Questions - Fundamental differences in Analog and Digital synths OR Intrinsic problems in each

[Aiynzahev]

Hi guys

I've come to realize that I only have a very basic knowledge of how synthesizers actually work, especially digital ones.

My question for now (while I do research) is there any short summary you can give of the basic, intrinsic problems with Digital syntheizers?

I know some of the Analog synths problems: Noise floor, the expense of components, heat, stability. Lots of these have been overcome in time.

For example, one intrinsic problem with EQ is warping near nyquist. Thats the kind of thing I am trying to find out.

What is the difference between digital and analog envelopes for example in how they work. And how are waveforms generated differently?

I am not able to phrase this very well so it would help if someone chimed in then I could build on that.

Regards

Sami

[antto]

fundamental differences?
one is analog, current flows thru a circuit and all kinds of things happen
the other is "virtual" nothing flows, it's all split in individual samples

[Aiynzahev]

fundamental differences?
one is analog, current flows thru a circuit and all kinds of things happen
the other is "virtual" nothing flows, it's all split in individual samples

It's probably too much of a loaded question to be answered here. That part I gather, its every detail about those two different processes I am wondering about!

Probably way too much info to get some condensed answers.

[antto]

why do you ask such a question?
is it for some homework? or do you just wanna know for yourself?

[Acid Mitch]

A google search will reveal a wealth of knowledge on the subject.
Some of it more in depth than most of the replies you'll get here.

[V0RT3X]

Wikipedia has some good articles

http://en.wikipedia.org/wiki/Digital_synthesizer

http://en.wikipedia.org/wiki/Virtual_analog

or are you asking about the actual programming?

[mystran]

Frequency warping is largely a non-issue actually. For LTI cases you can compensate the transfer function and for non-linear cases aliasing tends to dominate anyway (eg 4x oversampling largely hides frequency warping problems).

Aliasing is bigger issue; in the worst case (eg oscillators) you pretty much have to use special algorithms to construct band-limited results. In other cases you might get away with some oversampling, but oversampling has it's own problems too (higher CPU cost, resampling filters add phase-shift). Instant feedback loops can also be problematic (well, at least costly in terms of CPU).

[Aiynzahev]

Thanks guys.

I just want to know for myself.

Every now and then a developer gives some insight into what problems had to be overcome and it fascinates me.

How is a sawtooth waveform actually rendered for example? And why are some better than others?

I might be over my head at this point.

[Music Engineer]

How is a sawtooth waveform actually rendered for example? And why are some better than others?

maybe look at this thread:

http://www.kvraudio.com/forum/viewtopic.php?p=3298389

it's mostly the presence or absence of aliasing which makes the difference in quality. but there's more. in analog synths, the sawtooth often is not really a mathematically ideal sawtooth but has some other sawtooth-alike shape (mostly, it's an RC-loading curve which is reset periodically). moreover, the actual shape of the waveform itself changes as function of the pitch, which may (or may not) be an desirable attribute of the sound. the same is true for other waveforms.

[Music Engineer]

For example, one intrinsic problem with EQ is warping near nyquist.

...

Frequency warping is largely a non-issue actually. For LTI cases you can compensate the transfer function [...]

well, yes - but you have to actually do it and the most basic EQ algorithms (i.e. RBJ cookbook, which i assume to be widely used) don't. so, i'd say - it is an issue but one that can be solved.

[stratum]

well, yes - but you have to actually do it and the most basic EQ algorithms (i.e. RBJ cookbook, which i assume to be widely used) don't. so, i'd say - it is an issue but one that can be solved.

And.. I was wondering why my graphic eq band pass filters do not overlap well near nyquist... hmm

[wiwatti]

Hi guys
I've come to realize that I only have a very basic knowledge of how synthesizers actually work, especially digital ones.
My question for now (while I do research) is there any short summary you can give of the basic, intrinsic problems with Digital syntheizers?
Sami

Once very very common problem is that people how are using some digital synths think that they are using some analog stuff (because it's hardware you know). Actually they are using some gear with a very poor quality (usually a Roland Machine based on wavetables, made of low Q samples) and a very poor DA converter. They claim to to use some real analog gear but actually they are using some sample-based stuff with a standard of 2001 ( 16 bit- 44100 ), but it's better because it's a gear, yep a fluckin damnn gear....

So for me the basic problem with digital synths is that they are often some poor wavetable stuffs. I'm particularly mocking all the douches with a Roland MC from the 90's and such things... Most of the time they even don't know that their boxes are based on wavetables and samples.

But technically if a synth has to emulate a sound, a digitial synth will always fail
- the starting of a waveform is not the same as its ending
- the expression cannot be emulated on a keyboard

And that's why a synth has a particular playing style ( you know the mono leads like in hard rock etc). This is just the big massive problem with synths...when someone tells you that it doesn't sound lie the "real stuff"...then you are with a douche

Oh and sorry I know I'm a bit OT

[Aiynzahev]

How is a sawtooth waveform actually rendered for example? And why are some better than others?

maybe look at this thread:

http://www.kvraudio.com/forum/viewtopic.php?p=3298389

it's mostly the presence or absence of aliasing which makes the difference in quality. but there's more. in analog synths, the sawtooth often is not really a mathematically ideal sawtooth but has some other sawtooth-alike shape (mostly, it's an RC-loading curve which is reset periodically). moreover, the actual shape of the waveform itself changes as function of the pitch, which may (or may not) be an desirable attribute of the sound. the same is true for other waveforms.

Thats the kind of thing I was looking for.

It is interesting to me to read of the problems inherent in digital V
a synth design and to find out about their work arounds.

Thanks but anything more like this is useful. Especially about envelopes and voices.

[antto]

well IMO digital synths can deal with envelopes easier
most of the times analog envelopes are made out of RC segments i think
the same thing can be done digitally, and even taken further with response curves and other madness that would make a very expensive (or impossible) analog variant

tho, in many digital synths this might be compromised
recalculation of coefficients (like oscillator frequencies, filter cutoff/resonance and the-like) is expensive, so quite a number of devs probably do not recalculate them per sample but less often
sometimes it might be due to other factors like filters which explode when cutoff is changed too frequently (per sample)

[antto]

voices? it's obvous..

in an analog synth - each voice is basically physical copy of the same circuit, thus multiply the cost of each voice by the number of voices and you get a very expensive synth

on the digital synth it's easy, it's usually

Code: Select all

while (i < num_voices) { ... }

the big dissadvantage of digital synths (especially those which we run on our PCs) is the CPU consumption
because the more complex the algos are, the more CPU power they need
and sooner or later your CPU cannot cope with it, and it is no longer realtime.. then you increase your buffers (aka latency)

hardware digital synths are probably meant to cope with whatever they are running, so the latency is probably pre-determined, and you should never get into such a situation

analogs rule here, no CPUs, everything is "real" and almost "instant"