Masque de prsentation Communication Externe

1
Stockholm Workshop, June 16th, 2008
Source coding in feedback less bits, minimum
energy Carlos Canudas-de-Wit GIPSA-Lab,
CNRS Grenoble, FRANCE

• Communication needs
• Source coding in control
• An example coding/control co-design
• Perspectives

2
Needs from Communication /Control interaction

• Miniaturization low cost
• Coarse information
• Energy-limitations
• Large number of sensor units
• Collaborative information Management of
  connectivity
• Distributed use of computational resources

Needs
Difficulties

• Compressing data
• Minimizing transmitted bits
• Energy-aware transmission
• Reliable transmission
• Asynchronous transmission

• Heterogeneous
• Different clocks
• Energy-limited
• Varying topology,
• Large number measures

3
Feedback design for wireless networked controlled
systems
Feedback design for wireless networked controlled
systems
4C-E

• Innovations
• novel system concept control, communication,
  computation, complexity and energy (C4E)
  co-design framework for large-scale distributed
  control systems.

INFSO-ICT-223866
4
Consortium
Collaboration with CITRIS (at Bekeley).
5
The hidden revolution What you dont see will
need more careful watching

• Communication Efficiency increases
• Cost decreases
• Computation power increases

6
How to reduce the transmission cost
What cost means

• Cost in terms of bits numbers
• Cost in terms of energy per bit transmitted

Associated questions

• How to design a (constructive) coding strategy
  (source coding) close to the minimum
  (theoretical) bit rate ?
• How to make the coding method energy-efficiency ?
  
• How to make the coding algorithm usable for large
  number of sensors having a distributed location ?

7
Reservoir of source coding strategiesfor
potenital use in control

• Discrete Sources
• Fixed-length code words
• Variable-length code words (entropy coding)
• Huffman code minimizes the average numbers of
  bits for codes words that are uniquely and
  instantaneously (no delays) decodable.
  Probabilities of word occurrence is assumed
  known.
• Lempel-Ziv algorithm independent of the source
  statistics, and uses variable-to-fix length
  algorithm.
• Vector quantization
• Analogous Sources
• Temporal Waveform Coding
• Fixed and adaptive delta (differential)
  modulation (1-bit)
• Event-based coding (2-bits) energy-aware
• Spectral Waveform coding
• Subband coding waveform of each subband is
  encoded separately. Bits numbers depends on
  spectral energy concentration and/or noise
  distribution
• Adaptive transform coding more bits to the most
  important spectral coefficient

8
Example Differential CodingStandard
communication form

• Encode Differences
• Need good initialization
• Selection of is crucial
• Unclear properties when used in feedback set-up

Granulate error Too large
Slop overload Too small
9
Accommodation of Delta-modulation in Feedback
10
Benefits Limitations of the modified 1 Bit
delta modulation
Benefits

• Few bits
• Cascade structure between coding error and
  closed-loop equation
• Methodological constructive

11
Adaptive Delta modulation
Adaptive
Fixed
12
Ideas for Reduction of minimum energy per bit
transmitted
Event-based 3 level codingWord-by-word
transmission
13
Optimizing Transmission Energy by using entropy
VL coding package-based transmission
14
package-based transmission optimizing Energy use
by using entropy VL codingCont.
Mean transmission rate
Energy efficiency ratio
15
Regulation problems (set-point control ) are
suited situations for the use of VLE coding
16
Decentralized vs Centralized Coding for MIMO
systems

• Centralized coding encoder and decoder (and
  control) uses the full available information from
  ALL sensors.
• ALL sensed system outputs are collected in a
  central point

17
A global view pointTransform Coding
18
Achievements , Needs future directions

• Some Achievements
• Some understanding on fundamental limitations,
  i.e. Rate limits, logarithmic optimal
  quantization, etc
• New concepts, i.e. rate theorem, channel
  any-time capacity,..
• Partial results on control communications
  co-design, i.e. zoon-in zoom-out coding
  algorithms
• Some Needs
• Constructive control analysis Volume,
  Lyapunov, ISS,,..
• Integrated view of the problem conceptualize
  communication, computation energy. as a general
  control resources
• Tools methods
• Some Directions
• Bring and adapt existing communication
  concepts/ideas into the feedback loops
  transform coding,
• Asynchronous control/communication co-design