rePhase file attached.
Used a low noise floor windowing and adjusted centering for IP as suggested by Pos, otherwise nothing special.
EDIT: Also used same roll off values for LP and MP.
IP filter you mentioned has similar tradeoff (higher stopband floor). I hope somebody actually tests this for image attenuation and other criteria. I don't have any means to do so and time to set up one at the moment.
I'm testing the filter with mostly small band acoustic and electric music with plenty of acoustic or electric bass, drummers using cymbals a lot, clean guitars, female vocals, brass instruments, piano and electric keyboards and synths. Not tested with heavy guitars yet. And still too much albums to listen for a final verdict...
Of course, this verdict will be for my system and ears
Very excited that people are sharing and testing these.Cheers.
Last edited:
Interesting, as:
MB2b: dam1021,44100,2,1,1016,1
is not a valid option. Maybe it result in something close to non oversampled, but I have no clue....
Why IP for 44kHz sources.
OK, a post explaining my interest in IP filters. Statements are based on scientific claims. They may help somebody beginning their journey on this thread. My conclusion may be BS.
1. We distinguish timbre of different instruments mainly by "attack" of the sound envelope. Daniel Levitin mentions (1) a research where test subjects had difficulty distinguishing piano from guitar when attack portion of the sound was removed. Too long pre-ringing may affect attack of the envelope negatively. This eliminates maximum phase filters and warns about linear phase (LP).
2. We are more sensitive to pre-masking than post-masking in a phenomenon called temporal masking (2). This means we can tolerate more post-ringing than pre-ringing. This gives minimum phase (MP) and advantage over LP.
3. Duration of ringing is affected by the steepness of the roll-off curve. Longer durations affect items 1 and 2, and also transients negatively (3). A slow-roll off is desirable for shorter ringing. But as everybody knows DA artifacts also need to be attenuated. By how much? This should depend on person's hearing and listening environment (headphones, speakers, acoustic treatment, etc.). Although there should be a limit to this since we share the same evolutionary hearing system. Also manufacturers that adverise use of slow roll off filters are likely to deal with this after the dac by other means. We don't have them on dam1021, so we need to deal with this by filters. This implies using filters with steep-ish roll off and with low stopband floor. Unfortunately, too much post-ringing caused by this type filters may also render MP only filter undesirable.
4. We are trying to get best out of 44.1kHz sources. This will always be a compromise. There are many publications about benefits of high sample rate (both processing and hearing wise). I have to settle on a filter and move on.
5. [Non-scientific] Different types of filters said to perform better for different types of music and recording environment. My personal interest is to settle on one that is true to the timbre of instruments as much as possible, then transients, and finally headphone listening.
Hence my preference of IP for 44.1kHz as an optimisation between pre and post ringing given the necessary roll-off steepness. I also found the phase related experiment I mentioned earlier useful.
For the IP filter I'm using at the moment, I will try to bring down the stopband floor lower to at least -80dB at Nyquist while keeping ringing around 1ms.
Cheers.
PS: When configuring a filter for higher sample rates than 44.1kHz, is it just a matter of using correct number of taps Soren stated? Do we need any other adjustment to the filter itself? Thanks.
(1): Daniel J. Levitin, This is Your Brain on Music: Understanding a Human Obsession
(2): Auditory masking - Wikipedia, the free encyclopedia
(3): Music and the Human Ear
OK, a post explaining my interest in IP filters. Statements are based on scientific claims. They may help somebody beginning their journey on this thread. My conclusion may be BS.
1. We distinguish timbre of different instruments mainly by "attack" of the sound envelope. Daniel Levitin mentions (1) a research where test subjects had difficulty distinguishing piano from guitar when attack portion of the sound was removed. Too long pre-ringing may affect attack of the envelope negatively. This eliminates maximum phase filters and warns about linear phase (LP).
2. We are more sensitive to pre-masking than post-masking in a phenomenon called temporal masking (2). This means we can tolerate more post-ringing than pre-ringing. This gives minimum phase (MP) and advantage over LP.
