Title: Sensor Network Hardware Platform Design Andreas Savvides Embedded Networks and Applications Lab ENAL
1Sensor Network Hardware Platform Design
Andreas SavvidesEmbedded Networks and
Applications LabENALABhttp//www.eng.yale.edu/en
alabYALE EE CS DepartmentsApril 27, 2005
Research Supported by
2Hardware Platform Design
- Platforms in applications and deployments
- Computation requirements in applications and
design - Platforms vs. application needs
- Hardware design and interface issues
- Experiences with the platform development process
- Emphasis topic Hardware characterization
- Power characterization is discussed in SPOTS
papers/posters - I will pick on antenna behaviors in 3-D scenarios
- Algorithms and platforms should change together
- HW platforms can still change the way we think
about algorithms
3Hardware Sensing Platforms
HW Platforms
Experiment with unknown environments
Shrink the HW
NIMS Nodes _at_UCLA
UC Berkeleys Spec Node Smartdust
Intelligent Integrated Sensing Network Platforms
4Hardware Platform Priorities
HW Platforms
Experiment with unknown environments
Shrink the HW
Understanding unknown sensing phenomena
Power Cost Reduction
NIMS Nodes _at_UCLA
UC Berkeleys Spec Node Smartdust
Intelligent Integrated Sensing Network Platforms
5 Platforms vs. Application Needs
- Each application has different computation,
memory and interface requirements - Wide range of applications requirements
- Surveillance
- Medical care
- Structural health monitoring
- Traffic management
- Tracking fires
- Environmental exploration
- Child motion monitoring
- Hard to create a single platform for all
applications - Links to SPOTS Platforms - Pages 429 431 of the
proceedings
6Opportunities for New HW at Different Levels
- Processor core
- New instructions
- Support for different power modes
- Peripherals
- Need new custom peripherals
- Often running as different HW treads
- Sensors
- Create new sensing modalities
- Move computation and intelligence inside the
sensor - Still many tradeoffs and engineering challenges
to address
7When should you attempt to build a new platform?
- If you have a specific problem in mind for which
existing platforms wont suffice - If you plan to create a hardware component for
which you need tight control of the hardware - If cost and size become a limiting issue
- Need to consider
- What is the benefit of having own platform?
- Is this going to enable or handicap your research
effort?
8Plan your priorities
- What is your design objective?
- Avoid building new HW for the sake of building
- Target a specific feature or application
- Power consumption vs. proof of concept
- Which is more important to you?
- Proof of concept
- Over-design vs. under-design
- If the algorithm is known, size and power become
the focus - If the algorithm/application is not known you
need to relax the constraints
9Before you begin to build a sensor node
- Are the tool chains available?
- Make sure you have all the tools you need to
complete the cycle available - Flash programmer
- Compiler
- Debugger JTAG tools
- Is the processor chip you are using mature?
- If not, then dont use it unless you have
collaboration with the manufacturer - Get the development kit first and try to write
software before you start - Does the radio you are using have software
support/tools?
10Design Tools and Component Selection
- Try to use well established packages, ORCAD for
instance - Typically available from the CAD tools suite
- Easier to find/share component footprints
- This is one of the most time-consuming and
error-prone part of the process - Make sure you select the right components
- Components come in different packages
- Components have different cost and power
consumption - Good idea to purchase all the components before
the prototype PCB is sent to fabrication - If you plan to build large numbers, talk to
people who did it before first
11Design Considerations
- Design for manufacturability
- Take into account that you need to build more
- Plan for an economical way to do it
- Capitalize on the fabrication cycle
- Most companies have reduced rates for 4 week runs
- Assembly houses may have specific requirements on
assembly - Find out about this before you begin
- Talk to the assembly house before you finalize
your design - Some PCB boards and components will require
fiducial points for machine assembly - Some manufacturers may be able to suggest
alternative components - Put testpoints for debugging and power
characterization during operation - Be careful with radios they have specific PCB
requirements
12Developing your PCB
- Look around for existing designs first
- Investigate parts
- Availability, packaging, power consumption cost
- Get your tools together for the whole process
first - Schematic capture and review process
- Layout
- Double check your component footprints
- Talk to the manufacturer some places charge
less for 1 phase board - Odd number of layers does not save you money
- Make sure you follow the manufacturer directions
for radio laout - Make sure you wire the board for test
measurement - Plan for testing
13SmartKG iBadge Platform (NESL/UCLA)
- One of the most highly integrated sensor
platforms - Hard to build very small components 0201
components, difficult to machine assemble - Production and assembly costs is a limiting
factor - Lots of educational value!
