Explainers · 2026-07-11 · Patreon guide

Patreon for electronic music creators: VCO waveforms and Fourier series, Moog transistor ladder VCF resonance and self-oscillation, ADSR RC charge dynamics, Eurorack CV/Gate and 1V/oct standard, FM synthesis Chowning algorithm and Bessel function sidebands, wavetable aliasing, FDN reverb Schroeder allpass diffusion, and the Apple Tax

Electronic music and synthesizer Patreons retain when the creator delivers the physics layer that YouTube tutorials always skip: not “filter the cutoff down for that classic sound” but why four cascaded RC stages at −6 dB/octave each produce −180° of total phase shift and how that phase shift causes a transistor ladder to self-oscillate at exactly the cutoff frequency when loop gain hits 1.0. The patron who understands the expo converter thermal voltage equation, the Bessel function amplitude distribution across FM sidebands, and the Kappa-sigma analog of Winsorized Sigma in Schroeder comb-filter T60 math does not find that depth anywhere else; canceling the subscription means losing the technical layer they came for.

VCO waveforms and Fourier series: the harmonic mathematics of oscillator cores

Sawtooth, square, triangle, and sine: Fourier decomposition

Every VCO waveform is a sum of sine waves at integer multiples of the fundamental frequency — the Fourier series. The specific harmonic content is what gives each waveform its timbre before any filtering occurs. Sawtooth wave: contains all harmonics (fundamental, 2nd, 3rd, 4th, …) with amplitudes that decrease as 1/n, where n is the harmonic number. The Fourier series is x(t) = (2A/π) × ∑ ((−1)ⁿ⁺¹/n) × sin(2πnf₀t) for n = 1, 2, 3, …; at f₀ = 220 Hz (A3), the 2nd harmonic at 440 Hz has amplitude 1/2 relative to the fundamental, the 3rd at 660 Hz has 1/3, and the 20th at 4.4 kHz has 1/20. The 1/n rolloff is 6 dB per octave (halving amplitude per doubling of frequency), which makes the sawtooth spectrally rich up to the Nyquist limit. Square wave (50% duty cycle): contains only odd harmonics — fundamental, 3rd, 5th, 7th — at amplitudes 1/n. x(t) = (4A/π) × ∑ (1/n) × sin(2πnf₀t) for odd n only. The absence of even harmonics gives the square wave its hollow, reedy timbre; it is the Fourier series of a signal that is symmetric about its midpoint. Triangle wave: also only odd harmonics, but amplitudes fall off as 1/n² instead of 1/n. x(t) = (8A/π²) × ∑ ((−1)^(n−1)/2/n²) × sin(2πnf₀t) for odd n. The 1/n² rolloff means the 3rd harmonic is at only 1/9 of the fundamental amplitude, and harmonics above the 7th are largely negligible — the triangle sits midway between a square and a sine in spectral content. Sine wave: only the fundamental frequency, no harmonics at all; it is already a pure single-frequency sinusoid and contains no additional Fourier components.

1V/oct expo converter: thermal voltage and temperature drift

The 1V/oct pitch standard requires that each 1V increase in control voltage (CV) doubles the VCO frequency — one octave. Frequency doubling is a multiplicative relationship, which means the voltage-to-frequency conversion must be exponential: f = f₀ × 2^V or equivalently f = f₀ × e^V × ln2. The expo converter circuit exploits the exponential current-voltage relationship of a bipolar junction transistor (BJT): I_C = I_S × e^V_BE/V_T, where I_S is the reverse saturation current, V_BE is the base-emitter voltage, and V_T = kT/q is the thermal voltage. At room temperature (T = 300 K): V_T = kT/q = (1.381×10⁻²³ × 300) / 1.602×10⁻¹⁹ ≈ 25.85 mV ≈ 26 mV. The problem: V_T changes with temperature (V_T is directly proportional to absolute temperature T), which means the expo conversion gain drifts as the room temperature changes, causing the oscillator to go out of tune. Classic solution: matched transistor pairs thermally bonded together so both halves of the differential pair track the same temperature coefficient; the LM394 “super-matched” transistor array and the SSM2210 were designed specifically for this application. A linearizing resistor in the emitter circuit compensates V_T drift by scaling with temperature in a complementary manner. Even with matching, trimming the “temperature compensation trimmer” is required during calibration: the VCO is cycled between low and high pitch while the room is warmed or cooled (or a heat gun used deliberately on the chip), and the trimmer is adjusted until pitch is stable across temperature — the patron-level setup that no YouTube patch tutorial explains.

