[sdiy] Roland Alpha Juno DCOs

[Russell McClellan]

After reading the very interesting conversation regarding the new
novation peak synthesizer, I was made curious by something Richie Burnett
mentioned in that thread regarding the Alpha Juno oscillators.  This
led to some further investigation on my part and some may be
interested in hearing about the results.

I had always (incorrectly, it turns out) thought that the alpha juno
oscillators followed the same basic topology of the Juno 6, 60, and
106 synths - which, for those unfamiliar, is a complicated
digital/analog hybrid; basically an analog ramp wave hard-synced to a
digitally generated pulse.

However, it turns out that the Alpha Juno has a completely different
design, with much less of an analog component.  There is a custom
"DCO" chip which has 6 independent digital oscillators.  Each
oscillator takes the 12MHz master clock and divides it down by a power
of two based on the note being played.  For the highest notes the
effective clock is 6MHz, and for the lowest notes this seems to be
divided by 2048 to form an effective sample rate of 3kHz.  On each
divided clock, an increment is added to a 16 bit accumulator (I'm sure
the accumulator is at least 13 bits, but I'm not sure of the exact
number of bits).  Then, the top 8 bits of the accumulator are sent to
a digital waveshaper, and then to what looks like an R-2R based DAC
on-chip.  This signal is then sent straight to the fully-analog
filter.

In general, the output is "pretty" clean - certainly there's no frequency
drift since it's based on the crystal.  There are aliasing artifacts,
but since the sampling rate is always so high compared to the note
they are usually fairly quiet (I noticed some audible aliasing on the
lowest notes).  Since the DAC is only 8-bits, quantization noise is
also an issue.

Anyways, I was really stunned to learn that the alpha juno had so much
digital horsepower, and a bit surprised that I haven't seen this
oscillator design before.  Does anyone know of any synths with a
similar approach? (Other than the new novation, which seems to run at
a fixed sampling-rate, but is otherwise similar)

I think it would be a fun project to create a similar oscillator using
an FPGA and an R-2R dac - perhaps with some additional capabilities
like inter-oscillator FM and a sine lookup table shaper.

Thanks to Richie for mentioning this in the other thread and for
helping me off-list to investigate this.

Thanks for your time,
-Russell