[Currawong]

Most ADCs will work using a symmetrical filter. What this means is that for any digital recording, there will be (necessary) pre- and post-ringing present on the recording, as a result of the filter which was used. The key point to be made here is that all digital recordings will include ringing from the filters, even before they reach the DAC, but this is the best approach to take – provided the filters are correctly designed and implemented within the ADC.

Something I don't understand is, why do manufacturers use an illegal signal, an impulse response (a single-sample) to show that the filter will ring, when in an actual recording of music will not have such signals? Normal music will not cause ringing in a digital filter as I understand it, as it doesn't contain illegal signals. Am I missing something here?

Minimum phase filters, that ring less when presented with an impulse response, actually leave measurable ringing in the audible band, so does your description not have it backwards?

 

[A Jedi]

Something I don't understand is, why do manufacturers use an illegal signal, an impulse response (a single-sample) to show that the filter will ring, when in an actual recording of music will not have such signals? Normal music will not cause ringing in a digital filter as I understand it, as it doesn't contain illegal signals. Am I missing something here?

Minimum phase filters, that ring less when presented with an impulse response, actually leave measurable ringing in the audible band, so does your description not have it backwards?

Illegal signal? Is that like when you yell "fire" in a theater? Music is just an infinite series of impulse signals.

 

[Currawong]

Illegal signal? Is that like when you yell "fire" in a theater? Music is just an infinite series of impulse signals.

An impulse response (a single, positive sample surrounded by zeros) does not exist in music.

 

[Pondoro]

An impulse response (a single, positive sample surrounded by zeros) does not exist in music.

Tires, glues and aircraft all get extreme tests that may not represent actual usage, but are standardized. So a new product can be compared to predecessor products under standard conditions. For example a new passenger aircraft has it's wings loaded to 150% of the most extreme forces expected. This is a condition that should be impossible to achieve in actual use, but all the previous certified planes have passed that test. Waterproof glue gets boiled for an hour, it is more standard and cheaper than building a boat with it and waiting a few years. The NFL combines measure 40 yard dash speed, but no player ever runs 40 yards dead straight. A perfect impulse cannot be created, but testers come as close as they can. It is more standard than saying, "We played "Born to be Wild" and it didn't ring." Also in the old pre-computer days control systems were analyzed for impulse response because the math was easy, I'll bet it is easier to keep doing that test than to justify the decision to stop doing a test that has always been done. The experts are used to seeing the results and judge the new design against previous designs based on years of experience, which includes this test.

 

[Lerrens]

Something I don't understand is, why do manufacturers use an illegal signal, an impulse response (a single-sample) to show that the filter will ring, when in an actual recording of music will not have such signals? Normal music will not cause ringing in a digital filter as I understand it, as it doesn't contain illegal signals. Am I missing something here?

Minimum phase filters, that ring less when presented with an impulse response, actually leave measurable ringing in the audible band, so does your description not have it backwards?

The real question is, what would the impulse response of an ideal anti-aliasing filter look like?

Mathematically speaking, the 'ideal' anti-aliasing filter (a filter that removes all frequency components above a certain frequency, without affecting lower frequencies at all) would in fact be a 'sinc filter', which when viewed in the time domain, will have symmetrical ringing on both sides: sinc filter

When applying an 'ideal' sinc filter to an one sample impulse (which is not a bandlimited signal = containing frequencies higher than the sampling rate), the pre-ringing and post-ringing will show in the time domain and there's no problem with that, because it's exactly what it should look like in the first place when you remove higher frequencies from a non bandlimited signal while trying to keep lower frequencies enact.

When feeding music, being a bandlimited signal, the sinc filter will reconstruct the sequential samples and canceling out the ringing you see on the sinc filter, as long as the music is truly bandlimited. Another way to look at this is looking at how a step response will look like through a sinc filter: the ringing will be seen at the 'abrupt rise step' (transient state) because it contains non bandlimited information, but you won't find the ringing at the steady state where the information is bandlimited.

So the question would then be: which filters acts closer to an ideal sinc filter? So far linear phase filters (with symmetrical ringing) seems to act like a more closer representation of the sinc filters where both amplitude & phase response would be enact in the band limited signal. What I still don't understand is from a fidelity stand point, what are the benefits of minimum phase filters?

