Electron – Tau Pro oscillator question.

[Rabid]

I’ve read a few threads about the analysis and design put into the Tau Pro oscillators. Are the Electron oscillators based on the same concept or code?

Robert

[Muon Software Ltd]

Tau Pro's oscillators are alias-free waveshape oscillators. They use a bandlimited wavetable, hermite interpolation and 2x oversampling (the oversampling filter used is a 12th order polyphase filter with >100dB rejection). They rock, in other words

Electron on the other hand uses a technology better suited to a polysynth that needs to make the classic analog sounds. The oscillators use a technique called BLIT-SWS (bandlimited impulse train sum of windowed sinc) to create a basic pulse waveform without aliasing. This waveform is then mathematically integrated to produce the normal saw and square waveforms. The resulting noise floor and aliasing in Electron's output is well below -96 dB and there are no interpolation noise artefacts at all.

You can find out more about the technology used in Muon synths at:
http://www.muon-software.com/html/muon_quality.html

Kind regards
Dave

[helium]

As we talk about quality. Is it realy usefull to use 64Bit internaly. With 32Bit float the smales value that can be represented is -749,3dB. As the result is normaly rendered to 24Bit integer (-144dB) or even 16 Bit (-96dB) the quality of 32Bit floats is enough, isn't it?

[Muon Software Ltd]

Actually, it is normally accepted that the dynamic range of a 32-bit VST plugin (since the values are normalised to a range of -1 to +1) is roughly equivalent to 24-bit PCM audio.

The extra dynamic range afforded by using 64-bit is useful when mixing voices and performing amplitude scaling etc. It doesn't have any negative affect on performance, so it doesn't really make sense *not* to use it.

Regards
Dave

[Borogove]

The extra dynamic range afforded by using 64-bit is useful when mixing voices and performing amplitude scaling etc. It doesn't have any negative affect on performance, so it doesn't really make sense *not* to use it.

Good point; I'll just point out that if you're using wavetables, delay lines and other buffery things, that 32-bit floats are probably worthwhile there to reduce memory access bandwidth and cache violence. I could also ramble on about the compiler's tendency to optimize away temporaries, such that many of your calculations will automagically be done in 80-bit floats...

[Muon Software Ltd]

and other buffery things, that 32-bit floats are probably worthwhile there to reduce memory access bandwidth

Couldn't agree more. If you're using a lookup table to store the value of some calculated function, it's going to be quantised to N steps anyway and so holding the values at full precision isn't likely to do you a lot of favours.

Regards
Dave

[andrewsimper]

Hi Dave,

I like your synths and think they sound great. I am a little confused, however, at how you can actually do what is quoted from your tau pro web page:

64-bit oscillators with 11 waveforms & 2x oversampling for totally alias-free sound. Width modulation, sync & ring mod on all waves too!

Alias free ring mod is possible as you are double sampling you can contain the possibly double frequency generated (assuming the polyphase downsampling filters are post ring mod).

But how can you calculate alias free sync with a sine wavetable?

What about alias free pwm? (you can get up to 3 times nyquist with phase modulation if that's how your doing it).

Perhaps extreemly low aliasing is more accurate?

Andrew

[Muon Software Ltd]

Perhaps extreemly low aliasing is more accurate?

At what point does "extremely low" become "alias-free" though?
assuming aliasing is kept to below -100dB, and you render to a 16 bit output (dynamic range 96dB) there won't be any aliasing in the output. Either way, its just semantics.

Regards
Dave

[helium]

I checked what you said and it's true. 32Bit floatingpoint has exacptly the same quality as 24Bit integer (2/Epsilon_of_32BIt_float = 16777216 = 2^24)

and 64Bit floatingpoint would be 53Bit integer so it's realy worth using 64Bit.