Columns Retired Columns & Blogs |
After the first page, the review image is, shall we say, rather blurred...
So it's no surprise that you can buy very good Chinese-made DACs that measure very well, very cheaply. Those Chinese DACs are probably designed by first-rate engineers, and while extracting maximum technical performance from a good DAC chip requires care and attention, it isn't rocket science (footnote 1).
What, then, is the point in paying tens of thousands of dollars for a D/A converter?
It's a reasonable question, one that every DAC shopper must answer for themselves. Is extremely low measured jitter, noise, and distortion all that matters in a DAC? Is it sufficient assurance that it will sound "perfect," as good as a DAC can sound? Or is it possible to take this basic technology further, despite what the measurements show? It's easy enough to find people who are quite happy with their $1k DAC and smugly confident that they're getting the best possible sound. But in perfectionist audio (and certainly in this magazine), it's axiomatic that progress is always possible, that you can always do better, and that measurementsat least the easy and obvious measurements, such as S/N ratio, distortion level and profile, and Miller-Dunn J-Test jitterdon't tell the whole story. And if you listen with trained ears through topnotch audio systems well set up, it's frankly hard to miss the improvement in sound achieved by expensive DACs produced by companies committed to achieving the best possible digital sound.
And if you disagree? Then you just saved yourself a ton of money.
The CH Precision C1.2 D/A Controller
I'm sitting back in my lightly chewed IKEA chair, listening to Benjamin Grosvenor's performance of the Liszt B-minor sonata, S.178, recorded in Queen Elizabeth Hall at London's South Bank Centre. It's from Grosvenor's album Liszt, and it's streaming from Tidal (24/96 MQA, Decca). I'm listening on a system most would consider very good; it certainly isn't cheap. It includes the Wilson Alexx V loudspeakers, two Burmester 218 amplifiers (each bridged for mono, in for review, footnote 2), the Pass Labs XP-32 preamplifier, and not-quite top-level cabling by Nordost and AudioQuest.
The source of this music is the new CH Precision C1.2 D/A Controller ($43,000 as equipped), aided at the moment by a complete CH Precision digital front-end: the X1 power supply ($20,500), the T1 clock ($24,500), and the D1.5 transport ($49,500 but not currently in use). I've set the volume to what I'd expect to hear if I were sitting in the first few rows of the concert halland indeed, the sounds I'm hearing could be emerging from a Steinway on the stage of a good concert hall.
Well, to be completely honest: not quite. This is a very good performance and well-recorded, but, while the highs I'm hearing have an appropriate, crystalline "ping," the lower-midrange keystrokes seem ever so slightly dulled; a touch of transient bite is missing. There's also some congestion on the loudest passages, a sense that the piano's case is filling up with sound and distorting a little, some thing I've noticed in live performances but not this much. Despite these minor flaws, this system is delivering a spectacular experience. The piano has real gruntmore than makes it to my listening seat at most of the piano performances I attend (footnote 3)and lots of high-end sparkle. Decay, of notes high and low, is natural and even.
But what of those flaws I heard? Should we blame them on the CH Precision digital front-end? No. It's clear that the fault lies in the way the piano is miked, which trades transient clarity for low-end impact.
How to build a CH Precision DAC
If your goal is to make a DAC that's better than one you can make with a very good DAC chip, the way to do it is to start with a concept. You need a theory for how to proceed, or, as baseball commentators like to say about hitting, you need a thoughtful, fundamentally sound approach. It helps, of course, if your theory is correct, and if it's just plain wrong you're in trouble. But for reasons I think will soon become apparent, your theory need not be precisely on the money. CH Precision's approachshared generously with me by Florian Cossy and Thierry Heebis based on the notion that timing is everything. Getting the frequency part right is easy enough. What's hard is getting things right in the time domain.
Both Cossy and Heeb are engineers. Heeb is the digital guy. In addition to being the "H" in "CH," he's a senior researcher at the University of Applied Sciences and Arts of Southern Switzerland, specializing in DSP for audio. Cossythe "C" in "CH"is the company's CEO; his engineering expertise is on the analog side (footnote 4).