 

[Currawong]

Tires, glues and aircraft all get extreme tests that may not represent actual usage, but are standardized. So a new product can be compared to predecessor products under standard conditions. For example a new passenger aircraft has it's wings loaded to 150% of the most extreme forces expected. This is a condition that should be impossible to achieve in actual use, but all the previous certified planes have passed that test. Waterproof glue gets boiled for an hour, it is more standard and cheaper than building a boat with it and waiting a few years. The NFL combines measure 40 yard dash speed, but no player ever runs 40 yards dead straight. A perfect impulse cannot be created, but testers come as close as they can. It is more standard than saying, "We played "Born to be Wild" and it didn't ring." Also in the old pre-computer days control systems were analyzed for impulse response because the math was easy, I'll bet it is easier to keep doing that test than to justify the decision to stop doing a test that has always been done. The experts are used to seeing the results and judge the new design against previous designs based on years of experience, which includes this test.

Extreme testing of products has absolutely nothing to do with audio signals and their processing, sorry.

 

[Currawong]

What I still don't understand is from a fidelity stand point, what are the benefits of minimum phase filters?

I think that a lot of it has to do with the limited processing power of earlier DACs, which had issues producing clean attenuation at the stop band. Nowadays, we can apply very high quality digital filters to music, either with our computer or with an FPGA or similar inside a DAC. Minimum phase filters can sound subjectively better to people. For example, while the extremely linear Chord DACs produce an accurate soundstage in 2 dimensions, the more forward sound produced by minimum phase filters can be more engaging, if it does push all the instruments in a soundstage forward.

 

[Pondoro]

Extreme testing of products has absolutely nothing to do with audio signals and their processing, sorry.

If a more extreme transient creates stronger high frequency harmonics a filter that handles it well might be considered likely to handle easier transients well also. Nothing is better than extensive, real world, testing. But severe tests also show weaknesses. Weaknesses tend to be bad overall. No one can listen to a filter on all possible pieces of music. But if it sounds great on a few pieces of music and can also handle severe transients as well as all your previous DAC designs, that is a good thing. On the other hand a DAC that shows weaker performance on the harsh transient, relative to previous designs, might raise some cautions.

 

[mwilson]

I think that a lot of it has to do with the limited processing power of earlier DACs, which had issues producing clean attenuation at the stop band. Nowadays, we can apply very high quality digital filters to music, either with our computer or with an FPGA or similar inside a DAC. Minimum phase filters can sound subjectively better to people. For example, while the extremely linear Chord DACs produce an accurate soundstage in 2 dimensions, the more forward sound produced by minimum phase filters can be more engaging, if it does push all the instruments in a soundstage forward.

And when done with that, crank it up and add some negative-phase 2nd harmonic distortion in the analog stage and make the soundstage wider than the Grand Canyon

 

[Currawong]

If a more extreme transient creates stronger high frequency harmonics a filter that handles it well might be considered likely to handle easier transients well also. Nothing is better than extensive, real world, testing. But severe tests also show weaknesses. Weaknesses tend to be bad overall. No one can listen to a filter on all possible pieces of music. But if it sounds great on a few pieces of music and can also handle severe transients as well as all your previous DAC designs, that is a good thing. On the other hand a DAC that shows weaker performance on the harsh transient, relative to previous designs, might raise some cautions.

I'm not sure if you understand what I'm talking about here. I'm talking about an impossible audio signal -- an impulse response. An impulse response is not a transient, nor represents one. The effect an impulse response has on a digital filter is the opposite, in essence, from the result you get of the filter preventing the DAC creating high-frequency aliases of the in-band data. Nothing to do with stress testing, just mathematics.

 

[mwilson]

So basically like gazing into a laser pointer in a pitch black room vs looking at the real world.

 

[sajunky]

I'm not sure if you understand what I'm talking about here. I'm talking about an impossible audio signal -- an impulse response. An impulse response is not a transient, nor represents one. The effect an impulse response has on a digital filter is the opposite, in essence, from the result you get of the filter preventing the DAC creating high-frequency aliases of the in-band data. Nothing to do with stress testing, just mathematics.

This is now confusing, as your original question was about ADC part. At least it was my understanding, now I am considering I might be wrong. It is why these few words. In respect of ADC converting pulse response does not make sense. Simply like that, it is impossible. It is why filtering is in place to prevent aliasing.