Sawtooth core: Miller integrator with comparator reset

The most common analog VCO core for a sawtooth wave is the Miller integrator with reset. A current source (exponentially controlled by the expo converter) charges a capacitor; the capacitor voltage rises linearly with time because V(t) = I×t/C (a constant current on a capacitor produces a linear ramp). When the ramp reaches a threshold voltage set by a comparator, the comparator fires and closes a switch (MOSFET or BJT) across the capacitor, discharging it rapidly back to zero; the cycle repeats. The output taken from the capacitor is a sawtooth: a linear ramp with a sharp reset. The frequency is f = I/(C×V_threshold): doubling the charging current I doubles the frequency, which is why the exponentially-controlled current source from the expo converter produces the 1V/oct response. Frequency range: to cover 10 octaves (20 Hz to 20 kHz), the charging current must span a 1,000:1 ratio, which is e^10×ln2 ≈ 1,024; the expo converter must track this accurately over temperature. Triangle wave cores use a different approach: charge the capacitor upward until an upper threshold is reached, then switch the current source polarity to discharge until a lower threshold, alternating to produce a triangle; a waveshaper differential pair can convert the triangle to an approximate sine wave by soft-limiting the peaks of the triangle through the curved transfer characteristic of a bipolar transistor differential pair.

Pulse width modulation and formant-like timbres

A square wave with duty cycle D (the fraction of the period spent in the high state) has a Fourier series that changes character dramatically as D varies from 50%. At D = 50%, only odd harmonics are present. At D = 25%, both even and odd harmonics appear, but the 4th, 8th, 12th harmonics (multiples of 1/D = 4) are cancelled while the 2nd and 6th are restored. The general rule: harmonics at multiples of 1/D are cancelled; the resulting spectral notch creates a formant-like peak in the perceived brightness that rises in frequency as the pulse width narrows. Slowly modulating the duty cycle with an LFO (pulse width modulation, PWM) causes this formant to sweep up and down, producing the characteristic chorus-like beating of PWM pads. The beating occurs because the harmonic content is physically changing in real time; it is not a comb filter artifact but a direct consequence of changing which harmonics are cancelled by the zero-crossing in the waveform.

Moog transistor ladder VCF: pole-zero analysis, resonance, and self-oscillation

Four cascaded RC poles and the −24 dB/octave Barkhausen condition

The Moog ladder filter, patented in 1966 by Robert Moog, consists of four identical one-pole RC lowpass filter sections cascaded in series, with a voltage-controlled feedback path. Each RC section is a first-order lowpass filter with transfer function H(s) = ω_c / (s + ω_c), where ω_c = 1/(RC) is the angular cutoff frequency and s is the complex frequency variable. In the frequency domain, each section contributes −6 dB/octave above ω_c and a phase shift of −arctan(ω/ω_c) that ranges from 0° at DC to −90° at very high frequencies, passing through −45° at exactly the cutoff frequency. Cascading four such sections multiplies their transfer functions, producing a total rolloff of −24 dB/octave and a total phase shift that is four times the single-section phase: total φ = 4 × (−arctan(ω/ω_c)). At the cutoff frequency exactly, total phase shift = 4 × (−45°) = −180°. The cutoff frequency of each stage: f_c = 1/(2π × RC). In the Moog implementation, the R and C of each stage are realized by biased transistors operating in their linear region, making the effective R (and therefore f_c) voltage-controllable; the four stages are biased together from the same exponential current source so they all track the same f_c.

Feedback path and the Barkhausen self-oscillation criterion

A controlled fraction of the filter output is fed back to the input. Because the four-pole ladder has already produced −180° of phase shift at f_c, and the feedback path inversion adds another −180°, the total loop phase shift is −360° (or equivalently 0°) at f_c — this is positive feedback at that exact frequency. The Barkhausen criterion for sustained oscillation requires two conditions simultaneously: (1) total loop phase shift = 0° (or integer multiple of 360°), and (2) total loop gain ≥ 1.0. At low resonance settings, the loop gain at f_c is less than 1.0: the filter resonates (Q > 1) but decays. At the resonance control setting where loop gain = 1.0 at f_c, the filter sustains a continuous sine wave at f_c without any input signal — self-oscillation. The Q factor increases monotonically with the feedback gain; in typical ladder implementations, the feedback gain is controlled by a current set by the Resonance (or Emphasis) control voltage. Self-oscillation pitch tracks 1V/oct if the filter’s cutoff frequency control is well-calibrated, making the self-oscillating ladder a voltage-controlled sine wave oscillator that is spectrally pure at low signal levels and introduces pleasurable soft clipping (waveform flattening) at higher amplitudes due to the transistors entering saturation. The distinction from state-variable and Sallen-Key topologies: those designs achieve resonance via different feedback configurations and generally cannot achieve the specific phase behavior that produces Barkhausen self-oscillation at a clean, trackable pitch.

