Sphinx Channel

Congratulations

You just joined a very small group of engineers and producers who own one of the most comprehensively modeled analog channel strip plugins out there. Sphinx Channel is a channel summing plugin meant to be added onto every track: a component-modeled analog input chain with an input transformer, a discrete drive circuit, TrueRail power-supply grades, and a stereo width stage. It models the electronic components inside the hardware, wires them together in the real circuit topology, connects every plugin instance together through a shared virtual console frame, and lets the sound emerge from how they interact when summed together.

That's the hardware 3D sound.

Sphinx Channel instances in the same DAW session behave like real channel strips in a mixing console: they actually share supply loading, mains-ripple phase, and signal ground, so a full DAW session gains a subtle common electrical character from microscopic interactions between all channels. A single instance always sounds like one channel.

Sphinx Channels even warm up in their first moments and keep self-heating with the program, like real hardware.

This interconnection technology is called TrueRail™.

Knob Interactions: drag vertically to adjust its value, hold Shift while dragging for fine adjustment, double-click a knob to type a value directly, scroll the mouse wheel for fine or fast adjustment depending on scroll speed, and Alt+Click (Option+Click on Mac) to restore the value the knob held before your last edit. The Alt+Click undo is single-step per knob - a second Alt+Click does nothing until you make another change. The undo state persists when you save and reload your DAW session.

What's in the Box!

The Sphinx Channel audio plugin (AU, VST3, Mac/Win), two visual skins (Amok, Nevy), default factory presets, this manual, the signal test results, an uninstall script, and your serial key.

Installation

Requirements:

• macOS: macOS 11 (Big Sur) or later, any AU- or VST3-compatible host. Six macOS installers ship: two Universal builds (AU Universal and VST3 Universal), each carrying both Apple Silicon (arm64) and Intel (x86_64) code in one package so macOS picks the correct architecture at load time; and four single-format single-architecture builds (AU Apple Silicon, AU Intel, VST3 Apple Silicon, VST3 Intel) for cases where a lean single-arch payload is preferred. Pick the format your DAW loads. The Intel builds run natively on Intel Macs and on Apple Silicon via Rosetta 2; the Apple Silicon builds run natively on M-series Macs. Rosetta 2 is the Apple Silicon translator for x86_64 binaries and does not run on Intel Macs. Formats install to separate paths, so you can install AU and VST3 side by side without conflict.

• Windows: Windows 10 or later, x64 (any Intel or AMD CPU from 2011 onward), any VST3-compatible host

• RAM: 4 GB minimum, 8 GB recommended for large sessions
• Disk: approximately 30-50 MB
• Activation: one-time internet connection from any machine for web activation

macOS

Double-click the installer .pkg.

The AU installer places the plugin at /Library/Audio/Plug-Ins/Components/Sphinx Channel.component. The VST3 installer places it at /Library/Audio/Plug-Ins/VST3/Sphinx Channel.vst3. Both formats ship in two flavors: a Universal installer containing both Apple Silicon and Intel binaries in a single package (macOS loads the correct architecture automatically - the recommended choice for most users), and separate single-architecture installers for people who want to keep the payload minimal on a known target. You can install both formats side by side, as each format's installer only touches its own plugin path. The manual, signal test results, and an uninstaller .command are placed at /Applications/Analog Realism/Sphinx Channel/ regardless of format - visible in Finder, parallel to the Windows layout below. Your per-machine activation state (calibration.dat) lives in ~/Library/Application Support/Sphinx Channel/ and is preserved across upgrades.

Windows

Run the installer .exe.

The plugin installs to C:\Program Files\Common Files\VST3\Sphinx Channel.vst3 (VST3). The manual and signal test results are copied to C:\Program Files\Analog Realism\Sphinx Channel\. Your per-machine activation state (calibration.dat) lives in %APPDATA%\Sphinx Channel\ and is preserved across upgrades.

Open your DAW, and insert Sphinx Channel into all of your channels and/or groups.

Authorization

Demo Mode: Full functionality with a brief 2-second silence approximately every 2 minutes, synchronized across all unactivated instances in the session.

Click the key icon, paste your serial, click Authorize. A challenge code appears - copy it, visit the combined activation/deactivation web page shown, paste your serial and challenge on the Activate side, click Activate. Copy the response code back into Sphinx Channel, click Activate. Done - zero internet needed from the plugin itself. Your serial works on 2 machines. Re-activating after an OS reinstall is seamless - the system recognizes your machine.

