Most people make a category error when they hear "analog delay." They imagine a continuous signal flowing through a smooth circuit, the way reverb springs work or how a tape head reads magnetic particles. The reality of a bucket brigade device is stranger: it samples your signal thousands of times per second, holds each sample as a tiny charge on a capacitor, then passes those charges down a chain, one step per clock tick, like workers handing buckets of water.
It is analog in the sense that each bucket holds a real, continuous voltage. It is digital in the sense that time is divided into discrete slices. It is neither, exactly, and that ambiguity is the source of its character.
The Chip That Passes Buckets
F. Sangster and K. Teer at Philips Research Labs filed the foundational patent for the bucket brigade device in October 1967 (US 3,546,490, granted 1970). Their design was elegantly simple: a series of capacitors connected by MOS transistors acting as switches. A two-phase clock alternately opens and closes those switches. Each tick, every capacitor transfers its charge to its neighbor. The input sample gets loaded at one end; N clock cycles later, it exits at the other.
The delay time follows directly from that arithmetic:
t_delay = N / (2 × f_clock)Where N is the number of stages and f_clock is the clock frequency. The Panasonic MN3005, a 4096-stage chip, produces delays from 20.48 ms at its maximum clock down to 204.8 ms at minimum. The smaller MN3007, with 1024 stages, covers 5.12 to 51.2 ms. A single knob varies the clock frequency continuously between those endpoints.
Philips released the TDA1022 as the first commercial BBD. By the mid-1970s, Panasonic had refined the design significantly. Their MN3xxx series became the dominant family: MN3002, MN3005, MN3007, MN3009, each optimized for different applications.
Why Bandwidth Costs You Something
Sampling creates an unavoidable problem. If you feed a 10 kHz sine wave into a clock running at 15 kHz, you get aliasing. The standard solution is a low-pass filter before the BBD input, steep enough to remove any frequency above half the clock rate.
In practice, the filters used in early BBD designs had their -3 dB point around 3 kHz, a steep rolloff. The result: anything entering the delay line gets bandpass-filtered down to roughly telephone quality. High frequencies that entered your delay signal simply do not come out the other side.
This is not a bug that engineers failed to fix. It is a tradeoff that defines the sound. A 3 kHz-limited delayed signal sits differently in a mix than a full-bandwidth repeat. It recedes. It sounds like something heard through a wall, or across a room.
The effect compounds with each repetition. The original dry signal retains full bandwidth. The first repeat loses everything above 3 kHz. The second repeat, which passes the feedback signal through the chip again, loses another pass of high-frequency content. By the fourth or fifth echo, you have something genuinely warm and blurred. This is not the same as applying a single static lowpass filter with more gain; it is cumulative, and the rate of darkening is tied to the feedback amount and the filter characteristics.
The Clock Leaves a Fingerprint
A second artifact shapes the BBD sound in more subtle ways. When capacitors switch, they inject a small amount of the clock signal into the audio path. You end up with a periodic spike at the clock frequency, plus its harmonics, mixed in with the delay output. This is called clock feedthrough or clock noise.
Designers addressed it with notch filters tuned to the clock frequency. But the clock frequency changes with delay time, since the user is turning a knob that shifts the clock rate. A static notch filter only helps at one clock setting. Most designs ended up accepting a compromise: a fixed filter, carefully chosen resistor networks, and shielded PCB layout to minimize the artifact rather than eliminate it.
The resulting texture, a faint metallic shimmer riding under the repeats, became part of the signature of specific pedals. Players comparing an original Boss DM-2 with an MN3005 chip against a later revision with the MN3205 noted a difference in character. The MN3005 measured quieter (75 dB signal-to-noise vs 67 dB for the MN3205) while the MN3205 had slightly lower distortion (0.8% vs 1%). Many preferred the MN3005 version: its better noise floor outweighed the minor distortion advantage of the replacement.
Companding: Borrowing From Tape Noise Reduction
The noise floor of a BBD chip is significant. Running a dynamic guitar signal straight through generates noticeable hiss in quiet passages. The solution, borrowed directly from tape noise reduction, is a compander circuit: compress the signal before it enters the chip, then expand it after it exits.
The NE570N chip, an integrated compander IC, appears in the Boss DM-2 circuit for this purpose. It operates similarly to how dbx noise reduction works on tape: dynamic range compression increases the effective level going into the noisy medium, and expansion on playback restores dynamics while attenuating the noise floor. The Boss DM-2 (released 1981) used this combination of MN3005 and NE570N to achieve around 20 to 300 milliseconds of delay with usable noise performance.
Electro-Harmonix took a different path with the Memory Man, first released in 1976, which became the first widely available BBD delay pedal aimed at gigging musicians. Its circuit produced the characteristic warmth and mild instability that made it a fixture in post-punk and early ambient recordings.
The Chorus Trick
The clock-rate-equals-delay-time relationship has a second consequence that engineers turned into a feature. If you modulate the clock frequency with a low-frequency oscillator, you get time-varying delay. Short delays with mild modulation produce chorus. Faster modulation at the same depth produces vibrato.
This is exactly what the Roland Juno synthesizers do. The Juno-60 uses MN3009 chips, 256-stage BBDs with a delay range of 0.64 to 12.8 ms, running dual lines in opposing LFO phase. The chorus sits at the center of the Juno sound: the BBD bandwidth limitation and clock modulation add up to something that no static parameter set can fully replicate. The Boss CE-1 Chorus Ensemble from 1976, the first pedal developed by the Boss subsidiary, used the MN3002 chip for a similar effect and became a standard for keyboard players.
What Digital Emulation Has to Model
Modern plugin emulations of BBD delays must capture a specific set of behaviors:
The input lowpass filter response, including how it changes with clock rate. The stage-by-stage signal degradation that makes feedback progressively darker and not just quieter. The clock noise fingerprint, which varies by chip and by clock frequency. The compander behavior under transients, including the slight pumping when dynamics shift. The specific distortion character of the BBD stages themselves, which is not the same as simple saturation.
Plugins like EchoBoy in Memory Man mode and PSP BBDelay attempt to model these characteristics explicitly rather than just applying a static lowpass to digital delay. The more convincing implementations capture the progressive darkening of repeats and something of the clock noise texture.
What they cannot perfectly replicate is that the original hardware operates at the actual intersection of sampling theory and analog signal degradation, a regime where the artifacts emerge from physics, not from a parameter someone tuned to sound good. The original MN3005 chips have been discontinued since the late 1990s. Xvive now manufactures a re-issue, and modern pedals like the Chase Bliss Tonal Recall use two of them to achieve up to 550 ms of delay. The demand says something about what that specific physics produces.
The bucket brigade device was not designed to make music. It was designed to delay a signal cheaply and compactly. The sound it makes while doing that job, filtered, noisy, clock-stamped, progressively degraded, turned out to be exactly what a generation of musicians wanted.