Do you really trust the manufacturers that, prior to Wolfson? introducing this concept to the market, had not a single word to say about filter sound other than one was a short delay filter?
You don't think there is a marketing reason for this? For 10 years everyone gave up development on high-end DAC ICs after AD1955 and PCM1792/4 hit the market, until ESS basically reignited a paper spec war.
Also- note that the more conventional (reputable?) manufacturers of such DACs like Cirrus, TI, AD, etc. make no such claims. Cirrus offers five different modes on their newest DACs and could but declines to make any comments on sound quality.
The AKM datasheets also have no real comments about the sound quality of said filters.
Sometimes, the tail wags the dog.
Last edited:
Not exactly. A brickwall filter is a idealized model, no perfect brickwall filter exists. One can freely download data sheets for top of the line AKM 'velvet sound' A/Ds can D/As. Graphs of anti-aliasing filters for A/Ds are shown in the data sheets. For D/As there is much less filter information, but the numbers that are given kind of suggest they might look a lot like the A/D filters. You might find the information interesting if you take a look.
One thing to keep in mind is that modern high performance 24-bit A/Ds and D/As are typically oversampling designs. It means if the nominal sample rate is 44.1kHz, either type of data converter actually runs at several times the nominal sample rate and with much fewer than 24-bits. That being the case, analog anti-aliasing filter requirements do not include the need for analog brick wall filters. Since data conversion is done at a much higher sample rate than 44.1kHz, digital filters can be used for most of the anti-alias filtering as a part of the decimation process down to 44.1kHz sample rate (for the case of A/Ds). For D/A conversion, 44.1kHz CD digital audio data is upsampled to several times the CD sample rate during which or immediately after which interpolation filters are typically used to provide digital anti-alias HF image attenuation which should relieve much of the analog output filtering requirements.
So, hopefully it is starting to become more clear that most of the so-called brickwall anti-alias filtering is now done digitally, and in the case of one type of dac we have been talking about, those are the seven different filter choices we have in a Sabre dac that all sound different from each other. It is also the external FPGA anti-alias filter in Benchmark DAC-3 that accounts for a lot of its improved sound quality over typical Sabre dac designs, IMHO, since as a result of the external filter if does not need to use any of the seven built-in Sabre filters.
Last edited:
It would be interesting to duplicate the major brands (Sony, Panasonic, Yamaha etc) propriety noise shapings into the list of filter choices available. All filter/noise shaping curves will cause noise floor dynamic modulation and different frequency dependent phase delays...... the 'majors' have made claims of reducing the audibility of playback errors (signal noise) by shifting this noise away from mid band where it is subjectively strongly noticeable. Different frequency dependent phase delays are a byproduct of this type of processing and causes another set of subjective differences.
System ULF and VLF stability affects bass nature very strongly and is probably the major component of 'voicing'. As I replied above, noise shapings could cause VLF phase shifting that changes 'sense of power' and stability/solidness in the foundation of the reproduced sound. Upscaling in the Amplitude domain and non integer Upsampling in the Frequency domain are both artificial processes that derive 'apparent' increase in signal resolution BUT at the expense of time uncertainty.
Mark I think you quoted 211kHz sampling for the DAc-3...any reason why this specific frequency ?.
Dan.
Fits my aged ears.
Don’t have to worry about those “brick wall filter effects”
Small things like cartridges.
How many up to now Bill?
George
By "brickwall" I don't mean necessarily a 1980's analog filter, rather the Red Book requirement of 22 KHz anti-aliasing. However it's done, a brickwall must exist before sampling. This seems to me to be one of the few things not symmetrical between A/D and D/A, and entropy increases here.
I'm way behind the curve on what's the best of current playback (D/A) topologies (they all sound jest fahn to me) but I just wonder if the whole system A/D/A could be improved by knowing the specific details of the pre A/D (anti-aliasing) filter. At the D/A end of things, could this knowledge be used to back out of some part of their effects?
This question passes the "innocent eye" test, but what do cows know about music?
Thanks, and all good fortune,
Chris
I have met AKM's golden ear. he is really serious and not bound by either corporate requirements or classical measurements. He is a nice guy and has strong opinions on what sounds right (can you imagine. . .).
All of the different filters have rationales. Some are to support applications where low latency is important (movies, live sound etc.), others are different selections to meet someones desires. Cambridge Audio launched this 30 years ago with filter selections on their DAC. I suspect most users listen and select what sounds right or never explore.
