[sdiy] Electrosmith Eurorack DSP platform

Eric Brombaugh ebrombaugh at gmail.com
Fri Jan 28 17:34:49 CET 2022

Because I'm using a sigma-delta approach to generating the PDM signals, 
there is some noise shaping going on and this system performs better 
than merely 12MHz/48kHz bits. I'm actually getting SNR that's 
approaching 16-bit performance.

I found the FPGA-only ADC approach fiddly because it suffers from a lot 
of the same nonlinearity issues seen in the PDM DAC methods, is very 
sensitive to layout and component selection and requires a fair bit of 
post-filtering to reduce noise. It's kinda like the singing dog - 
amazing that it works but not what you'd call "good".


On 1/28/22 08:54, cheater cheater wrote:
> Thanks. 12 MHz 1-bit is equivalent to 48 kHz at 8 bit. What do you
> think would be necessary to bump this up to add 4 more bits? Seems
> like using a 1-bit approach you'd need an FPGA able to output at 200
> MHz... at which point layout becomes a problem. Do cost effective
> FPGAs with that capability even exist? (never mind availability)
> What made the ADCs fiddly?
> On Fri, Jan 28, 2022 at 4:24 PM Eric Brombaugh via Synth-diy
> <synth-diy at synth-diy.org> wrote:
>> On 1/27/22 23:20, cheater cheater wrote:
>>> On Thu, Jan 27, 2022 at 11:00 PM Eric Brombaugh via Synth-diy
>>> <synth-diy at synth-diy.org> wrote:
>>>> I was seeing harmonics down slightly better than 80dB with all the
>>>> mitigation measures active. Without it was more like 60-65dB. Not quite
>>>> as good as even an inexpensive sigma-delta I2S audio DAC, but a few SMT
>>>> passive parts beat that for cost & simplicity and I2S DACs don't support
>>>> 12MHz sample rates.
>>>> Eric
>>> Thanks for the write-up, that's some great info. The github repository
>>> was very helpful learning a bunch more about your approach. Really
>>> nice of you to share, and those are some interesting results. I've had
>>> a few questions if you don't mind.
>>> Regarding the results - really impressive. Is that 12 MHz at 1 bit?
>> Yes.
>>> What was the performance at DC?
>> Didn't measure that - I don't see any reason it wouldn't be equivalent
>> to the AC performance though. The back end is fully DC-coupled.
>>> How many channels can you run on that ICE5LP4K? Assuming the data
>>> comes from somewhere else and you want DC+Audio rates at enough bit
>>> depth for a synth signal (pre-VCA/ENV which reduces required bit rate
>>> but, say, it could be post filter, or any other module)
>> This particular FPGA (iCE40UP5K in the QFN48 pkg) doesn't have a lot of
>> I/O pins, so you'd be limited in that regard. Two pins / channel and
>> roughly 32 uncommitted I/O so 16 chls. For more you could go to other
>> families with larger pincounts - the cheapest ECP5 parts are available
>> with 197 pins free, so 98 channels max, but then you'd likely be limited
>> by the internal resources and would need to bump up to higher-capacity
>> devices.
>>> Do you think the same method could be used with a comparator or with a
>>> saw oscillator to create a similarly inexpensive and good ADC?
>> I've tried a few different types of FPGA-based ADCs - either with
>> sigma-delta techniques or with PDM or charge-timing methods. So far I've
>> found them to be fairly fiddly and not performing as well as inexpensive
>> dedicated devices like the SPI ADC I used in the oscillator project. It
>> may be possible to get good performance from such approaches but it's a
>> trade-off in R&D time. For me it was simpler/easier to just buy something.

More information about the Synth-diy mailing list