OTA filters and the CEM/SSM chip alternatives

The Curtis CEM 3320, CEM 3379, and SSI 2144 (and their predecessors CEM 3340 VCO, CEM 3360 VCA) implement voltage-controlled filters using Operational Transconductance Amplifiers (OTAs): a current-mode differential amplifier whose transconductance g_m (output current per input voltage, in siemens or amps/volt) is proportional to a bias current I_ABC. Setting I_ABC with an exponentially-controlled current source makes g_m track 1V/oct. The effective resistance for each filter pole R_eff = 1/g_m; f_c = g_m/(2πC). OTA filters achieve the same voltage-controlled cutoff as the discrete ladder but in a single integrated package, at the cost of a different character: OTA filters (Oberheim OB-Xa, Prophet-5 Rev 3.3) have a different self-oscillation and distortion character than the discrete Moog ladder, a distinction audiologically detectable at high Q and signal levels and the subject of much patron-level synthesis discussion.

ADSR envelope generator: RC charge dynamics, gate signals, and exponential perception

RC charge equations for Attack, Decay, and Release

The ADSR (Attack, Decay, Sustain, Release) envelope generator produces a time-varying control voltage that shapes the amplitude (and often the filter cutoff) of a synthesizer voice. In analog implementations, the core is a capacitor charged and discharged through resistors. Attack phase: the gate signal transitions high (0 V → 5 V); a switch connects the capacitor to a supply voltage V_s through the attack resistor R_A. The capacitor charges according to V(t) = V_peak × (1 − e^−t/τ_A), where τ_A = R_A×C is the attack time constant. The voltage asymptotically approaches V_peak but reaches 63.2% of V_peak in one time constant τ_A, 86.5% in 2τ_A, and 95% in 3τ_A. Decay phase: once the capacitor reaches V_peak, the charging target switches to the sustain voltage V_sustain (set by the Sustain potentiometer); the capacitor now decays from V_peak toward V_sustain through R_D: V(t) = V_sustain + (V_peak − V_sustain) × e^−t/τ_D. Sustain phase: the gate signal remains high; the envelope holds at V_sustain as long as the gate is held, because the capacitor has discharged to V_sustain and the circuit is in equilibrium. Release phase: when the gate goes low (5 V → 0 V), the sustain voltage target switches to 0 V and the capacitor discharges through R_R: V(t) = V_sustain × e^−t/τ_R. Time constant τ = RC for each stage; to cover the musically useful range of approximately 1 ms to 30 seconds, the time constants must span a factor of 30,000, which requires either switchable capacitor banks, logarithmic potentiometers with very large dynamic range, or voltage-controlled time constant circuits.

Exponential vs linear perception and gate signal types

Hardware RC envelopes are inherently exponential in shape. Perceptually, exponential envelopes feel more “natural” and musical than linear ones because human loudness perception follows a logarithmic scale (the decibel): a voltage that decays exponentially corresponds to a decay that sounds linear in perceived loudness. This is why the natural RC curve, without linearization, sounds musical. Digital ADSR implementations that produce linear voltage ramps (voltage changing at a constant rate in volts per second) produce envelopes that sound perceptually fast at high amplitudes and slow at low amplitudes — the opposite of what the ear expects. Some digital synthesizers explicitly offer exponential, linear, and logarithmic envelope curves for precisely this reason. The gate signal is a 5 V logic-level signal from a keyboard, MIDI-to-CV converter, or sequencer: the gate goes high when a note is triggered and stays high for the note duration (key-held time), then goes low when the note is released. The gate duration = sustain phase duration. A trigger is a short pulse (typically 5 ms) regardless of note duration; trigger-sensitive envelope generators re-trigger with each trigger pulse (multi-trigger mode), while gate-sensitive designs hold the sustain phase for the gate duration. Single-trigger mode: the envelope only re-triggers if the gate has gone fully low first; used when legato (overlapping) playing should not restart the attack. Multi-trigger (repeat) mode: any trigger pulse restarts the attack from wherever the envelope currently is.