14Study Case Building the XYZ
- Work with Cogent Computer
- Small single board computer company in Rhode
Island - Already has expertise and interest in embedded
ARM - Collaboration with OKI Semiconductor
- Make sure that all the peripherals are available
- Talk to Chipcon to make sure they would have an
IEEE 802.15.4 MAC available - Design prototype according to our specification
- Second pass design with Cogent Computer
- Identify inexpensive components
- Make the design easier to manufacture
- 1 side, 6-layer board
- Placement done to accommodate hand and machine
assembly
15Example XYZ Mobility Ultrasound Board
- Align components to make low production assembly
and debugging more efficient - Makes hand assembly or low end machine assembly
easier
16Lessons Learned
- Dont bother soldering everything by hand
- Look for places esp. local shops that can help
you - If the layout is too complex, outsource to an
expert - Cost is the same if you consider the lost time
and the possibility of bugs - Pace yourself
- Long, organized planning period
- Fabrication assembly cycle (2 to 6 weeks)
- Have a support strategy for the system
- How are you going to make more, distribute it,
test it, use it etc. - Plan for iterative implementation and
customization - After some field deployment you will probably
need to make some changes - Verify the software and programming cycle before
you finalize the hardware design
17Lessons Learned
- Go for the mainstream design tools
- Design for manufacturability and testability
- Be aware of what if already available
- Look into the community to see if there are
pieces you can reuse - Reconsider picking platform development as a
research topic if other companies are doing it - Ember OKI have IEEE 802.15.4 implementations on
radio chip - re-implementing the same MAC w/o a longer term
plan will have short half-life
18After Fabrication Completion
- Have a test strategy in mind for SW HW
- Write diagnostic code to check each subsystem
- Diagnostic code should become part of the runtime
environment - Treat your new platform as a new device.
Characterize it! - Characterize power consumption at different modes
- Characterize platform in a realistic environment!
- Push the platform to the limits, know where
things break down - Post your data, this is would be the most
valuable asset to the community - Example Antenna Characterization for CC 2420
- PCB design affects the antenna
- Characterize radio and antenna properties in 3D!
19Chipcon CC2420 Radio Power Levels
RSSI_VAL Computed by the radio over 8 symbol
periods (128us) RSSI_OFFSET Determined
experimentally, based on front end gain (around
-45dBm) Approx. Range at power level 6 in an
office corridor 30ft Antenna Length 2.9cm
20Radio Calibration for TX and RX
- Each radio chip is different
EPr29.94dBm s2.7dBm
40cm
10 Different Transmitters
10 Different Receivers
EPr26.375dBm s2.88dBm
40cm
21Orientation variations at ground level
- Repeat experiment for 4 different nodes, same
receiver - TX Power -15dBm
- 8 different positions, 4 orientations for each
position
22Indoor Path Loss Measurements
Floor measurements in a 24 x 20ft lounge no
obstacles
Same power level using suboptimal antenna
?3
23Indoor Path Loss Measurements
Floor measurements in a 24 x 20ft lounge no
obstacles
24Monopole Antenna Radiation Pattern
Side View
Top View
Communication range
Symmetric Region
Antenna orientation independent regions
Communication range
25RSSI at Different Antenna Orientations
- At the bad orientation, antenna has to be at
similar height to get proper results
263-D Radio Connectivity
27Link Asymmetry in 3D-scenarios
of one-way links
28 RSS Asymmetry at Different Power Levels
29Platforms in Undergraduate Curriculum Setting
up a lab
Capstone Project
EENG 460a Networked Embedded Systems S.
Networks
EENG 449 Computer Systems
- Embedded and Real Time OS
- Radio Technologies and MAC
- Routing for small devices
- Sensor network applications
- Self-Configuration
- Data Storage
- Mobility and Actuation
- Expect to have a research
- caliber project
- Undergraduates participate on
- research papers
- Computer Architecture
- Embedded Processors
- Assembly Language
Most important assets 1. Develop HW intuition
early on 2. Have fault diagnostic code for the
device
30Conclusions
- Building HW is a great learning experience and
adds to the diversity - Useful to uncover new ideas and concepts
- More insight, more prudent researcher
- Close consideration with software design is
crucial - HW changes faster than SW
- One of the biggest challenges
- Radio technology
- There is a large domain of problems for which the
radio may not be sufficient - Need to become more critical of radio
capabilities in applications - Try out different radios!
- Data traces and benchmarks are still missing
- Need better ways of reporting power and
performance - Utility value in terms of the application should
be factored in