The first step toward understanding why timing matters in a D/A converteror why it makes sense to assume it matters beyond mere 1s and 0sis to recognize, as Heeb told me months ago in a Zoom conversation, that in audio, a digital signal is best thought of as analog. "Even if the signals or the electrical signals are supposed to be digital, basically just two levels, a zero and a one, as soon as you get into an electronic board, they are actually analog signals, current or voltage flowing through components. That is especially true, for instance, for clock signals. If you just consider clock signals as being a shift between two values between zero and one, you don't really get what clock is. The most important point in clocking is in the time domain"well, duh"with finite resolution. Basically, it boils down to an analog signal again."
I'll just throw this in: In the physical world, music happens in the time domain. True, we do hear frequencyas pitch, and combinations of frequencies at chords, or as vocal or instrumental timbrebut, strictly speaking, those musical signals exist only as a function of time: In your ear canals, there is only one level of pressure at an instant of time, and it changes.
The frequency domain is, strictly speaking, a mathematical abstraction.
There are two things (at least) that lead to time-domain errors: timing randomnessalso known as jitter (footnote 5)and an intrinsic lack of precision in D/A conversion, which Heeb (and others) call time smearing. Time smearing is the same concept that MQA is intended to addressthey too call it time smearingand, indeed, CH Precision's approach to dealing with that phenomenon seems quite similar to MQA's approach. In a comment published in my review of the CH Precision D1.5 transport/player, Heeb said, "Time smearing is basically if you put a single pulse through the system, if you have a filter with a very long impulse response, that single sample will extend over a large number of samples." The goal, then, is to shorten the impulse response so that the musical content of an input sample extends over as little timeover as few samplesas possible. How is that achieved? With an approach to conversion that's quite different from the approach outlined by the foundational document of digital audio, Shannon's theorem.
Shannon's theorem says that if certain conditions are met, the output of an A/DD/A sequence can exactly match the input, mathematically. But that's not true in the real world, under any real-world circumstances, because the conditions are unphysical. They do not exist. For example, the basic mathematical function Shannon employed for sampling and reconstructionthe sinc(x) functiongoes from minus infinity to plus infinity in time, which in the real world never happens. ("There is no energy in the signal before the instant where the musician starts playing," Heeb wrote to me by email.) Anyway, CH Precision would not want to use a sampling/reconstruction "kernel" that's infinite in duration, because, well, that's a lot of time smear (footnote 6). "We prefer to use splines, which have a much more compact support (footnote 7), which makes it so that when the sample goes in, what comes out has, in our case, [no more than] 100Ês of pre-ringing and post-ringing," Heeb said. A particular spline can be used to represent music locally; a long series of overlapping splines can represent a whole song or symphony.
In my review of the D1.5, I found it to be a transport of obvious quality. I also found it to be, with its two monophonic D/A converter cards, an excellent player of CDs, SACDs, and MQA CDs (footnote 8). Good as it was, though, those DAC cards are limited implementations of the CH Precision conversion approach. The C1.2 is an outright assault.
The C1.2 upsamples everything (except, according to the Roon Signal Path display, MQA data, which makes sense) to 16 times the base rate: 44.1kHz data and its multiples are upsampled to 705.6kHz; 48kHz data and its multiples are upsampled to 768kHz. This, though, is not your mother's upsampling. In performing this upsampling, the C1.2 does something that was common in the early digital era but that's surprisingly rare these days (so maybe it is your mother's upsampling): It keeps all the original data points, interpolating new samples between them. Other approaches, most notably asynchronous sample-rate conversion, obliterate the original stream completely (except the very first sample) and replace it with a completely new datastream. The time series described by the new stream may be very close to the old stream; nevertheless, this strikes me as an interesting point, philosophically and perhaps sonically: How can you claim the original spectrum is perfectly recreated (it's not) when all the data have different values?
The "base"-model C1.2 doesn't include a USB input, but you can get one ($3000). CH Precision's USB input card is a bit different from others. While it does reclock incoming datathat's the advantage of an asynchronous, isochronous USB interface, in principleit does not resample. Even via the USB input, the original samples are preserved.
