AnsweredAssumed Answered

No DC convergence with HB Simulation for SDD

Question asked by LHFT on Jul 8, 2004
Latest reply on Jul 13, 2004 by ddshark

Like • Show 0 Likes0
Comment • 3

Hello,
I have to implement in ADS a component that is defined as voltage across this component depending on charge in this component: V=V(q). In reality V(q) could not be solved for q analytically.

I used a SDD 2-port as follows; in this simple example there is V=q**a:
I[1,1]=q
F[2,0]=-U+q**a (hoping that this equation is numerically solved for q)
with the synonyms U=_v1 and q=_v2; a is constant.
Port 2 is attached with a resistance of 1e+18 Ohm, so it acts as an open.

With a>0 the error message is as follows:
Matrix is singular (probably detected at port 2)

With a<0 the error message is as follows:
Warning detected by HPEESOFSIM in `X1.SDD2P1' during DC analysis `HB1'.
SYM warning: Negative x in x**y.
Complex values not supported.
Error detected by HPEESOFSIM during DC analysis `HB1'.
No DC convergence.
To search for a more robust convergence algorithm,
set ConvMode=1 on the DC analysis and perform the DC
simulation again. This will cause the simulator to vary the convergence parameters and report which parameter setting produces the fastest convergence.

What's wrong here?
Is there another way to implement this component?

Can anybody help me?
Thank you.

Stefan

No one else has this question

Outcomes

wrivas

Jul 8, 2004 1:51 AM

When you specify

F[2,1]=-U+q**a (1)

you are saying that

_v1=_v2^a (2)

also

I[1,1]=q (3)

means that

_i1=d(_v2)/dt (4)

combining 2 and 4, you have that

_i1=d(_v1^-a)/dt (5)

Equation (5) is saying that the current at port 1 is dependent on the derivative of the voltage of port 1 to the minus a power.

Is equation (5) what you intended?

Actions

LHFT Jul 11, 2004 8:16 PM

Hello and thanks for the answer.
The equation I wrote was only a simple example.
The equation I need to solve looks more like the following:
V(q) = a*q + sign(q) * [b*q^2 + c (1- e ^(d*|q|))]

This equation is not easily solvable.
Do you have any idea, how I can implement this component ?

Thanks in advance,
Stefan

Actions

ddshark Jul 13, 2004 8:23 AM

The following approach seems to work for a simple DC simulation. But be very careful. Models must be mathematically (and physically) valid. Models with multiple solutions, abrupt discontinuities, etc. are only asking for convergence problems. Also, numerical scaling should be incorporated so that any additional state variables (ie. charge in this case), have numerical values in a "reasonable" range for a voltage. For example, with a = 2, the equation becomes v1 = v2^2. There are two possible solution for v2. And an initial guess of 0.0 will result in a singular matrix since dF/dv2 = 0.0 (at v2=0).

I[1,1] = _v2
F[2,0] = _v1
F[2,0] = -(mag(_v2) + 1e-20)**a

The 1e-20 is used to make sure the 0^-exp case is never triggered, and hopefully bias the solution away from singular 0.0 value. mag(x) is used instead of abs(x), as it does not return a value of 0 for its derivative at x=0. The F[2,0] equation is broken up into two parts so that a useable AbsRHS is internally computed. The small offset hack is also probably needed with sgn() function. For example, -sgn(_v2)*(mag(_v2))**1 does not work above, even though this should be equivalent to a linear capacitor. This is due to the derivative computations [d_sgn(0) = 0] times an indeterminate derivate (d_abs(0)). But sgn(_v2+1e-20)*mag(_v2) appears to work.

Actions

3 Replies

wrivas Jul 8, 2004 1:51 AM
Mark Correct
Correct Answer
When you specify

F[2,1]=-U+q**a (1)

you are saying that

_v1=_v2^a (2)

also

I[1,1]=q (3)

means that

_i1=d(_v2)/dt (4)

combining 2 and 4, you have that

_i1=d(_v1^-a)/dt (5)

Equation (5) is saying that the current at port 1 is dependent on the derivative of the voltage of port 1 to the minus a power.

Is equation (5) what you intended?
Like • Show 0 Likes0
Actions
LHFT Jul 11, 2004 8:16 PM
Mark Correct
Correct Answer
Hello and thanks for the answer.
The equation I wrote was only a simple example.
The equation I need to solve looks more like the following:
V(q) = a*q + sign(q) * [b*q^2 + c (1- e ^(d*|q|))]

This equation is not easily solvable.
Do you have any idea, how I can implement this component ?

Thanks in advance,
Stefan
Like • Show 0 Likes0
Actions
ddshark Jul 13, 2004 8:23 AM
Mark Correct
Correct Answer
The following approach seems to work for a simple DC simulation. But be very careful. Models must be mathematically (and physically) valid. Models with multiple solutions, abrupt discontinuities, etc. are only asking for convergence problems. Also, numerical scaling should be incorporated so that any additional state variables (ie. charge in this case), have numerical values in a "reasonable" range for a voltage. For example, with a = 2, the equation becomes v1 = v2^2. There are two possible solution for v2. And an initial guess of 0.0 will result in a singular matrix since dF/dv2 = 0.0 (at v2=0).

I[1,1] = _v2
F[2,0] = _v1
F[2,0] = -(mag(_v2) + 1e-20)**a

The 1e-20 is used to make sure the 0^-exp case is never triggered, and hopefully bias the solution away from singular 0.0 value. mag(x) is used instead of abs(x), as it does not return a value of 0 for its derivative at x=0. The F[2,0] equation is broken up into two parts so that a useable AbsRHS is internally computed. The small offset hack is also probably needed with sgn() function. For example, -sgn(_v2)*(mag(_v2))**1 does not work above, even though this should be equivalent to a linear capacitor. This is due to the derivative computations [d_sgn(0) = 0] times an indeterminate derivate (d_abs(0)). But sgn(_v2+1e-20)*mag(_v2) appears to work.
Like • Show 0 Likes0
Actions

No DC convergence with HB Simulation for SDD

Outcomes

Related Content

Recommended Content