Experimental Research

1
Experimental Research

• Alberto Sangiovanni-Vincentelli
• UC Berkeley

2
Overview

• Experimental research is an essential component
  of CHESS
• Feedback on approach
• Inspiration for new theory
• Impact
• Wide range
• Industrial and Government test cases
• Automotive (safety-critical distributed systems)
  to be covered in the afternoon
• System-on-Chip (high-complexity platforms)
• Signal Processing Applications
• Hierarchical and Distributed Control
• Internal experimental test benches
• Wireless Sensor Networks (security, low power)
• UAVs (complex control, sensor integration)
• New domains
• Hybrid Systems in Systems Biology

3
Overarching Criteria

• An application should exercise
• Theory hybrid models, Models of Computation,
  control algorithms
• Tools and Environments
• Path to implementation
• An application should be relevant for industry or
  for government agencies

4
Some Applications Addressed
Automotive
Avionics UAVs
Systems Biology
Networked Embedded Systems
5
Outline

• Industrial cases
• System-on-Chip (high-complexity platforms)
• Signal Processing Applications
• Hierarchical and Distributed Control
• Internal experimental test benches
• UAVs (complex control, sensor integration)
• New domains
• Hybrid Systems in Systems Biology

6
Metropolis and Xilinx Characterization Environment
Real Performance Data
ML310
Abstract Modular Model
Narrow the Gap
Synthesis File
Xilinx Virtex II
Metropolis currently has a flow to automatically
generate sample architectures, extract
performance information, and use that information
dynamically during simulation.
D. Densmore, A.Donlin, A. Sangiovanni-Vincentelli,
FPGA Architecture Characterization for System
Level Performance Analysis, Design Automation
and Test Europe (DATE), 2006. (to appear)
7
Metropolis Xilinx Design Environment
Real Performance Data
ML310
Abstract Modular Model
Narrow the Gap
Synthesis File
Xilinx Virtex II
Metropolis currently has a library of Xilinx
based components which a designer can instantiate
as an architecture instance. When composed their
structure can be extracted for performance data
or structural synthesis flows.
8
Xilinx Example Designs

• Metropolis and Xilinx flow highlights
• Produces accurate simulation results with
  fidelity.
• Can capture structural effects like clock cycle
  and resource usage.
• Large portions automatic, independent, and one
  time cost operations.

9
Intel JPEG Encoder Application
Pre- processing
DCT
Quantization
Huffman
10
Intel MXP5800 Architecture

• Designed for Imaging Applications
• Highly Heterogeneous Programmable Platform
• Top Level 8 Image Signal Processors with Mesh

11
Design Space Exploration

• Replication of scenarios from Intel library
• Accurate Performance Modeling
• Easy implementation of additional scenarios

A. Davare, Q. Zhu, J. Moondanos, ASV, JPEG
Encoding on the MXP5800 A Platform-based Design
Case Study, Proceedings of EstiMedia 2005
12
Picoradio Baseband System-Level Design
Explored the different partitioning between
analog and digital
Early-Late Gate synchronization algorithm (timing
recovery)
13
Design Space Exploration for Integrator
Explore the Analog Platforms

• Define configuration space
• different biasing, different device sizings, etc.
• Impose constraints
• bounding ranges for devices size, biasing
  conditions, etc.
• Characterization framework
• Matlab client generates configurations, and the
  configuration space is statistically sampled
• Ocean server manages circuit simulation in
  Spectre and extracts performance figures
• Generate feasible performance space

14
Outline

• Industrial cases
• System-on-Chip (high-complexity platforms)
• Signal Processing Applications
• Hierarchical and Distributed Control
• Internal experimental test benches
• UAVs (complex control, sensor integration)
• New domains
• Hybrid Systems in Systems Biology

