Anyone wish to take a stab at this: what is the difference between what is called a 'brickwall filter' in d/a conversion, versus digital and analog filterless dacs like the 47 labs? Does this mean they don't have the brickwall filter?
1. Brickwall filter
Can only be implemented as a digital filter. Usually used with synchronous oversampling. You cut off all information from the the digital signal after c. half the (over)sampling frequency. Pitfalls: ringing - with a simple single pulse the pulse will start to sound before it is played, because the filter "sees the signal before it is played". Most implementations also ring after the pulse is played. Some call this smearing in the time domain. One could say that this implementation is more correct in the frequency domain.
2. Filterless DAC
You don't cut off (or even attenuate) the alias images predicted by Nyquist theorem by filtering after c. half the sampling frequency. You get alias images right after half the sampling frequency and multiples of that. With CDs these mean ultrasonic noise in the analog signal. If implemented properly, you don't have ringing like with digital brickwall filters, but you do get ultrasonic noise. One could say this implementation is more correct in the time domain.
So it's either frequency or time artifacts. Pick your poison.
More technically adept may correct any misunderstandings or inaccuracies I have made.
Your explanation is fine, except brickwall filtering **can** be done with an analog filter. In fact, the first Sony players used analog brickwall filtering. The major difference is that analog filtering is the "ringing" occurs only *after* the signal transitions. (Square wave traces with an analog brickwall filter look a lot like those from an underdamped phono cartridge.) There is also claims the average phase error for analog brickwall filters is worse than that with "sinc" interpolation digital filters. (The analog filters don't "interpolate"- the ringing is from the natural "high order" characteristic of a filter that requires a sharp frequency cutoff.)
Those analogue filters were hardly brickwall, only being
The ringing seen with digital filters is partly irrelevant because
the single impulse signal used to demonstrate it is on its
own an illegal signal in the sampled digital domain: it is not
possible to construct a real-world analogue signal that after
band-liminting and sampling manifests itself as a single impulse
in the digital domain.
"Those analogue filters were hardly brickwall, only being 6th-12th order."
What is the "order" requirement for an analog filter to be qualified as a "brickwall" filter? Is there a standard for such qualification?
I think the order of most analog filters were even higher anyway. I don't think there were many analog filters in early Redbook players that were only 6th order.
"The ringing seen with digital filters is partly irrelevant because the single impulse signal used to demonstrate it is on its own an illegal signal in the sampled digital domain: it is not possible to construct a real-world analogue signal that after band-liminting and sampling manifests itself as a single impulse in the digital domain."
You've indirectly described the reconstruction limitation of "sinc" interpolation.
I think the ringing is **very** relevant to our perception of the music off a Redbook CD- With typical digital filters, it illustrates the "time smear" (remnants of the transient both before and after the transient) of "sinc" interpolation, the predominant filter method in Redbook playback. And I personally can hear this quite easily when A-B'ed with either an analog signal or a "time correct" filter such as the Wadia "spline" interpolation.
The ringing IMO illustrates how corrupt the sinc interpolation algorithm is in the *time* domain, which I personally consider more-relevant to our perception of the music signal than how well the filter performs in the *frequency* domain.
The dirty secret of "sinc" interpolation is the FR "flatness" is from an "averaged" response, and the HF energy providing the "flatness" is in the form of *ringing*. (It's basically "cheating" to get a flat FR.) The rolloff of non-sinc filters is due to the fact such ringing is either not added or added to a far-lesser degree, hence the *averaged* frequency response appears to be "rolled off." But if you look at the signal in the *time* domain, the non-sinc filter is far closer to the pre-digitized signal.
Proper reconstruction by the convolution of a train of Dirac-like
samples with the Sinc(t) function, in other words PERFECT reconstruction, when viewed at with a single impulse, like you do,
would also show ringing. And yet it would be perfect, as it follows
Shannon's theorems completely. Mind, it can not be realised, it only
exists in mathematics.
Sinc(t) is the timedomain manifestation of the perfect brickwall
filter, the boxcar function. Sinc(t) rings like hell, but this
ringing is necessary to reconstruct the original bandlimited signal.
What appears to be 'better' temporal behaviour of Wadialike
filters is just the presence of (parts of) images above fs/2.
If there are real problems with digital filters at all, then these
stem from their necessary truncation, in time and in resolution,
of the Sinc(t) they try to approximate.
