[JohnnyCanuck]

With this in mind, I'm trying to get my head around products such as PS Audio's DirectStream Memory Player. Their literature claims to have figured out ways to make the decoding of a CD... more accurate.... more precise?

That sounds like marketing hype to me. An improved clock would provide the "improved performance" stated.

JC

 

[earnmyturns]

The various burrito filters up-sample the input to a higher rate (and possibly bit depth) used by the DACs. There is magic and subtleties in exactly how you do that to preserve all aspects of the waveform as described by the original samples. The DAC chip (in Schiit's case) takes the numbers from the filter as input and generates the corresponding waveform as output. I don't believe there is a stair-stepped waveform anywhere in the process, My understanding is that stair-stepped wave forms are a way to visualize digital sampling but never actually exist.

Oh yes there's a stair-stepped waveform, which you can see on an oscilloscope. Look around on the Web and you'll find such graphs. But that's in any case clear from the principles of multibit D2A processing: the incoming sample turns on and off a set of (transistor) switches to select a resistor combination whose total resistance is appropriately proportional to the encoded amplitude. When that switching happens, the output changes as fast as the circuit allows to the new level, and stays there until the next sample arrives. Of course there are high-frequency switching glitches, which in a well-designed DAC are low enough to not matter. A scope trace shows all of this clearly.

 

[earnmyturns]

With this in mind, I'm trying to get my head around products such as PS Audio's DirectStream Memory Player. Their literature claims to have figured out ways to make the decoding of a CD... more accurate.... more precise?

There are two different types of effects involved in waveform reconstruction. The first kind are the inevitable inaccuracies of clocks, switches, and other parts (such as resistors in multibit DACs), as well as switching transients that cannot happen totally instantaneously. The second kind is due to the mathematical fact that samples in the CD do not fully determine the output analog signal above the Nyquist frequency (22.05 kHz for CDs). In other words, the DAC is "free" to reconstruct those higher frequencies however it wants. But in practice, different reconstructions interact with inevitable circuit nonlinearities in different ways. That's why every DAC has digital and/or analog filters to tame that out-of-band energy. Schiit's (combo-)mega-burrito filters are one approach. PA Audio do their own thing by converting the incoming samples to DSD (rate coding) and doing their own magic. In the end, neither can get beyond the CD samples, but the different reconstructions can (through nonlinear effects) sound different, possibly more "faithful" to what one would hear if they were sitting in the studio. One final remark is that frequency-domain theory (Nyquist frequency, filter-based signal reconstruction) might not fully explain the subtleties of what and how we hear. At least that's what the experts I've worked with tell me... but that is way outside what I have the technical tools to understand in detail.

 

[Scott Kramer]

Spoiler

lol, oops I said that, was kidding -- seriously tho a day & it's good -- non issue re: the ADxxxx r2r chips -- my point is sometimes the yggy may be dismissed immediately 'cause it does sound average on cold startup– few hours in it's incredible. YMMVAATHHS

(referring to warmup in Jason's thread and here you guys talking about the DSD dac)

 

[rkw]

My understanding is that stair-stepped wave forms are a way to visualize digital sampling but never actually exist.

You may be referring to this video, which has been posted here before. At 6:00 to 8:40, he is saying that there are no stair steps after analog conversion. There is a misconception that D to A converters don't produce smooth waveforms, that there are stair steps in the output and that steps are reduced through higher sample rates. Actually steps are eliminated by the low pass filtering at input and output when following the Nyquist-Shannon formulas.

 

[Scott Kramer]

^ yep -- (please) don't wanna rehash it again

 

[earnmyturns]

There is a misconception that D to A converters don't produce smooth waveforms, that there are stair steps in the output and that steps are reduced through higher sample rates. Actually steps are eliminated by the low pass filtering at input and output when following the Nyquist-Shannon formulas.

