The mid-to-late 1980s marked the beginning of what could undeniably be considered the golden age of digital multi-effects studio processors.
The Yamaha SPX-90 and Lexicon's PCM 70, both released in 1985, successfully adapted algorithms from their flagship products into devices at a price point that placed high-quality reverbs, delays, and modulation effects within reach of musicians and smaller studios. Around the same time, Eventide released what would ultimately become a multi-effects staple, the H3000.
The H3000 was, in its designers' own words, "a multi-effects monster," combining Eventide's standard-setting, pitch-shifting algorithms (now fully diatonic and stereo), along with delays, a comprehensive library of modulation effects, and powerful reverb algorithms—all in a modular, upgradeable package.
Whether in its original yellow-on-black cosmetics, the iconic gray-on-blue, or the later D/SE and "squiggle font" D/SX incarnations, the H3000 is immediately recognizable in the racks of the finest studios and producers around the world. And while vintage units can be had at reasonable cost, rest assured that most of the H3000's algorithms have been ported to software plugins.
This article explores both the under-the-hood technology of the H3000 and the developments at Eventide that ultimately led to its creation, beginning with the truly groundbreaking H910.
Released in 1975, the H910—"H" for "harmonizer," "910," a play on The Beatles song "One After 909"—was the first product in what would eventually become the company's Harmonizer line. Even by this early stage, Eventide had already left a lasting impression on the audio industry with its Omnipressor (a dbx VCA–based RMS compressor) and the 1745 delay, but it is the Harmonizer series for which the company is best known.
It is worth noting that Eventide also had —and still has—its hands in a number of other markets, including profanity delays for broadcast. A profanity delay lives between a broadcast studio and the air waves, allowing broadcast engineers time to catch and "bleep" obscenities before they hit the air—and before the broadcaster is hit with a fine from the FCC.
They also developed a tape counter for the Ampex MM1000 2" tape recorder—essentially an "autolocator," or remote control —which was so successful that Ampex made it an OEM product. Eventide also created moving map displays for aviation and a number of accessories and expansion products for Hewlett Packard computers, including RAM expansion boards priced to undercut HP's OEM offerings and an ethernet card.
Up to this point, analog pitch shift effects were largely limited to an octave up or two octaves down, and results bore little direct harmonic or tonal correlation to the original signal. Octave-up effects pedals relied on full-wave diode rectification to "flip" the negative-going portion of a waveform up, crudely doubling the frequency of the incoming waveform.
Octave-down pedals relied on an op-amp comparator to "convert" the incoming signal into a square wave, whose frequency was then halved and/or quartered using CMOS flip-flops (like the 4013 chips found in the MXR Blue Box and Boss OC-2 pedals).
Earlier pitch-change techniques relied on tape manipulation. The simplest method was to play a tape back at a different speed from which it was recorded, a technique employed to great success by Ross Bagdasarian on his Chipmunks novelty records. An alternative method utilized arrays of two or four tape-playback heads mounted on a cylinder that could rotate with or against tape passing over the heads, resulting in upward or downward pitch shifting.
One such device, the Eltro Information Rate Changer cost a staggering $3,950 USD in 1967 dollars (for reference, an LA-2A cost $395 at the time) and was utilized by The Beach Boys in "She's Goin' Bald." Stanley Kubrick also used it in 2001: A Space Odyssey to "wind down" HAL 9000's voice as Dave deactivates him.
In contrast, Tony Agnello, the H910's creator and one of Eventide's earliest employees, used complicated arrays of off-the-shelf logic components to create a "largely analog device" that was able to shift pitch within a two-octave range. Pitch change on the H910 was expressed as a ratio, where a setting of "2.0" is an octave up, ".5" is an octave down, and "1.0" is no pitch change. To achieve the "micro pitch shift" effect for which the H910 is so well-known, users could shift pitch a few cents up or down by setting the Manual control to hover just above or below 1.0.
Interestingly, the ratio control wasn't entirely stable, and the pitch change ratio might occasionally move up or down by a few hundredths (say, between .99 and 1.01 when set to 1.0), unintentionally creating a more effective doubling effect. Eventide's H910 plugin faithfully recreates the hardware, imperfections and all. A stereo implementation of this effect (derived from a pair of algorithms from the later H3000) is also included as "Style I" and "Style II" on Soundtoys' Microshift plugin.
The H910 could also generate up to 112.5ms of delay digitally, all in the days before dedicated digital signal processing (DSP) chips were widely available. The first two iterations of Eventide's first delay product—the 1745/1745A—employed shift registers. These are logic circuits that pass individual bits within digital audio samples from one register to the next, and the next, and so on (like a digital, bit-level bucket-brigade device) to generate short delay lines. But the H910 used what is now the standard technique of storing samples in RAM to generate delay.
