Synth School: Part 5

Roland D50 multitimbral S&S synthesizer.

At the January NAMM show in 1987, Roland launched their D50, which mixed synthesis and sampled sounds in one package, a combination which has remained popular to the current day. Paul Wiffen examines how S&S evolved into the most widespread form of sound generation on the market. This is the fifth article in a 12‑part series.

Until Roland launched the D50, sampling and synthesis had been perceived as two wholly different disciplines, almost like competitive ways of doing the same thing. Some people favoured sampling because it gave you a more accurate representation of actual instruments (the holy grail of piano, strings and brass, for example), while others stuck to the various styles of synthesis because they offered greater expressivity and speed of use. There had been massive improvements in the sampling arena in the preceding few years. It was no longer just the province of rich stars and well‑paid programmers. The Ensoniq Mirage had solved the expense problem, the Prophet 2000 had made state‑of‑the‑art fidelity affordable (12‑bit linear as opposed to 8‑bit companded) and shortly thereafter the Akai S900 had made fidelity relatively easy to use as well (thereby drastically reducing my income, as I had been making a nice living out of operating first Emu and then Sequential samplers for people who found them difficult to use!). So sampler ownership was reaching a much wider market than in the early '80s.

Sampling's principal remaining drawback in the late '80s was the amount of time it took to load sounds. As a result, the majority of people playing live, and those who were frightened by the idea of using computer technology (RAM, floppy disks and hard drives), were still using the various competing forms of synthesis we have examined in previous instalments of this series, because even if less sonic authenticity was available from these forms of synthesis, they responded better to velocity and aftertouch and (most importantly, I suspect) you could switch sounds instantaneously. The great debate raged between the two opposing schools of thought, often with things getting a bit personal. The great irony was that the whole situation was about to be resolved, by these two supposedly conflicting technologies being merged together (a bit like this year's shock announcement that Bill Gates was putting money into Apple).

Before The D50

It has to be said that sampling and analogue synthesis were not existing in glorious isolation anyway; as early as the PPG Waveterm it had been possible to make a sample and then play it back on the Wave synth through analogue filters. The Emulator II added analogue filtering and enveloping to sampling technology, and this was carried over into more affordable samplers such as the Mirage, Prophet 2000 and Akai S900 (although many people never used the facility). And even on synthesizers there had been the odd attempt to increase the fidelity of certain sounds by using small PCM samples loaded into ROM. (This was how the Ensoniq ESQ1 provided its drum sounds.) But all these half measures meant that synthesis and sampling were seen as mutually exclusive fields — until the D50 came along.

The D50 used a much larger amount of PCM ROM (separate from that holding the operating system of the synth) to store a significant number of samples, allowing the expressive performance of a wide range of sounds previously only possible with any fidelity on a sampler. Although the D50 itself didn't have a sequencer, this approach paved the way for a new breed of instruments known as 'workstations', which were designed to perform a wide range of musical tasks — for example, playing drum, bass, piano and string parts simultaneously using an internal sequencer. These sounds were the ones which were the most difficult to make with analogue or digital synthesis, the ones which had previously only been possible by loading a disk or two's worth of data into a sampler. Although Ensoniq had already released the first instrument worthy of the workstation title, it was Korg who had the breakthrough success with the M1, not because its sequencer was notably easy to use, but because of the sheer size of its palette of sounds. The reason was that the samples in the M1 were larger than those in the D50, in the same way that the D50's samples had been larger than the percussion snippets in the ESQ1.

How Did They Do That?

It's still well worth looking at how the D50 generated sounds, because in the course of taking a few enforced shortcuts (dictated by the budget they were working to), the Roland engineers came up with some techniques which changed synthesis forever.

The D50 was actually a hybrid of three previously distinct technologies:

• Analogue (or subtractive) synthesis

• Digital sampling
• Digital signal processing (DSP) for effects

These core technologies met in the D50, perhaps not for the first time, but certainly in the most affordable and usable way. Although each on its own would not have been enough to make a viable instrument (the samples were too short, the synthesis too restricted and the effects too primitive), the combination of the three made an instrument people couldn't wait to get their hands on. And although its imitative capabilities have long since been surpassed, as a synthesizer it still has much to recommend it today.