Deauthorization: Click the key icon, click Deauthorize. Copy the removal code, visit the same combined web page, switch to the Deactivate side, enter your email and removal code, copy the response back, and click Confirm removal from the plugin itself. Your seat is freed. Need the removal code again? Just repeat - the system re-issues it.

Quick Start

Insert Sphinx Channel on all tracks and/or groups - and play. The defaults are the calibrated starting point: transformer L1544 driven at 120 percent, SLL circuit, TrueRail C, width on at +6 percent. Set the INPUT gain so the LED stack rides around the -6dBFS mark on the loudest channels, then choose the main circuit, transformer, and width that suit the song. Use the powerful "ALL" button to control all Sphinx Channel instances from just one plugin.

Why Sphinx Sounds Different

Many analog modeling plugins use "black box" modeling: measure what hardware does, build a filter that copies it. Sphinx models what the hardware IS - every transistor, tube, transformer core and coupling capacitor simulated as individual components wired in the real circuit topology.

The difference is profound. In black-box models, the components won't interact and breathe together. In Sphinx, everything connects through a shared power rail. When the channels work hard, they draw more current, causing voltage sag that shifts the bias of every other instance. They saturate differently. The transformers' operating point shifts. The entire circuit breathes as one organism.

TrueRail

TrueRail is Analog Realism's secret sauce used in the Sphinx Channel and the Sphinx 101 plugins. It is a family of interacting analog mechanisms producing complexity that no hand-tuned static algorithm can replicate, all verified by automated measurement. In Sphinx Channel the family spans the input transformer physics, the power supply, the active electronics, the environment - and the shared console that binds every instance in the session together:

Transformer physics

• Jiles-Atherton magnetic hysteresis with flux integration - the core is driven by integrated flux (V/f law), producing frequency-dependent saturation that matches real iron
• Four core material models with distinct saturation curves, permeability, coercivity, and remanence per material class
• Winding asymmetry producing even-order harmonics whose magnitude varies with transformer model
• Bias-DC flux walk - program-dependent DC magnetisation offset producing session-responsive even harmonics
• Remanence - the core retains a fraction of its magnetisation during silence, starting each new phrase from a non-zero magnetic state
• State-domain Bertotti dynamic core loss - eddy current and excess loss terms that resist rapid flux changes like physical friction in the iron
• Signal-dependent copper winding loss with thermal DCR rise over the session
• LC resonance from leakage inductance and winding capacitance, per model, entering the measurable band at high sample rates
• Signal-presence-gated demagnetisation with remanence target

Power supply

• Shared rail sag with asymmetric attack and release, soft-clamp asymptotes at the consumer sites
• Conduction-gated reservoir ripple with rectifier harmonics and mains frequency selection (50/60 Hz)
• Series-pass voltage regulator model with per-topology PSRR
• Signed ground-return injection carrying the signal fundamental through per-topology chassis impedance

Active electronics

• Ebers-Moll BJT with Class-AB crossover emphasis and thermal junction bias shift
• Koren vacuum tube with grid-current blocking at millisecond charge rate and hundred-millisecond discharge
• Capacitor dielectric absorption with multi-time-constant charge re-release per dielectric type
• Per-topology amplifier slew-rate limiting
• AC-coupled output path - gain-cell DC offsets never reach the output, blocked exactly the way hardware output coupling capacitors block them

Environment

• Cross-channel crosstalk with frequency-dependent shaping
• Per-instance manufacturing tolerance variation from seed
• Filtered-noise 1/f thermal drift, aperiodic and per-instance
• Program-responsive thermal self-heat with fanout to transformer permeability, transistor bias, tube transconductance, and core loop parameters (k, a, c)
• Physical component warm-up on instance creation
• Current-scaled shot noise on the modeled Class-A device stages
• Per-topology 1/f noise spectral shaping across the full audio bandwidth
• Channel-independent noise floors - each channel draws its own noise stream while the stereo pair shares matched component tolerances, so the floor sums to mono and nulls the way two real devices do
• True Gaussian noise statistics - the modeled thermal floor carries the statistics of real thermal noise rather than a uniform digital approximation

The shared console (Sphinx Channel exclusive - instances in one session behave like strips in one frame)

• Shared supply loading - every instance's current draw sags the common rail that every other instance reads, so a busy session breathes as one power supply
• Console-coherent mains ripple - all instances phase-lock their supply ripple to one shared console clock, summing the way strips in a single frame hum together
• Console channel bleed - each strip contributes a calibrated share of adjacent-channel bleed, held at "one console" level no matter how many instances the session runs
• Shared ground-return coherence - the signed ground currents of all strips sum, with real cancellation, into each strip's ground-lift path
• Console ambient heat - the frame warms with the whole session's work, and every strip's thermal state feels it

TrueRail Buttons: - the TrueRail A, B and C scales the per-instance manufacturing-tolerance spread. Same instance seed across all tiers - switching tiers does NOT re-roll the random draw, it widens or narrows the same underlying variation pattern.