Eurorack modular format: CV/Gate standards, power, and physical dimensions

Physical format: 3U, HP, power connector

Eurorack is the dominant modular synthesizer physical format, introduced by Doepfer in 1995 and now comprising thousands of module designs from hundreds of manufacturers. Physical dimensions: 3U format where U (rack unit) = 1.75” = 44.45 mm; 3U = 133.35 mm panel height (the height of a Eurorack module front panel). Module width is measured in HP (horizontal pitch): 1 HP = 5.08 mm (0.2”); typical module widths range from 2 HP (small utilities: attenuators, mults) to 42 HP (complex oscillators, large sequencers). A standard 84 HP rack holds approximately 84 × 5.08 mm = 426.7 mm of module width per row. Power: 16-pin IDC (Insulation-Displacement Connector) ribbon cable, with a 10-pin keyed version also common; the bus board carries ±12 V and +5 V regulated DC rails. The −12 V rail is required for nearly all analog signal processing (op-amps, OTAs, and VCAs need a negative supply). Module current consumption is specified in milliamps (mA) on +12 V, −12 V, and +5 V separately; a complex oscillator might draw 80 mA on +12 V and 60 mA on −12 V; a power supply is sized by summing all module currents per rail with a 20–30% headroom margin. Patch cables: 3.5 mm (1/8”) mono TS (Tip-Sleeve) jacks throughout; sleeve = signal ground, tip = signal. All jacks are mono; stereo patch requires two jacks.

CV and gate standards: 1V/oct, ±5V audio, 5V gate

Control Voltage (CV) conventions in Eurorack: 1V/oct pitch CV is the standard for pitch control of VCOs and filter tracking inputs. Each 1 V increase in CV raises the pitch by exactly one octave, doubling the frequency; 10 V range covers 10 octaves (approximately 27.5 Hz to 28.2 kHz). The 1V/oct standard is hardware-enforced by the expo converter in each module; calibration to 1V/oct requires trimming the expo converter so that the octave interval is exact across temperature. Gate signal: 5 V high / 0 V low logic; gate high = note held; gate duration = note duration in sequencer. Trigger: a short 1–5 ms pulse at 5 V, used for clock signals, percussion triggers, and envelope triggers where sustain phase is not meaningful; clock dividers and multipliers operate on trigger streams. Audio signal level: ±5 V peak-to-peak is standard for audio in Eurorack (10 V_pp); some modules output ±10 V (20 V_pp); mismatched levels require attenuation (attenuverters). LFO CV typically ranges ±5 V or 0–10 V depending on the module. MIDI-to-CV converters (Expert Sleepers ES-3, Intellijel MIDI, Befaco MIDI Thing) accept DIN-5 MIDI or USB MIDI and output per-note pitch CV (1V/oct), gate, and optionally velocity (0–10 V corresponding to 0–127 MIDI velocity) and aftertouch. The MIDI clock standard (24 pulses per quarter note, PPQN) is converted to trigger streams for analog sequencer sync.

FM synthesis: the Chowning algorithm and Bessel function sideband amplitudes

The FM equation and modulation index

FM synthesis was formalized by John Chowning at Stanford in 1973 and licensed to Yamaha, resulting in the DX7 (1983), the best-selling synthesizer in history. The core FM equation: y(t) = A × sin(2π×f_c×t + β × sin(2π×f_m×t)), where A is amplitude, f_c is the carrier frequency (the fundamental pitch heard in the output), f_m is the modulator frequency (the oscillator doing the modulating), and β is the modulation index. The modulation index β = Δf / f_m, where Δf is the peak frequency deviation — the maximum amount by which the carrier’s instantaneous frequency is displaced from f_c. When β = 0, the carrier is unmodulated: the output is a pure sine at f_c. As β increases from 0, sidebands at frequencies f_c ± n × f_m (for n = 1, 2, 3, …) appear and grow in amplitude. The DX7 has 6 operators (each a sine oscillator with an individually controllable ADSR envelope scaling its output amplitude) and 32 algorithms (hardwired routing configurations) specifying which operators are carriers (heard in the output) and which are modulators (modifying other operators); in some algorithms all 6 operators form a series FM chain (operator 1 modulates operator 2 which modulates … operator 6 which is the carrier); in others, three operators in parallel each modulate their own carrier, producing three simultaneous FM voices per note.