3. Duration of ringing is affected by the steepness of the roll-off curve. Longer durations affect items 1 and 2, and also transients negatively (3). A slow-roll off is desirable for shorter ringing. But as everybody knows DA artifacts also need to be attenuated. By how much? This should depend on person's hearing and listening environment (headphones, speakers, acoustic treatment, etc.). Although there should be a limit to this since we share the same evolutionary hearing system. Also manufacturers that adverise use of slow roll off filters are likely to deal with this after the dac by other means. We don't have them on dam1021, so we need to deal with this by filters. This implies using filters with steep-ish roll off and with low stopband floor. Unfortunately, too much post-ringing caused by this type filters may also render MP only filter undesirable.
4. We are trying to get best out of 44.1kHz sources. This will always be a compromise. There are many publications about benefits of high sample rate (both processing and hearing wise). I have to settle on a filter and move on.
5. [Non-scientific] Different types of filters said to perform better for different types of music and recording environment. My personal interest is to settle on one that is true to the timbre of instruments as much as possible, then transients, and finally headphone listening.
Hence my preference of IP for 44.1kHz as an optimisation between pre and post ringing given the necessary roll-off steepness. I also found the phase related experiment I mentioned earlier useful.
For the IP filter I'm using at the moment, I will try to bring down the stopband floor lower to at least -80dB at Nyquist while keeping ringing around 1ms.
Cheers.
PS: When configuring a filter for higher sample rates than 44.1kHz, is it just a matter of using correct number of taps Soren stated? Do we need any other adjustment to the filter itself? Thanks.
(1): Daniel J. Levitin, This is Your Brain on Music: Understanding a Human Obsession
(2): Auditory masking - Wikipedia, the free encyclopedia
(3): Music and the Human Ear
Maybe some info to pick here about digital filters ?
I don't like too much this webzine think it's biased, but mostly here in this review the Chord engineer is talking about him HUGOTT DAC and some choice about digital filters strategy : too complicate for me ! Maybe could help, dunno ! 6moons audioreviews: Chord Hugo TT
I don't like too much this webzine think it's biased, but mostly here in this review the Chord engineer is talking about him HUGOTT DAC and some choice about digital filters strategy : too complicate for me ! Maybe could help, dunno ! 6moons audioreviews: Chord Hugo TT
The dam1021 first FIR1 filter have 352.8K/384K sample rate as output, this is fixed in hardware. So when having 44.1K as input you per definition can only do 8 times oversampling.
There is no reason to want 2x oversampling at 44.1K input, except to save processing, which is not an issue in the dam1021 as it can do 1016 taps at 44.1K input rate.
So the oversampling rate have to be 8 when doing 44.1K/48K input, 4 when doing 88.2K/96K, 2 when doing 176.4K/192K and 1 when doing 352.8K/384K.
OK. Now one wonder what is going on with the MB2b filter in reality
Is best bet:
MB2b: dam1021,44100,8,1,1016,1
But would not the output be very low?
//
Depends on the coefficients, look at them.... I have noticed that most filter generators don't seems to take into account zero insertions, so the coefficients peaks at around 0.125, you then need the multiplier by 8 to scale them to 1....
x->x0 at 2x41kHzNo but 1->10 as it works with 0 instertions - and 88khz. But what Sören say, it seems undefined
I wonder in which dimension?
//
is the same as
x->xxxx0000 at 8x41kHz,
at least if that would be the last thing before output.
If you want to "continue with 82kHz" you have to make sure that you do the same operations on any four consecutive timeslots.
You can bring up John if you like...
John is a smart guy, but you have to pay close attention to what he says, and the context in which he says it. You also really need to keep an eye on his underlying assumptions, or you can easily be mislead.
One of his posts linked here was interpreted as saying that SoX is superior to chip based filtering when what he was saying - painting in very broad strokes - was that all DAC chips use cascaded filters, and that applying filtering in a single pass was audibly superior. As you point out John uses a FPGA to provide a 1200 tap FIR filter in his own hardware DAC, so not earth shatteringly different to the 1016 taps DAM1021 uses at x8 44.1.
cheers
Paul