Tier A - Factory Spec. Tightest tolerances (±0.5-1% typical, scale factor 0.0667). Closest match between L and R channels, smallest deviation from nominal component values. The "studio reference" sound - what a freshly-calibrated production unit would measure.

Tier B - Production Standard. Moderate tolerances (±1-2%, scale factor 0.1333). Adds natural character without demanding attention. The sweet spot for most work.

Tier C - Vintage Wide (default). Widest tolerances (±2-4%, scale factor 0.20) - hand-wired vintage character. Most L/R asymmetry, most deviation from nominal values, the most "alive" sound. The welcome preset ships on Tier C because the per-instance L/R variation it produces is the most audibly distinctive of the three tiers.

Every plugin instance generates its own unique set of component tolerances, just like no two pieces of real hardware are identical.

Signal Flow and Routing

Audio passes through: Input Gain → Input Transformer → Drive Stage → Width → Output.

Input ─► [Input Gain] ─► [Input TX†] ─► [Drive Stage*] ─► [Width] ─► Output

  *  Drive Stage = main circuit (SLL / Nevy / Amok) - always active
                   when the plugin is enabled. Processes the
                   signal AFTER it has passed through the Input
                   Transformer, so the drive nonlinearity acts on a
                   signal that already carries the transformer's
                   harmonic shape and LF resonance.
  †  TX = Input Transformer (optional, bypassable).

  ┌──────────── TrueRail™ shared infrastructure (always active) ────────────┐
  │  Power Supply Rail  │  Crosstalk Matrix  │  Thermal Drift                │
  │  Component Tolerance │ Instance Seed     │  Shared Console Link          │
  └─────────────────────────────────────────────────────────────────────────┘

Main Circuits - SLL, Nevy, Amok

The main circuit selector determines the core amplifier topology your audio passes through. This is the most fundamental choice in Sphinx Channel - it defines the electronic soul of the entire chain. Use a specific main circuit for all DAW channels, or mix them up to better complement the channels' individual sounds.

SLL - solid-state precision inspired by legendary British mixing consoles. Push-pull-style BJT topology modeled with the Ebers-Moll equations. Odd-harmonic dominant - the BJT's symmetric transfer curve and emitter degeneration suppress even-order nonlinearity, giving the clean, punchy odd-harmonic character of precision solid-state design. The cleanest of the three personalities at the -6 dBFS operating sweet spot, and the fastest transient response. Choose SLL when you want "finished" - polished and professional.

Nevy - single-ended Class-A discrete transistor with asymmetric off-centre bias plus a gentle low-shelf around 200 Hz modeling classic iron-core output-transformer coupling. The Class-A asymmetry enriches the even-order harmonic content, producing the blend of even and odd harmonics that gives this console class its dimensional warmth - a warm-transistor middle ground, sitting above SLL's cleaner solid-state character and below the tube path's depth. Choose Nevy when you want "expensive."

Amok - tube-flavored signal path modeled with Norman Koren's plate-current equations on a 5670 dual triode (a mastering-grade vari-mu valve, not a 12AX7), full Miller capacitance, single-ended Class-A. The rest of the personality is a mastering-console scaffold: ±17.5 V op-amp rails with a regulated 225 V tube B+ supply feeding the triode. Complex even-harmonic distortion, the slowest rise time of the three (Miller capacitance effect), the deepest saturation. Enriched even-order harmonic content from the triode's asymmetric transfer curve - dramatically richer even-order character than solid-state. Choose Amok when you want "alive."

Each circuit changes the entire chain - not just the drive stage, but the Transformer and Width interaction also.

Transformers

Four distinct transformer models. Each represents a different core material with its own magnetic character. The transformer uses full Jiles-Atherton magnetic hysteresis - it remembers its recent magnetization history, producing saturation that depends on what came before. Use a specific model for all DAW channels, or mix them up to better complement the channels' individual sounds.

M1166 - grain-oriented silicon steel core. Warm, smooth saturation. Low-frequency resonant peak in the deep sub band adds subtle low-end weight. The classic "console sound." Best for: warmth and body.

C9049 - nickel alloy core. Authoritative, controlled saturation. The lowest LF resonant peak of the iron cores - it sits deeper in the sub than the M1166's, so the audible low end reads tighter and more defined. Best for: precision and punch.