At the end of this chain of conversion technologies is something surprising: a DAC chip. Not just any DAC chip, but one that was an important step forward for digital audio when introducedin 1998. It is Burr-Brown's PCM1704 R-2R ladder DAC chip, four per channel. Why do it this way instead of laying out an actual R-2R ladder with resistors, as several much cheaper, excellent-sounding Chinese imports do? I asked that. "The fact that it is a monolithic chip makes it both consistent and wonderfully accurate to work with, something that a discrete ladder cannot achieve even with the highest precision resistors," Cossy answered. He also wrote, "Even though it is an 'old' chip, it more than meets current requirements."
You wouldn't expect CH Precision to use a boring old volume control (footnote 9), would you? Well, they don't. Instead, the C1.2 utilizes a hybrid analog/digital control, which combines three large analog steps (via relays) with smaller digital domain steps.
The C1.2 from the outside in
Everything I've written up to now was true of the earlier C1 DAC (except maybe the part about the volume control; I'm not sure about that). So, what's new in the C1.2? What has changed?
First, though, an aside on naming. Why name two products released so close together so differently? The D1.5 came out months before the C1.2. Why not call them both "1.2" or "1.5"?
Footnote 2: According to Stereophile policy, reviews must be performed in a well-known room, mainly on well-known equipment, so I have already listened extensivelyfor several weekson my reference Pass Laboratories XA60.8 amplifiers. See my review here.
Footnote 3: Although not, I'm thinking, at Manhattan's newly rebuilt Geffen Hall. I've attended two shows there now. Though a little bit dry, that hall has serious grunt.
Footnote 4: Also, of course, "CH" stands, in Latin, for "Confoederatio Helvetica," or Swiss Confederationfor Switzerland.
Footnote 5: I've been hearing for years, from digital designers, that jitter can affect sound at far lower levels than previously thoughtand that the effects of jitter are manifold: It's not just the edginess heard, for example, on the jitter tracks on Stereophile Test CD 2 that affect imaging precision, subjective tonal balance, and other aspects of musical presentation.
Footnote 6: Modern sampling theory long ago abandoned the idealized notion of perfect reconstruction. An example of this is the use of a reconstruction kernel (a spline function, say) that differs from the one used for sampling (perhaps a sinc(x) function). "The key question is, how do the sampling and reconstruction kernels combine?" Heeb wrote in answer to another question. "In other words, what is the result of a reconstruction kernel applied to a sampling kernel on a unit pulse? If the result is close enough to identity (in a given frequency band and a given time space), then different kernels can be used with no apparent drawback." So, wise designers long ago stopped being slaves to Shannon's theorem, favoring instead an approach that attempts to minimize error and to shift error to where it does the least harm. This, I believe, is why there's more than one legitimate approach to D/A conversionand why it remains an unsolved problem. There are various legitimate approachesvalid assumptions as to where the inevitable error does the least sonic harm.
Footnote 7: "Compact support" is a mathematical term that means that, outside a certain finite range, the value of the function is zero.
Footnote 8: Although as an MQA-CD player, I had nothing to compare it to. It was the only MQA-CD player I've ever auditioned.
Footnote 9: The volume control can be difficult to locate among the C1.2's many menu options. It's hidden in the "Factory" menu, presumably because it's such a fundamental choice: whether to use the C1.2 just as a DAC or also as a preamplifier.
After the first page, the review image is, shall we say, rather blurred...
"Not just any DAC chip, but one that was an important step forward for digital audio when introduced—in 1998. It is Burr-Brown PCM1704 R-2R ladder DAC chip, four per channel. Why do it this way"
Have a listen to an old Naim CD555 to hear how good R2R ladder can sound.
The R2R PCM1704 was a killer converter chip. (just got too expensive to produce, having to laser trim all the minute R2R resistors Barrie Gilbert (AD) told me) Delta Sigma conversion good for DSD, ESS, Wolfson etc etc, doesn't have this laser resistor trimming problem, then they don't sound as good converting PCM 16 or 24 bit recordings either, they can only do a "facsimile" of it, where R2R ladders convert PCM "Bit Perfect"
If they were easy to get, even the discrete R2R dac manufacturers would use them. The "potential" (depending on the design) for far better specs are always there for a far smaller package of the same thing.