15
Signal Processing Platform (SPP)
ToolchainSupported Activities (1)
System Modeling
Goal Component-based development of large-scale,
hard real-time embedded signal processing systems
Analysis/Simulation Translation
Component Modeling
Functional Validation
Latency
Design Space Modeling
Component Core Modeling
Timing
Generative Modeling
Analysis Interchange Format (xAIF)
Metamodel Verification
Data-Type Dependency
Platform Integration Modeling
Model Components
Dataflow Dependency
Configuration Translation
Platform Modeling
CoActive Platform Configuration
Generate Configuration
Used by Raytheon, for embedded DSP
applications Available via ESCHER
Platform Wrapper Synthesis
HW/SW Partitioning
Build
Component Allocation
Test
Structural Optimization
Instrumentation
SPML/GME
Translators
SPML/GME
Builder,Translator
Modeling Environment
16
Signal Processing Platform (SPP) ToolchainTool
Components (2)
SPML/GME System Design Space
Optimization Tools
DESERT Design space exploration
S2D
D2S
Analysis Tools
MATLAB Functional Validation
Signal Flow Modeling
SPML/GME Point-Design Configuration
Simulink/ Stateflow
S2A
AIRES Schedulability
Ptolemy
S2C
CO-Active Execution Platform
VHDL
CONF
Comm Interf
Libraries
Target HW
17
Large-scale, real-time embedded system
architecture modeling and analysis
Safety Models
C4ISR-FP ?FT
Safety Models
DES Models
Performance Analysis
C4ISR ?SIM
Goal Architectural modeling and analysis of very
large-scale, distributed real-time embedded
systems.
Used by Boeing and SAIC, for analysis of
embedded systems architectures. Anticipated code
size 30M SLOC
Scenario Net Conditions
ARINC 653 Partitioned Processor Utilization
Network Utilization
18
Model-based Tools for Embedded Fault Diagnostics
and Reconfigurable Control

• Visual modeling tool for creating
• Physical models of the plant
• Controller models (incl. reconfiguration)

Controller Models
Strategy Models
Hybrid


Diagnostics



Active

• Modular run-time environment contains
• Hybrid observer and fault detectors
• Hybrid and Discrete diagnostics modules
• Controller model library
• Reconfigurable controller

Model

Failure Propagation

Controller

Diagnostics

Selector

Fault Detector

Plant Models


Hybrid Observer

Interface Controllers
Reconfiguration


Manager


Run-time Platform (RTOS)
Used by Boeing for autonomous vehicles.
19
Outline

• Industrial cases
• System-on-Chip (high-complexity platforms)
• Signal Processing Applications
• Hierarchical Distributed Control
• Internal experimental test benches
• UAVs (complex control, sensor integration)
• New domains
• Hybrid Systems in Systems Biology

20
Hierarchical Distributed Control

• Model-based approach using Limited Lookahead
  Methods
• Application Complex systems made up of
  interacting subsystems Challenge Hierarchical
  control of Advanced Life Support (ALS) Systems
  for NASA regenerative systems
• Problem Specification
• Dynamic model of subsystems expressed as hybrid
  discrete-time equations
• Controller input discretized to finite number of
  values, i.e., control input finite space
• There exist buffers (real or virtual) between
  subsystems
• Individual independent controllers for
  subsystems, interactions handled through higher
  level controllers
• Modeling abstractions that focus on buffer
  input/output relations provide the framework for
  building models at higher levels.
• Design model-based controllers with limited
  look-ahead schemes that search for optimal
  control input in finite space.

21
Distributed Control applied to Advanced Life
Support (ALS) Systems
Constraint-based Distribution of resources Weekly
crew schedule
Global Controller
Supervisory Controller
Utility-based Optimize performance
AES Controller
WRS Controller
Crew Scheduler
Controller
Controller
AES System
Power Generation
Crew Chamber
WRS System
WRS System
ARS System
SABATIER
CDRA
RO
AES
OGS
BWP
LC-SAB
LC-CDRA
LC-OGS
LC-AES
LC-RO
LC-BWP
22
ALS Data flow Control
Measurement Command Mass Flow
water_level
System Resources Monitor
Supervisory Controller
O2_level
power_level
ARS_mode
WRS_mode
week_schedule
H2T_L_ARS
ARS Controller
O2T_L_ARS
Estimation module
CO2T_L_ARS
CDRM_mode, CDRM_time
OGS_mode, OGS_time
Crew Controller
eCW_FI_CRW eWW_FO_CRW
eCW_FO_ARS eWW_FI_ARS
HCA_FO_CRW
CRW_state
CO2_FI_ARS
day_schedule
WRS Controller
CDRA
WW_L_WRS
CW_L_WRS
CCH_state
LCA_FO_ARS
Crew
CO2
RO_mode, RO_time
Crew Chamber
AES_mode, AES_time
CO2_FO_ARS
SABATIER
O2 Reg.
PA_FI_CRW
AES
RO
H2_FI_ARS
O2T
H2T
CW_FI_CRW
WWT
CWT
PPS
BWP
CW_FO_WRS
WW_FI_WRS
O2_FO_ARS
OGS
CW_FI_ARS
H2_FO_ARS
WW_FO_CRW
WW_FO_AES
23
Results
Potable water Initial 650 liters End 200
liters

• Evaluate controller performance for 90 day
  challenge mission 4 astronauts in lunar habitat