K1166 - silicon steel variant. Tight, fast saturation with minimal overshoot, highest LF resonant peak of the iron cores. Best for: transient-heavy material that needs transformer color without smearing.

L1544 - amorphous strip core. The most transparent. Produces an odd-harmonic-leaning character consistent with the symmetric saturation behavior real amorphous strip cores exhibit. Essentially flat - no LF resonance peak, no audible saturation. Best for: transparency with a hint of iron.

Knob Reference

INPUT adjusts the level feeding the whole chain, -12 dB to +12 dB. The meters read the signal right after this knob.

DRIVE sets how hard the transformer core is driven, range of 0 to 200 percent. 100% is the published real-world response of that specific core model.

ON engages the transformer stage.

TRANSFORMER four input cores (M1166, C9049, K1166, L1544), selected with the rotary switch or the model names.

CIRCUIT (SLL, Nevy, Amok) chooses the drive-stage topology.

TRUERAIL (A, B, C) selects the power-supply grade; each grade changes how the supply sags and recovers under load.

WIDTH adjusts stereo width from -100 to +100 percent; ON engages the module. On mono tracks the width section stays off regardless of the setting.

ON engages the width stage.

ALL broadcasts any new knob changes, preset loads, window size, and skin changes to every other Sphinx Channel plugin in the DAW session. This is the remote control button. Preset saves stay local though. ALL button always starts "disarmed" when a session opens.

OS selects the oversampling mode. See more in chapter "Oversampling and Antialiasing".

ENABLE is the strip's master bypass. Lit means the strip processes; click it and the entire chain steps aside, passing the input through untouched with a short crossfade. The reported latency never changes with the switch, so the track stays in time either way. A bypassed strip stays powered in the console frame and neighboring instances keep sensing it as an idle channel.

Top Bar

Oversampling and Antialiasing

Sphinx Channel oversamples the entire signal chain. One minimum-phase polyphase IIR half-band filter at the input upsamples the audio to the user-selected rate (2× / 4× / 8× / 16×); one matching half-band at the output downsamples back to the host's sample rate. Because every nonlinearity is anti-aliased by the same filter pair, there are no per-module latency mismatches in the chain.

All four oversampling rates are fully functional and selectable in real time. Higher rates reduce the chain's fold-back contribution further but use proportionally more CPU.

Metering

The VU needle shows the stereo sum, calibrated so a channel riding -18 dBFS RMS sits on the printed 0 VU mark, in the style of a console channel alignment. The scale runs from -20 VU at rest to +3 VU at the end stop. The input LEDs light at -18, -12, -6, -3, and 0 dBFS per side. On a mono track the left column carries the level and the right column stays dark.

Skins

Switch instantly between skins by clicking the skin selector switch button above the VU meter.

Resize the window by dragging the corner handle in the lower right - the interface stretches smoothly to any size, and the chosen size is remembered per plugin instance.

Nevy (default) - dark-styled layout.

Amok - alternative aesthetic, same controls, light-styled metal with industrial accents.

Gain Staging

Sphinx operates best at -6 dBFS peak. This is where the nonlinear components produce their most musical harmonics. Every channel and group should still be gain-staged relative to the loudest channels, so that Sphinx Channel's processing gets to affect different levels of channels in their own unique ways.

Too hot (peaking at 0 dBFS): harsh highs, pumping, lost definition - reduce Input Gain 3-6 dB.

Too cold (below -12 dBFS): thin and lifeless - increase Input Gain.

The TX Drive knob pushes transformer saturation independently: 100% anchors the harmonic balance to the model's nominal core-flux operating point - the calibrated reference where saturation begins to color; above 150% extrapolates the modeled physics into obvious warmth; below 100% backs off into cleaner-than-hardware operation.

Components

Sphinx Channel models analog circuits at the individual component level. Each electronic part is computed from published component physics - Ebers-Moll BJTs, Koren tubes, Jiles-Atherton transformer cores - augmented with light per-model harmonic shaping so each modeled unit keeps its own distinct audible character. None of it is a generic saturation curve or static waveshaper. The componenttest suite verifies every model directly against its expected behavior; the sections below name the model and list what is measured.

Vacuum Tube - Koren Model

The Amok circuit's triode stage uses Norman Koren's plate current equations, applied to a 5670 dual triode (a mastering-grade vari-mu valve - the low-mu type used by classic tube mastering hardware, distinct from the higher-mu 12AX7 found in guitar/preamp circuits). The model computes plate current as a function of grid voltage, plate voltage, and tube-specific parameters (mu, kp, kvb, Ex). This produces the asymmetric transfer curve that gives tubes their characteristic even-harmonic warmth - even-order content rises naturally with signal level. Miller capacitance is modeled for accurate HF rolloff, and a cathode-bypass capacitor sets the operating bias.

