RF and high-speed calculator review

Reviewed against the actual HTML/script implementations on 2026-11-09. Equation checks establish a bounded analytical calculation, not a measured board model. All restored fragments are self-contained; reopening their scripts is exercised.

Tool	Decision and exact basis
RF unit translator	Restore. dBm/dBW power references, dBV/dBµV/dBu voltage references and P = Vrms²/R. Removed field-strength wording; sine-wave peak conversion only. Rejects empty, nonfinite, negative linear quantities and nonpositive resistance; zero power gives −infinity dB.
Friis transmission	Restore. FSPL = 20 log10(4πdf/c), c = 299792458 m/s; received dBm = transmitted dBm + antenna gains − FSPL − explicit system loss. Far-field free space and matched polarization only. Does not predict real coverage.
Cascade noise figure	Restore after fixing missing stage-1 DOM wrapper and the skipped-middle-stage bug. Friis cascade in linear factors/gains for matched same-frequency linear stages; passive loss assumes physical temperature 290 K. Does not model mixers or compression.
PDN target impedance	Restore as a voltage-budget ratio, not a decoupling design. Ztarget = allocated dynamic droop / actual entered current step. Removed implicit 50% current assumption and arbitrary 100 MHz capacitor estimate.
Attenuator	Restore only equal-real-impedance Pi and Tee networks. Standard matched resistive equations; resistor dissipation calculated from loaded node voltages/currents. Removed unequal-impedance and bridged-T/L-pad modes with inconsistent power calculations. No RF bandwidth, tolerance or component rating claim.
Skin effect	Restore only after narrowing to skin depth and DC resistance (handled in the catalog integration). Removed ad hoc AC-resistance interpolation and radius-as-skin-depth at DC. No AC current rating or conductor temperature prediction.
Insertion loss	Defer. Conductor loss uses an unqualified 0.1 sqrt(f)/(W/5) rule; neither roughness nor actual return geometry is represented. Not an S-parameter calculation despite its title.
Basic impedance calculation	Defer. Mixes IPC-style log fits with an explicitly demonstration-only CPW expression; lacks valid geometry/range checks and masks invalid inputs with defaults or zero.
Differential impedance	Defer. Legacy microstrip/stripline fits lack documented applicability bounds; effective permittivity and delay are inconsistent approximations. Invalid geometry can be clipped to zero impedance. Needs a separately bounded implementation.
Stackup solver	Defer. Inverts the same uncontrolled log/exponential fits, offers protocol presets as general requirements, and binary search can return a boundary even when a requested solution is infeasible.
Crosstalk estimator	Defer. NEXT = 1/(1+(D/H)²); FEXT uses an explicitly heuristic 0.1 multiplier and dimensional L/Tr ratio. Not defensible as a percentage prediction.
Comprehensive via	Defer. Lumped impedance misses the nH/pF conversion factor; current uses an unqualified ×1.5 multiplier and trace-fit thermal assumptions. Thermal output remains excluded.
Via impedance	Defer. Unit conversion is consistent, but a lumped LC ratio is presented as via impedance/delay without geometry applicability or extracted-field comparison.
High-speed design	Defer. Useful delay/quarter-wave equations coexist with unqualified 3W/5W spacing guidance; global event dependency and fallback input masking need cleanup. Source-specific DDR/Ethernet/SerDes/LVDS sheets are a better replacement.
Smith chart	Rebuilt and restored. Self-contained bilinear Z↔Γ mapping, passive reflection circle, mirrored normalized-admittance grid, VSWR/return loss/reflected-power readouts, click-to-select load, and editable series/shunt R/L/C matching. Frequency recalculates ideal component reactances; load remains an explicitly entered fixed complex impedance. No real transmission-line, parasitic or measured-frequency behavior claim.
RF signal suite	Defer. Large shared visualization suite includes unexplained Eb/N0 lookup values, arbitrary EVM noise scaling, malformed 32-QAM geometry and mismatched decision-grid keys. Sound pieces need separation from unsupported estimates.
Balun calculator	Defer. Topology diagrams and current-balun/common-mode claims need circuit validation; 1:9 uses a simplified 1:1 diagram. Ideal turns/impedance ratios alone do not defend bandwidth or power capability.

References and verification

ITU-R P.525, free-space attenuation.
Analog Devices, system noise-figure analysis, including the distinction between linear cascades and mixers.
TI SWPA222A, PDN analysis, section 4. This tool uses an explicit current step instead of treating that document’s particular 0.5 × maximum-current assumption as universal.

node --test tests/rf-tools.test.cjs executes the actual fragment scripts. Coverage includes reference values, unit conversions, zero where meaningful, empty/nonfinite/out-of-domain inputs, stage skipping, reopened fragments and independent loaded-network calculations for Pi/Tee input match, attenuation, per-resistor power and conservation of power. Browser layout and measured RF performance are separate checks.

node --test tests/smith-tools.test.cjs additionally checks match/short/reactive limits, the Analog Devices 100+j50 Ω → Γ = 0.4+j0.2 example, round trips across 100 passive impedance/reference combinations, known SWR/return-loss ratios, series reactance cancellation, an independently derived shunt-C match, changed frequency, resistor networks and invalid input rejection. The drawing samples the same complex transformations; the trajectory is the actual intermediate series impedance or shunt admittance path rather than a straight line.

Smith chart reference: Analog Devices, impedance matching and Smith chart impedance.