Bessel functions, sideband folding, and C:M ratios

The Fourier expansion of the FM equation using the Jacobi-Anger identity yields: y(t) = A × ∑_n=−∞^+∞ J_n(β) × sin(2π×(f_c + n×f_m)×t), where J_n(β) is the Bessel function of the first kind of order n evaluated at β. The Bessel functions J_n(β) have several key properties: J₀(0) = 1 and J_n(0) = 0 for n ≠ 0 (only carrier at β = 0); J_n(β) = (−1)ⁿ J_−n(β) (symmetry); for fixed β, J_n(β) decreases rapidly for n ≫ β (only sidebands with n ≤ β + 2 or 3 have significant amplitude, which gives the rule-of-thumb bandwidth estimate: FM bandwidth ≈ 2×(1 + β)×f_m, Carson’s rule). Worked example: β = 2; J₀(2) ≈ 0.224 (carrier), J₁(2) ≈ 0.577 (1st sideband pair), J₂(2) ≈ 0.353 (2nd pair), J₃(2) ≈ 0.129 (3rd pair), J₄(2) ≈ 0.034 (4th pair, negligible). Negative sidebands: for sideband index n where f_c − n×f_m < 0 (a negative frequency), the component folds back to the positive frequency axis with a 180° phase inversion, adding to (or partially cancelling) the positive sideband at f_m−f_c; this folding explains why FM spectra with large β are asymmetric and why the carrier is not at the center of the sideband power distribution. The C:M ratio (f_c/f_m): integer ratios (1:1, 1:2, 2:1, 3:2) place all sidebands at integer multiples of a common fundamental, producing harmonic spectra with a recognizable pitch. Non-integer ratios (1:1.41, 2:2.786 for bell-like inharmonicity) scatter sidebands at non-harmonic locations, producing metallic, bell-like, or noise-like textures that are the signature FM timbre of DX7 electric pianos and bells. Operator feedback: in DX7 algorithms, any operator can self-modulate (its output is added to its own phase input with a scalable feedback coefficient); at low feedback, this adds subtle odd-harmonic content; at maximum feedback, the operator output approximates a square wave (odd harmonics only), a consequence of the FM equation approaching a pulse-position modulation waveform at high self-modulation depth.

Wavetable synthesis and sampling: Nyquist theorem, bit depth, aliasing, and interpolation

Sampling theory and quantization noise

The Nyquist-Shannon sampling theorem states that a signal can be perfectly reconstructed from its samples if and only if the sampling rate f_s ≥ 2 × f_max, where f_max is the highest frequency in the signal. At f_s = 44,100 Hz (the CD standard, chosen to exceed 2 × 20,000 Hz = 40,000 Hz with margin for anti-aliasing filter rolloff), frequencies up to 22,050 Hz can theoretically be represented. Professional audio rates: 48,000 Hz (broadcast and production standard), 96,000 Hz (commonly used for synthesis processing to give anti-aliasing filters more headroom), 192,000 Hz (mastering and specialist applications). Bit depth: each sample is quantized to 2^N discrete amplitude levels. The quantization error (the rounding error from digitizing a continuous amplitude) has maximum magnitude ½ LSB (Least Significant Bit); its RMS power in a full-scale signal produces a Signal-to-Noise Ratio (SNR) of approximately SNR = 6.02 × N + 1.76 dB for uniformly distributed quantization noise. At N = 16 bits (CD standard): SNR ≈ 6.02×16 + 1.76 = 97.8 dB. At N = 24 bits (professional DAW): SNR ≈ 6.02×24 + 1.76 = 145.8 dB. At N = 32-bit float (DAW internal bus): the effective bit depth is approximately 24 bits for magnitude representation plus the 8-bit exponent providing headroom without clipping; 32-bit float is more than adequate for internal synthesis processing.

