Circuit Atlas · HUSH M4 · Rev 2027.4 draft
5 Circuit description
One short table per stage. Full drawings live on the schematics pages.
IN DEVELOPMENT · REV 2027.4 (DRAFT) · DESIGN TARGETS, NOT MEASUREMENTS
HUSH M4 is in development. This draft is published on zakkhutoretsky.com; it has no price and no ordering. Figures marked design target are goals for the finished product. Figures marked calculated come from the published formulas. No value on these pages was measured on a built unit. Interviews are cited as the design brief, not as an endorsement.
See this control Signal path at a glance Play the panel
5 Circuit description
One short table per stage. The full schematics are in Appendix B and on /mixer/schematics.
S1 RIAA phono preamp (moving magnet)
| Item | Value / formula |
|---|---|
| Standard | Playback curve with time constants 3180, 318 and 75 µsR1, R2 |
| U1 | Non-inverting, G = 1 + Rf1/Rg1 = 1 + 2.21k/220 = 11 (+20.9 dB); Cg1 = 470 µF puts the stage high-pass corner at 1/(2π·220·470 µF) = 1.5 Hz |
| U2 | Shelf: pole at Rf2·Cf2 = 20k × 160 nF = 3.2 ms (3180 µs), zero at Rf2·Cf2/(1 + Rf2/Rg2) = 318 µs. Rf2/Rg2 = 20k/2.21k ≈ 9 |
| R3·C3 | Passive low-pass, 5.76k × 13 nF = 74.9 µs (75 µs) |
| U3 | Flat gain 1 + 6.19k/1k = 7.19 (+17.1 dB): total at 1 kHz = 38.1 dB, a design target of 38 dB for a 5 mV moving-magnet cartridge |
| Noise | U1 should be the lowest-noise op-amp in the design (NE5534-class, 3.5 nV/√Hz typ)R7, R6; the 220 Ω gain resistor contributes about 1.9 nV/√Hz (√(4kTR)) |
| Result | Calculated RIAA error with these E96/E24 values and ideal op-amps: ±0.017 dB, 20 Hz to 20 kHz. A real build adds op-amp bandwidth error, part tolerance and cartridge loading; the 20 Hz IEC rumble corner is not included |
S2 Line input and trim
| Item | Value / formula |
|---|---|
| Trim | Inverting stage, G = −Rf/Rin, Rin = 10k, Rf = 1.8k (fixed) + P1 50k: -14.9 dB to +14.3 dB |
| Polarity | U5 is a unity inverter so the line path has the same absolute polarity as the phono path |
| Filters | RF low-pass 1/(2π·1k·220 pF) = 723 kHz; input coupling 2.2 µF into 47k = 1.5 Hz high-pass |
S3a Isolator building block
| Item | Value / formula |
|---|---|
| Section | Second-order Butterworth Sallen-Key, equal resistors and equal capacitors, f0 = 1/(2πRC), K = 1 + Rf/Rg = 3 − 1/QR3 |
| LR4 | Two cascaded sections give a Linkwitz-Riley 24 dB/oct filter, each stage Q = 0.71R4 |
| Values | C = 10 nF. 300 Hz: R = 53.6k (f0 = 296.9 Hz). 3 kHz: R = 5.36k (f0 = 2969 Hz). Rg = 10k, Rf = 5.9k gives K = 1.59, Q = 0.709 |
S3b Three-band isolator (master section)
| Item | Value / formula |
|---|---|
| Split | LP4 at 300 Hz, HP4 at 300 Hz then LP4 at 3 kHz, HP4 at 3 kHzR4 |
| Band gain | ×2 band amp and a 10k linear pot: minimum = full kill, centre = unity (the pot divides by 2), maximum = +6 dB |
| Sum | Inverting sum, 10k per band. This is the master isolator. The channel EQ (S13) reuses the same split with the low crossover at 180 Hz and a fixed middle leg |
| Open task | Allpass delay trim between the three paths so the bands recombine flat in phase; to be done on a prototype |
S4 Channel filter
| Item | Value / formula |
|---|---|
| Sections | Butterworth Sallen-Key (12 dB/oct), one low-pass and one high-pass, Q = 0.71R3 |
| Low-pass | C = 4.7 nF, R = 1.87k + P5 500k dual-gang log: 18,108 Hz down to 67 Hz |
| High-pass | C = 10 nF, R = 2.67k + P6 250k dual-gang log: 63 Hz up to 5,961 Hz |
| Control | One knob: low-pass counter-clockwise, high-pass clockwise, with a centre dead zone |
| True bypass | A window comparator on a third wiper track drives relay K1 across the dead zone: IN goes straight to OUT and neither section is in the path. At the dead-zone edge both sections sit at their open ends (LP 18,108 Hz, HP 63 Hz) |
| Resonance | Butterworth, Q = 0.71: low resonance, no peakR3 |