Let's first analayse the strengths and weaknesses of the D50's three component technologies:

• Analogue synthesis took electronically generated waveforms and used filtering to shape the harmonic content of the sound over time. While such a process was excellent for creating rich and interesting sound timbres, its imitative capabilities were limited, especially for inexperienced users. In addition, polyphony was usually limited, due to the need for discrete circuitry for each voice.
• The digital sampler used the technique of digitising sound and storing it in computer memory to allow real instruments to be recorded and played back from the keyboard. This provided an instant realism unavailable from traditional synthesizers, but with a loss of expressivity (the only nuances available being those 'frozen' in the recording). The amount of sample recording time available was very limited, due to the high cost of computer memory. In addition, instruments had to be sampled every few notes along the keyboard for authentic reproduction, which used up the available memory even more quickly. Looping (repeating an unvarying section of the recorded sound for as long as the key was held down) helped reduce memory usage enormously. Techniques such as fading or switching between samples depending on how hard the key was hit increased expressivity, as did the introduction of analogue components such as filters and envelopes, to allow the timbre of the recording to be changed by playing style.
• Digital signal processing, the third component, had reached the stage where a single DSP chip could be programmed to imitate many different analogue effects, such as chorus, flanging, reverb and echo, and even combine two or more of these effects simultaneously.

Interestingly, the D50's PCM loops pre‑date the use of sampled drum loops, now omnipresent in modern recordings.

Roland used the term Linear Arithmetic Synthesis (or LA Synthesis, for short) to promote this combination of technologies, although the irony is that the term gives very little clue as to how the D50 works, based, as it is, on one often‑overlooked part of the process which determines how two waveforms are combined. The options are to sum the waveforms together (like two oscillators in analogue synthesis) or multiply them (as in ring modulation or various digital synthesis techniques). It will become clear that this part of the process played little or no part in the D50's phenomenal success. It seems likely that the term was coined more for its two‑letter acronym value (like FM, PD, and so on).

So what were the key components which gave LA synthesis its appeal?

The Sampled Attack

What let analogue synthesizers down more than anything in terms of imitation was that they could not create the extremely complex sets of harmonics present at the beginning of most acoustically produced sounds. The first few milliseconds of sound, when a piano hammer strikes a string or a bow begins to move on a cello, have a huge harmonic content not available in traditional electronic waveforms. If we do not hear these short‑lived frequencies, we perceive the sound as lacking in authenticity.

The Roland engineers realised that if they could use a digital recording to produce the initial attack, this would go a long way towards creating realistic instrument sounds. In addition, very little computer memory would be needed to store these very short 'attacks'. Pulse Code Modulation was used to record the attacks, and they were stored on Read Only Memory (ROM) chips, which did not lose their contents when the power was shut off, unlike the Random Access Memory (RAM) chips used in samplers. This did away with the need for floppy disk drives (although a memory card could be used to make further PCM samples available).

The Sustain Loop

The Roland engineers turned to waveforms that had more in common with analogue synths in order to produce the sustained portion of the sound. The looped portions of samples often sounded very similar to traditional synthesizer timbres. They were known as single‑cycle loops, as they contained only one of the repeating patterns which make up the timbre of an electronic oscillator. The small amount of data contained in such loops meant that they took up very little room in the ROM chips, which meant that notes could be held indefinitely without using up valuable RAM space.

Partials

Obviously, the fact that different parts of the sound were being created by different PCM waveforms meant that it was necessary to control these sources separately. The solution that was devised to allow this was called the Partial. There were four Partials available for each sound program in the D50. Each Partial could be loaded with a PCM waveform and then combined with its fellows. Each Partial used one voice of the D50's polyphony, so it was often better not to use all four Partials unless really necessary (more complex sounds would result in less polyphony). Most of the more realistic sounds used at least two Partials (one for the sampled attack and one for the loop section). However, some of the sounds had four Partials, using two pairs of two to create the effect of two sounds layered together (piano and strings, for example). Perhaps the sounds which characterised the D50 the best were those which used all four Partials independently, to create complex evolving timbres. These did not require much playing, but simply sustaining a note or chord while different elements faded in and out. Such sounds were almost unknown to the average musician before the D50, being only possible on professional systems such as the Synclavier or PPG.

Add Or Multiply?

Although I've played down the importance of the way the Partials were combined in the success of LA synthesis, it is actually one of the things which makes the LA process of interest as a synthesis type today, now that it has been superseded by systems with more memory. The standard way of combining the different elements of sound since the beginning of synthesis was simply to mix them — ie. sum them in a linear fashion. This was just one way of combining the sounds on the D50 (admittedly the one used by the majority of its sounds). The other was to multiply the waveforms together, which tends to create metallic or bell‑like timbres, as the normal harmonic series is supplemented by less usual frequencies. This is where (for me at least) the really interesting sounds from the D50 were created. Although the success of the instrument can be attributed more to the new level of authenticity it brought to conventional keyboard sounds than any revolutionary new timbres it created, it is in this latter area that purchasers of second‑hand D50s today may want to look for its unique character.

The combination of the four Partials on the D50 was determined by Algorithms. Although there were nowhere near as many as offered by 6‑operator FM, they were illustrated on the front panel of the instrument, just like on the DX7 (see picture above). These algorithms allowed you to determine which sounds were added together and which multiplied. You could, for example, multiply two pairs of Partials and then add the results, or multiply one pair and add the result to two other Partials. However, this feature of the machine was perhaps the least exploited, with 95% of sounds created by simply adding together whichever Partials were selected.