Verified: THD-rises-with-level curve, even-order content present, output level still alive at hot drive, per-instance THD spread, per-instance seed variation in tube parameters.

Bipolar Junction Transistor - Ebers-Moll Model

The SLL and Nevy circuits use BJT gain stages based on the Ebers-Moll equations. The model computes collector current from base-emitter voltage using the transistor's forward current gain (hFE), Early voltage, and thermal voltage. The SLL topology uses push-pull configuration (symmetric, odd-harmonic dominant). The Nevy topology adds a DC bias offset for Class-A operation, producing the asymmetric transfer curve that enriches even harmonics. Two real part numbers are modeled - BC184 (NPN, low-noise audio) and BC214 (PNP complement) - with the Ebers-Moll equations populated from their datasheets.

Verified: BC184 THD at -6 dBFS, BC214 THD at -6 dBFS, NPN vs PNP transfer-curve differential (they should not be identical), THD responds to level, output level alive under drive, per-instance seed variation.

Per-Instance Component Tolerances

Every plugin instance receives a unique seed that draws its component parameters - transistor gain and bias, tube parameters, transformer core constants - within the TrueRail tier's tolerance range:

• Tier A (tightest): ±0.5-1% - precision matched components
• Tier B: ±1-2% - standard professional grade
• Tier C (default, vintage wide): ±2-4% - aged, drifted components

No two instances measure identically, and the L/R halves of one instance carry the matched-but-not-identical relationship of a real stereo pair.

Verified: per-instance seed variation is non-zero on every component model above, and the per-instance variation suite confirms every seed still lands in the same harmonic-ordering regime.

Transformer Cores

Four transformer core materials are modeled using the Jiles-Atherton magnetic hysteresis equations (see also section "Transformers"). The JA model computes magnetization from field intensity using five physical parameters: saturation magnetization (Ms), anhysteretic shape (a), domain coupling (alpha), pinning/loss coefficient (k), and reversibility (c). Each core has its own JA parameter set plus a per-core harmonic-shaping term that gives each model its distinct voicing - M1166 silicon steel produces warm harmonic character with even-order enrichment; L1544 amorphous produces a transparent, odd-harmonic-leaning character consistent with symmetric core saturation.

Verified: M1166 shows more even-order content than L1544, measurable spread in harmonic balance across core types, per-instance JA-parameter tolerance variation non-zero.

Shared Infrastructure (Always Active)

Beyond the per-component models, a set of "chassis" subsystems is always on - they are how the components interact, not what they do individually:

• TrueRail Power Supply (RailState): shared virtual rail. Every feature pulls current from it in proportion to its current workload, AND every feature reads the resulting rail-voltage deviation to adjust its own operating point - transformer saturation headroom, output PSRR coupling, inter-channel crosstalk, and per-module bias all respond. Rail sag is personality-dependent along two dimensions.
• Depth (how far the rail drops under steady load): SLL is the shallowest, Amok the deepest, Nevy sits in between - the classic "SLL has the most headroom, Amok pumps hardest, Nevy is warm-and-cohesive" ordering, matching the reference hardware.
• Recovery time (how long the sag lingers after a transient ends): sag onset is fast in every personality (the supply drops at whatever rate the circuit demands), but recovery has to go back through the supply's impedance, and the release side is personality-scaled. SLL releases nearly as fast as it sags (matching modern tight regulation), Nevy holds sag the longest (matching its unregulated vintage supply, and where the characteristic "glue" comes from - the sag lingering musically between transients), Amok sits in between (its mixed op-amp / tube-B+ topology contributes moderate recharge lag). On loud passages the headroom lingers slightly reduced after the transient before returning to nominal - the "spongy" transient feel a real console produces. A per-personality full-wave-rectified supply ripple runs continuously on top of the sag at the selected mains frequency (100 Hz at 50 Hz mains, 120 Hz at 60 Hz mains, selectable via the top-bar MAINS toggle), at amplitudes that vary by topology (tightest on SLL, strongest on Nevy). At shipped levels the ripple sits well below the analog noise floor and contributes to the organic baseline rather than audible hum.
• CrosstalkMatrix: frequency-dependent L/R coupling through virtual ground impedance.
• ThermalDrift: slow, never-repeating 1/f temperature wander modulating component values over time.
• Chain bandwidth: the drive stage's finite transistor transition frequency and tube Miller capacitance give the signal path a frequency-dependent gain rolloff.
• InstanceSeed: deterministic per-instance random draw of component tolerances, scaled by TrueRail tier.
• Shared console link (ConsoleSegment): the cross-instance connection that lets every strip in the session share supply loading, mains-ripple phase, channel bleed, ground return, and ambient heat - see the TrueRail chapter's shared-console family.
• Per-Module Bypass: every module uses a smooth 20 ms crossfade for click-free enable/disable. When fully bypassed (enableGain = 0), output equals dry input - true passthrough - and the module stops drawing current from the shared rail.