As high-quality A/D converter chips were unavailable at the time, the H910 used custom, discrete A/D conversion developed by Agnello (built on an earlier design by Richard Factor for the 1745 delay) and dbx 303 compander cards. These cards, which allowed the H910 to achieve an impressive >90dB dynamic range, are also a common culprit of non-functioning H910s. (D/A conversion appears, at least, in my own H910, to have been handled by Analog Devices AD7530 10-bit DACs.)
When pitch-shifting with a pedal or on a tape machine, you have likely noticed that changing your delay time will also change the pitch. On the H910, two delay lines were constantly modulated with ramp-wave LFOs. By crossfading between them, the processor could maintain a constantly changed pitch.
Tony Agnello says, "The H910 used simple cross fading between two varying delays that were offset in time, so that when one delay reached its limit (either zero or max delay), the other delay was mid-way through its excursion."
The technique was not without imperfections and limitations—some of which, like the constantly wavering pitch ratio, can be employed to creative effect—but the H910 was readily and eagerly adopted by the audio industry, including heavy-hitters like Brian Eno, Tony Visconti, Frank Zappa, Tony Bongiovi, Kevin Killen, and Flood.
The H949 succeeded the H910 in 1977, with the ALG-3/LU-618 "de-glitch" card as an available option (later models included the card). The ALG-3 card corrected the often-audible pop and click artifacts generated by the H910's wave-splicing pitch shift method.
This intelligent de-glitching system was made possible by the development of the Reticon 5105, an early charge-coupled device-on-a-chip. This was the same type of CCD used in digital imaging, radar applications, and—as it was similar to bucket brigade devices—audio delays.
According to Agnello, "I … designed the H949 using the Reticon [5101] chip as one step in my intelligent splicing method. The other steps remain proprietary."
The 5101 allowed auto-correlation operations to be performed. Auto-correlation, in a nutshell, compares two waveforms for similarity.
"Circa 1972–3 [when the H910 was in development] auto-correlation was not possible," Agnello says. But, with the H949, Agnello was able to use the new technology to look for optimum splice points between the crossfaded delay lines, reducing glitches.
A more in-depth technical breakdown of the technique can be found at the Valhalla DSP website and in Agnello's patent for the process, filed in April 1981.
The H949 also included a flange effect, extended delay times, infinite repeat (of the last 400ms' worth of samples) reverse and random delays, feedback tone-shaping options, and a "micro pitch-shift" function. This µPC function restricts the range of pitch change ratio to about 0.93–1.07, depending on if the device is set to flat or sharp mode.
The H969 followed in 1983. However, according to Eventide founder Richard Factor, "It was a lot more expensive than the H949, and wasn't that much better. Also, we started making it during a time of turmoil during our move to NJ and it probably didn't get promoted as well. Therefore, it wasn't a big seller and there was less interest."
Over the course of a telephone conversation and email correspondence, Agnello emphasized the significance of an oft-overlooked Eventide product, the SP2016.
Released in 1981 (again, before dedicated DSP chips were available), the SP2016 was a signal processor with 20kHz bandwidth and 16-bit conversion. It was truly the audio world's first software-based, multi-effects computing platform.
To be clear, other digital audio processing devices with user-adjustable parameters existed at the time (including the Lexicon 224 and the EMT250), but the SP2016 was designed more as a fully programmable computing platform for digital audio processing tasks.
In addition to running factory algorithms, third parties—including users—could write their own software algorithms in SPUDSystem, a dedicated SP2016 algorithm development environment written in Pascal for HP computers. By using this program, users could burn reusable EPROM memory chips and plug them in to the SP2016's available ROM slots. "Remove Panel to Access Plug-In EPROMS," a small label on the unit's top panel says.
This is why Eventide rightly claim first usage of the term "plug-in." This was absolutely unheard of at the time and remains an innovation largely overlooked by the larger audio engineering culture, though a little bit of internet sleuthing quickly reveals much love for the SP2016 from a who's who of engineering royalty.
SP2016: Under The Hood
The SP2016 employed an array processing architecture using a 16x16-bit multiplier and 32-bit accumulator microprocessor by TRW (whose own history is quite interesting). All digital audio processing occurred at 24 bits, fixed point. Array processors, also known as vector processors, execute a single instruction at a time but on an array (think matrix or table) of data.