| Click | The relay switching click and the step at the dead-zone edge on the real circuit are prototype measurement items; nothing is claimed here |
S5 Channel fader (control-voltage VCA)
| Item | Value / formula |
|---|---|
| Law | VCA control constant 6.1 mV/dBR8: 100 dB range needs 0.61 V |
| Control path | P7 10k linear fader on a 5 V reference → buffer → 47k / 6.49k divider → 0.607 V maximum |
| Audio path | Rin = Rf = 10k sets 0 dB at full travel; the audio never passes through the fader |
| Crossfader | Not fitted, as stated on the /mixer page (left out on purpose); a second control bus is a footprint option only |
S6 FX send / return
| Item | Value / formula |
|---|---|
| Send | 10k pot, follower, 10 µF and 100 Ω to the jack; nominal 0 dBV, unbalanced |
| Return | 2.2 µF coupling, 47k bias, follower, 10k return pot, SW2 into the master mix |
| Bypass | SW2 removes the loop from the circuit entirely: no effect sits in the signal path when it is off |
S7 Master, booth and cue
| Item | Value / formula |
|---|---|
| Sum | Inverting summer: gain per input = −Rf/Rin = −10k/10k = −1 |
| Output | Impedance-balanced driver: follower plus unity inverter, 47.5 Ω in each leg, Rout = 2 × 47.5 = 95 Ω |
| Level | +28 dBu = 19.46 Vrms balanced, 13.76 V peak per leg. On ±18 V rails with 16.5 V usable swing per leg the balanced limit is about +29.6 dBu (1.6 dB above the +28 dBu design target). On ±15 V rails (13.5 V swing) it would be only +27.8 dBu: this is why the design uses ±18 V |
| Cue | Channel pre-fader taps into an inverting sum, then a cue / master blend pot into the headphone stage |
S8 Headphone amplifier
| Item | Value / formula |
|---|---|
| Drive | P = V²/R. 250 mW into 32 Ω needs 2.83 Vrms (4.00 V pk, 88 mA rms); 250 mW into 64 Ω: 4.00 Vrms; 100 mW into 600 Ω: 7.75 Vrms (10.95 V pk) |
| Topology | Op-amp with a complementary emitter-follower stage inside the loop, 0.47 Ω emitter resistors, 10 Ω series output resistorR12 |
| Damping | 32 Ω / 10 Ω = 3.2: a design choice for short-circuit safety, not a high-damping design |
S9 Power supply
| Item | Value / formula |
|---|---|
| Rails | ±18 V, LM317 / LM337: Vout = 1.25 × (1 + R2/R1) = 1.25 × (1 + 3240/240) = 18.29 VR5 |
| Rectifier | 2 × 22 Vac, 50 VA toroid, bridge, 29.7 V DC nominal; 4700 µF reservoirs, ripple = I/(f·C) = 1.06 V pp at 0.5 A, 100 Hz |
| Headroom | 10.4 V above the regulated output at the ripple trough, 7.3 V with mains −10% (LM317 asks for 3 V)R5 |
| Heat | Each regulator dissipates about 5.7 W at 0.5 A: needs a heatsink |
| Budget | Estimated 124 op-amp sections (62 dual packages) × about 8 mA = 0.53 A per rail against the 0.5 A design current. An estimate, not a measurement |
S10 USB-C interface (optional tap)
| Item | Value / formula |
|---|---|
| Purpose | Record the four channels and the master over one USB-C cable, and play host audio back to rear line outputs. A tap: it listens to the analogue path and is not part of it. The mixer works with the board removed or switched off |
| Converters | Two ES9842PRO 4-channel ADCsR14: eight record channels (CH1–4, master L and R, two spare). Datasheet figures: dynamic range +122 dB in 4-channel mode (+128 dB mono), THD+N -116 dB, PCM / TDM words up to 32 bit. The ES9843PRO is an alternativeR15 |
| USB device | XMOS XU316 running USB Audio Class 2.0, asynchronous, as in the XK-AUDIO-316-MC reference (8 in / 8 out, up to 384 kHz, 32-bit)R16. The playback DAC part is not chosen in this draft |
| 32-bit float | A file and transport format. The converters output integer words; the host converts to 32-bit float when it writes the file. Float has a 24-bit significand (about 144 dB), which holds the 122 dB converter range (about 20.3 effective bits) without rounding. The real dynamic range is whatever the datasheet and the analogue noise floor give; float does not add range to the analogue path |
