[sdiy] Generating a large number of CV outputs

[brianw]

Exactly.

My paragraph followed a question about minimal reconstruction filtering, which is what removes images after a DAC (anti-aliasing filters prevent aliasing during ADC). Since I've seen commercial products that failed to remove images of their DAC output, I wanted to ensure that my designs don't suffer the same mistakes.

The challenge is that reconstruction filters can be expensive to do correctly, unless there's significant oversampling (which is expensive on the digital processing side). It's easy enough to plop down a single-pole low-pass, but that's not good enough in nearly all cases.

On the one hand, I'd like to agree that CV does not need reconstruction filtering. In fact, I've gotten into seemingly unsolvable arguments around the design table because DC is *always* below the Shannon-Nyquist frequency, but some folks want to blindly filter anyway. I'm not quite sure of the mathematical terminology to use, but producing a clean impulse response - although equivalent to frequency response - is a very different consideration. i.e. Even if you don't need to reproduce periodic signals with a high frequency, you might get into the same challenges if you want a clean impulse response.

On the other hand, it's so very tempting and useful to generate periodic signal on CV channels, so it's at least beneficial to consider what solutions (oversampling, higher-order reconstruction filtering, or perhaps both) would be best.

Brian


On Dec 10, 2023, at 4:47 PM, cheater cheater wrote:
> brian said you have to record the output at over 44.1 to see it,
> meaning it's above fs/2, meaning it can only be imaging due to a bad
> reconstruction. aliasing from algorithms is below fs/2. case closed.
> 
> On Sun, Dec 10, 2023 at 11:32 PM Mike Bryant <mbryant at futurehorizons.com> wrote:
>> I'd vote for it being digitally generated aliasing - all too common a mistake.
>> 
>> ________________________________
>> From: Gordonjcp
>> Sent: 10 December 2023 21:33
>> 
>> On Sun, Dec 10, 2023 at 11:44:11AM -0800, brianw wrote:
>>> I've recorded enough live electronic music performances to know that there is a surprisingly large number of commercial products out there with digital-to-analog conversion that has insufficient reconstruction filtering, as evidenced by the horrible aliasing. Granted, you have to record higher than 44.1 kHz to see this, since most DAC hardware is running at least that sample rate or higher.
>> 
>> How can you tell that's down to poor reconstruction filtering, as opposed to generated aliasing from the numerical side?
>> 
>> --
>> Gordonjcp