Skip to content

Latest commit

 

History

History
293 lines (249 loc) · 17.3 KB

File metadata and controls

293 lines (249 loc) · 17.3 KB

Character voicings — research record

Every number in Loopex's MEQ_DEF table, where it came from, and how far it can be trusted. Written so that a future change can tell a documented figure from a judgement call without re-doing the research.

Read the tags strictly:

Tag Meaning
DOC Manufacturer manual, service manual, spec sheet or patent
DOC-REISSUE Manufacturer documentation, but for a modern reissue standing in for vintage hardware
MEASURED Independent bench measurement or peer-reviewed analysis
COMM Forum, blog, dealer copy, magazine editorial
CHOICE Our decision. Not a fact. Never present it as one.
GAP Not findable. Deliberately left unmodelled rather than estimated.

Plugin-emulation marketing is not hardware documentation. Where a figure exists only in a plugin manual it is tagged COMM and named as such.


Where Character sits

Char is the output machine — what the loops reach the world through. It is the last coloured stage before the converter:

… → PUNCH FX → Drift → Pump → [ gSat → LoCut → HiCut → CHARACTER → Glue → Out → Limit ] → dither → hardware

Input Tape is the complementary stage: media formats, printed into the loop at record time. Char is named machines, applied live at playback. Formats in, machines out.

Table columns

loHz loDb · midHz midDb midQ · hiHz hiDb · sat · crushHz · makeupDb · grit · bits · clip · asym · compand · bwHz · poles · glueAtk glueRel glueKnee glueDist · limCeil · thrust

Notable ones: grit morphs the loop reader Hermite → linear → drop-sample. bwHz is the reconstruction-filter ceiling. poles sets LoCut/HiCut order. glueDist is distortion that rises with gain reduction. thrust is API's sidechain tilt.


The thirteen voicings

Off

Bypassed. master_character returns before touching anything.

962 — Studer 961/962 broadcast desk · default

Service manual supplied directly; the EQ was fitted by digitising the printed curve family at 600 dpi and least-squares fitting RBJ shelves against it.

  • Response +0.5 / −1 dB; −3 dB at 4.5 Hz and 45 kHz; THD <0.03 % at +6 dBu; EIN ≤ −125 dBu — DOC (§1.7.2)
  • Channel EQ is a Fächerentzerrer (fan equaliser). The published "20 Hz ±15 dB" and "20 kHz ±15 dB" are measurement frequencies — the gain reached at the edge of the band — not shelf corners. Reading them as corners is what made Bass and Treble inaudible for a year — DOC
  • Fitted: low shelf 100 Hz S=0.64 (0.43 dB RMS), high shelf 4750 Hz S=0.50 (0.32 dB RMS) — MEASURED (our fit to their plot)
  • Presence filter Q = 1, 150 Hz–7 kHz, ±11 dB — DOC
  • Master limiter: ratio to 1:20, threshold +6 dBu, stereo LINK, PDM/VCA — DOC
  • Voiced as near-nothing, because the desk is transparent by design.

Air — Focusrite ISA 110

The EQ Rupert Neve designed in 1985 for the Focusrite Studio Console at AIR Studios Montserrat — hence the name, and hence the nickname its top end still carries.

  • Parametric bands 40–400 Hz and 600 Hz–6 kHz (×3 ranges) — DOC
  • HPF 36/60/105/185/330 Hz and LPF 3.9/5.6/8.2/12/16 kHz, both 18 dB/oct — DOC
  • THD 0.003 % line, 0.0008 % mic — DOC
  • Transformer-coupled mic in, line in and output — DOC
  • Shelf frequency points never published by Focusrite. Mix and SOS reviews give LF 33/56/95/160/270 and HF 3.3/4.7/6.8/10/15/18 kHz and disagree with each other — COMM. Our 95 Hz / 15 kHz carry that caveat.
  • Correction: we first used 16 kHz for the HF shelf, having confused the unit's low-pass positions with its shelf.

