A question on anti-aliasing filter

[adrianywu]

With Redbook CD standard, the Nyquist limit is close to the audible frequencies and a steep anti-aliasing filter is required. I wonder how they approach this with high rez PCM ? Do they use a steep filter close to the Nyquist frequency, or use a gentler filter slope at a lower cutoff frequency ? Are there qualitative differences in these approaches ?
Some LPs are now cut from high rez digital sources and the mastering process necessarily imposes an HF cutoff that is lower than that achievable with high rez digital. I wonder if the removal of supersonic frequencies can explain why some people prefer listening to LPs made from digital files to listening to the digital files directly.

 

[analogsa]

With Redbook CD standard, the Nyquist limit is close to the audible frequencies and a steep anti-aliasing filter is required.

This is only true of non oversampling dacs, which although existing are a bit exotic. The difficulties with brick wall filtering predispose many designers to skip any form of analogue filtration altogether. Zanden being a notable exception.

Hi res pcm, similarly to oversampled redbook, usually gets some minimalistic first or second order filtering.

 

[Ian B]

NOS R2R DACs do actually roll off the highs to reduce aliasing, especially at 44.1 and 48khz, because they are using analog filters and will cause lots of phase distortion if they are too steep. Also it is just really hard to make an analog filter steep enough.

But outside that class of DAC, different manufacturers use different filter designs, and even different chips manage high sample rates differently. In pro audio, typically the chip has a brick wall filter at half the sample rate, but some chips relax this significantly at 192khz and above. Most high end manufacturers have custom slope filters, or multiple selectable filters. A number of DACs upsample to 352khz or DSD which allow more gentle filters at the output.

These are technically called "reconstruction" filters which are part of the modern design that uses oversampling to reduce the steepness of analog anti-alias filters. But in reality, the steep filtering still needs to happen and is simply moved from the analog to digital domain.

It is actually very uncommon for any standard default reconstruction filter to dip into the audio band. It is generally considered a no-no, but some of the filter modes on Chord, DCS products and others do reduce the audible high frequencies and increase aliasing a little bit to get a punchier sound.

There is also specialized digital reconstruction filter used in some DACs called an "Apodizing" Filter, which was originally designed by Meridian and is involved in MQA processing. This filter is another digital brick wall filter, but the cutoff is actually lowered below half the sample rate so that it removes any ringing caused by the ADC used in recording. For example, at 44.1khz the Apodizing filter is set at 19khz instead of 20. The same is true with higher sample rates.

 

[DonH50]

With Redbook CD standard, the Nyquist limit is close to the audible frequencies and a steep anti-aliasing filter is required. I wonder how they approach this with high rez PCM ? Do they use a steep filter close to the Nyquist frequency, or use a gentler filter slope at a lower cutoff frequency ? Are there qualitative differences in these approaches ?
Some LPs are now cut from high rez digital sources and the mastering process necessarily imposes an HF cutoff that is lower than that achievable with high rez digital. I wonder if the removal of supersonic frequencies can explain why some people prefer listening to LPs made from digital files to listening to the digital files directly.

As has been said, virtually all DACs today are delta-sigma designs that sample at a much higher rate and thus are able to use filters that are either very sharp (high-order) in the digital domain and/or softer analog filters at much higher frequencies since the Nyquist frequency is much higher. This is especially true of high-resolution DACs since achieving even 16-bits in a standard segmented "R2R" design is difficult.

FYI it is an anti-imaging filter at the output of a DAC; an anti-alias filter goes before the input of an ADC. There are some WBF articles linked in my signature that walk through sampling and all that jazz.

HTH - Don