For reference, some widely known examples include the PlayStation 3's Cell processor and the IBM PowerPC Altivec co-processor found in G4 and G5 Macintosh computers. Most contemporary graphics processing units (GPUs) and computer CPUs employ some form of vector processing.
The TRW microprocessor was not designed specifically for audio use. It was essentially designed for general-purpose computing applications (array processors, incidentally, dominated the supercomputing industry in the 1980s and 1990s) and optimized in the SP2016 specifically for audio processing applications—it was, truly, an audio computing platform.
This differs from contemporary devices in a few significant ways:
- There was no proprietary silicon used.
- The SP2016 platform's open architecture allowed it to run many types of algorithms—it wasn't limited to any single type like reverbs or delays.
- The SPUDSystem opened the platform to third-party developers and users.
AD/DA conversion was handled by Sony CX20018 and Burr Brown PCM75JG ICs, respectively. Interestingly, audio Input and Output level sliders were not simply analog faders. These sliders carried DC voltages that were sent to Analog Devices AD7524 Multiplexing D-A converters, through which analog audio passed. The 7524 varied the level of any audio passing through it, depending on the DC voltages received, much like a VCA.
The same method of using DACs to control analog levels was carried over into the H3000. This is also the same "multiplying DAC" technique used by SSL in all of their current moving-fader console offerings.
SP2016: Algorithms
Agnello hired Robert Belcher and Ken Bogdanowicz (who later left to create audio plugin company Soundtoys) to develop algorithms for the SP2016. While the most famous programs on the SP2016 are probably the reverbs—Room (mono in, stereo out), Stereo Room (stereo in/out), and Diffuse Plate—several other algorithms were available on EPROMS: Vocoder, Band Delays, Shimmer, Crystal Echoes, and Time Scrambler, among others.
"People had trouble understanding what it was and what it could do. Digital reverbs, like the EMT, Lexicon, and Sony, were very popular products, and we confused a lot of people who just wanted a great reverb. We would have been better off if we had limited the box to just reverb," Agnello told Reverb in this 2018 article.
If some of these algorithms sound familiar, it's because many of them would find themselves another home on the H3000.
For the reverb geeks out there, all reverb algorithms were of the feedback delay network (FDN) variety. From Agnello, "The 2016 reverbs were FDNs, but that terminology wasn't used by me. I was calling it a delay matrix but never published."
While the SP2016 certainly hasn't maintained the level of visibility of its Eventide stable mates, its significance to later Eventide developments—let alone its contribution to the advancement of audio signal processing—cannot be understated. Agnello explained in an email:
"As is sometimes the case, wildly successful products are designed by a team that designed the previous generation. The success of the H3000 cannot be understated, but the SP2016 is the reason that the H3000 design was possible, and, from a historical/academic perspective it's a story that has never been told. The SP2016 was the ground-breaking product. The H3000 was its commercially successful descendent."
For those interested in hearing the SP2016, it can be found all over the work of a number of high-profile engineers and producers. Agnello points to Mick Guzauski, George Massenburg, Allen Sides, Dave Pensado, and Jack Douglas.
"It was the go-to reverb for them and many others, so it's on thousands of records from 1983 on. Allen Sides still uses the hardware but many people use the plugin version."
And while the SP2016 is probably best known for its reverbs, Agnello points out that “Suzanne Ciani credits Band Delays at being at the heart of her landmark album, Seven Waves.”
Given the rarity of a functioning SP2016, they don't come up for sale often. Fortunately, the recently-released SP2016 Reverb finally captures the mojo of the cherished Room, Stereo Room, and Plate algorithms. Long-time SP2016 user Dave Pensado sings the plugin's praises in the following video:
The design brief for the H3000, according to an AES paper by Belcher and Bogdanowicz, "was to build a stereo pitch shifter and to do it quickly and inexpensively ... its audio quality needed to be better than any of our previous products. Additional effects, such as reverb, were thought of as possible features."
To keep costs down, Eventide again chose to use off-the-shelf silicon, specifically the Texas Instruments TSM 32010 digital signal processing chip. Agnello states, "The TI 32010 chip introduction convinced Eventide to migrate our development efforts to DSP chip-based hardware."
The H3000's I/O and Conversion
Conversion was to be "at least 16-bit and processing with at least 44.1kHz sample rate," according to Belcher and Bogdanowicz. To this end, the H3000 used the same converters employed in the SP2016: Sony CX20018 ADCs and Burr-Brown PCM53JP-V DACs.
The Sony ICs were successive approximation converters—no oversampling here! And where most contemporary ADC/DAC ICs have on-board anti-aliasing/reconstruction filters, converters at the time lacked these faculties. The H3000s used Murata ALF89WB 9th-order low-pass filters with a 20kHz bandwidth on both ADC and DAC stages—large silver "hybrid ICs" in a SIP package.
