SY77 - The Power behind the Buttons! (Part 2)
RCM Synthesis explained
In this second part of our in-depth review of the SY77 synthesizer, Martin Russ delves into Yamaha's ultra new synthesis method: Real-time Convolution and Modulation.
Yamaha's new Real-time Convolution and Modulation (RCM) synthesis is the basis of the sound generation in their new flagship synthesizer/workstation, the SY77. Last month I looked at the performance and design aspects of the instrument, and this month I will show you how RCM produces the sounds which make it something rather special.
Yamaha tend to be technical in their descriptions - some people might have shied away from the complexities of using Frequency Modulation as a name for a synthesis method. Real-time Convolution and Modulation synthesis is much the same - hardly the name a 'plain English' campaigner would choose! Before FM appeared on the scene and was forced into keyboard players' vocabularies, analogue synthesis was the only practical way to make synthetic sounds. The DX7 offered the first truly affordable way to utilise fully digital synthesis, and the mid-Eighties saw FM become a huge success.
In the post-DX7 period, we find ourselves in a world populated with LA, AI, PD, Wavetable, Additive, Resynthesis and many other ways to produce sounds, almost exclusively using all-digital techniques and often mixing samples with synthesis into composite forms. Against the realism and immediacy of current techniques, FM has recently started to be perceived as clinical, slightly boring, and definitely not fashionable - although the V50 breathes new life into FM, with its complex and versatile stacking of FM sounds.
Unfortunately there is a problem for the current generation of 'Sample+Synth' (S+S) instruments: samples stored inside the machine are permanently 'frozen', and there is very little which can be done to alter their intrinsic sound. When I reviewed the Korg M1R a while ago, I was struck by the way that the extra plug-in sound cards had the same sort of 'feel' as the internal presets - something in there was not changing. The Roland U20 goes some of the way to supplying more useful variations in the sounds by providing envelope control, but I still feel that this is only a partial solution.
Combining sampled sounds with FM, as in the SY77, might seem to be prone to exactly the same problem of fixed sounds, were it not for the fact that FM is a powerful synthesis method in its own right. Few people would say the same about the synthesis parts of most other S+S implementations. In fact, one of the reasons that FM took the world by storm was its intrinsic superiority over analogue VCO/VCF/VCA based synthesis, which forms the basis of the synth part of many S+S instruments. As if this was not enough, RCM provides several ways to change the samples themselves, including making them part of the FM synthesis. All of this lets RCM put a great deal of sonic distance between the original samples and the final sound. So what exactly does RCM do?
Digital Signal Processing is all about real-time operation. Unless you can alter things without any perceptible time delay, you lose the advantage of immediate feedback, and this is especially important in a synthesizer where 'live' editing is the prime method of producing sounds. Unfortunately for synthesizer designers, two things are difficult to achieve in real time: smooth and fast sweeping resonant low-pass digital filters, and meaningful alterations to sampled sounds. This is why you normally find simple filtering and enveloping on most S+S instruments, but little in the way of more complex processing.
Yamaha have considerable experience in the design of digital filters (the DEQ equaliser and TX16W sampler, for example) and recent advances in the mathematics and implementation of filters have allowed the SY77's filter to generate the same resonant filtering effects as the old analogue VCFs. Further, the addition of samples to FM allows the samples to be used as modulators within the FM algorithms themselves, and this allows very complex changes to be made to the sounds, far beyond simple enveloping. These two techniques are known as 'Convolution' and 'Modulation'.
In everyday use, convoluted means 'twisted' or 'complex in form', but in the world of Digital Signal Processing it has a precise mathematical meaning - which is quite difficult to grasp if you are not familiar with the subject. Rather than attempt to describe all the necessary stages, I will give an overview of the basic concept.
Most things which are difficult to do in one way can be made much easier by first converting them into another form, processing them and then reconverting them back. Convolution is one example of something which can be done in exactly this way, and so achieves filtering by a roundabout or 'twisted' route, where simple digital processing produces complex filtering.
Yamaha use real-time convolution to make the complex task of filtering as simple to use and familiar as the analogue equivalent. What it boils down to is that you can replace the word 'convolution' with 'filtering', and although it is technically less descriptive of what is going on, for most purposes it is sufficiently precise.
The Modulation part of RCM refers to the ability to use AWM2 sampled waveforms as modulators in the AFM synthesis section. As with ordinary FM, when you increase the level of the modulating waveform, the resulting sound becomes more and more spread out across the audio frequencies. When you use an AWM2 sample, this has the effect of gradually blurring the sample by adding in extra harmonics to the original sound. At low levels the sample can remain virtually unaffected and can be pitch-shifted by using carrier ratios other than 1.00, but at high levels the extra harmonics can be interpreted as distortion - with a whole range of sounds in between these two extremes. Conventional S+S instruments often allow you to filter the samples, to remove some of the component harmonics already present, but the SY77 also allows you to add in new harmonics based on the sample itself!
When FM was first released, the four-segment envelopes were amongst the most complex available. Since then, envelopes with more segments and break points have come into common use, with the result that the DX7-type envelope has begun to look a little underpowered. The SY77 brings Yamaha bang up to date with six-segment envelopes with looping, delayed start, velocity rate scaling, and other useful additions.
The new envelopes allow you to specify separate levels for the start and end points, as well as loop any of the segments before the sustain period - which makes repeated percussive sounds simple to produce. The start of the envelope can be delayed with the Hold Time parameter, which makes echo and flam effects easy. Two release segments mean that better control can be achieved over the tail-end of sounds, and the velocity rate scaling enables playing style to alter the relative timing of the envelopes, such that hard playing reduces the attack time but gentle playing increases attack; very effective on string and brass sounds. This is in addition to the rate scaling, which makes envelopes shorter for higher notes (or longer, as is now possible in the SY77).
The SY77's Advanced FM uses 24-bit precision arithmetic to improve the audio quality of the sounds produced. New parameters provide finer control over the detail of frequencies and levels, as well as opening up further possibilities. The individual operator pitch modulation, for example, opens the door to ring modulation effects, which can sound very clangorous if over-used, but can enhance many sounds if employed subtly. When combined with AWM2 modulation, some really unusual timbres can be created.
The extra waveforms provided by the SY77 are an extension of the eight additional waves first seen on the TX81Z, and more recently in the DX11 and V50. Each is a mathematically derived variation on a basic sine wave (waveform 1) and a chart is supplied in the owner's manual detailing their harmonic content. The waveforms are arranged with a gradual increase in the harmonic content as you move to the higher numbers. As these can form the basis of complex waves from just a couple of operators, they are a great asset to the FM synthesis power of the SY77 - in fact, this is the first time that six-operator FM has used anything other than a sine wave.
The Noise waveform and AWM2 waveforms (especially the looped ones) can also provide extra input to the AFM operators. By using the Pitch EG to change the pitch of just one operator (a new feature exclusive to AFM), you can produce ring modulation type effects. This opens up a whole host of new and different sounds, especially sound effects and bizarre modifications to recognisable sounds. The ability to use AWM2 inside AFM, and vice-versa, tends to blur the boundaries between the two methods of tone generation in a way which is not matched by any competing synthesizer, and makes the SY77 much more than just another S+S synth.
There are 45 arrangements of the six operators in AFM. Of these, 14 are equivalent to 30 of the 32 algorithms on the DX7 (because you have more control over the placement of the feedback loops, one AFM algorithm can be the equivalent of more than one FM algorithm). The algorithms are arranged into an order determined by the number of carriers and maximum height of the stack - so the first algorithm (1) is a stack of five modulators on top of a single carrier, whilst the last algorithm (45) is just six carriers.
Two DX7 algorithms have no obvious direct equivalent on the SY77, algorithms 19 and 20: each of these have three modulators and two carriers both modulated by one modulator. None of the AFM three carrier algorithms has a modulator which modulates two carriers. By way of compensation, the four and five carrier algorithms in AFM have a very flexible structure, where one modulator can affect up to five carriers at once (these are the only two AFM algorithms which directly implement multi-carrier modulation). Conversely, up to five modulators can be programmed to modulate one carrier in the single carrier section, and up to four modulators can modulate one carrier in the dual carrier section (three modulators controlling one carrier was the maximum number available on the DX7).
Some 31 of the AFM algorithms are new, and have no direct equivalent on the DX7. On several of these the feedback function is fixed on one of the operators, which partially 'locks' the topology of the algorithm, but for the remainder the feedback can be around any operators. A graphical representation of the algorithm diagram is shown in the SY77's display by pressing the 'Alg' soft key (f8) whenever the element you are editing is an AFM one, although the diagram does not show the currently active feedback loops - you have to go to the first of the AFM display pages to find out about the feedback. Feedback and the special inputs to each operator are all tied together on these pages, and they make algorithms much more flexible and powerful.
The feedback loops are also the answer to the 'missing' equivalents to algorithms 19 and 20 on the DX7: by routing 'feedback' so that it goes from a modulator to a carrier, you can produce exactly the same structure as a single modulator connected to a maximum of three carriers (the cases of four and five carriers are already covered, see above). So the missing algorithms are actually not only present, but in a much improved form where you can edit them to suit your own requirements. More importantly, you can also produce much more complex algorithms than those provided by using the three available feedback loops - which can also be thought of as 'feed-forward' loops when used in this manner, to add extra connections of your own choice. One constraint on placing these extra connections between operators arises when the feedback loop is already specified as part of the algorithm itself (there are only five out of the 45), and in this case you are left with one less input and thus one less connection for you to control.
An additional constraint on feedback placement involves the use of AWM2 waveforms as modulation sources. Because there are only two special inputs to each operator (including those used by the connections which determine the algorithm topology itself), some inputs are not available for external modulation by AWM2 or the Noise waveform. In practice, this is not a major problem, since when you are using an AWM2 waveform, it will normally be the only modulator of the chosen carrier, and so at least one input will be available. The same argument applies to the more unusual case of using an AWM2 waveform at the input to a modulator - where adding another operator as a modulation source can produce very 'rough and dirty' sounds, not the squeaky clean FM you might be used to.
Because a 'white noise' type waveform (or should I say spectrum) is available as a standard input to an operator, it is no longer necessary to use feedback and high operator output levels to produce noise-like sounds - which leaves the six operators free for other uses within a sound. Combined with the additional waveforms, instead of just a sine wave, the overall effect is more like having eight or more operators at your disposal. And the flexibility of the algorithms means that you have almost complete control over their structure. The 'Advanced' description of AFM is certainly well deserved!
The SY77 has 112 AWM2 waveforms, sampled in 16-bit linear form at either 32 or 48kHz (the sampling rate is invisible to the user). Additional waveforms can be accessed from ROM cards inserted in the waveform card slot.
The first 33 supplied preset waveforms are multi-sampled and looped (there is some cyclic timbre modulation, but no glitching that I could hear). They include a varied mix of conventional Western instruments, traditional Japanese instruments, vocal sounds, and sampled analogue synthesizer sounds. The next 31 waveforms are single samples taken from a different set of instruments, but covering the same broad areas - again, they are looped (with one exception: the orchestral stab) and mostly contain both the attack and the steady state sounds. There are four non-pitched waveform samples, where the majority of harmonically related frequencies have been removed, leaving the inharmonic residue - the clunks and clinks. Eight attack and transient samples provide bow scrape, vocal consonants, blown and plucked sounds.
There are 15 special effect sounds (all looped, except the 'Bamboo' sound) which provide a range of noise-like timbres, bells, and additional percussion sounds. Some 20 drum sounds complete the roster, with a combination of traditional Yamaha drum samples and some more interesting gated reverb snares and bass drums to liven things up a little.
The complete absence of any Latin percussion is probably a pointer to the contents of at least one of the add-on AWM2 waveform ROM cards. Although a flute sound is included, the shakuhachi (over-used in TV adverts!) is thankfully not present, and the orchestral stab is a little disappointing. Everything else sounded very nice indeed, even to these jaded ears!
At first sight, there are only three ways to use the AFM and AWM2 Elements in an SY77 Voice: AFM only, AWM2 only, and both together. In other S+S synths this is indeed the case, but the power of RCM becomes apparent when you start to investigate the other ways of using Elements in combination. Rather than just going through the possibilities of using the Voice Modes, I will give examples of the sort of instruments which they might emulate:
1. AFM only - the equivalent of up to four DX7s stacked! And don't forget the creative potential of splitting the keyboard, velocity switching, and dynamic panning.
2. AWM2 only - the equivalent of up to four EMT10 sounds simultaneously. Alternatively, the Kawai K4 or Ensoniq VFX can be thought of as just waveform playback, and so this configuration produces similar sounds. All the tricks are possible: like fixing the pitch of a sample to give a background 'clunk'; transposing it down to give strange low-pitched groans; or running looped sounds at high pitches so that they take on new pitches based on the loop time.
3. AWM2 with filtering - classic analogue synths (Oberheims, Moogs, plus the modern S+S equivalents: Roland D50, Korg M1, Kawai K4 etc). Remember that up to eight filters can be present in one Voice, so the range of timbral variation is wider than with many other synths.
4. AFM with filtering - SY77 only! This configuration gives you classic DX7 sounds, but with filter sweeps, band-pass filtering, subtle timbral changes and dynamic modulations. The best of performance FM with added rapid control overtone!
5. AFM using the extra waveforms - Phase Distortion. The typical Casio PD sound uses FM operator-like pieces with waveforms other than sine waves - the latest Casios let you alter the arrangement of these pieces as well. In SY77 terms, using the additional FM waveforms should provide PD-type sounds without any problems - the higher numbered extra waveforms being particularly suitable for this purpose.
6. AFM with filtering and EG looping - New Age 'rhythmic' sounds. Many S+S synths use looped samples to provide cyclic rhythmic accompaniment to ordinary synth pad sounds. By using the SY77's EG loop facilities, similar FM type percussive emulations can be produced and added to other FM pad sounds, producing sounds which are not related to the AWM2 waveforms in anyway, unlike any other S+S synth. Since you have total control over the AFM sound as opposed to a conventional sample, this also avoids many of the problems which can be encountered in some S+S synths with looped repetitive samples ('BambooTrem' on the M1, for example), where you can not change its pitch without changing the repeat rate.
7. AWM2 plus AFM - equivalent to other S+S synthesizers (eg. up to four Roland D50s or two Korg M1's). The SY77's envelope delay can be used to create Kawai K1/K4 type 'evolving' sounds, and LA type scaled PCM samples can be achieved by using micro-tuning.
8. AWM2 plus AFM, with filtering - S+S synthesizers, plus classic analogue synths. This is the most widely used Voice Mode in the preset Voices, because it can provide 16-note polyphony with two separate and distinct sounds - the equivalent of a Combination (or two at once) on a Korg M1.
9. AFM plus AWM2, with AFM as the waveform source - this gives Kawai K5 type additive synthesis with up to 32 harmonics, and eight separate filters. The filters can be used to make broad changes to the harmonics whilst the individual EGs can be used to control the fine detail. Unlike the K5, in this mode, the SY77 has individual envelopes for each harmonic!
10. AWM2 plus AFM, with AWM2 as a modulation source - SY77 only! The possibilities here are huge: from simple AWM2 modulating a single carrier, via AWM2 modulating a whole stack of operators, to complex additive AWM2 structures with the AWM2 modulating several carriers.
11. AWM2 plus AFM, where each is the source for the other - SY77 only! By assigning the AFM output as the AWM2 waveform, and the AWM2 output as a modulator in the AFM, you can create wild digital feedback loops running between the two types of Elements - completely uncharted territory for the brave explorer/programmer!
12. Mono Modes - equivalent to multiple mono synths. Do not overlook the SY77's monophonic modes, which also provide 'fingered' Portamento control instead of the fixed 'follow' mode available in the polyphonic modes. As well as making emulation of real monophonic instruments more realistic, using Mono Mode can help preserve notes when using a sequencer by preventing you from wasting polyphony on melody lines etc. Detuning Elements within a Voice, velocity switching, note zoning and other controls can help produce some very powerful sounds, from acidic basses to flowing solos.
13. AFM with selective Pitch EG control - Ring Modulation. A return to the modular analogue synth, but with the tuning stability advantages of digital. This offers a wide range of possible timbres, from subtle but complex modulation (as used extensively on CS80s by Vangelis) up to complete dismemberment of timbre. A powerful tool when used carefully!
There are still more ways of using the Elements. Unlike the Korg M1, the SY77 can place a huge sonic distance between the original AWM2 samples and the final sounds appearing at the output sockets. The combinations of Elements within Voices can provide a truly astounding scope for emulation of many instrumental sound producing architectures. As with all S+S synths, the combination of two separate sound sources can appear much 'bigger' than either on its own, but in the SY77 this is multiplied further by the interaction between AFM and AWM2, and by the modifications which can be made to the AWM2 sounds. Here's an example of what you can do...
Whilst reviewing the SY77, I spent some time editing and exploring the features (I had no manual at the time, so this was the only way to find out what was happening) and creating some of my own sounds. One of these takes the basic AWM2 alto saxophone sample and splits it into two frequency bands - the high frequencies are enveloped, velocity sensitised and post-processed with an early reflection effect, whilst the low frequencies are used to modulate an operator with a ratio of 0.5 (producing an output one octave lower than the incoming AWM2 sample), and then further enveloping and effects were added. The two sounds were also separately dynamically panned, with the final aural effect being of a 'treated' saxophone split into two different parts, moving around in an ambient space. The end result, 'Russ LoSax', is hardly the sort of sound you would encounter on a conventional S+S synth!
Internal inspection revealed that the SY77 is quite similar to the inside of the DX7 II, with five main printed circuit boards (PCBs, not to be confused with poly-chlorinated biphenols!) in three layers: one containing the switches and other controls on the front panel; the I/O board with all the sockets; the main processor and control board; the DSP board; the power supply (linear, 28W).
The processor board seems to be a typical Yamaha design: a main processor and a subprocessor (the Hitachi 63701 microprocessor continues their use of the 63xxx series products), with a Yamaha 68-pin custom chip, and a Western Digital floppy disk controller. The usual EPROM-based Operating System chips and static/dynamic RAM mix complete the roster.
The DSP board houses 12 Yamaha custom chips: four 40-pin DIL; six 80-pin ceramic surface mount; two 128-pin ceramic surface mount; and eight Toshiba 4 Mbit ROMs for waveform storage.
It is interesting to see that Yamaha still produce their own custom chips - there is certainly a lot of signal processing power in the SY77, and most of it probably resides in these few chips. The mix of surface mount and conventional plated-through holes is typical of current design philosophy - pure surface mount only designs still tend to be restricted to items like mobile phones, where size is most important.
The extrusion and the keyboard itself form the main structural basis for the design, with additional folded steel shielding hidden underneath the plastic end cheeks. The overall construction was neat, tidy and remarkably well finished for a prototype. According to the labels on the EPROMs, the Operating System was a mixture of versions 1.00 and 1.03, whilst the internal sequencer software was version 1.00. If Yamaha follow the usual software practice of using numbers before 1.00 for pre-release versions, then these are probably quite close to the final release versions.
I could not see the components on the I/O board - it was sandwiched between the DSP and the front panel PCBs, and I did not want to pull the SY77 apart completely! What follows is best described as informed speculation...
I assume from the information given in the specifications about a 22-bit Digital-to-Analogue Convertor (DAC) that the SY77 uses one of the recently released 18-bit DACs with four added bits of volume information, in a similar way to the previous Yamaha FM synths and the Roland D50 etc. The audio output was very quiet, with a good signal-to-noise ratio (SNR) - comparable with most domestic Compact Disc players (about 90dB), although the maximum output level (MOL) was not up to the conventional -10dBm standard.
The ever-helpful Jim Corbett at Yamaha-Kemble UK told me that final production models will be about 6dB louder, bringing the specification into line with other instruments. This will also improve the volume in the headphones, which proved adequate under normal conditions but not loud enough for monitoring in a noisy environment.
The highest unwanted signal present in the output was at 88.2kHz and was more than 80dB below the MOL, suggesting that the sample rate is based on the CD rate of 44.1 kHz. Since the clock breakthrough in DAC output circuits is normally at half the sampling rate, this probably means that the SY77 has a four-times oversampling output running at 196.4kHz.
With the Effects section in circuit (the above measurements were taken in Effect Bypass mode) the signal-to-noise ratio was degraded by about 7dB, which indicates that the stated 24-bit precision is probably restricted to the AFM and AWM2 sound generation sections, the effects processors being the usual 16-bit format.
The overall sound of the SY77 is remarkably clean and noise free when it needs to be, and disgustingly dirty when using the distortion! I could detect only the faintest quantisation noise on the piano sample, and the bass sounds in particular seemed much improved over previous FM incarnations. The audio quality is very high and the bandwidth extends well beyond my hearing's upper limit of about 16kHz. In quiet conditions with headphones on, I could not detect any noise in the output, although this might change with the increase in output level on production models. There seems to be some sort of gating at work in the audio output section, although you have to listen very carefully to the reverb tails to spot it.
The SY77 is a formidable instrument. On the surface, the mixture of FM and sampled waveforms may seem to be just another S+S synthesis method, but the strength of FM and the flexibility of the combination make this a synthesis method with wide application, superb sounds, and huge scope for exploration. Realtime Convolution and Modulation is bigger than the sum of its component parts, and it should prove to be a big success for Yamaha in the 90s.
£1999 inc VAT.
Yamaha-Kemble Music (UK) Ltd, (Contact Details).