SSL — SSL 4000 bus

  • E vs G is a documented behavioural difference: E holds bandwidth constant with gain; G varies it (more boost/cut → more selectivity) — DOC (X-Rack manual)
  • '242 vs '02: ±18 dB vs ±15 dB, 18 dB/oct HPF — DOC, though SSL's own documents conflict on shelf slope (6 vs 12 dB/oct)
  • Original-console band ranges exist only in licensed plugin manuals — COMM
  • G-series bus compressor: 3 ms / 0.3 s / 4:1 — DOC
  • No original-console THD or transformer data exists anywhere — GAP

Neve — Neve 1073

  • LF shelf 35/60/110/220 Hz ±16 dB; mid bell 360/700/1.6k/3.2k/4.8k/7.2 kHz ±18 dB; HF shelf fixed 12 kHz ±16 dB; HPF 50/80/160/300 Hz, 18 dB/oct — DOC
  • THD ≤0.07 % at +20 dBu — DOC
  • Third-harmonic dominant, bench-measured — MEASURED (Sound On Sound, 1073N)
  • Mid-band Q: AMS Neve state only "fixed Q". No number exists — GAP
  • Dynamics from the 33609 (the 2254 diode bridge as a stereo bus box): comp attack 3/6 ms, release 100–1500 ms plus Auto1 100 ms/2 s and Auto2 50 ms/5 s, ratio 1.5–6:1, soft knee reaching full ratio ~6 dB above threshold — DOC-REISSUE
  • THD per mode: <0.075 % bypassed, <0.2 % compressing, <0.45 % limiting — DOC. The only unit in this entire table whose maker publishes distortion under gain reduction, and the sole basis for glueDist.

Trap: a "0.3 % THD at 6 dB GR, third-harmonic dominant" figure circulates near 2254 discussions. It is from an SOS review of the Rupert Neve Designs 5254, a different modern unit. Do not apply it here.

Trident — Trident A-Range

Weakest-documented row in the table. Treat every figure with suspicion.

  • LF shelf 50/80/100/150 Hz; low-mid 250/500/1k/2k; high-mid 3k/5k/7k/9k; HF shelf 8k/10k/12k/15k; ±15 dB; Q 1.3 — DOC for the reissue; Trident's own handbook for the original console prints no dB range and no Q
  • The two sources disagree on the lowest LF point (50 vs 80 Hz)
  • Transformer-coupled in and out; inductor mid bands — DOC
  • Harmonic order: GAP. Do not assume 2nd or 3rd. Our asym is a CHOICE.
  • Dynamics from the CB9146 (FET): attack 20 µs–200 ms, release 25–500 ms, ratio to 20:1 — COMM only, and specifically a forum user reading a photograph of a front panel. No manufacturer documentation for this unit exists.
  • Correction: the CB9066 is an equaliser, not a compressor. That pairing was wrong.

Studer — Studer A800 tape machine

Distinct from the 962 desk. Different machine, different character.

  • 60 Hz–18 kHz ±1 dB at 15 ips; 60 Hz–20 kHz at 30 ips — DOC
  • W&F 0.04 % DIN peak weighted; operating level 510 nWb/m; bias 240 kHz — DOC
  • THD specified explicitly as 3rd harmonic, 1 % max at operating level. Studer publish no even-order figure at all — DOC
  • Head bump: not documented by Studer. The word does not appear in 1.1 MB of service text — GAP. Our ~60 Hz comes from Jack Endino's measured A80 MkII and A827 siblings (+1 to +1.8 dB, 30–62 Hz at 15 ips) — MEASURED, third party
  • The structurally useful law: the bump is a damped ripple, not one peak, and it shifts one octave up per speed doubling — MEASURED