We are talking about different steps in the process. The chips used in the Schiit multibit DACs, or the discrete resistor arrays used in other multibit DACs, will produce (approximate, because of switching noise) stairsteps at the incoming sampling frequency. However, the circuitry downstream from that stage will filter out most of the out-of-band frequencies from a piecewise-constant conversion and thus smooth the signal. And of course transducers are band-limited for physical reasons, which is what some NOS DACs take advantage of. Nyquist-Shannon has nothing to do with this. It just says that a band-limited waveform can be exactly replicated with a sampling rate twice the top frequency of the band. But that does not tell you anything about what an actual physical D2A will do above that top frequency, even if it is absolutely accurate within the band. That's why digital and/or analog filtering above the target top frequency are needed in DACs around the actual D2A switching circuitry.

Here's a nice discussion with oscilloscope traces to make it all clear:

http://archimago.blogspot.com/2018/11/nos-vs-digital-filtering-dacs-exploring.html

 

[winders]

Has PS Audio figured out a way to harvest that hidden data and make it perceivable?

Of course they have not. All players have to read all the data to either run the digital to analog conversion process or pass the digital data on to a DAC to do that. Any company that says anything else is, at best, stretching the truth and, at worst, out and out lying.

 

[rkw]

I just learned a fascinating tidbit. Why does the compact disc have a 44.1 kHz sample rate (rather than something else such as 44.5, 45, or 43.6)?

In the late 1970's, only video cassette tapes had sufficient data capacity for recording digital audio. The Sony U-matic video tape recorder was used, and wrote bits as black and white areas in the video raster scan lines. 44.1 kHz was chosen as an optimized fit that could store 3 audio samples per scan line while being compatible with both NTSC and PAL video formats.

 

[US Blues]

A question to the elders wrt my DAC's "multibit feature & combo-burrito-thing": when I read back-issues Mr. Stoddard's & Mr. Hunter's musings, I have a nagging question. If I call my Modi DAC's multibit feature a digital filter, or it produces a tuned output it gets frowned upon. Am I confusing "multibit" (where the Modi's firmware takes the digital input and converts it into accurate analog audio signal [that more closely resembles the original recording session]) with digital filters... like the plug-ins found in software such as Foobar2000 and VLC ?

{{{{edit: perhaps more concisely, if I call my Modi-Multibit tuned to a desired output... am I unwittingly cheapening Schiit Audio's efforts?}}}}}

This question keeps coming up. I regret it causing eye-rolling among my peers.

The "Mr. Hunter" referred to in the title of this thread is Robert Hunter, he wrote the lyrics to Jerry Garcia's songs.

@Baldr, aka. Mike Moffatt, is the fellow from Schiit who designed the MegaComboBurrito Filter.

 

[PolarBehr]

Caution: I'm not a card-carrying electrical engineer, I escaped from EE into pure math and CS before I took the signal processing classes so this is all stuff I acquired on the side, and could be quite wrong. PCM audio represents the original waveform as a sequence of regularly spaced samples. Each sample is encoded as a fixed-point waveform amplitude value. For example, the standard CD (Redbook) is 44.1/16, meaning 44100 samples/second where each sample is a 16-bit number. If that sequence of numbers at that rate is directly fed into the Modi's DAC chips, the resulting waveform will look like a staircase on an oscilloscope because the chip maintains a fixed output amplitude for the whole 1/44100 fraction of a second for that sample. To reduce that effect and produce a smoother (smaller stair steps) waveform, the Schiit multibit DACs upsample the PCM stream to a higher sampling frequency by interpolating new synthetic samples between the original ones. For example, 44.1hKz would be upsampled to 176.4kHz by interpolating 3 regularly spaced synthetic samples between each of the original 44.1kHz samples. What should be the value (amplitude) of each of those new samples? That's the job of the combo-burrito filter. It applies a Schiit-proprietary computation to the original samples to create the new ones in a way that optimizes the "smoothness" of the upsampled stream according to criteria decided by them. Since this is a proprietary design, I don't know the exact mathematical optimization objective they use in their upsampling algorithm, but it sure sounds good! Finally, why is this called a "filter"? Well, that's the jargon of digital signal processing: anything that transforms a sample stream is called a filter, because if you look at it through the lens of Fourier analysis, it changes the decomposition of the signal into periodic components (similarly to an analog filter). Of course these filters are designed to not affect periodic components with audible range frequencies, but ultrasonic components affected by these filters can still mater audibly if they interact with nonlinearities in the circuitry to create aliasing, where energy at a higher frequency leaks into lower frequencies in the audible range.