Aliasing, anti-aliasing, and wavetable transposition artifacts

Aliasing occurs when a signal contains frequency content at or above the Nyquist frequency (f_s/2): these components are “folded back” into the audible spectrum at the alias frequency f_alias = |f_s − f_component|. A 30 kHz harmonic in a signal sampled at 44.1 kHz aliases to |44,100 − 30,000| = 14,100 Hz, which is audible and inharmonic — a clanging, digitally harsh artifact. Anti-aliasing filter (AAF): an analog lowpass filter applied before the ADC with cutoff below f_s/2, removing components that would alias; the sharpness required (the “brick-wall” ideal) demands high filter order; Chebyshev and elliptic filter types are common because they achieve sharp rolloff at the cost of passband ripple or stopband ripple. Wavetable synthesis: a single cycle of a waveform is stored as a digital array (the wavetable); playback reads through the table at a rate that determines the output pitch. The problem: when a wavetable recorded at a base pitch is transposed upward (played back faster), the high harmonics in the waveform approach and exceed the Nyquist limit of the output sample rate, causing audible aliasing — the characteristic metallic shimmer of early wavetable synthesizers (PPG Wave, Waldorf Microwave). The solution used in modern wavetable synths: store multiple versions of the wavetable, one per octave zone, where each zone’s version has harmonics band-limited to stay below Nyquist for that octave range; the synthesizer selects the appropriate band-limited version as the pitch is changed (PPG Wave 2.3 and Waldorf instruments from the mid-1990s onward). Oversampling: running the synthesis engine internally at 4× or 8× the output sample rate (e.g., 176.4 kHz or 352.8 kHz internally for 44.1 kHz output), computing all oscillators and filters at the oversampled rate, then applying a high-quality decimation lowpass filter and downsampling to the output rate; this pushes all potential alias products above the oversampled Nyquist frequency, which the decimation filter removes before they can fold into the output spectrum.

Interpolation methods: nearest-neighbor, linear, cubic Hermite, and sinc

Reading a wavetable at a non-integer table index (which occurs whenever the playback pitch is not a simple integer multiple of the base pitch) requires interpolation between stored sample values. Nearest-neighbor: returns the value of the table entry nearest to the read position; introduces a staircase distortion that adds high-frequency harmonics and is audibly coarse at low playback rates. Linear interpolation: weighted average of the two adjacent table entries on either side of the read position; eliminates the staircase artifact but introduces a mild high-frequency rolloff (the interpolated signal is convolved with a triangular function in the time domain, which is a sinc² function in the frequency domain, attenuating high frequencies). Cubic Hermite interpolation: fits a cubic polynomial through four surrounding table entries, matching both value and slope at the boundaries; significantly better high-frequency accuracy than linear, at roughly 4× the computational cost; used in most professional software synthesizers. Sinc interpolation (ideal reconstruction): samples the ideal reconstruction formula y(t) = ∑_n y[n] × sinc((t − n) / T), where sinc(x) = sin(πx)/(πx) and T is the sample period; this is the mathematically perfect reconstruction of a band-limited signal from its samples, as proven by the sampling theorem. In practice, sinc interpolation must be windowed (truncated to a finite number of surrounding sample points, typically 16–64) and windowed with a Kaiser or Blackman-Harris window to suppress the Gibbs phenomenon; this windowed-sinc or “polyphase FIR” interpolation is the highest-quality resampling method and is used in professional DAWs and Ensoniq/E-mu and later sample-based instruments.

Reverb and effects: FDN algorithm, Schroeder allpass, and diffusion

Schroeder comb filter and allpass reverb (1962)

Manfred Schroeder’s 1962 paper “Natural Sounding Artificial Reverberation” introduced the fundamental building blocks of algorithmic reverb still in use in digital effects today. Comb filter: a delay line of N samples with feedback gain g. Transfer function: H(z) = 1 / (1 − g×z^−N). Time-domain difference equation: y[n] = x[n−N] + g × y[n−N]. The delay N (chosen as a prime number of samples to avoid correlated reflections) sets the echo period; the feedback gain g (0 < g < 1) determines the decay rate. The comb filter name comes from its frequency response: peaks (resonances) at integer multiples of f_s/N with the peak magnitudes determined by g; the frequency response resembles the teeth of a comb. Multiple comb filters in parallel, summed, provide the dense multi-echo tail of a reverb. T60 decay time (the time for reverb to decay by 60 dB): T60 = −3 × N / (f_s × log₁₀(g)) seconds; at f_s = 44,100 Hz, N = 2,250 samples (51 ms delay), g = 0.84: T60 = −3 × 2250 / (44,100 × log₁₀(0.84)) = −6,750 / (44,100 × (−0.0757)) ≈ 2.02 seconds. Allpass filter: combines a comb filter with a feedforward path so that the magnitude of the frequency response is flat (|H(e^jω)| = 1 for all ω) while the phase response varies; this diffuses reflections without coloring the spectrum. Schroeder’s original reverb: four parallel comb filters summed → two series allpass stages.