Cheers George
The question is, is an 80K+ Dac (with power supply and clock) also 8 times (800%) better than a 10K Dac (with everything integrated) or still only 0.2 (20%) better or still only 0.1 ( 10%) or even just 0.05 (5%). "The best sound I've heard" ... yes, I believe that ... but how much in proportion? That's not what the review says. Better but how much better? Let's say audibly better and yet really 0.2 (20%) (which seems a lot to me, but okay) for an 8 times (800%) more expensive. Is that proportional? The answer is of course subjective. And for many here, even a 10K is way beyond the available budget. Always nice to read about the Bugatti Chiron while driving an Audi A1, or looking for it and so to speak.
And if you listen with trained ears through topnotch audio systems well set up, it's frankly hard to miss the improvement in sound achieved by expensive DACs produced by companies committed to achieving the best possible digital sound.
Jim, hard to miss how? And why has no one anywhere ever been able to demonstrate this? Perhaps you can show us the way.
...but your remark prompts a tangential question anyway: Would you like to tell me what I hear?
See, the Editor already included the answer to your question. He heard the sound of it. I'm not sure how else it would manifest if not through hearing, that being the point.
From that I think it's fair to ask if you missed that or if you're alluding to denying another listener their impressions. Hence, would you like to tell anyone else what they hear too.
Would you then expect someone to somehow demonstrate these things to you? How? By what means? And who are you or I that it matters when you effectively admit you cannot or will not hear it anyway? What proxy is this you propose that not hearing a thing, you are interested in a demonstration of its presence anyway?
The biggest problem with the objectivist line is that taken to its conclusion it's a tacit admission that s/he's not capable of hearing a thing. At which point then why bother with better audio at all?
Not being able to admit one hears great musical presentation when listening through a $1M system at an audio show remains the last vestige of proof objectivists willingly deny.
Many have walked out of MQA demos without listening to 1 minute of music before reaching their summary judgment.
They are a literally 'not bothering with better audio' at all. They are the very definition of fool.
Some patience and willingness to hear the difference, in spite of their theoretical dogma, is the key.
...argue online about anything and given anonymity, do so with reckless abandon. Audio is no exception. With this in mind it's no leap to predict they'll make meaningless noise about things they've never heard, don't understand, couldn't technically comprehend, and deny others the effect of.
I hope writers and readers realize the pop-objectivity movement only purports to have scientific interest. They're malcontents. They're actually coming for your enjoyment.
Is that many are unwilling to set up an unsighted test, because they will find out that they don't have golden ears and can't hear many of the differences they claim to hear.
Doesn't mean no one can; just means many make the false assumption that they can. They don't want to face the reality that they can't.
... no subjectivist line. I hear. The problem is the incivility and intolerance you harbor because I do.
I also don't need an unsighted test. Again, I hear. In effect I "set up" an unsighted test every time I plug something in, almost always finding that what I thought it would do it won't. It goes and does what it actually is because it is. I allow for that and I enjoy that. There I learn; I learn because I'm a scientist and not what you'd like to think I am, your complete stranger.
That's the difference. That's the intellectual asymmetry. This drives you nuts and your recourse is to lash out. You're postmodern audio-woke. You despise that classical structure and you've invented a new pseudo-ethical system you wish you controlled.
So. You're projecting and further, you're doing it among your superiors, not the degenerates you seem to think number these subjectivists you think you're on about. For your dysfunction to work in your mind you have to invent adversaries that don't exist. So you do and here we are.
Your formulation is logically incomprehensible and I suspect you'll deny it forever.
Total nonsense.
One, where did I lash out? I think reading posts, it's obvious who's projecting and who's inventing adversaries. It isn't me.
And the fake intellectual approach is just you hiding ad hominem arguments behind sophisticated language instead of discussing.
