table_model_examples

table_model_examples — Verilog-A $table_model (Enhancement-16)

Demonstrates the $table_model lookup-table system function, using version11's own openvaf-r and ngspice-46. $table_model interpolates a value from a tabulated grid; the interpolation is differentiable, so it works in the main device equations (its slope becomes the Jacobian conductance/capacitance).

File	What it shows
table_xfer.va	Transfer function V(out) = $table_model(V(in), '{x0,y0, x1,y1, ...}) — inline data array, constant (clamped) extrapolation.
table_res.va	Nonlinear resistor I(p,n) = $table_model(V(p,n), "diode_iv.tbl", "1L") — file-based data, linear extrapolation.
diode_iv.tbl	Two-column V I data file for the nonlinear resistor.
verify_table.py	Compiles both, runs them through ngspice, checks interpolation and its derivative across DC, AC and transient against references.
plot_table.py	Runs the same simulations and writes the PNG plots below.

Plots

python3 plot_table.py

• table_dc.png — the interpolated transfer function V(out)=table(V(in)) and the file-based I-V curve, with the tabulated grid points marked; shaded regions are the extrapolated ranges (clamped for the transfer table, linear for the I-V).
• table_ac.png — the AC small-signal conductance g vs bias lands exactly on the analytic piecewise-constant table slope dI/dV (the Jacobian the interpolation supplies).
• table_tran.png — a large-signal sine through the transfer table; V(out)(t) tracks table(V(in)(t)) instantaneously (the piecewise-linear kinks appear as V(in) crosses the grid points).

Run

python3 verify_table.py

Expected — the interpolation and its derivative are exercised across DC, AC and transient:

transfer function (inline table, clamp)      max err 0.00e+00  PASS
DC: nonlinear op-point via table Jacobian    max err 4.8e-10 V  PASS
AC: small-signal g = table slope             max err 1.7e-18 S  PASS
Transient: V(out) tracks table(V(in))        max err 2.0e-08    PASS
ALL PASS

The derivative checks are the important ones. In DC, the nonlinear resistor is driven through a series resistor, so ngspice solves (vin - V)/Rs = I_table(V) by Newton iteration — converging to the analytic answer only because the table supplies the correct per-segment slope dI/dV to the Jacobian. In AC, that same slope is the small-signal conductance (it matches the analytic table slope exactly at every bias). In transient, the table is re-evaluated each timestep and V(out) tracks table(V(in)) instantaneously. All three work identically because $table_model lowers to plain differentiable MIR arithmetic.

Usage & notes

out = $table_model(x, <data>[, "control"]);

• <data> is either an inline real array of flat {x0,y0, x1,y1, ...} pairs, or a data-file name (two whitespace-separated columns x y; blank lines and #////* comments ignored, resolved relative to the source file). Points are sorted by x; duplicate abscissae are dropped.
• Interpolation is piecewise-linear (degree 1) and fully differentiable — usable directly in V(...) <+ / I(...) <+ contributions.
• Extrapolation outside the grid: constant (clamp to the endpoint value) by default; pass a control string containing L for linear extrapolation (the end segments' slopes continue). Constant clamping produces a zero slope outside the grid, so prefer L when the operating point can land there.
• Scope of this enhancement: 1-D tables, linear interpolation. Multi- dimensional tables and higher-degree (spline) interpolation are natural follow-ups (as 1-D arrays in Enhancement-14 were extended to N-D in Enhancement-15).