Commitment to quality
Muon Software Ltd is keen to maintain its reputation for delivering software synthesisers with the highest possible sound quality.

There are special challenges encountered when trying to emulate an analogue synthesiser digitally. The quality of a software synthesiser depends on how well those challenges are met.

Challenge #1 - Aliasing
Sampling theory tells us that it is possible to recreate an exact copy of a signal by sampling the amplitude of the wave at discrete intervals and then playing back through a special low pass reconstruction filter. This is what your sound card does to play back a wave file.

If the gap between samples is too long (i.e. sampling rate or frequency is too low) the signal will be changing faster than we are sampling and we will miss some information.

The diagram on the left shows the original signal (yellow line) and the effect of sampling at too slow a rate (white markers). When played back through the reconstruction filter, the output waveform (green line) is at a lower frequency than the original.

Aliasing sounds extremely unpleasant. In an aliased musical tone, the harmonics in the sound lose their musical relationships to each other and the sound becomes atonal. This is because some harmonics will be mathematically related to the sampling frequency and not to the fundamental frequency of the note. With non-musical tones the effect is less noticeable.

Aliasing can occur in many kinds of digital sound generating and processing algorithms. In a software synthesiser, the most likely place aliasing will occur is inside the oscillator section.

Muon has developed a number of technologies to stop aliasing. The effectiveness of these techniques can be demonstrated with the use of a spectrogram. This kind of diagram shows time (across the horizontal axis) frequency (vertical axis) and amplitude (brighter colour). We generated the spectrograms shown on this page in Cool Edit Pro.

This diagram shows a Muon Electron playing a steady sawtooth wave for a couple of seconds. The coloured lines represent the harmonics in the sound (from bottom to top). Noise in the signal would show as faint patterns in between the lines. Electron’s noise floor is better than -96dB!

This diagram shows the same tone, but this time modulated with the LFO set to a depth of +/- one octave and a rate of 1hz.

As you can see, as the upper harmonics approach the top of the chart they fade away. This is what should happen!

This is the sawtooth wave, at the same pitch, from a competing software synthesiser launched recently.

As you can see. the lines of the harmonics are fuzzy and there is noise clearly visible inbetween them.

When the LFO is activated, this becomes an almost perfect example of aliasing in action.

As the upper harmonics (top of the picture) approach the highest frequency possible, they alias and “bounce back”. This sounds awful!

Challenge #2 - Calculation Accuracy
Computers have a reputation for being extremely accurate in mathematical calculations. As any scientist will tell you however the way floating point (fractional) maths is represented in binary leads to inaccuracies in the results. The more bits you use to represent your floating point fractions, the more accurate the results.

This diagram shows the first 25% of a sine wave cycle at two very different bit depths.

Although this is exaggerated so you can see more clearly on this small diagram, quantisation affects the accuracy of calculations throughout a software synthesiser and degrades sound-quality significantly

Some commonly-used DSP chips in hardware (especially those with Motorola 56000 processors) utilise only 24-bit integer accuracy. This gives a potential signal-to-noise ratio of 144dB. 32 bit floating-point DSP (as used in most software synthesisers and digital audio workstation software is roughly equivalent.

Muon uniquely use 64-bit floating point for all internal calculations within their DSP algorithms. This extra headroom for calculation accuracy means that Muon synths work at a higher level of accuracy than the host software itself! The extra headroom, when mixing voices for example, means that subtle details are not lost, and the overall noise floor is much lower. The result is a cleaner, brighter, sharper sound.

Challenge #3 - Audio Rate Modulation
Software synthesisers with complex modulation routing possibilities like Electron and DS404 have to perform many thousands of calculations before finally outputting a single fraction of a second of sound. Each oscillator, filter, LFO and envelope generator in the synthesiser has to be calculated individually, and this places a strain on the CPU of the host computer.

One of the most common tricks to reduce the CPU load of a software synth is not to calculate modulation sources at the output sampling rate. Instead, the LFOs and EGs are calculate much less  often (referred to as the “control rate”), thus saving CPU. Reducing the control rate can mean that the output of envelopes and LFOs can become “stepped”, and take longer to respond.

The alternative to a low control rate is a more efficient algorithm within the synthesiser. More efficient algorithms take less time to compute and can therefore be performed more often in a given time.. Muon’s technology is so efficient we can calculate the output of every single component in the synthesiser (for all running voices) each time an output sample is created. This is called audio-rate modulation and gives sharper, faster envelopes and a more convincing analogue emulation.

Here we’ve greatly exaggerated the effect of a low control rate on a filter sweep so you can see more clearly how it affects the sound. Instead of a smooth envelope, it has become a “stair step”.

(source: Reaktor)

 

This is a similar filter sweep recorded from CM101.

The filter closes perfectly smoothly as the envelope is calculated at the output sample rate - 44,100 times per second in this case!

If you have any questions regarding software synthesis, talk to our resident experts anytime at 64bit@muon-software.com

All text, images, programs and multimedia data associated with this website are © 2002 Muon Software Ltd and may not be copied, altered or reused in any form. All rights reserved.

Cubase and VST are registered trademarks of Steinberg Media Technologies AG. All other trademarks are the property of their respective owners and are referred to here for informational purposes only.

The technical performance characteristics, features and prices of products featured on this website may be subject to change without notice. Purchases made through this website are subject to our standard terms and conditions of business (available on request).

All software made available here requires the user to accept the terms and conditions of the license included in the download package. It is not permitted to redistribute ANY of our products in any form without prior written permission.

Contact the webmaster