API — API 550A, with API 2500 dynamics

  • LF 30/40/50/100/200/300/400 Hz; MF 200/400/600/800/1500/3000/5000; HF 2500/5000/7000/10k/12.5k/15k/20k; ±12 dB stepped; 12 dB/oct — DOC
  • Proportional Q published only as a graph. The two community figures disagree by ~2× — GAP, deliberately not modelled
  • Correction: we first used 900 Hz and 6 kHz. Neither is on either switch.
  • Dynamics from the 2500: attack .03/.1/.3/1/3/10/30 ms; release .05–2 s plus variable 50 ms–3 s, no auto; ratio 1.5–∞; HARD/MED/SOFT knee; TYPE OLD = feed-back, NEW = feed-forward; THAT 2180 VCAs with a 2252 RMS detector — DOC
  • THRUST (US Patent 5,170,437): a sidechain filter before the RMS detector, +10 dB/decade, −15 dB at 20 Hz to +15 dB at 20 kHz, pivot ≈ 632 Hz, inverse of the pink-noise curve. Bass therefore stops driving the gain reduction — DOC, with a plotted curve in API's manual and a five-stage RC ladder in the patent. We implement a first-order tilt, an approximation of that ladder.
  • Max output +32 dB, response ±0.5 dB 20 Hz–50 kHz, S/N 120 dB, clips +28 dBu — DOC
  • Knee width in dB: GAP. API's HARD/MED/SOFT plot has no scale on either axis and nobody has measured it. Ours is a CHOICE.
  • THD under gain reduction of the 2500 as built: GAP. No measurement exists publicly. But the THAT 2180 VCA inside it (four per channel, per SOS's inspection) publishes plots — DOC, component-level, not whole-unit:
    • 0.005–0.010 % at unity gain → 0.020–0.030 % at ±15 dB gain. Distortion triples to quadruples once the VCA actually pulls gain.
    • Datasheet FFT (Fig. 13): 2nd harmonic −80.1 dB, 3rd −104 dB — strongly even-order dominated, third some 24 dB below.
    • Second mechanism THAT document: 2252 ripple modulating the VCA, worst at low frequencies under heavy GR. Their alignment nulls it at 100 Hz / ~20 dB compression. So API carries a small even-order grit (−0.05) against the Neve's larger odd-order 0.35. In MEQ_DEF the sign of glueDist carries harmonic order, the magnitude carries amount. The 2510/2520 op-amps and the output transformer remain GAP.
  • Caveat on the pairing itself: 550A is 1967, 2500 is 2000s. This is matched by function — the API bus compressor people actually use — not by era, unlike Neve→33609 or ISA 110→ISA 130 which are genuine contemporaries.

Ampex — Ampex ATR-102

  • ±0.75 dB 200 Hz–20 kHz at 30 ips; ±2 dB 35 Hz–28 kHz. At 15 ips, ±0.75 dB 100 Hz–15 kHz, ±2 dB 20 Hz–20 kHz — DOC (Table 1-4)
  • 3rd harmonic <0.3 % at 370 nWb/m, <3.0 % at 1040 nWb/m (+9 VU). Even-order <0.1 % at reference — DOC. That 1:3 ratio is what sets its asym, and the 0.3 → 3 % progression over +9 dB is why it has the lowest Limit ceiling.
  • W&F ±0.03 % peak weighted; bias 432 kHz — DOC
  • HF −3 dB point: GAP — Ampex publish only ±0.75 and ±2 dB windows
  • Head bump: GAP, same as the A800. Ampex's spec structure implies a bound of ≤±2 dB below 200 Hz, but that is an inference from the table's shape, not a figure.
  • Ampex never made a compressor or limiter. Catalogue OCR-searched; two circulating listings ("Ampex 350 + Triad opto", "AM-10 with Fairchild") are third-party racks.

MPC60 — Akai MPC60

  • 40 kHz fixed; response 20 Hz–18 kHz; "12 bit sample resolution, special non-linear format, reduced noise"; tuning +½/−1 octave — DOC (service manual §I)
  • 16-bit ADC and DAC with companded 12-bit storage — DOC (operator's manual)
  • The companding law is never named. "Special non-linear format" is a deliberate non-disclosure, not an omission — GAP. µ-law (µ=255) is our CHOICE: period- correct and behaves as described.
  • It has a real reconstruction filter, unlike the E-mus. Its character is the companding quantiser, not imaging — hence low grit and bwHz = 18000.
  • Service manual parts list uses internal Akai part numbers (EF-…), so filter component values are not derivable from it — GAP

S950 — Akai S950

  • Sample rate 7.5–48 kHz; bandwidth 3–19.2 kHz — DOC
  • fc = fs × 0.4, annotated on the voice block diagram. Rate and bandwidth are locked at 2.5:1 — DOC
  • 6-pole, 36 dB/oct MF6CN-50 switched-capacitor filter on both record and playback (IC7 + IC109 ×8), plus a fixed 2-pole 18 kHz stage — DOC
  • 12-bit linear. No companding — DOC. Do not port the MPC60 assumption across.
  • Pitch by variable sample clock — no interpolation, no drop-sample — DOC
  • Frequency response, S/N, THD: never published by Akai — GAP
  • Correction: independently adjustable bandwidth vs sample rate is a feature of Inphonik's RX950 plugin, not the hardware. We recommended this unit on a feature it does not have.

