Full engineering specification for the Ghost Spring Tank — transformer coupling rationale, complete signal flow, component selection, output buffer, enclosure, and build rules.
One spring. One voice. Transformer-coupled for the organic bloom that no direct-drive circuit can replicate — the same mechanism that made the Fender 6G15 Reverb Unit sound like nothing else.
Every stage in the signal chain with component values inline. Left to right: input from line source, through the spring and back out to the power amp.
Line In (from Lehle Parallel M) → C_in 1µF film · DC-blocking coupling cap at input jack → Input Buffer U1 OPA2134 · R1 1MΩ from U1(+) to GND (bias return + input Z) · unity gain · R2 100Ω series output → Dwell Pot RV1 10kΩ linear · C1 1µF film coupling cap · controls drive level → BD139 Driver Q1 Class A · R3b 6.8k + R4 1k bias divider · R5 68Ω emitter degen C2 100µF bypass · ~18mA quiescent · D3 1N4148 flyback clamp → REB3S Xfmr T2 primary (from Q1 collector) → 8Ω secondary → 9AB3C1B Tank RT1 · 3 springs · long decay · 8Ω input / 2550Ω output → C3 Coupling 470nF film · blocks DC from tank output terminals → Recovery Preamp U2 OPA2134 · non-inverting · tank → (+) input · Ri=470Ω / Rf=100kΩ → gain=214× (~46dB) Rbias=100kΩ at (+) input to GND · 1–5mV → ~1Vrms → 300Hz Wet HPF C4 100nF film + R6 5.6kΩ · design f = 1/(2π×5600×100n) ≈ 284Hz; measured ≈312Hz in full sim wet signal only — dry path bypasses this stage → Tone Pot RV3 100kΩ · high-shelf on wet signal only → Mix Pot RV2 100kΩ audio taper · 3-terminal passive blend (NOT a volume knob) dry (U1 via Rdry 10kΩ) → CCW lug 1 · wet (Tone wiper) → CW lug 3 · wiper lug 2 → U3 C_bright=47pF silver mica across the FULL pot (lug 1 ↔ lug 3) → Output Buffer U3 OPA2134 · voltage follower · R7 100Ω series · <100Ω output Z → Line Out (to McIntosh MC100)
This is the part that makes the Ghost Spring sound like a spring reverb and not a plate sim. Most DIY spring reverb designs drive the tank directly from a transistor or op-amp output. That works — but it doesn't sound like a Fender 6G15.
The REB3S transformer's inductance forms a resonant circuit with the tank's input impedance (8Ω). This resonance creates:
The Accutronics REB3S is designed specifically for this application. It's used in boutique spring reverb units and is correctly matched to 8Ω tanks. Using a generic transformer (or no transformer) changes the resonant character fundamentally.
Three springs give a denser, more uniform reverb tail than two-spring tanks. Two-spring designs can produce a metallic "ping" on hard attacks because the two resonant modes are more distinct. Three springs blend into a closer-to-continuous reverb tail — the same reason the original Fender 6G15 and most boutique spring reverbs use 3-spring tanks.
Most DIY reverb designs filter before the driver — a HPF or shelving filter before the spring to prevent low-end boom. This unit takes the opposite approach: the full-bandwidth signal enters the spring, and the HPF is applied to the wet output after the recovery preamp.
The result: the physical attack transient (the full-frequency "thump") enters the spring intact, creating a realistic reverb bloom. The reverb tail is then filtered. Low-end boom clears up as the note decays, which is more natural than a pre-filtered attack.
HPF values: C4 100nF film + R6 5.6kΩ → first-order corner 284Hz (~300Hz); the corner measured in the full simulation lands at ≈312Hz (the small shift is loading on the R6/C4 node). The bench pass window is 250–320Hz — expect ~312Hz on the meter. Use 5.6kΩ exactly — 4.7kΩ pushes the corner to 338Hz (too aggressive), 6.8kΩ drops it to 234Hz (insufficient mud rejection).