Technical Specifications

Formats: Audio Unit (AU) and VST3, macOS and Windows.

Platform: macOS 11 (Big Sur) or later - six macOS installers ship: AU Universal and VST3 Universal (both architectures in one package), plus AU Apple Silicon, AU Intel, VST3 Apple Silicon, and VST3 Intel as single-arch alternatives. The Intel builds also run on Apple Silicon via Rosetta 2. Windows 10 or later (x64).

Sample rates: 44.1-192 kHz (verified identical character).

Buffer sizes: 64-2048 (verified bit-identical output).

Internal precision: 64-bit.

Oversampling: 2× / 4× / 8× / 16× user-selectable, applied globally - one minimum-phase polyphase IIR half-band at the input upsamples to the elevated rate, one matching half-band at the output downsamples back. The minimum-phase design eliminates the pre-ringing a linear-phase filter would produce.

Optimal operating level: -6 dBFS peak input material.

Bypass transparency: plugin bypass crossfades directly between the processed and unprocessed signal (~12 ms) with no silent gap, so switching in and out never clicks or drops audio. In steady state, bypassed output is bit-exact to the input - zero processing, no residual coloration. The TrueRail chassis coloration is only present while the plugin is enabled.

Polarity: non-inverting at all frequencies.

Latency: sample-accurate DAW compensation.

Parameters: All automation-compatible except the oversampling selector - a latency-changing preference is deliberately kept out of automation's reach. The window size is a per-instance view state outside the parameter system entirely, so resizing mid-session never leaves automation footprints.

Authorization: one-time web visit, serial + C/R, 2 seats per serial.

Demo mode: 2 sec silence in approximately 2 minute intervals, synchronized across unactivated instances.

Signal Verification

Every copy is built from the same codebase that passes automated signal measurements on every release: full signal chain integrity (includes envelope-preservation, phase-integrity, and real-music material integrity tests), transformer verification (four iron cores: M1166 / C9049 / K1166 / L1544), module isolation, parameter sweep verification, oversampling and anti-aliasing, component-model verification with edge-case robustness tests, configuration audit, noise-floor regression gate (verifies every component model's noise source is live across personality × buffer-size combinations), transition-click verification (every enable, selector, preset, and bypass transition gated for click-free behavior), HF alias regression (per-element alias floor at 15 kHz / -6 dBFS probe), Plugin-Path alias regression (the same probe run through the actual processBlock at every oversampling factor, with a separate extreme-settings case), and per-instance variation (Hammerstein rank-stability and THD spread across 16 seeds × 3 topologies × 2 tiers so every instance lands in the same harmonic-ordering regime). The shipped test reports list the exact per-suite measurement counts for the version installed.

Beyond the synthetic test signals (sine, multi-tone, noise), the battery also runs a real-music level section: a calibrated commercial master is pushed through the actual welcome preset and metered the way a DAW meter does. This is the authoritative loudness check - mono and synthetic tones can mislead on level (a mono tone collapses the stereo side and hides the chain's true behavior), so the design's unity-loudness invariant is confirmed against real, bass-heavy, stereo program material alongside the synthetic measurements.

Signal Analysis

Sphinx is built to be measured. If you open it in an analyzer such as Plugin Doctor and inspect its harmonic distortion, Hammerstein kernels, or phase response, this section explains exactly what you might see and why - with the detail of a hardware technical sheet. None of the behaviors below are flaws; each is the physical consequence of modeling real console circuitry at the component level.

The three console personalities model three classes of legendary large-format console hardware. We use original names - SLL, Nevy, Amok - out of respect for the trademark holders, but the topologies and the measured harmonic signatures below identify the sonic lineage to any engineer who knows the genre:

• SLL - the transparent, solid-state British console class (push-pull BJT drive).
• Nevy - the transformer-coupled, discrete Class-A British console class.
• Amok - the valve (vacuum-tube) console class (single-ended Class-A triode).

