Rev A Clean Rebuild — Tank Driver / Recovery Board

Branch: rev-a-clean-rebuild. Board 1 of 6, taken end-to-end first per plan.

What exists now

Artifact	Status
../tank-driver-recovery.kicad_sch	Full electrical capture, 97 components, ERC = 0
../sim/tank-driver-recovery-sim/	Simulation testbench project, ERC = 0, runs in ngspice with no errors
../tank-driver-recovery.kicad_pcb	Netlist-synced board, 90 footprints placed by stage, outline + M3 holes + AGND pours, DRC errors = 0 (13 silk warnings), routing not yet done
gen_tdr.py	Generates both schematics from one circuit definition — edit this, not the .kicad_sch
gen_tdr_pcb.py	Generates the board placement from the same circuit definition
run_sim.py	Runs the testbench through KiCad's bundled ngspice.dll and sanity-checks levels

Verified simulation results (0.5 Vpk 1 kHz drive, ±15 V rails)

• Primary send: unity gain, 0.52 Vpk into the 8R tank model (~65 mA pk), op-amp pre-driver swings ±1.16 V to correct the class-B follower dead zone in-loop
• Primary recovery: 20 mV tank return -> 0.69 Vpk out (gain 34)
• Secondary send: gain 2.2 into the 800R tank model; recovery -> 0.86 Vpk
• Left/right channels identical; no convergence errors over 60 ms

Engineering decisions made during capture (vs. the written spec)

1. Feedback point: R108/R128 feedback is taken from PRI_OUT_L/R (after the BD139/BD140 followers) instead of the op-amp output pin. With the bias diodes DNP per spec, the followers would otherwise run open-loop class-B with gross crossover distortion. In-loop feedback is the standard fix and simulation confirms clean output. D101/D102/D121/D122 stay as DNP bias-spreader reserves exactly as specced.
2. Tank cable landings (J101-J108): spec defers RCA vs shielded-wire. Chosen: JST XH 2-pin board landings (stock footprint, cheap, crimp harness to the tanks' RCA plugs). Swappable later without net changes.
3. Recovery input coupling: direct-coupled by default via 0R jumpers R183-R186, with DNP film caps C175-C178 in parallel positions, so the bench choice in the spec stays a solder-jumper choice.
4. Decoupling: C299 + C300 bulk (one per rail) — spec reserved only C291-C299 but intent lists two bulk parts.

Tank models

sim/tank-driver-recovery-sim/tanks.lib models what the electronics see: input coil L+R load, band-limited coupling (~80 Hz-4 kHz) with realistic insertion loss, output coil source impedance. Spring delay/reverb tail is intentionally NOT modeled — that is what the real tanks are for.

How to run the simulation in KiCad

1. Open hardware/kicad/sim/tank-driver-recovery-sim/tank-driver-recovery-sim.kicad_pro
2. Inspect -> Simulator -> Run (the .tran 20u 60m directive is on the sheet)
3. Probe PRI_OUT_L, PRI_RET_L, SEC_RET_L, etc. For frequency response, comment .tran and uncomment .ac dec 50 10 100k on the sheet text.

Still to do (after review sign-off)

• Route the board (power first, then signals), refill zones, DRC to zero including connectivity, then Gerber/drill/BOM/position exports
• Replicate this workflow for the other five boards
• Order: fab/assembly quote packet per hardware/rev-a-manufacturing.md