The Mix pot acts as a voltage divider at audio frequencies. At low mix settings, it attenuates high frequencies more than low frequencies — the reverb tails lose their shimmer before they lose their body. A 47pF silver mica capacitor bridges the full pot, end to end (CCW lug 1 to CW lug 3), bypassing it for high frequencies and keeping the reverb "glassy" as the wiper approaches full-wet. Note: the bright cap goes across the two end lugs, not from a lug to the wiper.
Mix pot is a 3-terminal passive blend, not a volume knob. Dry signal (from U1 via Rdry 10kΩ) enters the CCW end (lug 1); wet signal (from the Tone wiper) enters the CW end (lug 3); the wiper (lug 2) taps the blend and feeds U3. Wiring both signals to one lug — or the wet source onto the wiper — collapses the blend (this was a real netlist bug). See the build sequence and the builder guide for per-lug wiring.
Why silver mica: Silver mica is stable to ±1% over temperature. A ceramic disc capacitor will add high-frequency distortion that's audible in reverb tails. Do not substitute.
R5 (68Ω emitter resistor) provides thermal stability and sets quiescent current. Without C2, R5 degenerates the BD139's gain at audio frequencies — the spring is underdriven and loses high-frequency drive. C2 (100µF low-ESR Nichicon) bypasses R5 at audio frequencies, restoring full gain while keeping the DC stability of R5 intact.
The spring tank's output is 1–5mV at low-to-moderate playing levels. The recovery preamp (U2 OPA2134 in non-inverting configuration) brings the signal up to line level: tank signal enters the (+) input via C3 (series coupling cap); Rbias (100kΩ) shunts the (+) input to GND for DC stability; Ri (470Ω) and Rf (100kΩ) form the feedback divider at the (−) pin → gain = 1 + (100k/470) = 214×. This is not unusual — spring tanks are low-output transducers, and this gain is normal for the stage.
Headroom note: 214× gain is sized for the low end of tank output. Hard pick transients can produce 10–20mV at the tank output, giving 2–4V at U2's output — well within the OPA2134's ±13V swing. Set Dwell conservatively on first use and verify U2 output stays clean on peaks. If clipping is heard at moderate playing levels, Ri can be raised slightly (e.g. 680Ω gives gain 148×) to add headroom.
DC offset at U2 output (bench reality): a real OPA2134's input offset (Vos up to ~500µV), multiplied by 214×, leaves a steady 20–150mV DC at U2's output. This is normal and is blocked by C4 before the output — do not chase it as a fault. The final output (after C4, at U3/J2) settles to <5mV.
Cf — strongly recommended on perfboard: fit a 10–22pF capacitor across Rf. On point-to-point construction, stray capacitance at U2's inverting input forms a pole with the 100kΩ Rf that can cause HF gain peaking or ringing in this 214× stage; Cf (22pF → ~72kHz corner, above the audio band) rolls the loop off cleanly. It is a cheap optional build-time part, not a BOM line — but recommended insurance against layout-dependent instability.
Every component was chosen for a specific engineering reason. These notes are the difference between a build that sounds right and one that doesn't.
| Stage | Ref | Why OPA2134 |
|---|---|---|
| Input Buffer | U1 | FET input (10¹³Ω) places virtually no load on the upstream device. Unity-gain stable. |
| Recovery Preamp | U2 | FET input is critical here — U2 must load the tank's 2550Ω output without signal attenuation. THD+N=0.00008%. Slew rate 20V/µs eliminates slew-limiting distortion on pick transients. |
| Output Buffer | U3 | Voltage follower — drives the MC100's RCA input and any cable capacitance without treble loss. <100Ω closed-loop output impedance. |
Do not replace OPA2134 with NE5532 or TL072. Both have significantly higher noise floors and lower input impedance — the NE5532 is bipolar-input (low impedance) and will attenuate the tank's output signal at the recovery stage. The OPA2134 is specified throughout for consistency; you need only one part number and the builder can swap a section if one fails.
High-current NPN bipolar rated 1.5A / 80V. At 18mA quiescent current it runs well within ratings with negligible heat. The TO-126 package mounts flat against the chassis for passive heatsinking if needed. BD139 has excellent hFE linearity at low currents — critical for low distortion in the driver stage.