All I wrote was most audiophiles don't actually test themselves and find out if they can really hear differences. There's no illogic there. It's a fact. How many audiophiles have ever actually taken a listening skills test like the Harman one to see how golden their ears are?
Most ignore the role expectation bias plays in human perception.
A good scientist would acknowledge all of that and deal with it instead of pretending it doesn't exist and exaggerating his abilities.
...that you have an opinion.
What I have is a demand: Where's the statistical proof that seeing a big walnut box alters the way it sounds versus a small black one. Or a shiny amplifier versus a utility enclosure. Or a fat cable versus lamp cord.
Isn't that your sighted bias canard in a nutshell? Then for it not to be a canard I expect tables of cited evidence of these sighted biases and not the usual driveby youtoob mcgurks.
Where's your statistical proof a commercial for-profit organization has captured the essence of all natural auditory phenomenona per all recreated sound from all conceivable reproduction systems and has interpreted it strictly for that benefit and across all use cases?
Where's the statistical proof the unnatural structure of your forced listening test is utterly transparent, not just technically but mentally, psychologically, and auditorily? Remembering your foregoing sighted bias, that is.
These are not rhetorical questions and I expect you to answer each of them with scientific rigueur and depth and references. You haven't. What you have done instead, not just throughout your work here but in this one comment alone, is
-Denied obvious reason;
-Expressed an assortment of frankly risible opinions based on generalities;
-Denied an obvious truth with unfounded, accusing opinion;
-Based on an assumption, issued a completely ambiguous demand;
-Denied it;
-Asserted an arbitrary standard (which I now demand you confirm per statistical proof that it speaks for all audio, all hearing, and all experience);
-Asserted another unmoored generality without evidence;
-Closed with doublespeak, rhetoric, and an uninformed lie, accusing your interlocutor himself of lying.
You're excellent at open opinion, DH, and frankly you're very good at the character defect so many of us have seen for so long among your kind. Now let's see your body of research supporting a more factual basis.
Try checking the Harman research on sighted listening bias.
Or the massive amounts of studies on all types of human perception, including audio perception that show expectation bias. It's really not hard to find.
Not long ago it was shown that humans even hear different volume levels (where none exist) and that the perceived volume level changes according to the color of the volume knob they are presented with. Just one of many examples.
It's massive and undeniable, except by people like you who refuse to see if their listening skills actually stand up to scrutiny.
My position (that not everyone has golden ears and that expectation bias in listening exists) is backed by science. Are SOME people skilled listeners and able to hear small differences? Yes. But a minority. And certainly not all those who ASSUME they are.
You are the one taking an extreme position - that both of those ideas mentioned above are false, and that you and other audiophiles all have golden ears and are immune the the basic workings of the human mind.
My position is backed by science. Yours is backed by ego and wishful thinking.
The onus is on you to prove your a-scientific assertation.
...Google it. Like one of the local writers said to a fellow traveler in another thread recently, that's a system of belief:
Do you have research to back up your assertion, David? ... Because, if not, it's simply your belief.
To which I'll add, now you admit it's a belief; that neither he or you or anyone else can begin to mind-read the subjects or even conceive of the technical granularity needed to validate the claim, the belief. You invoke science in a decidedly unscientific, even anti-scientific rhetorical ploy, and you further admit no acuity or appreciation for the ways and means of the people you troll, scientific or otherwise.
Again quoting Jason:
As I learned a long time ago from my cursory exposure to A Course in Miracles, there is nothing to defend. What there "is" to do is pay very particular attention to set-up, upgrade my system to make it a fine tool for discerning differences large and small, listen closely, and write with integrity and dedication to what I perceive to be the best interests of readers of Stereophile.
Bravo. As it should be. And those best interests are reciprocal, DH.
None of this was a mystery before and with each passing comment you drive that point home. "Google it. Because science and commercial interest." And any extrapolation of this basis in your purported body of objective fact shall not and may not apply, while any expectation it apply will be not just refused but baldly denied.
Like I just said, driveby mcgurk the thing anyway. Because science. But really because you really can't get to the granularity needed, can you, and you can't imagine what it might be. You're left with, as you just did again, projecting a coarse generality borne by a meager, unqualified belief about physical systems, systems you didn't engineer and systems you, evidently, cannot hear.
