Cyber Physical Systems –
Solutions for Design Challenges
Part 1- Energy and Architecture
Sumanth Bhat
7th July 2015
Recap: Design Challenges for CPS
1. Abstraction issues
2. Timing issues
3. Architecture models
4. Miniaturization and Energy efficiency
5. Precision
6. Security & Privacy
7. Standardization
Low Power Designs for CPS
• Can be optimized at various layers
• Low energy VLSI design techniques - Near threshold Vdd operation
Low power communication
• “Communication is more expensive than computation”
• Edge mining techniques
1. Spanish Inquisition Protocol
2. Bare Necessities
Edge Mining
• “Data mining that takes place on the wireless, battery-
powered, smart-sensing devices that sit at the edge of
the IoT”
• Aim : To reduce number of sensing messages (packets) ,
thus achieving lower energy consumption and storage
area.
Data mining at the edge
• Instead of streaming raw data,
introduce state estimation and event
detection.
• Ex : In case of just streaming the data of
whether a new person has entered a
room, send out aggregated data such
as how many people are there in the
room etc.
Edge mining and privacy
• Due to introduction of states and event detection in edge nodes, raw
data is not stored anywhere.
• This makes it impossible to use available data for using it to any other
purpose than intended purpose.
• Example : Edge mining with GPS data may send out data such as “At
home”, “At work”, “In transit” rather than sending out raw GPS co-
ordinates. Thus, tracking your activities at minute level is not possible.
Introduction to SIP
• Spanish Inquisition Protocol
• Edge node transmits only that which the data sink (central
database server) does not expect
• Models can be fit in for the measured parameter, and hence only
the measured values that are not expected / outliers can be
transmitted.
Algorithm
1. Sensing : Obtain vector of sensor readings, and current time
2. Filtering : Estimation of new state using current sensor readings and old
state and time values
3. Prediction : Predict the sink state using old sink state and timing values,
and new timing value
4. Innovation Calculation : This is the difference between the new state
estimate and the predicted sink state
5. Conditional Transmission : Transmit new state and sequence number to
server if,
• innovation is greater than some threshold, or
• time since the last transmission exceeds a threshold
6. Acknowledgement: Post this stage, local copy of sink state and timing is
updated.
Performance
• Packet transmission reduced by around 95.5 %
Time-discounted Histogram Encoding (BN)
• Bare Necessities (or BN) is used for summarising relative
time spent in given states.
• Example, for determining how long is spent in a certain
modality (walking, driving, standing) in a given day.
• BN reduces the number of transmissions by 99.98%
Algorithm..
• If a sensed value is within a certain bin, then the bin count
B− is incremented
• A discount factor is also applied so that more recent states
are given more weight than older states.
• The counts are then converted to a normalised distribution
Architecture
• Needs to cater for heterogeneous
network performance
• Real Time Control
• Context-Aware Security
• Hard to come up with unified
CPS architecture!
3-tier Architecture overview
Service Oriented Architecture

Internet of things - 3/4. Solving the problems

  • 1.
    Cyber Physical Systems– Solutions for Design Challenges Part 1- Energy and Architecture Sumanth Bhat 7th July 2015
  • 2.
    Recap: Design Challengesfor CPS 1. Abstraction issues 2. Timing issues 3. Architecture models 4. Miniaturization and Energy efficiency 5. Precision 6. Security & Privacy 7. Standardization
  • 3.
    Low Power Designsfor CPS • Can be optimized at various layers • Low energy VLSI design techniques - Near threshold Vdd operation
  • 4.
    Low power communication •“Communication is more expensive than computation” • Edge mining techniques 1. Spanish Inquisition Protocol 2. Bare Necessities
  • 5.
    Edge Mining • “Datamining that takes place on the wireless, battery- powered, smart-sensing devices that sit at the edge of the IoT” • Aim : To reduce number of sensing messages (packets) , thus achieving lower energy consumption and storage area.
  • 6.
    Data mining atthe edge • Instead of streaming raw data, introduce state estimation and event detection. • Ex : In case of just streaming the data of whether a new person has entered a room, send out aggregated data such as how many people are there in the room etc.
  • 7.
    Edge mining andprivacy • Due to introduction of states and event detection in edge nodes, raw data is not stored anywhere. • This makes it impossible to use available data for using it to any other purpose than intended purpose. • Example : Edge mining with GPS data may send out data such as “At home”, “At work”, “In transit” rather than sending out raw GPS co- ordinates. Thus, tracking your activities at minute level is not possible.
  • 8.
    Introduction to SIP •Spanish Inquisition Protocol • Edge node transmits only that which the data sink (central database server) does not expect • Models can be fit in for the measured parameter, and hence only the measured values that are not expected / outliers can be transmitted.
  • 9.
    Algorithm 1. Sensing :Obtain vector of sensor readings, and current time 2. Filtering : Estimation of new state using current sensor readings and old state and time values 3. Prediction : Predict the sink state using old sink state and timing values, and new timing value 4. Innovation Calculation : This is the difference between the new state estimate and the predicted sink state 5. Conditional Transmission : Transmit new state and sequence number to server if, • innovation is greater than some threshold, or • time since the last transmission exceeds a threshold 6. Acknowledgement: Post this stage, local copy of sink state and timing is updated.
  • 10.
    Performance • Packet transmissionreduced by around 95.5 %
  • 11.
    Time-discounted Histogram Encoding(BN) • Bare Necessities (or BN) is used for summarising relative time spent in given states. • Example, for determining how long is spent in a certain modality (walking, driving, standing) in a given day. • BN reduces the number of transmissions by 99.98%
  • 12.
    Algorithm.. • If asensed value is within a certain bin, then the bin count B− is incremented • A discount factor is also applied so that more recent states are given more weight than older states. • The counts are then converted to a normalised distribution
  • 13.
    Architecture • Needs tocater for heterogeneous network performance • Real Time Control • Context-Aware Security • Hard to come up with unified CPS architecture!
  • 14.
  • 15.