How many taps?
I think the latest Chord DAVE has over 1 million. Chord thinks that # of taps is important. Other companies (mbl, Germany, IIRC), claims there is a "sweet-spot" Goldilocks number.
See this thread:
16x Digital interpolation filter - drive PCM56, PCM58, AD1865 and so on up to 768 kHz
With the Parks-McClellan algorithm you can make filters with a few millionth of a dB passband ripple and 140 dB stopband suppression using a few hundreds of taps, or maybe a few thousands if the transition band is very narrow and the interpolation factor very large, so why anyone would need a million of taps isn't clear to me.
16x Digital interpolation filter - drive PCM56, PCM58, AD1865 and so on up to 768 kHz
With the Parks-McClellan algorithm you can make filters with a few millionth of a dB passband ripple and 140 dB stopband suppression using a few hundreds of taps, or maybe a few thousands if the transition band is very narrow and the interpolation factor very large, so why anyone would need a million of taps isn't clear to me.
Last edited:
Apparently, the reason is empirical: It subjectively sounds good to many people as implemented.
It seems like nobody who has heard Chord Dave and other competing very high end dacs believes Dave produces the most accurate representation of what was recorded, rather they tend to opine that it brings out great detail and sounds exceptionally musical at the same time.
I also felt DAVE is a breakthrough for 44.1K playback. A really good sounding DAC, for sure.
Just for my understanding: are we talking about a DAC with interpolation by a simple integer factor or about asynchronous sample rate conversion? An equivalent number of taps in the millions (most of which multiply their weight by zero and need not be calculated) is not unusual for ASRC, because you try to closely approximate a continuous-time filter (infinite oversampling).
Okay, thank you.
It says Dave has 164,000 taps in the following article: https://www.moon-audio.com/mwdownloads/download/link/id/159/
However, in another article it says a million taps is fed into DAVE from Mk.2 :Chord's Million-Tap Digital Filter | Stereophile.com
Last edited:
Could be (even all together), but you shouldn´t forget to mention that it might be also a matter of sonical reproduction quality if evaluated by human listeners.
It surely depends on the target and the filtering method, if they want to address the 44.1 kHz material, it is a difficult task and if one wants to implement for example the "original" reconstruction method by using a sinc-filter approach (which unfortunately requires to handle "infinity") it surely gets a bit more elaborate.
Last edited:
Sure, if you want to do something in a strange way, then I can see it. There are more highly regarded products that don't have filters with nearly this many taps, so I am inclined to guess it's not some secret weapon of quality that Chord has.
I find bragging about the length of your filter very strange. It's the output that's important, not how many operations it takes. The latency this filter must have is huge. Forget using it with a PC or video playback device.
I'm curious how you found it. I tested many different length of FIR tap by myself, and I found longer tap is not nonsense at all. The difference between 6000 and 60000 should be clearly audible by anyone who can hear the difference between mp3 256 and WAV. The difference between 100,000 to 1,000,000 would be extremely small, though.
A technical advantage of very long FIR filters is that you can make the transition band very narrow. This DAC can probably play back a 22045 Hz sinewave properly at 44100 Hz sample rate, while the much shorter FIR filters in my valve DAC would suppress it almost completely. The halfband filters used in most audio equipment would play it with about a factor of two attenuation and with an almost equaĺly strong image at 22055 Hz. An NOS DAC without filter (except for the zero-order hold) would play it with an attenuation of about 2/pi and with an almost equally strong image at 22055 Hz.
Does anyone know whether the typical CD has useful audio content or mainly aliases between 20 kHz and 22.05 kHz? Hardware anti-alias filters in ADC chips typically don't suppress aliases between 20 kHz and 22.05 kHz well at 44.1 kHz sample rate.
Does anyone know whether the typical CD has useful audio content or mainly aliases between 20 kHz and 22.05 kHz? Hardware anti-alias filters in ADC chips typically don't suppress aliases between 20 kHz and 22.05 kHz well at 44.1 kHz sample rate.
I was wondering about that question a couple of weeks ago too and gave it a short look.
I´ve also noticed the quite low attenuation of ADC chips around and above 22.05 Khz when used with 44.1 kHz sampling rate.
The search wasn´t successful (up to now) - although quite a few studies examining the spectral distribution found on music records, but unfortunately the graphs stopped at 20 kHz or used octave filters where the highest center frequency was at 16 kHz. Interesting stuff including comparisons for the decades from 1950 - 2000, though.
Back in the old days the Philips guys argued that 50 dB attenuation at Fs/2 (i.e. 22.05 kHz) would be sufficient, presumably based on the reasoning that high frequencies in "normal" music are usually already 30 - 40 dB down and that the ADC antialiasing filters provide additional attenuation and the anti-imaging filters during replay as well.
Looking at the graphs of typical antialiasing filters of that era it seems a bit to optimistic (could be that they weren´t able to realize more attenuation being under pressure due to the already postponed start of their cd players).
Todays antialiasing filters are surely better as the ADC are usually using oversampling and for downsampling the mentioned better digital filters could provide more attenuation if done properly.
When recording and editing are done at a higher sample rate and in the end everything is converted to 44.1 kHz using software, it's the quality of the used software that determines how much aliasing occurs - but I haven't a clue how good or how bad professional software sample rate converters are (and whether they do it by software at all).
Marcel,
WRT Dave, yes, very steep. If you check out the Stereophile measurements, the filter certainly lives up to the name 'brick wall'.
WRT CD having content between 20 and 22.05, it depends what ADC is used and what filters are employed / selected. Most pro ADC's historically used 1/2 band brick wall, typically CS5381 or AK5394 on the higher end pro gear.
Nearly all pro recording engineers don't really think about filters, most don't even know what they are. ADC's are generally chosen purely on a subjective basis.
Interestingly I recently read of a pro ADC that was raising quite a bit of interest WRT it's subjective transparency and natural sound. It has selectable decimation filters, almost everyone subjectively preferring the 'short' slower roll off filter.
It's also worth noting that as we move forward, more higher end pro converters will most likely have filter options available to the user in these new AKM based ADC's.
T
Nearly always software. Most engineers will just use the SRC in Protools or a plug in. I do see an increased awareness of various SRC options and the subjective quality offered.
I also think there is a slow increase in awareness of ADC filters, see previous post.
T
I thought pre-ringing was the audiophile devil? There is a large group that claims the very opposite and are saying the best filters are short minimal-phase.
My opinion is that the filters in recent hardware are audibly transparent if there is not a lot of content in the transition band, since they allow for some aliasing to hit the marketing passband specs.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.