This emulation is about the 1984 Yamaha CX5M and
1985 Yamaha CX5M-II. (Almost identical)
The CX5M incorporates a classic Yamaha FM sound
process, using 4 operators and 8 algorithms.
It also includes the Yamaha-type feedback, which
routes the output of an operator to its own input.
Referring to Fourier Series, an important feedback
will generate a Sawtooth spectrum, but with
important Gibbs distortion.
Due to some approximations, the feedback effect is
hard to reproduce. SynthEdit – that was used to
develop this software - cannot run instant feedback
loops, so I had to use a special third party module
from Elena Design.
This module uses Lagrange polynominal cubic
interpolations, so we can’t obtain a perfect sine.
In FM, every single harmonic plays an important role
as it sometimes modulates to high coefficients other
Yet as at the beginning we don’t have a perfect
sine, the result differs from a true DX-type
Though I tried to correct the feedback spectrum
using butterworth filters, that better fits the
Anyway, because of this issue, some differences can
be heard when comparing this software with an
original unit, as feedback rate goes over 50%. This
specific module isn't available for Mac, so that's
why this software is Windows-only (And 64 bits).
The original CX5M runs on a 3.58 Mhz processor, in 8
bits. As a result, strong aliasing can be heard on
I have recreated this distortion the very same way
it sounds on an actual CX5M. Since it runs on 8
bits, the envelopes are not perfectly smooth. On
long attacks/releases, you can hear steps – this was
recreated in this plugin too.
Regarding envelopes, the CX5M had an ADSDR outline,
that I changed to ADSR + Delay, to my mind more
The CX5M also provided a weird HF
attenuator/envelope shortener parameter, that I
changed into a low pass gatted filter, as the
original idea was too complicated for pretty useless
The volume adjustment parameter was deleted (it
wasn’t even used in factory presets).
Same for a FM noise generator thing on OP1 – it
wasn’t used again (And I never succeeded to activate
it – but the results are identical to max feedback
Finally, LFO was simplified from a useless
distinction of PMD/PMS and AMD/AMS.