reserved for GPIO and interrupt lines between modules.

Applications

There are several planned applications of this hardware. Currently, the sensor modules are in use, stand-alone, as sensors connected to PC-104 class systems in the SCADDS testbed. This testbed is being used for experiments with "directed diffusion" and related protocols, and the sensor modules are providing data for the experimental applications.

The Radio modules will be useful as a follow-on to the Berkeley Mote hardware. The SCADDS testbed is currently migrating over to use Berkeley Motes as programmable radio network interfaces. Ideally, devices on the scale of the motes would be able to implement a scaled-down version of "diffusion"; however, the limitations on the program size and memory size inherent to the mote design may pose a problem. The Radio module, with its larger memory capacity, might be a good way to migrate, especially because most of the hardware is compatible.

The speaker output capability of the Sensor modules will also be useful in acoustic localization experiements. By sending a coded acoustic signal, it may be possible to estimate the location of the tag. This localization work is described here.

Related Work

• "Smart Dust Motes"
• Micro amps project
• The PicoRadio group at the BWRC