Energy stored Min 200 kW-hour Max 1300 kW-hour
CO2 tank Initial 0 kg Max 2.6 kg Min 1.4
kg
Oxygen tank Initial 9.9 kg Max 10 kg Min
9.9 kg
24
Stability Analysis for Limited Lookahead Control

• System Dynamics
• Single-Mode Discrete-Time
• One-step online control policy

B(r,xs)
Q set of all states from which a control action
is available to move the system closer to xs
xs

• Technical Results
• To find B(xs)
• find (NLP)
• where
• Theorem B(r,xs) is the minimal containable
  region of xs
• To determine finite reachability
• Theorem ?B(r,xs) ? Q ? B(r,xs) is
• finitely reachable from x?Rn

• Objective
• For a domain D and an initial state
• xs?D, decide if there is a neighbor-
• hood B(r,xs) ? D of xs such that
• B(r,xs) is finitely reachable from any point in D
• The system remains in B(xs) under the online
  control law

25
Outline

• Industrial cases
• System-on-Chip (high-complexity platforms)
• Signal Processing Applications
• Hierarchical and Distributed Control
• Internal experimental test benches
• UAVs (complex control, sensor integration)
• New domains
• Hybrid Systems in Systems Biology

26
Time-Triggered Software for UAV

• Real-time systems, e.g., automobile control
  system, flight control system, air traffic
  control system etc, must produce their results
  within specified time intervals.
• Real-time systems can be classified to
  event-triggered systems and time-triggered
  systems.
• In the event-triggered system, all tasks are
  initiated by an event which can be sensor inputs
  or results of other tasks etc. It may be hard to
  specify precise time for any action due to
  variance of time of an event, which results in
  jittering of the system.
• In the time-triggered system, all tasks are
  initiated by predetermined points in time.
• A missed instant of any action can result in a
  catastrophe, possibly including the loss of human
  life, in hard real-time system.
• A hard real-time application demands a
  predictable, reliable and timely operation which
  a time-triggered system is able to guarantee.

27
Plant Berkeley Autonomous Helicopter

• Radio controlled helicopter from YAMAHA
• Control software was originally designed based on
  an event-triggered architecture
• We have decided to design and implement
  time-triggered embedded control software for the
  UAV as above

28
Time-triggered executing sequence

• Reading sensor inputs, writing actuator outputs
  and changing mode are happening at points of
  predetermined real time
• The time-triggered embedded architecture provides
  predictable (deterministic) operations of software

29
Test Results Hovering and Cruising
30
Summary

• Time-triggered embedded control software was
  designed and implemented for the Berkeley
  autonomous helicopter system
• Embedded control software was implemented with
  modularity in mind to keep the software clean and
  make it easy to read and enhance
• Software is structured to have multi-mode and
  mode switches among modes. New modes can be added
  and the current mode can be modified or removed
  with relative ease
• Designed software was mounted and tested on the
  safety critical helicopter system

31
Outline

• Industrial cases
• System-on-Chip (high-complexity platforms)
• Signal Processing Applications
• Hierarchical Distributed Control
• Internal experimental test benches
• UAVs (complex control, sensor integration)
• New domains
• Hybrid Systems in Systems Biology

32
Antibiotic biosynthesis in Bacillus subtilis
SpaI
SigH
discrete states (with randomness)
input
modeling with hybrid system
continuous states
SigH
output
SpaRK
SpaS
spaS
spaRK
S2
S1
33
Planar cell polarity in Drosophila

• Simulations
• Parameters estimation
• Study of mutants

phenotype
cell model
proteins feedback network
34
Box Invariance for biological reactions systems
A dynamical system is said to be box invariant if
there exist a box-shaped invariance set around
its equilibrium point(s)

• Concept of Set Invariance around the system
  equilibrium/a
• Naturally prone to describe biological systems
  (modeled via rate equations)
• More flexible than classical notion of
  (Lyapunov) stability
• Yields itself to describe robustness properties
• Closely related to lots of concepts from linear
  algebra and systems theory
• Can specify logical conditions for verification
  purposes

Claim most of the stable biological
reactions systems are indeed box
invariant Very descriptive concept.
In Collaboration with the A. Tiwari, SRI
International
35
Quantitative and Probabilistic Extensions of
Pathway Logic

• Pathway Logic (SRI Int.)
• tool for symbolic modeling of biological pathways
• based on formal methods and rewriting logic
• Protein functional domains
• and their interactions
• Queries performed through formal methods

• Extensions
• reasoning with quantitative data
• probabilistic interactions between different
  domains

In Collaboration with the PL team, SRI
International