Enveloping

The combination of sampled attack and single‑cycle loops meant that realistic strings, brass and other sustained sounds could be easily created. However, pianos and other timbres produced by striking do not sustain forever, but die away gradually. A single‑cycle loop, therefore, has to be 'faded out' to simulate this effect. The more sophisticated samplers had already borrowed envelopes from analogue synthesis to deal with this, and the D‑series engineers followed suit. Because the D‑series used sampled attacks along with single‑cycle loops, a more complex envelope system than the conventional ADSR was used, with six independently adjustable times and levels was employed. In this the D50, once again, had more in common with the DX7 than conventional analogue synthesizers.

Because the D50 featured analogue‑style filtering (albeit digitally controlled, hence the term DCF), these envelopes could be used to adjust not just the volume of the loop over time but also the harmonic content. This was absolutely critical to the realism of LA synthesis, as you might not want the looped part of the sound to retain its full harmonic content throughout a decay. Without the ability to vary harmonic content, any decaying sounds, from pianos and accoustic guitars, through to tuned percussion, would have had little realism.

Blending Partials

Even very careful adjustment of the basic envelopes could not prevent the attack and looped segments sounding like two different sounds being triggered at the same time. What was needed was something to 'fuse' the sounds together so that they became one instead of two distinct timbres. To achieve this, the Roland engineers used DSP effects such as reverb and chorusing to blend the parts together. Reverb tends to make the point where a sound ends difficult to perceive, so this was perfect for hiding the fact that the sampled attack had suddenly stopped and the looped portion was the only part remaining. Chorus was also good for adding some timbral movement to the single‑cycle loops, which can be a little static to the ear.

As a result of all the M1's features, Korg found themselves with the best‑selling synth of 1988/9.

Because the sounds which needed this 'smearing' really did need it, in order to be usable at all, the Roland engineers made the effects setting part of the basic patch, so that it was always automatically selected with the program, becoming one with the sound and therefore a fundamental component of the synthesizer for the first time. Previous synths had often featured a chorus unit, but usually as a separate item (it was not usually tied in permanently and selected simultaneously with the timbre, except in the case of the Roland Jupiters and Elka Synthex). Although the inclusion of full‑scale DSP effects may originally have been decided upon to mask problems with this style of sound creation, it changed the face of synthesis forever. Today it is virtually impossible to sell anything but the most basic of monosynths without a built‑in effects capability. This is because once a programmable DSP effect section has been added to a synth, there is no reason not to make available as many different effect algorithms as possible.

The Wider Effect

Once the DSP chip was inside the unit, there was no reason to limit its use to disguising the shortcomings of the D50's synthesis system. Reverb and chorus had a very pleasing cosmetic effect on any sound and were used on virtually every patch. Effects like distortion and ring modulation could take the most bland source waveforms and turn them into complex, expressive sounds. As a result, the D50 caused an even bigger change in the world of synthesis than the processing of sampled sounds through analogue‑style synthesis. With the exception of a few professionals, who would pointedly ask to hear synths on demo with the effects bypassed, the majority of purchasers simply accepted that this was an improvement in the final sound they could obtain from a synth, without the need to hook‑up expensive external effects. Soon built‑in effects were the norm, not the exception, for synthesizers.

The Korg M1

The next refinement of the PCM‑based synthesizer was the Korg M1, which burst upon the world one year later at the 1988 NAMM show in Los Angeles. The price for memory had come down since the launch of the D‑series, allowing Korg to increase the amount of memory within their new instrument. The major advantage of this was that instead of having to split samples into attack portions and single‑cycle loops, they could use samples which moved naturally from the attack into a longer looped section, in exactly the same way as in a sampler. This meant that it was no longer necessary to disguise the join between the attack and the loop, because there no longer was one.

Another advance the M1 made was that only one voice of polyphony was required to play back each entire sound. This meant that polyphony did not vary from one sound to the next quite as dramatically as on the D50 and, at the same time, the sounds were not so reliant on the built‑in effects to make them sound natural. Sound‑stacking could be used to make very complex timbres, rather than being necessary just to create authentic simpler ones.

Of course, this did not mean that all the M1's sounds were perfect reproductions of the instruments from which they had been sampled. With hindsight, a lot of the original sounds in the M1 used perhaps too short a segment of attack sound, and the loops came too early for authentic reproduction of timbres like pianos, guitars and other sounds which die away gradually (although anything which could sustain indefinitely, like strings and brass, was extremely authentic). As a result, the M1 produced a 'compressed' sound which became very popular in certain styles of dance music. The M1 Piano, in particular, became a staple of house remixes because it was artificially bright and 'in your face', and the organ sound on it was a similar staple for garage music. Whilst you would rarely use an M1 piano today for a 'straight' piano sound, at the time the M1 brought an unprecedented level of authenticity to sample‑based synths.