FDN (Feedback Delay Network) and the Jot algorithm

The Feedback Delay Network (FDN), formalized by Jean-Marc Jot (1991), generalizes the Schroeder design into a more flexible architecture. An N-channel FDN consists of N delay lines in parallel, each of length d_i samples (typically N = 8 or 16 delay lines, lengths chosen as mutually prime numbers to maximize mode density); the outputs of all delay lines are fed through an N×N feedback matrix A before being summed back into the delay line inputs. The feedback matrix A determines the mode density and decay character of the reverb. Stability requirement: all eigenvalues of A must satisfy |λ| ≤ 1; an orthogonal matrix (where A^T×A = I) has all eigenvalues on the unit circle |λ| = 1 and is energy-preserving (lossless feedback), meaning decay is provided only by the absorption filters in each delay loop, not by the matrix. Householder matrix: A = I − (2/N) × 1×1^T, where 1 is the column vector of all ones; easily computed with O(N) operations per sample regardless of N. Hadamard matrix: a matrix of ±1 entries with rows mutually orthogonal; also O(N log N) via the Fast Hadamard Transform. Each delay loop contains a frequency-dependent absorption filter (typically a simple one-pole lowpass with fc at 4–10 kHz) that models the air absorption of high frequencies in a real room: high frequencies decay faster than low frequencies, producing the characteristic warmth of natural reverb tails. The T60 target at each frequency band is set by adjusting the absorption filter gain, which can be derived from the target T60 and the delay loop length: g_loop = 10^{(−3×d / (f_s×T60))}. LFO modulation of each delay loop length (small random fluctuation of ±1–5 samples at 0.1–3 Hz) breaks up the metallic ringing and spectral coloring that fixed-length FDNs exhibit; this modulation effectively smears the comb filter resonance peaks over a small frequency range, producing the chorus-like shimmer of high-quality algorithmic reverb.

Convolution reverb and FFT efficiency

Convolution reverb captures the impulse response (IR) of a real acoustic space: a brief impulsive sound (starter pistol, balloon pop, sine sweep) is recorded in the target space; the resulting recording contains the full room’s time-domain response (direct sound, early reflections, diffuse tail). Convolving any audio signal with this IR produces a simulation of the signal played in that room. Direct convolution (time-domain multiplication): computing each output sample requires summing N multiplications for an N-sample IR; computational complexity O(N²) per second of output — practical only for very short IRs (<500 ms at 44.1 kHz). FFT convolution: the convolution theorem states that convolution in the time domain corresponds to multiplication in the frequency domain; compute the FFT of the input signal and the FFT of the IR, multiply pointwise, then IFFT the result. The FFT of length L has complexity O(L log L), making FFT convolution O(N log N) total — feasible for IR lengths up to several seconds. Overlap-add and overlap-save block processing: input audio is divided into blocks; each block’s FFT is multiplied by the pre-computed FFT of the IR (loaded once at initialization); the block IFFT output is overlap-added with the previous block’s tails. This enables real-time convolution reverb with arbitrary-length IRs in software synthesizers and DAW insert effects.

iOS rates and the Apple Tax

Electronic music and synthesizer Patreon creators distribute content across YouTube (tutorials, patch breakdowns, synthesis deep-dives), Instagram (live performance clips, modular rack photos), TikTok (short synthesis demos, patch ideas), and Discord (patron community for patch sharing and real-time feedback). iOS distribution varies by platform: YouTube synthesizer and electronic music tutorial content: 62–72% iOS — somewhat lower than purely aesthetic content because synthesizer enthusiasts who are actively patching or recording use laptops running Ableton, Bitwig, or VCV Rack, but casual discovery of synth content and mobile binge-watching raises the iOS fraction above desktop-primary categories. Instagram modular synth and electronic music: 75–85% iOS — the visual format and mobile browsing patterns of Instagram push iOS use strongly. TikTok electronic music: 78–88% iOS — TikTok is functionally a mobile-first platform.

Frequently asked questions

Related: Patreon for electronic music creators · Patreon for musicians · Patreon for musicians guide · How the Apple Tax works · All explainers