Do not use a TO-92 small-signal transistor (2N3904, BC547, etc.). Insufficient current capacity will clip drive transients and harden the reverb attack character. The BD139 TO-126 is specifically sized for this application.
All resistors are metal film, 1% tolerance. Carbon film resistors have higher noise (current noise 10–100× higher) and temperature drift that degrades the signal floor, especially in the 214× gain recovery stage. Order 5 of each value — metal film resistors are cheap and you want spares for the build.
| Ref | Value | Function | Why This Value |
|---|---|---|---|
| R1 | 1MΩ | U1(+) → GND — bias return and input impedance | Shunt to GND, placed after C_in. Provides the DC bias return path for U1's FET input — without it, U1(+) has no reference and drifts to a rail. Also sets 1MΩ input impedance (negligible load on upstream device). |
| R3b | 6.8kΩ | Upper bias divider | Sets base voltage to ~1.92V → Ve~1.22V → Ic~18mA |
| R4 | 1kΩ | Lower bias divider | Divider current ~1.92mA. R5 emitter degeneration is the actual bias-stabilizing element — verify by measuring Ve ≈ 1.22V, not the divider math. |
| R5 | 68Ω | Emitter degeneration | Ic = Ve/R5 = 1.22V/68Ω ≈ 18mA. Primary bias stability and thermal runaway protection. |
| Ri | 470Ω | Gain set — (−) pin to GND | Gain = 1+(Rf/Ri) = 1+(100k/470) = 214× ≈ 46dB. Tank signal enters (+) pin; Ri forms the lower leg of the feedback divider at the (−) pin. |
| Rf | 100kΩ | Recovery gain (feedback) | 100kΩ keeps thermal noise below OPA2134's own input noise |
| R6 | 5.6kΩ | HPF corner | First-order f = 1/(2π×5600×100n) = 284Hz; measured ≈312Hz in full sim (bench window 250–320Hz). Use exactly 5.6kΩ. |
| Rbias | 100kΩ | U2 non-inv (+) input → GND | Sets recovery stage input impedance. 100kΩ ensures 97.5% signal transfer from the 2550Ω tank (470Ω would lose 84%). Sources from same 100kΩ stock as Rf — no extra BOM line. |
Ceramic capacitors have piezoelectric microphonics and voltage-dependent capacitance (distortion) at audio frequencies. All signal-path capacitors must be WIMA MKS2 film. This is non-negotiable in a hi-fi circuit with 214× gain in the recovery stage.
| Ref | Value | Type | Why |
|---|---|---|---|
| C1 | 1µF/63V | WIMA MKS2 film | Coupling before driver. HPF corner = 1/(2π×1µ×10k) = 16Hz — well below guitar fundamentals. |
| C3 | 470nF/63V | WIMA MKS2 film | Tank output coupling. Blocks DC offset from tank terminals. Corner at ~3Hz with Rbias = 100kΩ — purely DC-blocking, no audio attenuation. |
| C4 | 100nF/63V | WIMA MKS2 film | 300Hz HPF with R6. Must be film — ceramic drifts with temperature, shifting the cutoff. |
| C_bright | 47pF | Silver mica | Bright cap across Mix pot. Silver mica ±1% stability, lowest HF distortion. Do not use ceramic disc. |
| C2 | 100µF/25V | Nichicon UKW audio grade | Emitter bypass. Low-ESR is critical — high-ESR cap won't bypass R5 at high audio frequencies. |
| C5–C8 | 100nF/63V | WIMA MKS2 film | Op-amp supply decoupling — one per supply pin, placed as close as physically possible to IC. 4 total: 2 supply pins per DIP-8 package × 2 packages. |
Military-grade cermet element, gold-plated wiper, stainless shaft, rated 10,000+ cycles. These are overkill for a guitar rig but they will outlast everything else in the build and will never develop contact noise.
Note on taper: Mix (RV2) and Tone (RV3) are specified as audio taper — Bourns PDB181-GTR01-104A0. Audio taper matches how ears perceive level changes and gives the blend and tone controls an even feel across the full sweep. A 100kΩ linear pot in a passive blend against a 10kΩ dry resistor compresses most of the audible range into a narrow arc near minimum. Dwell (RV1) is correctly linear.