Analog stages were TL07x-series operational amplifiers. At the time, these were substantially less expensive than NE553x ICs, while offering both a higher input impedance (due to JFET inputs) and a higher slew rate, with higher distortion and noise being the trade-off. Analog input/output levels were software-controlled using Analog Devices AD7524 (input) and AD7528 (output) MDACs to vary analog audio levels, again like the SP2016. Both Agnello and Richard Factor credit the analog I/O design to Dave Derr, best known as the brain behind Empirical Labs.
The H3000's Signal Processing Architecture
The H3000, like the SP2016, was designed to be software-based. That is, audio processing tasks weren't directly tied to a limited set of instructions built into a hardware-based device, like they were in the H910. The H910 could not be re-programmed later via a software update to perform additional audio tasks like flanging or phasing. The SP2016 and the H3000, on the other hand, could have additional algorithms and presets loaded into them by adding ROMs containing additional software. The SP2016 and H3000 were computing platforms made for audio processing—not just hardware devices designed to perform a finite set of audio tasks.
The way the SP2016's processing works is similar to the way UAD audio plugins are processed by dedicated DSP cards (or Thunderbolt boxes) attached to a computer: The plugin's user interface accepts input (parameter changes), but the instructions are ultimately passed by the "host processor"—in this case your computer's CPU—to the UAD hardware, which then applies an effect to audio in a DAW that is running on your computer (again, the host processor in this scenario).
The core of the H3000 actually consists of three Texas Instruments TSM 32010 digital signal processing chips, with the software user interface operating on an 8-bit Motorola 6809 host processor. Here's how a parameter change would play out as a user manipulates delay time via the H3000's front-panel encoder knob:
- User turns encoder knob.
- The parameter change instruction is passed to the 6809 host processor.
- The host processor then delivers the delay change instruction to the TSM 32010s, which alter the audio digitally, based on user input.
- The altered audio signal is then sent to the D/A converters via a high-speed bus.
- The 6809 host processor then provides visual feedback to the user via the backlit two-line LCD display.
At the time, no single-chip DSP option possessed the processing power required by Eventide's design spec for the H3000. So two possibilities were considered: "bit-slicing," where, for example, the first eight bits of a 16-bit word are processed by one DSP, and the remaining 8 bits are processed by another. This required expensive high-speed parts and would have made software development more difficult.
The second option—and the one ultimately chosen by Eventide—was to distribute audio processing tasks among a number of off-the-shelf DSP chips (in the case of the H3000, three TSM 32010s). This ended up being both more economical and easier to program, with TI offering excellent software support.
Different Versions of the H3000
One of the more interesting things about the H3000 is that the various versions released over the device's 10+ year lifespan are based on largely identical hardware.
Later versions added more features like digital I/O and the optional sampling card (standard in the H3500), but the analog section, conversion, and the three TSM 32010s remained unchanged. The differences ultimately come down to the included algorithms and presets. The upshot is that even the oldest H3000s can be updated with the last OS version and every algorithm/preset ever developed.
Again, this ultimately comes back to Eventide's forward-looking decision to design the H3000 as a software-based audio computing platform.
Curious about which H3000s came with which algorithms and presets? Consult this definitive Eventide H3000 Algorithm and Preset Chart.
And here are a few you might want to check out:
- 231 - Micropitchshift and 519 - Micropitchshift: The two are different, based on two different algorithms. They are "Style I" and "Style II," respectively, on Soundtoys' Microshift plugin.
- 217 - Dual H910s: Another micro-shifting favorite.
- 114 - Dense Room: A favorite of Chris Lord-Alge.
- 101 - Layered Shift: Vance Powell's H3000 lives on this preset.
- 246 - Shimmerish: F. Reid Shippen is a fan of this preset based on the Swept Reverb algorithm.
- 700–747: Steve Vai used the H3000 so prolifically that the H3000S and later models included an entire bank of presets authored by the guitarist.
Eventide's approach to building modular, expandable products first pioneered with the SP2016 and then the H3000 is reflected across their current product lineup, from guitar pedals like the H9 to their flagship multi-channel audio processor, the H9000.
Eventide continue to develop new algorithms for the former, while the latter is designed around a multi-card ARM architecture designed with future upgradeability in mind. The H3000's longevity is not just a testament to the unit itself, but to Eventide's forward-looking and future-shaping use of the new technology of the time.
Ian Anderson is an assistant professor of music at Kent State University, where he teaches audio recording.