| Data rate | Eight channels of 32-bit words: 24.6 Mbit/s at 96 kHz, 49.2 Mbit/s at 192 kHz (calculated); USB 2.0 high speed carries 480 Mbit/s |
| Grounding | A digital isolator between the USB side and the audio ground (design target). Converters run from the board’s own audio clock |
| Not claimed | No measured figure for the assembled interface; the datasheet figures are the converter alone |
S11 Texture (optional)
| Item | Value / formula |
|---|---|
| Placement | After S4, before the fader, switched by relay K2. Off is the analogue strip with no converter in it |
| Option A, analogue | Four-pole ladder or OTA filter with a drive stage in front, resonance by feedback, 24 dB per octave. The Moog ladder is analysed by Stilson and SmithR17 and modelled non-linearly by HuovilainenR18. Component values are not set in this draft |
| Option B, DSP card | ADC, DSP and DAC per strip or shared and assignable. Modes: Clean SVF, Ladder (zero-delay-feedback form, after ZavalishinR19), Comb, Bitcrush; 4× oversampled (2× in the efficient mode); drive before the filter; resonance; LFO or envelope-follower modulation of the cutoff. Latency in the browser model: 51 samples |
| Controls | TEXTURE knob (amount; zero is off), MODE button with LED. The same signal flow runs in the browser model on /mixer |
| Not part of this revision | Formant and Noise / air modes |
S12 Spatial FX send / return (optional)
| Item | Value / formula |
|---|---|
| Status | Design target. Nothing here is built or measured. The browser model on /mixer runs the same four modes in floating point so the idea can be heard; it is not the analogue design |
| Topology | Four post-fader taps through 10k resistors into an inverting sum (gain −Rf/Rin = −1), a send level pot, the FX engine, a return level pot and a hard-bypass switch SW3 into the master sum of S7. With SW3 out the engine is out of circuit, as for S6 |
| Option A, BBD ensemble | MN3207, 1024 stagesR22: delay td = 1024 / (2 × fclock); the datasheet gives 2.56 ms at 200 kHz to 51.2 ms at 10 kHz and a +4 to +10 V supply. A 6.4 ms voice needs 1024 / (2 × 0.0064) = 80 kHz (calculated). Typical THD 0.4 % and S/N 73 dB under the datasheet’s own test conditionsR22. Three voices with clocks 120 degrees apart give the ensemble; a clock generator and anti-alias / reconstruction filters are needed and are not drawn. Companding is a design choice for noise, with no part chosen |
| Option B, digital delay | PT2399R21: 44 kbit RAM, delay set by the VCO frequency through one external resistor (10 kΩ recommended, up to 50 kΩ allowed in the datasheet note), THD 0.5 % and SNR better than 90 dB typical, +5 V. Echo feedback would be an op-amp mix of the output back to the input with a pot; wow and flutter from modulating the VCO resistor current is a circuit idea of the designer, not a datasheet claim, and is untested. The datasheet’s delay-time table is on a page not included in the copy read, so no delay range is claimed |
| Option C, spring and plate | Real springs and plates are not modelled by a short BBD. A DSP card (ADC, DSP, DAC) or a dedicated reverb IC would be needed; none is chosen. The plate algorithm in the browser model follows DattorroR20: input diffusers 142, 107, 379, 277 samples, tanks 672 + excursion / 4453 / 1800 / 3720 and 908 + excursion / 4217 / 2656 / 3163 samples at 29,761 Hz, decay-diffusion 0.70 and 0.50, 0.6 output taps from the paper’s Table 2. The spring is the designer’s own: 30 first-order all-pass stages per line for dispersion plus a delay and damping loop; it is not taken from a source |
| Controls (model) | Per-channel send; time / size, feedback / decay, tone, return; mode; tap tempo or tempo sync for the echo. Feedback is capped inside the model: echo 0.85, spring 0.95, plate decay 0.93, BBD regeneration 0.5. Return at 0 adds exact zeros |