Builder's note — taper is a preference call: If MIL-PRF-39023 certification is required throughout, substitute Vishay/Spectrol 296UAL104B2 (100kΩ linear) for RV2/RV3. The circuit is electrically identical; only the feel of the controls changes. Some builders prefer linear for more predictable feel in studio use. Order what suits the intended application.
The output buffer (U3) drives the McIntosh MC100's RCA input and any cable between the unit and the power amp. Without a buffer, the Mix pot's output impedance varies with rotation — at center position it's at maximum (half of 100kΩ = 50kΩ), which would roll off high frequencies through cable capacitance and interact unpredictably with the MC100's input.
The power supply is internal to the unit. No wall wart, no switching regulator — a toroidal transformer, bridge rectifier, and pair of linear regulators. This is the correct approach for a hi-fi reverb unit and the reason the circuit doesn't need external power management.
| Stage | Part | Why |
|---|---|---|
| Transformer | Triad F-219X, 30VA dual-primary toroidal (2×115VAC in / 2×15VAC out) | Toroidal has ~10× lower magnetic field leakage than E-I laminate. Critical — the spring tank is a sensitive magnetic transducer and will pick up 60Hz hum from a nearby transformer. Keep transformer as far from the tank as the chassis allows. Primary wiring: Wire the two 115VAC primary windings in parallel for 120VAC mains (see build sequence). Secondary wiring: Wire the two 15VAC secondary windings in series — the junction of the two windings is the 0V/GND reference (forming 15-0-15), and the outer ends connect to BR1's AC inputs. Wiring the secondaries in parallel gives a single rail with no negative supply — the op-amps and LM7915 will not receive −15V. |
| Rectifier | W04G bridge, 2A/400V | 400V is derated 10× from the 42V peak. 2A rating gives 4× headroom over typical load — reliable even with inrush. |
| Filter | 2× 1000µF/50V Nichicon UKW low-ESR | ~350mVpp bus ripple at the ~40mA worst-case load — the regulators' ~68dB 120Hz rejection turns that into <1mV on the rails. Downsized from 2200µF (#95): low-mains trough still clears the regulator dropout floor by ~0.75V. 50V rated for headroom; low-ESR minimizes heat. |
| +15V Reg | LM7815CT (TO-220) | Linear regulation = zero switching noise. Switching regulators inject kHz noise that passes through op-amp power supply rejection. LM7815 is proven and inexpensive. |
| −15V Reg | LM7915CT (TO-220) | Same rationale. Provides negative rail for op-amp negative supply pins. |
| Reg output caps | 3× 25µF per rail (parallel, 75µF) + 2× 100nF film | Electrolytic for stability per LM78xx/79xx datasheet (LM7915 floor ≥25µF aluminum); parallel wiring, no balancing resistors (#94). Film in parallel for HF suppression above ~100kHz. |
The TO-220 tab is electrically connected to the output pin. Without mica insulating pads between each regulator and the chassis, mounting both regulators to the same chassis creates a short between +15V and −15V through chassis ground. Use TO-220 mica pads + M3 nylon screws or an insulated shoulder washer.
The 2U form factor is required — 1U is too tight to mount the 9AB3C1B tank horizontally with spring clearance. 2U gives comfortable room for the tank, PCB, power supply, and ventilation.
Hammond 1455T2201 — 2U aluminum rackmount. Aluminum is mandatory: steel would interact magnetically with the toroidal transformer. Rack ears are included with the Hammond chassis.
Rear Panel ─────────────────────────────────── Front Panel
│ │
│ ┌──────────────────────────┐ │
│ │ │ │
│ │ Spring Tank (9AB3C1B) │ ┌──────────┐ │
│ │ horizontal, open down │ │ Audio PCB │ │
│ │ grommet-isolated │ │ (Vector │ │
│ │ │ │ T44) │ │
│ └──────────────────────────┘ └──────────┘ │
│ │
│ ┌──────────┐ │
│ │ PSU PCB │ ← toroidal transformer here │
│ │ (Triad │ as far from tank as possible │
│ │ F-219X) │ │
│ └──────────┘ │
│ │
────────────────────────────────────────────────────
↑ IEC C14 inlet · Fuse · Rocker switch ↑
| Control | Type | Function |
|---|---|---|
| DWELL | 10kΩ linear pot, ¼" D-shaft | Drive level into transformer — spring saturation and reverb density |
| MIX | 100kΩ audio taper pot + 47pF bright cap (Bourns PDB181-GTR01-104A0) | Dry/wet blend |
| TONE | 100kΩ audio taper pot (Bourns PDB181-GTR01-104A0) | High-shelf EQ on wet signal only |
| Connector | Type | Notes |
|---|---|---|
| Input | Switchcraft 112A ¼" TS | From Lehle Parallel M output (future state) or directly from Alembic FX-1 |
| Output | Switchcraft 112A ¼" TS | To McIntosh MC100 RCA input (with 1/4" TS → RCA adapter cable) |
| IEC Power | Schurter 5110.1052 (EMI-filtered IEC C14 + fuse holder) | 500mA slow-blow fuse. Integrated EMI filter suppresses mains-borne noise before it reaches the transformer. Slow-blow essential — inrush current at power-on blows fast-blow fuses. |
| Power Switch | TE 1825232-1 SPST rocker, 6A/250V | Interrupts mains before transformer primary |
Custom 2U aluminum panel with laser-drilled holes and engraved labels. Full hole list:
Use Front Panel Express designer software to spec all hole positions to fit the Hammond 1455T2201 chassis dimensions.
These five rules are the difference between a working build and one that hums, oscillates, or drifts. Non-negotiable.
FR4 fibreglass is mandatory. The cheaper brown phenolic perfboard absorbs moisture and increases leakage current between pads — at the recovery stage gain of 214× this manifests as audible noise. Use 4× M3 nylon hex standoffs (not metal) to mount the board — metal standoffs can accidentally create a second chassis ground connection.
Build in this order to make each stage testable before adding the next:
| Test Point | Expected | Fault if wrong |
|---|---|---|
| U4 output pin | +15.0V ±0.5V DC | Regulator or rectifier issue |
| U5 output pin | −15.0V ±0.5V DC | Regulator or rectifier issue |
| PSU rails ripple | <50mV p-p at idle | Filter cap ESR or wiring issue |
| Q1 emitter (no signal) | 1.0–1.4V DC (sim 1.09V) | Bias resistors (R3b/R4/R5) or BD139 fault — this is the correct node to measure for bias verification |
| Q1 collector (no signal) | ≈ +15V DC | T2 primary is near-DC short to rail — a reading well below +15V means T2 is open or Q1 is saturated |
| U2 (+) input (signal) | 1–5mV at normal playing level | Tank not driven, C3 open, or Rbias fault |
| U2 output (signal) | ~200–1000mV at normal playing level | Op-amp fault, Ri/Rf values wrong |
| U3 output (idle) | <5mV DC offset | DC offset from upstream; check Rbias and Ri |
| Chassis ground to star point | 0Ω | Ground lift switch in wrong position, or open connection |
Spring tank polarity varies between Accutronics batches. If the reverb sounds hollow, thin, or "phasey" when mixed in at 50/50, the tank output is out of phase with the dry signal. Fix: on the output Molex KK connector at the tank's 2550Ω side (the RCA that goes to U2's input), swap the two wires. The input connector (8Ω driver side) is not involved. No design change needed — this is a 10-second wire swap on first commissioning.
After all testing is complete and the build is verified, apply MG Chemicals 422B acrylic conformal coating spray to the PCB to protect against humidity and oxidation over time. Mask or skip the following before spraying — coating these will cause problems:
Apply in two thin coats, allowing full dry time between coats. Hold the can 8–12" away and use short strokes — thick single-pass coats pool and crack.
Kester 44, 63/37 tin/lead, 0.031". Kester 44 is the standard for hand-soldered audio. 63/37 eutectic alloy — snaps solid with no mushy semi-solid phase, reducing cold joints. 0.031" diameter is correct for through-hole perfboard work. Do not use lead-free — higher melting point increases heat stress on sensitive components and produces higher-resistance joints.