There is a difference between aliasing during ADC and imaging in DAC. During AD conversion aliasing signal folds (using MQA terms) to the acoustic band and it cannot be removed. It is why removing most of aliasing spectrum power before conversion is critical. On the other side during DA conversion we can assume that a recording do not carry frequences above Nyquist. As long it is true (in my experience it is a common case), images created during DA conversion do coexist with our audio signal as there are placed above acoustic band, there is no aliasing, but intermodulation products can be created on non-linearities of our analog chain.

 

[dCS James]

The crucial point to consider from part 5 is that the digital audio has already been through a filter in the ADC before it reaches the DAC. The ADC is exposed to the real, analogue world, which isn’t bandlimited. Audio above 44.1kS/s Nyquist may well contain musical information (cymbals, trumpets etc.), but will also contain non-musical information (fluorescent lights, PC monitors, TVs and so on).

Given this, the signal coming into an ADC needs to be filtered to a finite band for sampling. Hence, all the effects of digital filtering will already be present in recorded digital audio before we even consider the DAC in a signal chain. In the case of most ADCs, this takes the shape of a half-band filter, often with a fair amount of aliasing back down into the upper end of the audible band.

These effects in the recording have implications in terms of the filter design in a DAC – it is not simply a case of designing the best DAC filter possible in isolation. The entire signal chain, from recording to playback, needs to be taken into account for the best real world performance. We will cover this particular topic in more depth in our next post.

A few additional points to clear up based on some of the comments from the last post:

• Even in a single-bit DAC (which the Ring DAC isn’t), the wideband SNR is 6dB, not 3dB.
• The digital filtering in a DAC cannot reduce the noise from a noise shaper as in a DAC the filtering happens before noise shaping.
• Not filtering in a DAC actually causes a droopier frequency response (-3dB at 20kHz). This happens as some of the Nyquist images are out of phase with the main signal, causing cancellations.
  Similarly, not filtering can result in overloading at lower frequencies, where Nyquist images in phase with the signal increase the amplitude at said frequencies.
• The diagrams shown in part 5 are to demonstrate the coefficients in a filter. If you feed a filter with a single full scale sample, you get an impulse which shows the coefficients of the filter. This concept hasn’t been fabricated to demonstrate a point, it is how we design filters, and how they work.
  To give an example, within the Bartók, filters 5 and 6 (playing at 44.1k) have absolutely identical frequency responses, but different impulse responses.

 

[sajunky]

Quality feedback response, thank you. Yes, it is 6dB, not 3dB. I don't care about 3dB drop @ 20kHz, it goes unnoticeable. Boosted low frequency can be welcome. Noise shaper is also the last stage on the Analog Devices AD1955 (block diagram attached), but nothing is shown on the block diagram you posted here. I mention this chip, as it is also 'multibit' DSD thermometer encoded and is using something called noise-shaped scrambling. I assume placement of the noise shaper is the same on dCS DAC. I understand a mapper in dCS does a different work (true randomisation of resistors value). A noise-shaped scrambling (judging from words used) is not the same, right? And finally, may I ask, which order loop is dCS modulator? Thanks.

 

[Ragnar-BY]

@dCS James Thank you for posting all that. It is very interesting to know how dCS see things.

Would like to know few more things:

1. What do you think about different filters from subjective side? Do they have a specific signature that most listeners perceive? For example, one filter sounds slower, but it has better scene depth. With the other, the sound attack is reproduced well, but decay is perceived worse, etc.

2. What is the difference between the DAC device in different dCS models (Bartok - Rossini - Vivaldi)? It is clear that there will be different power supplies, output stages and so on. What about the DAC itself and its filters?

4. I see that current dCS DACs do not support DSD256. Is there any specific reason for that? I understand it's not the most popular format, but some really nice classical albums recorded in native Quad-rate DSD are already available to public. Is there any chance that current generation Bartok or Rossini will support DSD256 after some software update, or it's not possible?

5. What do you think about DXD vs DSD? When buying classical music sometime there is a choice between same album in DXD and DSD. 2L even offers it's Test Benh for DSD vs. DXD comparison. What format will sound better on dCS DAC?