If reconstruction is necessary for fidelity, then the way is faster DACs, more oversampling, and much longer digital filters.
If reconstruction is not necessary, accepting that our ears do that job, then only a gentle analogue filter is required to protect
amps/speakers from the raw DAC output.
The intent of the brickwall filter is to filter out spurious signals above the sampling half-frequency (22.05 kHz), which consist of reflective "images" of the digital signal in the ultrasonic band. I've never listened to a DAC which does not filter out these images, so I cannot comment on it sonically, but I'd speculate the time resolution would be superior (most filters use a "sinc" interplation, which smears time resolution), but long-term listenability would most-likely be worse. Since ideally, the alias images go all the way into the RF band, which is IMO the single-worst artifact of digital sound reproduction in the first place. And no filtering means a direct dose of RF from the DAC's output. (As opposed to indirect from RFI permeation via the common ground.)
I think the filter-less DAC *could* be the ticket with electronics which naturally filter out RF artifacts. But I would not use such a DAC in an all-solid-state high-resolution system.
No one uses Brickwall filters these days , so you have to compare modern filters
Brickwall filters would allow all signal under a certain freq to pass and then attenuate TOTALLY above that , thats actually impossible to do and leads to serious problems , thus there is oversampling and MUCH gentler filters are used.
Technically , filterless or non oversampling DACs are *incorrect* - however sonically , some folk swear by them (same as Valves which are "technically inferior" to solid state , yet sound wonderful
I think you're confusing filters for Redbook CD playback with those for higher-resolution digital media- Your statement is true only with the use of higher-resolution media.
One misconception I've seen about upsamplers is it enables the use of "gentler" filters, but that is only true past the upsampler- The upsampler *itself* does the "brickwall" filtering. (This is why an upsampler does *not* gain anything relative to "older" interpolation methods.) Using a "gentle" filter *without* the brickwall filter with Redbook playback will allow the near ultrasonic alias images to pass through to the resultant analog signal.
The non-filtering DAC's are only "incorrect" in regard to passing relevant signals (more-accurately- frequencies) from the initial recording session.
Personally, I'd like to see some company attempt to use **new** filter algorithms that I personally think would be more-faithful to the original pre-digitized signal. "Lanczos3" is one algorithm I would *love* to see at least tried for digital audio. If the resultant audio signal from this interpolation is superior in the way it is with pictures and video, this would quench the need for a higher resolution digital medium than the maligned Redbook CD. (Unless RFI is addressed with digital playback, I think Redbook will be around for quite a long time anyway.) It would combine most the advantages of both classic "sinc" interpolation and "filterless" applications, without most of the disadvantages of these two implementations.
Fall in Capetown?
Actually pretty much all digital filters in use (be they 'over'
or 'up') employ brickwall filtering at the source signal's
fs/2. There are a few exceptions, notably the Wadia/Pioneer/...
and the BB DF1704 in slow-rolloff mode.
Whether filterless DACs are incorrect can be argued. Theoretically
proper reconstruction requires convolution with the Sinc function.
Digital oversampling filters approximate this, some better than others. Analogue output filters don't approximate this at all.
But with most source material available at 44.1kHz, and fs/2 at
22kHz, one might wonder if the human ear itself is not the dominant reconstructor. In other words, the additional components in the output signal put there (actually: left there) by theoretically incorrect reconstruction all are
sitting above 22kHz, and if we accept that we only hear up to 20kHz at best, then filterless DACs are not incorrect.
The 47 Labs has no brickwall filter after the DAC chip, actually there is no digital filter at all. The I/V conversion is done by a resistor.
Brickwall filters are steep cutoff digital filters. Because all DAC chips throw off out of band noise, some people feel the use of a filter to cutoff such artifacts is necessary.
Others, like Audio Note, 47 Labs, and others before those companies and currently, feel that the digital filter is the root of CD sound problems and either use passive filtering or no filtering at all, leaving the out of band information to be filtered naturally by your ears.
Yamaha had a 8x oversampling CD player in the late 80's that allowed the user to turn the digital filters on or off. Does anyone remember this piece? I think it was a CD 910 or something like that. I listened to one extensively at the time and thought it sounded better with the filter turned off. I can't recall any other company ever offering this feature.
Anyone else seen this on a machine?
This post is made possible by the generous support of people like you and our sponsors: