Talk:Wavetable synthesis

Definition

Wavetable Synthesis as a concept is aiming to simplify the simultaneous control problem you have when trying to do fully additive synthesis and also address the gross inefficiency of computing many partials in realtime even though the waveform changes only slowly. The solution is simply to define the control trajectory and sample the resulting waveform at (equidistant) spaces on that trajectory, which leaves only a single control parameter (the wavetable index) and a very efficient implementation for the playback mechanism. Wavetable synthesis as defined by PPG pre-computed to single-cycle (odd symmetrical) waves and added the interpolation between waves in a table as a mechanism to further save the computation and storage of unnecessary data. The wavetable is completed and the waves expanded to full-cycle in RAM on the PPG, so these are really details of the implementation.

So, the PPG definition of wavetable synthesis is: a table of 64 single-cycle waves addressed by a single parameter (the wavetable index). Scanning the wavetable with an appropriate index sequence produces the desired timbral changes over time. The index sequence does not need to be continous nor monotonic (in contrast to sampling, where the index would be strictly linearly ascending). Later incarnations of the technique have provided smooth interpolation between wavetable entries (i.e. you can use fractional index values) and reduced artifacts (aliasing, quantization and scanning noise) inherent to the implementation details of the original implementation. -- 87.170.196.95 (talk) 19:45, 6 January 2008 (UTC)[reply]

???

I think declaring Wave's "jump from one wave from to another" system as inferior to modern samplers is somewhat disingenuous; for one thing, PPG's system had to work with the limitations of late 1970s/early 1980s technology, and memory was expensive then. With more sample memory and looser limits on sample length, you can build a single sample that has attack, loop and decay stages instead of jumping from one digitized single-cycle to another, and you can also use multiple DCOs to stack sounds together.

Sample based synthesis is really something else entirely. There are no samples in the sense a ROMpler uses the term in wavetable synthesis.-- 87.170.196.95 (talk) 19:45, 6 January 2008 (UTC)[reply]

Also, the page does a poor job of saying why the PPG and others like it are different from samplers; it also leaves out mention of non-PPG systems (except for a token mention of Roland's LA system), and the paragraph bashing sound cards toward the end is muddy and borderline POV. I can't sort all this out myself, but I hope someone who knows better can. -lee 21:28, 20 Jun 2005 (UTC)

Roland LA does use a non-looped sample for the attack and a looped sample for the sustain phase, which has no direct relation to wavetable synthesis. -- 87.170.196.95 (talk) 19:45, 6 January 2008 (UTC)[reply]

Pronunciation

Is it pronounced as "wavetable" (as in "edible") or "wave table"? - 99.247.136.86 07:35, 12 November 2007 (UTC)[reply]

The latter: it is a table consisting of waves. -- 87.170.196.95 (talk) 19:45, 6 January 2008 (UTC)[reply]

Wavetable vs Vector Synthesis

Does anyone know the difference? I'm not too clear on vector synthesis, and there is no entry..although there should be! I'm inclined to think there's not much difference...would that be incorrect?

Yes it would be incorrect, although you can conceivably combine the two. Vector synthesis uses n-dimensional parameters to navigate to different locations in the parameter space. As implemented in the ProphetVS the boundary of that parameter space is defined by four single-cycle waves and the navigation is done by linear interpolation between these four waveforms. Other than the fact that the four waveforms are single-cycle, there is really no relation to wavetable synthesis. -- 87.170.196.95 (talk) 19:45, 6 January 2008 (UTC)[reply]

Wavetable synthesis is not PCM sample playback.

i'm happy for the article to be cleaned up, Lee, but not for the meaning to be changed to what is essentially PCM sample playback or Sample-based synthesis for which articles exist. one of the most annoying misappropriations of a technical term was what the commercial sampling keyboard industry (E-mu) and chip and soundcard manufacturers (Creative Labs and many others) did to jazz-up the term "sample playback" (which is not really any kind of "synthesis" at all) to the sexier "wavetable synthesis" when such a term already had a meaning and use that was not equivalent. probably some of the early references to Computer Music Journal regarding wavetable synthesis and group additive synthesis should be made in an article like this. but other than to dispel the notion that wavetable synthesis is PCM sample playback, i would not wish to see any confusion written in that would cause people to think these methods were the same. r b-j 02:00, 21 Jun 2005 (UTC)

Okay. I hadn't known about the Computer Music Journal articles, and coming from the world of PC soundcards myself, it's somewhat confusing to read that the term is somehow wrong or inaccurate. I do think it's being pedantic to strictly separate PCM playback and wavetable synthesis when the techniques are related so closely; indeed, even the Wave itself could play straight through a sample if told to, and the Waveterm add-on had a sampler in it. I'm not saying that the techniques are identical, by any means, but that one can emulate the other pretty easily given the right programming, and while E-mu's use of the term may have been inaccurate, I think it could also be stated that early samplers just didn't have the level of sophistication to emulate a Wave (whereas pretty much any synthesizer in 2005 can at least try). Also, I think a comparison with frequency modulation synthesis would be interesting, since (from what I've read) the original Wave's sounds and capabilities were akin to what the Yamaha DX7 could do. -lee 01:10, 27 Jun 2005 (UTC)

hi Lee. i s'pose you could say that a sampling keyboard could "emulate" any synthesis technique by either sampling the output of the other synthesizer or by, in an "off-line" process creating the same waveform in a memory buffer and, during the real-time playback, outputting that buffer when a Note On event is received. but i would hardly call that "synthesis", even though sampling keyboards can often be very cool instruments. i like to compare a sampler to the old Mellotron. a Mellotron is a sorta analog sampler.

"Wavetable Synthesis" as the term was originally used (these old CMJ articles from the late 70s or early 80s aren't online that i know of, but if you go to the AES site and look at the table of contents of the latest journal, you'll see an article by Andrew Horner which is about wavetable synthesis as used here) has at least one issue in common with PCM sample playback synthesis and that is about clocking through the samples at a different rate than one wavetable sample per output sample and interpolating between samples since that ratio is normally not an integer. but there really are fundamental differences. like a Numerically-controlled oscillator, there is only one cycle of the waveform stored in the wavetable, but it is normally not a single cycle of a sine wave. it's the complex waveform of some musical instrument (or some completely synthetic musical waveform). also there are more than one wavetable that are used to make the note evolve in time, from the attack through sustain and the release portions of the note.

coincidentally, here the term "sample" has a sorta dual meaning that can lead to confusion. when i use it, i mean the electrical engineering DSP definition: a single PCM data point. a single number representing a physical quanity (usually a voltage) at a particular instant of time. the other use of the word is that a "sample" of a note or a sound is the whole note or sound, usually a buffer or a block of memory or file space with many tens of thousands of PCM samples. the common, but misappropriated, usage of "wavetable synthesis" is the playback of that buffer at different rates, depending on which MIDI note is hit.

E-mu and Ensoniq sampling keyboards do the latter and Waldorf and the old Prophet VS ("vector synth") does the original kind of wavetable synthesis.

The ProphetVS does not do wavetable synthesis. It has four static NCO, and between two pairs of them a linear interpolation is done according to the X-Y-Vector parameter. -- 87.170.196.95 (talk) 19:46, 6 January 2008 (UTC)[reply]

my problem (and that of a few others that have worked on or with the "real" wavetable synthesis) is that while there are at least two other accurate and descriptive terms for sampling "synthesis" or sample playback or whatever you want to call it, there is only one good term for the wavetable synthesis that we're talking about here. no other term is as concise and accurate for it, but it's confusing when that "trademark" has been (mis)appropriated by soulless marketers of soundcards for something that is functionally different. r b-j 03:01, 27 Jun 2005 (UTC)

Okay, that makes sense. One of the things that was confusing me at first was the difference in terminology between PPG's/Waldorf's manuals and your paper; PPG uses just plain "waveform" for the waveform data (and to them, a "wavetable" is a list of how those waveforms play back over the duration of a note, what everyone else calls a patch), and your paper uses "wavetable" for the waveform data.

I also see your point about soundcards, and in my latest edit, attempted to offer an explanation; most of the soundcards in question were only ever meant to play back General MIDI files, and since GM only specifies a list of names (with no insight as to how to generate the sound), most people took the easy way out and used one long PCM sample per note (with maybe a keyboard split for higher/lower registers), even if the underlying hardware was capable of much more than what the manufacturer's Windows drivers let you do with it. On cards without a built-in MIDI controller, like the Live!, you can go directly to the DCOs and do pretty much whatever you want with them; having multiple waveforms per note is easy enough then (using the "wavestacking" method you describe), though it would most likely require custom programming to work. -lee 04:09, 27 Jun 2005 (UTC)

Confusing opener...

Being someone who has owned and worked extensively with analog, FM, wavetable, LA, samplers, ROMplers and vector synthesis machines, I feel it a disservice to lump or (perhaps) to even mention comparisons to Rolands LA synth engine with relation to a true WaveTable synth. They are really very different methods with their own strengths and uses.

Though it is true that one can somewhat mimic the other (as can be done with many methods of sound synthesis), the actual process of creating sounds is gone about quite differently altogether. A true wavetable synth can allow for both subtle and extreme effects that would be utterly impossible to create using Roland's LA system. Here are some additional problems below:

"Different wavetables can be used for the attack phase (at the beginning of the sound, as the volume increases) and release phase (at the end of the sound, as the volume decreases) of the sound. These are normally very difficult to synthesize with other synthesis techniques, but because these are stored as samples very realistic sounds can be produced with little processing power. The Roland Corporation's series of "Linear Additive" synthesizers such as the D-50 made use of a combination of digitally sampled attack phases."

This statement has numerous problems in addition to the unnecessary mention to Roland. What is difficult to synthesize? An attack phase? A release phase? Most wavetable synths sound anything but 'realistic' (accoustic would be a better term here) and it's rather clear from their general design that they were not intended to emulate such sounds - particularly when compared to the D-50 or a number of ROMplers on the market.

"Because single samples are somewhat limited for synthesis of new sounds, modern wavetable synthesizers can combine multiple samples or even change the sound with filters."

The very definition of a wavetable synth combines multiple (albeit very short) samples. Any synth featuring filters can alter it's sounds with them - this is not specific to a wavetable synth. More importantly, it is HOW a wavetable synth "combines" samples (animates would be more accurate) that gives them their unique qualities.

"Because of the low processing power required early synthesizers imitated filters and other expensive synthesis methods by rapidly playing successive wavetables in sequence."

Early synthsizers were analog (not digital or with CPUs) and therefore not wavetable synths at all. Early synths did not 'imitate' any process - they either did something, or they didn't. Most synths had analog filters, though some very early ones had none whatsoever. This statement could be helped by stating that this only pertains to digital-waveform based synthesizers - some of which may have lacked analog filters due to the expense of adding them into the sound chain. But this was certainly not limited to wavetable machines - and to the best of my knowledge, all the wavetable synths I'm aware of until very recently actually HAD analog filters. What they did lack was variable pulse-width waveforms, which could be easily emulated using a wavetable.

There are some other minor problems as well, but I feel I've made my point. There is a general lack of clarity in the opening statements. I could re-write much of this, but feel it would be foolhardy to do so without first pointing out some obvious problems and getting feedback.

Antfactor 22:38, 15 September 2005 (UTC)[reply]

i think nearly everything you've written here is right on. i am much less familiar with all of the market so i have not commented on products even though some i know do true wavetable synthesis such as the Waldorf and predecessors and the old Prophet VS. i suspect there are sampling keyboards listes as wavetable synthesizers and that is something i would like the article to get away from. r b-j 00:01, 16 September 2005 (UTC)[reply]

Actually, that "low processing power" bit was something I was working on a while back...what I meant to say was that PPG's own systems had to do it that way back then because the technology available back then didn't allow anything more complex (in particular, Palm wanted to use wavetables to replace filters, but couldn't get crossfading to work so used an analog filter instead). I don't remember if I wrote it, or if someone before me did. As for the other problems, I can go ahead and clean those up; the D-50, from reading about it, looks more like my Casio CTK-611 at home (which definitely is a ROMpler) than a wavetable synth. -lee 19:12, 9 January 2006 (UTC)[reply]

It was actually Wolfgang Dueren (sales manager and later a founder of Waldorf GmbH) who insisted on the inclusion of analog filters. The early PPG Wavecomputers didn't have filters. -- 87.170.196.95 (talk) 19:45, 6 January 2008 (UTC)[reply]

Lee, we gotta work on this. For some reason, even though this is on my watchlist, i missed these last changes you made which have the effect of moving the definition of wavetable synthesis back to the PCM sample playback definition which was usurped by chip manufacturers (like Creative Labs) and E-mu as a fancy term for sampling and it is not the same as what the original use of the term was, which is more along the lines of Waldorf or Palm (i think) and the old Prophet VS. We gotta work this out because I really do not want to consign the term "wavetable synthesis" to that term was usurped from its original meaning to something else not quite the same. r b-j 01:40, 3 February 2006 (UTC)[reply]

Hmmm. I was under the impression that the rewrite I did to the '"wavetable" sound cards' section made the difference more clear. Maybe it would be better if the sound cards were not mentioned at all? -lee 02:22, 5 February 2006 (UTC)[reply]

Also, what was wrong with the beginning? Most of the rewrite I did there was directly from PPG's own documentation for the Wave 2.3, and someone else stated that the D-50 doesn't really count here (it's a halfway approach I've seen Sound On Sound calling 'S&S'), so I made sure to edit that out. -lee 02:25, 5 February 2006 (UTC)[reply]

Ah...I went back and read all the documentation I could find for the Wave 2.2/2.3 (and found Hermann Seib's excellent WaveSimD on the Waveterm C site), and it seems that I've been missing the point...it's not the samples, it's how the CPU can move through the table and fill in waves that don't actually exist in ROM (that Palm could do that on a 6809 is pretty amazing, but then again the Fairlight designers used the 6809 as well and it's quite a bit more involved). I see what you meant by the additive synthesis bits, as well; most of the tables move from a simple wave to something far more complex, almost like pulling out the drawbars on a Hammond. There's even a few that are just function generators, completely done in software. This thing is absolutely mad, and it's no wonder I couldn't understand it at first. -lee 04:40, 5 February 2006 (UTC)[reply]

do you mean by "function generator" an ADSR thing or do you mean a (static) waveform producing thing? r b-j 04:45, 5 February 2006 (UTC)[reply]

Static waveform producing thing (I'm used to the old-style function generators used in analog electronics, which aren't all that different from analog synthesizers). The PPG docs call that particular wavetable "phased saw waves", and the table I was reading said that particular list of waves was calculated by the CPU instead of being stored in ROM somewhere. -lee 07:45, 5 February 2006 (UTC)[reply]

Also, what was wrong with the beginning? Most of the rewrite I did there was directly from PPG's own documentation for the Wave 2.3, and someone else stated that the D-50 doesn't really count here (it's a halfway approach I've seen Sound On Sound calling 'S&S'), so I made sure to edit that out. -lee 02:25, 5 February 2006 (UTC)[reply]

okay, let's work on this a little. here is the last version before the last revert - that is your intro with some other small changes made by third parties:

Wavetable synthesis is an algorithm used in early digital musical instruments (synthesizers) to produce sounds based on either a natural sound (such as a piano or a guitar), or pre-determined waveforms like the ones generated by analog synthesizers. The sound of an existing instrument or function (a single note) is recorded and parsed into a set of samples having one period or cycle each. These samples or waves are then combined together using lists called wavetables. Upon playback, the wavetable controls which waves to play and where to crossfade or loop. Special effects can be achieved by selecting a wave at random, sweeping through all of the waves in a table in sequence, changing waves on a tempo-tick from a sequencer, or using the modulation wheel, LFO or envelope generators to jump through the wavetable. This creates effects that are very "digital"-sounding, similar to FM synthesizers like the Yamaha DX7.

A patch can consist of a single wave (producing the same effect as an analog synthesizer), or an entire wavetable composed of many different waves. Early wavetable synthesizers had all of their wave samples referenced in order in their wavetables and could only sweep or seek inside a given table, while later ones (such as the Waldorf Wave) keep the wave samples and wavetables separate, making it possible to create new tables from existing sounds.

i might make these initial changes in the first:

Wavetable synthesis is an algorithm used in early digital musical instruments (synthesizers) to produce sounds based on either a natural sound (pre-existing instruments such as a piano or a guitar), or pre-determined waveforms like the ones generated by analog synthesizers. The sound of an existing instrument or waveform (a single note) is recorded and parsed into a set of wavetables having one period or cycle of the waveform each. The waves generated from wavetables are then combined together by some defined means, possibly mixed or possibly sequentially with cross-fading. Upon playback, the synthesizer controls which wavetables to play and how and where to mix with or crossfade to other wavetables. Special effects can be achieved by selecting a wave at random, sweeping through all of the waves in a table in sequence, changing waves on a tempo-tick from a sequencer, or using the modulation wheel, LFO or envelope generators to jump through the list of wavetables. This creates effects that are very "digital"-sounding, similar to FM synthesizers like the Yamaha DX7.

you see, the wavetable is the memory buffer that contains one cycle or period or "wave". e.g. the Prophet VS had 4 wavetables each could be arbitrarily defined but they all had 128 samples. the list or collection of these wavetables another term (i believe Korg might have called that a "wave-sequence", but i am not sure that the WaveStations were wavetable synthesizers in this sense or if they were more like samplers. it looks to me that this usage from PPG is calling a "wave" what us technogeeks call a "wavetable" and the list or string of waves, they are calling a wavetable (and i think the term i would use is a wave sequence or wavetable list or similar). r b-j 04:45, 5 February 2006 (UTC)[reply]

From what I can tell, the Korgs were the follow-ons to the Prophet-VS, and one page I found refers to the list of waveforms as "wave sequences" (and says the Wavestation could string together up to 255 waves and chain sequences together for even more). And yeah, PPG did indeed call the sequence lists "wavetables"; that's one of the things that was confusing me when I first found this page. It'd be good to say something about the usage of the term in the article (and I did write something like that in the edit I was working on before I re-read this page). -lee 07:15, 5 February 2006 (UTC)[reply]