SP-1200 — E-mu SP-1200

  • 26.04 kHz fixed (shared with the Drumulator and SP-12); 12-bit linear, no companding — DOC
  • No reconstruction filter. Output is bare zero-order hold; the imaging is the sound — DOC
  • Drop-sample pitch shifting: fractional index, truncated, no interpolation. Pitching down repeats samples outright — MEASURED (Yeh, Nolting & Smith, CCRMA/ICMC 2007)
  • No dedicated audio ADC — a 12-bit SAR loop around the playback DAC (AD7541) — DOC
  • Output channel filters: ch 1–2 dynamic SSM2044, ch 3–6 fixed, ch 7–8 unfiltered — DOC
  • Exact fixed-filter cutoffs: GAP

The archive.org SP-1200 service manual scan is image-only; an early automated fetch returned confident "quotes" that were actually SP-12 text. Treat any online claim of quoting it with suspicion.

Emu SP-12 — E-mu SP-12

  • Same 26.04 kHz clock, same 12-bit linear format, same absence of a reconstruction filter as the SP-1200. The published "27.5 kHz" is a spec-sheet error; measured at 26 kHz — MEASURED (CCRMA)
  • Differentiated from the SP-1200 by its darker fixed output filters, which is what our hiHz/hiDb carries
  • Anti-alias filter measured as needing an order-11 digital fit, attenuating above ~15 kHz — MEASURED
  • The manual's "42 dB per octave" is in a generic tutorial sentence — it is not a spec for this unit and is commonly misquoted
  • SSM2044 self-oscillation trim 1.3 kHz — DOC

Era-matched dynamics

Most of these consoles have no compressor. Each voicing borrows the unit the same maker built in the same era, rather than being given invented timings.

Voicing Unit Topology Pairing
Neve 33609 diode bridge Solid — only source publishing THD under GR
Air ISA 130 VCA Solid — same ISA rack, full spec table
SSL G-series bus comp VCA Solid
962 its own desk limiter PDM/VCA Solid — in-unit
API 2500 VCA + RMS By function, not era
Trident CB9146 FET Weak — zero manufacturer documentation
Studer, Ampex none exists — class-appropriate timings
all four samplers none exists — neutral Glue, hard rail on Limit

What the research overturned

  1. "Transformers and tape mean even harmonics" is false. Contradicted three times independently: SOS measured the Neve 1073 third-harmonic dominant; Studer specify the A800's THD as 3rd and publish no even-order figure; Ampex document the ATR-102 at <0.3 % 3rd against <0.1 % even-order. The asym column was built on folklore and is no longer used as a differentiator.
  2. "Juno chorus-console" was not a real machine — and the voicing contained no chorus.
  3. The S950's independent bandwidth control is a plugin feature, not Akai's.
  4. CB9066 is an EQ, not a compressor.
  5. Crush-rate corrections were sonically inert. eqCrush is an integer hold count, so MPC60 never crushed at 30 kHz or 40 kHz, and 22 kHz and 26.04 kHz both give ÷2.
  6. Nobody publishes harmonic order behind a THD figure — Neve's 33609 excepted.

Standing gaps

Head bump for both tape machines · Neve mid Q · MPC60 companding law · API knee width · API 2500 THD under GR · S950 frequency response and THD · SP-1200 fixed filter cutoffs · Trident's entire spec · harmonic order for every unit except the Neve.

Sources

Studer 961/962 service manual (supplied) · Studer A800 service notes · Ampex ATR-100 manual · MRL equalization fundamentals · Jack Endino, measured recorder curves · AMS Neve 1073 · AMS Neve 33609 · SOS Neve 1073N · Focusrite ISA110 manual · Focusrite ISA 130/131 · SSL X-Rack EQ · SSL Origin · API 550A · API 2500 manual, THRUST + knee plots · US Patent 5,170,437 · THAT 2252 datasheet · Trident A-Range handbook · Akai MPC60 service manual · Akai S950 operator's manual · E-mu SP-12 service manual · Yeh, Nolting & Smith, SP-12 modelling, CCRMA 2007