| Browser model limits | No hiss, clock noise or aliasing of a real BBD is modelled; tape echo has no head or tape model beyond a saturating soft clip and low-pass filters. CPU: measured in Node on the test runner only (see tests/dsp_test.js); no phone was measured |
S13 Channel EQ (four-band, per channel)
| Item | Value / formula |
|---|---|
| Status | Design target. Nothing here is built or measured. The browser model on /mixer is floating-point DSP that follows the same control law; it is not a simulation of this circuit |
| Bands | HI shelf 3 kHz: +6 dB to full kill. HI MID bell 1.1 kHz: +10 dB to −27 dB. LO MID bell 350 Hz: +10 dB to −27 dB. LO shelf 180 Hz: +6 dB to full kill. Every knob is exactly 0 dB at 12:00 (centre detent). Where these figures come from is stated in the manual, section 5, S13 |
| Strip order | Trim, LO MID bell, HI MID bell, three-way LR4 split with LO and HI band pots, then the filter (S4) and the fader (S5). The cue tap stays pre-fader |
| Bells (two stages) | Each bell is an inverting op-amp stage with a 10k linear pot and a band-pass filter tuned to the bell frequency, the virtual-earth boost / cut arrangement described by ESPR23. ESP lists multiple-feedback, twin-T and state-variable band-passes as options; this design picks multiple-feedback for part count (a design choice). At the pot centre the band-pass has no effect and the stage is a unity inverter, so it is flat. Two inverting stages in series keep absolute polarity. Bell Q about 0.9 at full boost (design choice). Component values are not calculated in this draft |
| Shelves (kill) | A plain Baxandall shelf cannot reach full kill, so LO and HI reuse the S3a LR4 split insteadR3, R4: crossovers at 180 Hz (R = 88.7k, f0 = 179.4 Hz) and 3 kHz (R = 5.36k, f0 = 2969 Hz) with C = 10 nF. LO and HI bands each get a ×2 band amp and a 10k linear pot (minimum = kill, centre = unity, maximum = +6 dB); the middle leg gets ×2 then a fixed divide-by-2; an inverting sum recombines |
| Kill depth | Design target: beyond 60 dB in the stop band of the LR4 split, as for S3b. Not measured |
| Model versus circuit | The browser model uses second-order shelf and peaking sections (the textbook biquad forms, not cited to a page in this draft) with the same corner and centre frequencies and the same plateau levels, and a −80 dB floor for kill. The LR4 split has steeper transitions near 180 Hz and 3 kHz, so the two differ in the transition regions |
| Headroom | Boosts stack: with every band up the strip can add well over 10 dB at some frequencies. Headroom for that has not been analysed in this draft. The model on /mixer keeps its limiter on the master output only |
| Op-amp sections | Bells: 2 sections each, 4 per channel. Split: 12 sections per channel (see the budget in S9) |
Where the S13 figures come from. The four corner and centre frequencies and the boost and cut ranges are the ones Allen & Heath publish for the stereo channels 1–4 of its Xone:96 mixerR24 (LO 180 Hz, +6 dB to kill; LO MID 350 Hz, +10 / −27 dB; HI MID 1.1 kHz, +10 / −27 dB; HI 3 kHz, +6 dB to kill). They are used as a published reference for a familiar DJ EQ layout. The circuit, the shelf and bell shapes (shelf Q 0.71, bell Q 0.9), the knob design and the panel drawing are this design’s own; the guide gives no EQ schematic, and no part of that maker’s design, artwork or logo was copied. No connection with, or endorsement by, Allen & Heath is implied. Relation to DVS1: he is quoted as saying he does not need a 4-band EQ (q6), so this four-band EQ departs from his statement; it was added at the site owner’s request. The only related account is a writer reporting him ducking a high-mid on an Xone:96 (q26), a band a three-band isolator does not have. See Appendix C.