Data processing: artificial intelligence – Fuzzy logic hardware
Reexamination Certificate
2000-09-05
2003-10-21
Siek, Vuthe (Department: 2825)
Data processing: artificial intelligence
Fuzzy logic hardware
C703S004000
Reexamination Certificate
active
06636839
ABSTRACT:
BACKGROUND OF THE INVENTION
The present invention relates generally to the analysis of the steady-state behavior of circuits driven by a periodic input signal.
The rapid growth in the field of wireless RF telecommunications has provided significant motivation for the development of simulation tools that can quickly and accurately analyze both the steady-state behavior and the response to multiple-frequency excitations of nonlinear circuits.
FIG. 1A
depicts a generic circuit
10
that is frequently analyzed in engineering applications. Circuit
10
is driven by a periodic input signal
11
, u(t), of arbitrary shape, and has an output
12
, v(t), that is periodic with the same frequency as u(t) when circuit
10
has reached a steady state. There is a great need for computationally efficient techniques for determining the steady-state behavior of a circuit such as circuit
10
, particularly given the fact that frequently the circuit behavior must be characterized over a wide range of input signal frequencies and amplitudes.
One traditional technique for determining a periodic steady-state solution (PSS) is transient analysis. This technique involves determining respective states of the circuit at a series of timepoints (starting at time zero with a set of initial conditions), the state determined for the circuit at any particular timepoint depending on the state determined for the circuit at the previous timepoint. The technique terminates upon detecting that a steady-state has been reached. Transient analysis requires several timepoints per period of the input signal, and thus is often impractical in circuits having a time constant many orders of magnitude greater than the period of the input signal. For example,
FIG. 1B
depicts a self-biased amplifier circuit
20
that is driven by a periodic input signal
21
, u(t). The time constant of circuit
20
might be chosen to be several orders of magnitude larger than the largest possible period for u(t), thereby requiring transient analysis over thousands of timepoints.
Another prior art technique for obtaining the PSS solution for a circuit involves the use of the shooting-Newton method. This method uses a Newton iterative technique to compute a series of estimates for the PSS. The difference between each estimate in the series and the previous estimate is determined by solving a respective linear system of equations via Gaussian elimination. Obtaining the solution for each linear system by Gaussian elimination involves factorization of a dense N by N matrix, where N represents the number of nodes in the circuit, and thus requires a computation time that increases with the third power of N. For this reason, prior art shooting-Newton techniques for obtaining a PSS solution are often impractical for circuits with a large number of nodes.
Another prior art technique for obtaining a PSS solution involves the use of a finite-difference method. Such a method discretizes the circuit differential equations at a set of timepoints spanning an interval of the input signal to obtain a set of difference equations. Solution of this set of difference equations provides a PSS solution for each of the timepoints simultaneously. Like the previous PSS methods discussed, PSS techniques based on finite-difference methods typically have a computational cost that grows as the cube of the number of circuit nodes, and thus also may be impractical for large circuits.
FIG. 1C
depicts another type of generic circuit
30
that is frequently analyzed in engineering applications. Circuit
30
is driven by two periodic input signals
31
-
32
, u
l
and u
s
. u
s
is a small sinusoidal signal of much smaller amplitude than u
l
. u
l
is periodic with arbitrary shape. One approach to the solution of circuit
30
is transient analysis. Transient analysis may be a very expensive option, particularly where one of the input signals is of much higher frequency than the other. For example,
FIG. 1D
depicts a switched-capacitor filter circuit
40
driven by a periodic signal u
s
to be filtered and a high frequency clock signal u
l
that controls the state of a transistor
41
. To ensure accurate results, transient analysis must cover several timepoints per period of the high frequency clock u
l
, even though the frequency of the filtered signal u
s
is much lower.
Another approach to studying the behavior of circuit
30
, known as Periodic Time-Variant Analysis (PTV), treats the small input signal, u
s
, as a small perturbation to the periodic steady-state (PSS) response of circuit
30
to u
l
when u
s
=0. In particular, typically, a small-signal steady-state response is determined at a set of time points spanning a period of u
l
, by linearizing the differential equations for circuit
30
about the PSS response, and then discretizing the resulting system at the set of time points. Such a technique has a computational cost that grows with the third power of the number of circuit nodes, and for this reason, is impractical for large circuits.
FIG. 1E
depicts another generic circuit
50
of a type frequently analyzed in engineering applications (e.g., mixer circuits). Circuit
50
is driven by d periodic signals
51
of unrelated frequencies, and typically has a steady-state response
52
that is quasiperiodic in d frequencies. One approach to determining a quasiperiodic response involves the use of mixed frequency-time methods as described in detail in chapter 7 of the text “Steady-State Methods for Simulating Analog and Microwave Circuits” by Kundert et al. However, such methods typically have a computational cost that grows with the third power of the number of circuit nodes, and for this reason, are impractical for large circuits.
SUMMARY OF THE INVENTION
As discussed above in the background section, various prior art techniques for steady-state and small signal analysis have a computational cost that grows with the third power of the number of circuit nodes, and are thus impractical for large circuits. By contrast, the present invention provides efficient techniques for steady-state and small signal analysis having a computational cost that is approximately linear in the number of circuit nodes. These techniques are, thus, practical for large circuits that cannot be feasibly handled by prior art techniques.
In particular, the present invention provides an efficient method for determining the periodic steady state response of a circuit driven by a periodic signal, the method including the steps of 1) using a shooting method to form a non-linear system of equations for initial conditions of the circuit that directly result in the periodic steady state response; 2) solving the non-linear system via a Newton iterative method, where each iteration of the Newton method involves solution of a respective linear system of equations; and 3) for each iteration of the Newton method, solving the respective linear system of equations associated with the iteration of the Newton method via a matrix-implicit iterative technique.
In one embodiment, the iterative technique is a matrix-implicit application of a Krylov subspace technique, resulting in a computational cost that grows approximately in a linear fashion with the number of nodes in the circuit, in contrast to typical shooting methods used to obtain a periodic steady state response whose computational cost grows with the cube of the number of circuit nodes. Thus, the disclosed method renders feasible periodic steady state analysis of circuits much larger than those that could be analyzed via previous methods. In addition, the invention provides an efficient finite-difference technique for obtaining a PSS solution, the technique having a computational cost that also grows approximately in a linear fashion with the number of nodes in the circuit. An efficient technique that is almost linear in the number of circuit nodes is also provided for determining a quasiperiodic steady-state response of a circuit driven by two or more signals of unrelated frequencies.
The present invention also provides an efficient method fo
Kundert Kenneth S.
Telichevesky Ricardo
White Jacob K.
Cadence Design Systems Inc.
James Lin Sun
Siek Vuthe
LandOfFree
Method for determining the steady state behavior of a... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Method for determining the steady state behavior of a..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method for determining the steady state behavior of a... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3130356