Another thing to note is that the “Mega-combo-burrito” filter keeps (uses) the original samples (the actual amplitude value stored in the original data at the sample time). As far as I know this use of the original samples is unique in oversampling filters. Others (particularly delta sigma DACs) just use interpolated amplitudes and the original samples are lost in the process.

 

[winders]

Another thing to note is that the “Mega-combo-burrito” filter keeps (uses) the original samples (the actual amplitude value stored in the original data at the sample time). As far as I know this use of the original samples is unique in oversampling filters. Others (particularly delta sigma DACs) just use interpolated amplitudes and the original samples are lost in the process.

No, not unique. You are talking about the filter being “closed form”. HQPlayer has 3 closed form filters in its filter array.

 

[ScubaMan2017]

There are two different types of effects involved in waveform reconstruction. The first kind are the inevitable inaccuracies of clocks, switches, and other parts (such as resistors in multibit DACs), as well as switching transients that cannot happen totally instantaneously. The second kind is due to the mathematical fact that samples in the CD do not fully determine the output analog signal above the Nyquist frequency (22.05 kHz for CDs). In other words, the DAC is "free" to reconstruct those higher frequencies however it wants. But in practice, different reconstructions interact with inevitable circuit nonlinearities in different ways. That's why every DAC has digital and/or analog filters to tame that out-of-band energy. Schiit's (combo-)mega-burrito filters are one approach. PA Audio do their own thing by converting the incoming samples to DSD (rate coding) and doing their own magic. In the end, neither can get beyond the CD samples, but the different reconstructions can (through nonlinear effects) sound different, possibly more "faithful" to what one would hear if they were sitting in the studio. One final remark is that frequency-domain theory (Nyquist frequency, filter-based signal reconstruction) might not fully explain the subtleties of what and how we hear. At least that's what the experts I've worked with tell me... but that is way outside what I have the technical tools to understand in detail.

Control-C, then Control-V into my Post-Teacher-PD document. Thank you!

We are talking about different steps in the process.......{snip}.......Here's a nice discussion with oscilloscope traces to make it all clear: http://archimago.blogspot.com/2018/11/nos-vs-digital-filtering-dacs-exploring.html

10-4, @earnmyturns . Archimago's blog is very civilized. I'll revisit it.

 

[porchwizard]

You may be referring to this video, which has been posted here before. At 6:00 to 8:40, he is saying that there are no stair steps after analog conversion. There is a misconception that D to A converters don't produce smooth waveforms, that there are stair steps in the output and that steps are reduced through higher sample rates. Actually steps are eliminated by the low pass filtering at input and output when following the Nyquist-Shannon formulas.

Thanks for finding that. That's what I was thinking of. Thanks @earnmyturns and @rkw for the in-depth info.

 

[bosiemoncrieff]

I heard Marek Janowski twice this weekend in his Mendelssohn/Bruch/Wagner concerts. Oddly, I was disengaged for the Tannhauser overture/venusberg music and especially the prelude and liebestod from Tristan. First night I was in premier orchestra, which sounded great. Second night I was in center terrace and then second tier. The sound was definitely more brass heavy from center terrace, but I can't say that premier orchestra was 8x as good, and the tickets are roughly that much more expensive. Thank god for friends in high places.

I was more engaged listening to Wagner on KSE1500 just now.

Just bought N90Q. They should be arriving tomorrow. Stay tuned.