loadpull_examples

Load-pull / source-pull analysis (Enhancement-234)

Load-pull is a staple of power-amplifier design: sweep the load impedance the device output sees over the Smith chart and contour output power, gain, PAE, and drain efficiency to find the optimum load. ngspice had no such command; open-source load-pull is genuinely scarce. This adds one, riding the existing .tran engine and alter mechanism (as optimize/sweep do), with contours rendered by pyplot -contour (E-218).

Usage

loadpull -load <R> <L> <C> -out <node> -drive <Vsrc> -f <freq>
         [-supply <Vsrc>] [-z0 50] [-n 15] [-gmax 0.85] [-nper 20] [-npts 50]
loadpull -source <Rs> <Ls> <Cs> -out <node> -drive <Vsrc> -f <freq> ...   (source-pull)

• -load R L C — three series elements (out → ground) that present the swept load. For each Γ the target Z = z0·(1+Γ)/(1−Γ) = R + jX is synthesized: the resistor is set to R, and the inductor/capacitor give +X/−X (the unused reactor parked near a short, so the branch is an AC-coupled R+jX at f0).
• -out — the output node; the fundamental power delivered to the load is measured here.
• -drive — the input drive source (for Pin → gain).
• -supply — the DC supply source (for PAE and drain efficiency). Optional; omit for a passive/linear load-pull (just Pout / gain).
• -source Rs Ls Cs — sweep the source impedance instead (source-pull).
• -n grid points per axis, -gmax max |Γ|, -nper periods integrated, -npts timepoints/period, -z0 reference impedance.

Each Γ point runs a large-signal .tran, extracts the fundamental by a direct DFT over the last nper periods, and computes Pout, gain, PAE = (Pout−Pin)/Pdc, drain efficiency = Pout/Pdc. Results are stored as gamma_re, gamma_im, pout_dbm, gain_db[, pae, eff] in a loadpull plot; the optimum (max Pout) load and its impedance are reported.

loadpull -load RL LL CL -out out -drive Vdr -supply Vdd -f 1e9 -n 21
pyplot -contour gamma_re gamma_im pout_dbm      ; power contours on the Smith grid
pyplot -contour gamma_re gamma_im pae

Verify

python3 verify_loadpull.py

Four checks: (1) a linear Thévenin source (Vs, Zs = 50+j30) peaks Pout at Γ_L = conj(Γ_s) with the analytic Pmax = |Vs|²/(8·Rs) = 3.979 dBm; (2) a behavioral PA yields physical gain/PAE/efficiency (0 < PAE < eff < 100%); (3) source-pull runs. Check (2) also guards the E-234 fix — a use-after-free in the source-name lookup was silently corrupting the branch-current metrics.