One sample impulses create all types of ringing etc. 3 sample impulses seem pretty benign. I'm not sure how you could ever record a 1 sample or even a 3 sample impulse. The anti-aliasing filters etc. would prevent it.
Before getting too wound up in phase shift in the electronics look at how coherent or incoherent the waveforms are in real spaces. At shorter wavelengths (above 250 Hz) in any real space they will get pretty mangled. Exploring impulse testing of speakers makes this pretty clear, still its pretty easy to recognize almost any instrument in a 100 piece orchestra in a hall with 1000 other people. If it were critical most speakers would be unlistenable (some would say they are. . .).
The research I have seen suggests a stable phase relationship even if its seriously shifted is not audible. Changing phase relationships dynamically is quite audible. There is a classic video test for this; differential gain and differential phase, which would show these issues quickly. Been discussed here long ago. Never went anywhere since it did not show 8 legs bad. Not sure about no feedback amps however.
All of the different filters have rationales. Some are to support applications where low latency is important (movies, live sound etc.), others are different selections to meet someones desires. Cambridge Audio launched this 30 years ago with filter selections on their DAC. I suspect most users listen and select what sounds right or never explore.
One sample impulses create all types of ringing etc. 3 sample impulses seem pretty benign. I'm not sure how you could ever record a 1 sample or even a 3 sample impulse. The anti-aliasing filters etc. would prevent it.
Before getting too wound up in phase shift in the electronics look at how coherent or incoherent the waveforms are in real spaces. At shorter wavelengths (above 250 Hz) in any real space they will get pretty mangled. Exploring impulse testing of speakers makes this pretty clear, still its pretty easy to recognize almost any instrument in a 100 piece orchestra in a hall with 1000 other people. If it were critical most speakers would be unlistenable (some would say they are. . .).
The research I have seen suggests a stable phase relationship even if its seriously shifted is not audible. Changing phase relationships dynamically is quite audible. There is a classic video test for this; differential gain and differential phase, which would show these issues quickly. Been discussed here long ago. Never went anywhere since it did not show 8 legs bad. Not sure about no feedback amps however.
Mark --- How does one know which DAC filter to use that corresponds to same type/characteristics in recording ADC?
If the DAC builder does not know the ADC filter characteristics used, the DAC people will provide several filters for you to find closest to what had been used in ADC.
THx-RNMarsh
If the DAC builder does not know the ADC filter characteristics used, the DAC people will provide several filters for you to find closest to what had been used in ADC.
THx-RNMarsh
Last edited:
John, you are welcome. Hope you find someone to work with on a JC signature dac one of these days.
DAC-3 only has one filter. I agree with Demian that there are different reasons for different interpolation filter choices when they exist. Sometimes one or another filter can sound better depending on the particular dac board implementation. In that case fixing any problems with the board may change which filter sounds most 'right.' The goal isn't really to undo the effects of a decimation filter in the ADC or somewhere else along the way. Ideally, if there was some true reference dac at NIST or somewhere, perhaps dac filters could be calibrated for accurate reproduction. As it is now, we don't have any way of knowing exactly how accurate an ADC was which used to digitize audio, and once digitized there is no way to know exactly what reproduction should sound like when converted back to analog. Of course we can synthesize very accurate digital test tones, but modern dacs are so complex with multiple less than ideal interactions within and between subsystems, I would hesitate to trust simple synthesized test tones as sufficient for the purpose. Interesting conundrum, in a way.
Last edited:
The DAC3 has one filter -- they make an ADC also. So, assuming you used their ADC, they only need one filter in thier DAC. If other ADC is used, other filter would be needed.
We need to know the GD of the filters. I would choose the one with lowest GD.
THx-RNMarsh
It doesn't matter unless you are doing live monitoring, they are linear phase filters. It's constant. They only offer one, because a linear phase sharp roll-off is the right filter to use. The fact that it's "their" ADC wouldn't even help because their ADC uses AK5394A which has its own built in filter which cannot be defeated. Regardless, just about every piece of information you could want on the ADC filter is in the AKM datasheet. It's also a linear phase FIR filter.
Last edited:
What if a lot of the dac sound is because of insufficient attenuation of aliasing in its filter? Improving the anti-aliasing filter in the ADC is unlikely to cancel that filter problem.
