trajectory-directed discrete state space modeling for formal verification of nonlinear analog...
TRANSCRIPT
Trajectory-Directed Discrete State Space Modeling for Formal Verification of
Nonlinear Analog Circuits
Presented by Valeriy Balabanov
Outline
Introduction
Problem description
Main algorithm
Experimental validation
Summary and discussion
Introduction
• Digital circuits vs Analog circuits– Digital circuits
• Operate with discrete signals• High level of automation• Many model and equivalence checking tools
– Analog circuits• Operate with continuous state space• Model and equivalence checking still needed• Deal with nonlinear differential–algebraic equations (DAE)
– Analytical approaches are not feasible – Good discretization methods are needed– Discretization error
Analog state space
• First order nonlinear DAE– x – vector of variables– x’ – first derivative (vector of velocity vectors)– u – input variables
• State space is spanned by a linearly independent subset z• Extended state space
• Candidates for state space variables can be identified in the DAE by their occurrence as first-order time derivatives– Example (capacitor):
Discrete analog transition structure
Problem description
• State space need to be partitioned (discretized)
User specified bounds
Partition of state space Z into R1 .. Rk
Maximum length error
Number of partitions
Overall mean-out degree error
Overall mean successor relation error
Maximum direction error
Main algorithm
• Discretization shall be rotation invariant – State space intersections cannot be axis-parallel
• Over-approximation of the successor relation significantly weakens expressiveness of verification algorithms– Geometric structure of partitions should follow the flow of state space
dynamics– Intersections should be either parallel or orthogonal to the state space
trajectories
• Use time step control algorithm to ensure homogeneity of the enclosed state space dynamics
• -> Trajectory directed discrete modeling algorithm
Main algorithm (example of partition)
Main algorithm
Coordinate transformation to centralize/normalize vectors
Random starting point that is not a DC-operating-point
Gram-Schmidt procedure
Control discretization error
Find new points by
Control the structure of the new points in order to avoid overlapping with existing points
Main algorithm
Main algorithm
• Mapping the trajectory-directed partitioning to DATS
Main algorithm
Experimental validation
• TDD (trajectory-directed discretization method) vs HBD (hyper box discretization)
Experimental validation
Experimental validation
• Model checking case study
• Has been tested and widely manufactured• Only lately found that under certain conditions
has critical behaviour
Experimental validation
Experimental validation
Summary and discussion
• Paper presents a completely new approach for state space discretization
• New algorithm outperforms existing one in partitioning strength
• There are many application in modern mixed (analog-digital) designs
• The material in paper is well presented• No visible improvements are needed
Thank you!