18.1 Total Harmonic Distortion (THD)

The three personalities produce distinctly different, deliberate THD - this is musical harmonic content from the modeled circuitry, not digital distortion:

For exact current-release values, see the shipped Signal Chain Verification and Component Verification reports. The relative ordering (SLL < Nevy < Amok by orders of magnitude) is the architectural fact; the absolute percentages drift slightly between releases as the analog models are refined.

The harmonic profile is also visible in the Hammerstein-kernel display, where each topology shows a distinct fingerprint: SLL - odd-order kernel content is prominent, characteristic of push-pull-style BJT topology where circuit symmetry cancels even-order nonlinearity. Nevy - even-order kernel content is enriched, characteristic of single-ended Class-A operation. Amok - even-order kernel content is the most prominent, characteristic of the 5670 triode's asymmetric transfer curve. For exact current-release harmonic levels, see the shipped Signal Chain Verification report; the relative THD ordering and architectural character described here are stable across releases, while absolute levels and individual harmonic amplitudes may shift as the analog models are refined.

Plus, the ThermalDrift and TrueRail rail-sag models modulate circuit parameters at sub-audio rates, which an FFT analyzer with finite frequency resolution renders as a slight broadband smearing around each harmonic spike.

The transformer cores' low-frequency resonance peaks (e.g. the M1166 resonance around 19 Hz on the SLL personality) apply equally to signal and to the upstream noise floor; this is the same behavior real iron-core transformers exhibit and is part of the per-core sound. If you are evaluating transformer character at high TX DRIVE and high monitoring gain and the bass region floor sounds elevated, that is the LF resonance amplifying the upstream device noise floor through the modeled resonance - try the L1544 core (flat, no peak) for a quieter floor in that band.

What to look for. Clean harmonic spikes at exact integer multiples of the test frequency, rising above a smooth floor.

18.2 Hammerstein Kernels

A Hammerstein decomposition separates the nonlinearity into frequency-dependent polynomial kernels - G1 is the linear (fundamental) response, G2 the 2nd-order term, G3 the 3rd-order, and so on. Each personality has a distinct fingerprint:

• SLL: Odd-harmonic-dominant fingerprint. The push-pull-style BJT topology's circuit symmetry suppresses even-order kernel content in the kernel decomposition, characteristic of precision solid-state design.
• Nevy: Enriched even-order fingerprint. The Class-A topology's single-ended operation lifts even-order kernels closer to the odd-order levels - the warm, dimensional character of transformer-coupled discrete amplification.
• Amok: Even-harmonic-dominant fingerprint. The 5670 triode's asymmetric transfer curve produces strong second-order kernel content - the deep, rich character of tube amplification working on the knee of its transfer curve.

For current-release per-personality Hammerstein levels and audible-spectrum harmonic amplitudes, see the shipped Signal Chain Verification report.

Note on kernels vs. spectrum. Hammerstein kernel levels show the strength of each nonlinear MECHANISM in the chain - they describe the structure of the nonlinearity, not the audible harmonic amplitudes directly. When you measure the output with a normal FFT spectrum analyser at a single test tone, the visible harmonic amplitudes combine contributions from multiple kernels via Chebyshev expansion of the input waveform. A higher-order kernel being strong doesn't mean the corresponding harmonic line is the loudest in the spectrum - most of its energy redistributes into the lower-order odd harmonics. Sphinx's input transformer's Jiles-Atherton hysteresis adds memory that produces particularly strong higher-order kernels; the audible harmonic spectrum reflects each topology's circuit physics rather than the raw kernel amplitudes.

Low-frequency behavior. Below ~50-200 Hz the higher-order kernels roll off, and on the Amok personality they show a characteristic oscillation. This is the interaction of the coupling capacitors with the valve stage's plate-decoupling and cathode-bypass networks at sub-audio frequencies - the same behavior measurable in real tube equipment with iron coupling, and absent from the solid-state SLL and transformer-coupled Nevy paths.

High-frequency behavior. Above ~5-10 kHz the kernels converge into the noise floor. This is physically correct: real transformers and amplifiers are bandwidth-limited, which naturally suppresses high-frequency harmonic generation.

18.3 Phase Response

Sphinx's phase response is dominated by its transformer models. The chain's global oversampling uses a minimum-phase polyphase IIR half-band - it contributes a small, monotonic high-frequency phase rotation that rises with frequency in the same direction the transformer already rotates. The phase contribution from the OS is well inside the cumulative analog-chain budget the modeled transformer produces, and it sits at the opposite end of the band from the dominant LF rotation. The DriveStage (main circuit) contributes effectively zero phase rotation in the audio band. The transformer contributes approximately one full rotation ("wrap") from its reactive elements: the coupling capacitor, winding inductance and resistance, and shunt capacitance that form the transformer's linear shell model. With the transformer enabled, the full chain shows roughly one wrap - monotonic downward rotation with no direction reversals.