Arf! Yes, the whole mindset where we start with illegal values seems wrong-headed to me. Given enough computation time, only the A/D/A system "curves" matter. Isolated D/A is currently operating blindly, unknowing of the pre-A/D filters. And designing based on illegal values is just goofy.
But, would designing based on the actual pre-A/D (anti-aliasing) filters give a better reproduction? IOW, could a playback (D/A) benefit from detailed knowledge of the pre-A/D filter's characteristics (given whatever computing juice needed etc.) ?
Maybe more theoretical than engineering, but folks transfering old 78-ish records are doing much more interesting stuff than this.
All good fortune,
Chris
"Before"? Where have you been? Almost every time the conversation is gently moved towards recording, mixing, speakers, rooms and all the real world nonlinearities, people resist and say things like, "don't matter, that's different", really? How so?
Maybe you're shooting the messenger, but this thread is broad enough to include systemic issues, surely? Wade into the deep end; there's lots here. And don't call me Shirley.
All good fortune,
Chris
Selling them far faster than buying them (luckily). I need to cull the MM collection a little once I decide which ones work best with with arms I have.
The downside of this is that I like the SME mount Signet Mk112E on my hand me down kenwood so much that I hardly use my my better turntable any more. I make no claims for accuracy of this combo, just a measure of pure late night pleasure with a glass of port. Total cost to me of that front end runs around £150.
In the case of 44.1 kHz that essentially always means it is " brickwall filter" just because people wanted to get a bandwidth up to 20 kHz +-0.1 dBr.
Philips speculated (while using oversampling to get better numbers for the cd-players despite the 14 Bit DA-Convertors) that ~50dBr attenuation at the Nyquist frequency of 22.05 kHz would be sufficient.
An assumption that seems to be justified when recording real acoustical instruments, due to the spectral distribution of the sources.
But a brickwall filter prior to sampling isn´t needed in the case of high sampling rates, as a much gentler filter type will do.
Downsampling to 44.1 kHz presents the "brickwall" filter problem again,due to the short transitionband from 20 kHz - 22.05 kHz. An offline filtering solution might be much better for the task.
In fact most people knew already back then that 44.1 kHz wasn´t the best choice; 48 kHz sampling frequency is already easier to realize (might be the reason why in modern ADC datasheets you´ll often find diagrams starting with 48 Khz as lowest sampling frequency, although 44.1 kHz is of course supported) .
As stated before, while that is true, nobody ensures that something of this kind is available even on a audio CD, for the same reason as nobody prevents us from using a one sample impuls on a CD for measuring purposes.
When working with music samples in the digital domain it would first be up to the software programmers to prevent such "nyquist violations", but we knew from the intersample overs example that it happened nevertheless.
Re-read yourself, Destroyer OS!
You use the same terms and methodology as those you criticize. That surprises me from you. ;-)
An attitude, usual in our forum from those that I call "objectivists", that consists of not checking the subjective observations that are reported by others with a clear bias against them and a final value judgment on their honesty (voluntary or not) or their ability to listen.
To "believe" that our scientific knowledge is such that it covers the whole field of pshycho-acoustics? (Anti-scientific attitude of believers of a new religion they call "science".)
The first thing to do is to observe by ourself to check the observation that is reported.
If we can not reproduce it, we can then question first our own listening ability or the performance of our system. And report the negative result of our own observation by modestly specifying that it is only valid for ourself in the conditions in which they were made.
The second thing to do is to check whether the explanation given by others to the phenomenon or to the solutions they bring does not contradict scientific knowledge that has been verified a thousand times. If so, we can talk about snake oil. Otherwise the subject remains open and our own opinion is not worth more than that of those who are criticized.
Do not make me say what I do not say, I have no conclusions on this subject of the "SOUND of the DAC", that personally, I consider as open. Especially since the few measures that I have been able to do clearly show differences in the measurements of square waves between different performance DACs close to the paper, and that, by myself, I noticed very slight differences in the way they reproduce music. Slight enough to not to put this problem in head of the things to improve on * my personal system *.
Last edited:
Stay there and don’ t move an inch.
If you would make a claim for accuracy for a combo at any price, you think that you would be able to provide any proof when asked for?
Just a measure of fitness to purpose for a vinyl playback combo.
All right but still no answer to the question
George
- Status
- Not open for further replies.