The other thing which the M1 offered, over and above the authenticity of its sounds, was multitimbrality (the ability to play numerous different timbres at the same time). This could be done from the on‑board sequencer or via MIDI, assigning each timbre to a different channel for triggering. Multitimbrality wasn't new; Sequential circuits had introduced it at the end of 1983 in the Six‑Trak and Ensoniq had made it a major feature of all their products since the ESQ1 in 1985. However, the M1 was the first instrument with a good range of really authentic sounds to offer this facility. As such, it was perhaps the first synth whose on‑board demos sounded like complete pieces of music, because it had everything from authentic drums and basses, to piano, strings and brass, guitars and synthesizer sounds, all in one unit.

Many previous keyboards had featured sequencers, but the usefulness of these was limited by the number of timbres they could produce simultaneously, or by the limited range of the synthesis type they featured. The full PCM multitimbrality of the M1 meant that the sequencer became much more than a sketch‑pad or demo facility. It was a compositional tool which was hooked directly to the sounds. This meant that people who had no computer sequencing facilities or knowledge of MIDI could sit down and play something, record it, and overdub more tracks, with different sounds on each one. Whilst those who had mastered MIDI and computer sequencing would find nothing remarkable about this, it was a real revelation to those who had never experienced the power of MIDI sequencing.

The term 'workstation' was borrowed from the computer industry to market this concept, as well as a floppy disk drive as standard, so that M1 sounds and sequences could be recorded, saved and loaded back into the machine. As a result of all these features, Korg found themselves with the best‑selling synthesizer of 1988/9.

The DSP Effect

The only area which caused a bit of difficulty for M1 users (and still does today, as one in every 10 calls to the Korg technical helpline still bears out) was the allocation of effects in multitimbral mode. Like the D50, the M1 features built‑in DSP for a wide variety of effects, which are also memorised and selected by the individual sound Programs. So whilst using a single timbre, it is possible to obtain sounds which are dramatically altered by the DSP. However, when Combi mode (which allows multiple timbres to be available simultaneously) is selected, the chances are that these sounds will suddenly become very flat and uninteresting. This is because the DSP circuitry in the M1 can only produce one effect setup at a time. Unless the Combi setup has exactly the same effects and routing setup as the individual Program, there's bound to be a noticeable difference to a Program when it's selected as just one of the sounds in Combi mode. The degree of difference is determined by two factors.

The first of these is how close the settings of the individual Program are to those selected in the Combi. If reverb is selected for both, and only the amount of early reflection or size of space has been changed, the difference will be subtle and not noticed by many novice users. If the Combi and Program settings have different effects selected, such as chorus and echo, the difference will be far more noticeable even to the most untrained ear.

Until Roland launched the D50, sampling and synthesis had been perceived as two wholly different disciplines, almost like competitive ways of doing the same thing.

However, the second factor is usually what alerts the novice user to the problem. If, in Combi mode, the user selects a Program where a complex DSP effect is actually creating the timbre from a very simple source (such as a distorted guitar sound), suddenly all the character of the distorted guitar or synth disappears and is replaced by a thin plucked sound. This often happens after three or four backing tracks have already been recorded and the user wants a really exciting lead instrument to play over the top. The general reverb used for the other sounds just does not work as a substitute for the distortion the guitar Program uses. The change in some timbres is so radical that some users at the time even contacted their dealers to report that their synth was not working properly. Many dealers received M1s back for repair, only to find that there was nothing wrong with them, except for an inability to faithfully reproduce several different Programs simultaneously in Combi mode. This is because the M1 does not have multiple effects processors (as do most other multitimbral workstations). M1 users have to try and make Combi setups for multitimbral sequencing which can share the same effects setup. This may mean backing off the reverb on that string sound, so that the bassline doesn't disappear in woolly mush, or leaving the chorus off the guitar, because it makes the piano sound like a honky‑tonk.

The best way to deal with this situation is to try and plan which sounds you want to use simultaneously in advance. It is usually obvious which sounds really need the effects to retain their inherent character and which ones only use the DSP for a little sugar‑coating. Then you need to reach a compromise between the amount of effect needed to give one sound its character and the amount which will not render the others unrecognisable.

Ideally, each M1 sound would have effect 'send' amounts, but this capability was not introduced by Korg until much later, on workstations such as the X‑series. The result was not the same as if separate effects were available on each part individually, but it did give the best compromise available. This effects limitation was only finally resolved last year on the Trinity, which does have enough DSP horsepower to allocate 'insert' effects separately to each part during multitimbral operation, and also to offer an overall master effect, such as reverb, which is available to every part in amounts determined by effect 'sends'.