Because the phase rotation comes primarily from the transformer rather than the DriveStage, the three main circuits (SLL, Nevy, Amok) produce very similar phase responses. Subtle personality-specific differences appear at low frequencies from each circuit's bias point and gain structure interacting slightly differently with the transformer's reactive load.

Even with modest phase rotation, the frequency-dependent phase shifts from the transformer models reshape the waveform of complex program material. On music this can raise instantaneous peak levels by a few dB while RMS (loudness) stays at unity - exactly what happens when you run audio through a real analog console with iron transformers. It is not a gain error, a DAW peak meter will read higher, but a loudness (LUFS) meter will read the same as the input. The peak rise is the physical consequence of frequency-dependent phase rotation on a complex waveform.

18.4 Oversampling and Anti-Aliasing

The entire signal chain runs at the user-selected oversampling rate (2× / 4× / 8× / 16×). One minimum-phase polyphase IIR half-band at the input upsamples the audio; one matching half-band at the output downsamples back. Anti-aliasing is uniform across the chain because the same filter pair handles every module's harmonic content.

Two test reports document the measured fold-back:

• HF Alias Regression (per-element, isolated-module probe): each nonlinear element is driven in isolation by a 15 kHz / -6 dBFS probe and its alias level is pinned as a per-item regression baseline with +3 dB headroom - a future code change cannot worsen any element's alias level by more than 3 dB without failing the suite.
• Plugin-Path Alias Regression (full-chain probe): the same 15 kHz / -6 dBFS probe is run through the actual plugin processBlock at every oversampling factor (2× / 4× / 8× / 16×) plus a separate extreme-settings case. Per-OS-rate regression gates catch any regression in the chain's full fold-back behavior.

• At default 2× oversampling, the full-chain fold-back products sit comfortably below the music. The named alias bins (3 / 6 / 9 / 12 / 18 / 21 kHz, the canonical harmonics of a 15 kHz tone folded back into the audio band at 48 kHz session rate) drop substantially compared to running base-rate. At 2×, the chain still has the per-module nonlinearities anti-aliased, but the single cascaded half-band has less HF stop-band suppression than the deeper cascades - fold-back products are noticeably higher than at 4× and above.
• At 4×, 8× and 16×, fold-back drops further and lands in the noise-floor region. Use 4× or higher when pushing the saturation stages.

Under typical mixing content the music itself fills the spectrum and most harmonic spurs are masked into the modeled analog noise floor - the device-level noise the NOISE toggle gates (the Width section's tube plate shot when engaged, plus the console channel-bleed floor neighboring instances contribute), per-topology PSRR coupling, and the sub-audio parameter modulation from ThermalDrift and TrueRail rail sag. Each source carries a per-topology spectral shape - silicon paths sit close to flat, tube paths tilt toward the low end - at unchanged total noise power. Higher oversampling factors reduce the alias contribution further; the analog floor itself is largely independent of the OS factor because it's the modeled analog character, not the digital conversion.

FAQ

Does the plugin have a guide inside the DAW? Yes, click on the book icon on the lower section of Sphinx GUI to open a popup guide.

Transfer license? - Deauthorize old machine, activate new.

Can't access old machine? - Email support.

Lost response code? - Repeat the activation or deactivation steps - it's re-issued.

Internet required? - One-time web activation only, from any computer. Plugin does not call home, ever, because everyone deserves privacy.

Server down? - Existing installs unaffected.

Mac + Windows? - Both for AU and VST3.

96/192 kHz? - Verified identical character.

CPU? - Around the usual industry top-dogs.

Multiple instances? - Yes, each gets unique component tolerances. Multiple instances never null due to their analog, unpredictable nature.

Troubleshooting

Plugin not found: - rescan plugins.

Clicks on DAW transport location changes: - increase buffer.

High CPU: - drop oversampling. 2× is the default and the lightest setting. Higher OS factors process more samples per block and use proportionally more CPU.

Sounds congested: reduce input 1-3 dB.

Challenge/Response rejected: - copy exactly, it's case-sensitive. Or email support with your serial and email.

Max seats reached: - deauthorize old machine or email support.

Air-gapped studio: - you can use any internet-connected computer with the web activation page to get the codes.

Credits and Legal

Sphinx Channel developed by Analog Realism. Component models based on published scientific equations: Koren (1996), Jiles-Atherton (1986), Ebers-Moll (1954). Made in Finland. (C) 2026 Analog Realism. All rights reserved.