Nope... realized quickly that this is mostly a problem of signal processing which falls firmly in EE... I’m more of a math and CS person...
Is there actually potential to create an all round better filter than the 4K stock linear? I recall someone said the custom filters each only work well for a specific type of frequency/timbre. That’s fine and dandy but is there potential to create something actually better in everything, I.e. theoretically better? I know it may be an impossible goal as even MSB seems to have multiple filters for different purposes....
Actually on a second thought, it would be really awesome if we can just figure out which filter works best when, e.g. one for string instruments and another for vocal, though there are certainly subjective differences which would require every person to experiment for himself to get the best results. But I wouldn’t be surprised if we can arrive at some strong conclusion applicable to most people 😉 which would be super super great.
Signal processing is math - so it should be right up your alley!
Math don't do distinction between strings and vocal. Nor should a good D/A.
//
Math don't do distinction between strings and vocal. Nor should a good D/A.
//
Absolutely DSP is math - an FIR filter is performing a time-domain convolution. Agree that a good DAC will do strings and vocals well without the need for a change of filter in between.
Hmmm. Maybe I should take a deeper look!
Ideally yes but that should be much more difficult than creating something that specializes, intuitively. MSB said something about the existence of different ways of guessing the true waveform based on the quantized information, which would intuitively justify the comparative difficulty in improving absolute D/A performance than improving a certain aspect of it (maybe at the cost of others). I think it would be really cool and useful either way.
That’s probably way too high level an argument to be useful. I’ll look into more details but probably won’t have time to make a ton of progress in the next couple of weeks...
Last edited:
I'd be very wary of taking MSB's marketing blurb as giving you an accurate picture of what's involved in a DAC, notwithstanding they do make good-sounding boxes (at least so I believe, I've not auditioned one). In particular there really isn't any 'guessing' going on in filtering the quantized information, the interpolation involved in oversampling is very tightly constrained by the need to band-limit the incoming signal.
You're right. And it seems super intriguing... Though my research is in CS and has little to do with continuous signals directly, this might be a lot of fun in itself... Did you know Fourier transform has to do with group theory?🙂 This might turn out to be too distracting...
Thanks! That was a great explanation... Alas, I thought MSB is better than this...
I would be interested in measurements of the stock 4k linear and minimum phase filters vs the best 2k filter. Especially WRT impulse response.
Isn't the skr format known, so a txt file could be extracted back from it?
Isn't the skr format known, so a txt file could be extracted back from it?
The skr file is a binary file. I have never seen it described.
What is the best 2k filter(s)?
//
What is the best 2k filter(s)?
//
Last night began taking measurements of some filters and almost immediately noticed something weird: the filters in post 1342 by Oneoclock and in post 1379 by Bambadoo appear identical. Am i being silly, or just mistaken?
https://www.diyaudio.com/forums/digital-line-level/269776-filter-brewing-soekris-r2r-27.html#post4924291
https://www.diyaudio.com/forums/digital-line-level/269776-filter-brewing-soekris-r2r-28.html#post5113505
https://www.diyaudio.com/forums/digital-line-level/269776-filter-brewing-soekris-r2r-27.html#post4924291
https://www.diyaudio.com/forums/digital-line-level/269776-filter-brewing-soekris-r2r-28.html#post5113505
Perhaps they are both made with rephase using almost identical parameters? BTW: I have only made filters using 44.1khz and 48khz samplerate. The rest are copied in the .txt file and compiled.
Does this play? Only 44.1 by me (and rePhase). Its for 1.20 FW (4k).
Warning - I could not try it as my DAC is not available here...
//
Warning - I could not try it as my DAC is not available here...
//
This is what i thought after looking at the impulse response, but no, did bit by bit file comparison: identical. There is also the possibility i've done something wrong...
Just to clear things up, rev 1.19+ firmware can use 1.05 filter files, and filters created to that specs, the 4K taps are max number of taps.
Interesting. I have had problems with the filters mentioned above by oneoclock/bambadoo. It seems they have abnormally high gain which leads to clipping at 0db in 1.19 but no clipping under 1.06. Does this make sense?