And that, DH, is actually the very core of anti-science. That is your impediment to discovery.
Great review. Actually, the best review of anything I've read regardless of the topic. Engaging, factual, informative, fun with the added spicy of honesties. The very best review I've ever read. Kudos, Jim Austin!
Hi Jim,
Your question, "What, then, is the point in paying tens of thousands of dollars for a D/A converter?" resonated with me. I had been very happy for several years with my BorderPatrol DAC (I know it caused some controversy in these pages). When shopping for a DAC, it clearly offered, to my ears, the best sound quality to cost tradeoff in my price range. Then last year I was at my local hi-fi shop where I heard a system similar to mine playing through a Chord Dave (about 10 times the price of the BoarderPatrol DAC). It ruined me. It was better in all ways that matter to me. So, I guess case in point. Thanks.
--Monty
"What, then, is the point in paying tens of thousands of dollars for a D/A converter?"
My first post, if designed right it "could" compete or even better the MSB R2R Ladder Discrete dac, similar pricing.
Not to mention trying to get the PCM1704k dac chips, as they are as rare as rocking horse **** to find, and big $$$$$ if you do.
Cheers George
Since when did frequency and time-domain performance exist as separate dimensions rather than intercorrelated in digital sampling?
Since when did we accept this as true?
"Time smearing is basically if you put a single pulse through the system, if you have a filter with a very long impulse response, that single sample will extend over a large number of samples."
That shorter impulse response has anything to do with improved sound quality because "time smearing" is supposedly better!?
Some of the best time-domain performance comes from designs like the Chord DACs. I agree with MontyM that a Chord DAVE is superior to his BorderPatrol (but no need to think this is about the price of the device!). Look at the impulse response of those Chord devices - and corresponding frequency response.
A long impulse response is not a measure of problematic audio "smearing". It's just a reflection of the device's filter when fed a poorly or non-low-passed signal like that single-sample "unit impulse".
Subjectivists like to believe that "measurements don't matter". So why believe the impulse response correlates to anything that subjective-leaning audiophiles should look at and think makes a difference?!
I'm sure the CH DAC sounds fine; although clearly the measurements show limitations like linearity of the old DAC PCM1704, and right-left variations as JA noted. Worth thinking about given the asking price.
...subjectivists that believe that "measurements don't matter". Rather they believe that measurements have not captured either the magnitude or ranked importance of all audible phenomena. This is because they hear; they do audio for the experience and find that data must serve that end.
There is copious evidence, including in regular drive-by remarks in these threads from the ASR swamp, that Objectivists like to believe that measurements are all that matter, notwithstanding that those measurements are meaningfully incomplete. This is because they do not listen; they do audio as a cancel culture and also find that data must serve that end.
"...would you still need to smoke a cigarette?" No. I would take two aspirins instead.
"And does this equate good comments with sex?" Let's just say that the best comments can give me a spontaneously rapturous release.
Reading thi review I have the sensation that the Stonehenge layout reveals the secrets of Betelgeuse.
Archimago: "I'm sure the CH DAC sounds fine; although clearly the measurements show limitations like linearity of the old DAC PCM1704, and right-left variations as JA noted."
That's just implementation, not the fault of the PCM1704-K dac chip itself, as shown here with JA doing the same measurement on the Mark Levison 30.6 which also uses 4 x PCM1704-K chips.
JA: "With its superb linearity and low noise floor, it comes as no surprise that the '30.6's reproduction of an undithered 1kHz tone at -90.31dBFS is essentially perfect."
Cheers George
The very expensive DACs may sound a little better.
I contend that the vast majority of audiophiles won't be able to tell 2 very good DACs apart in an unsighted test.
Most don't have golden ears - they just like to tell themselves they do, and never test themselves to find out if they actually do. They often rely on imagined differences brought about by expectation bias.
...set of projections, if by interesting we mean boring as hell. Would you like each fallacy of your contention exploded in order or as a lump sum?
Don't need your "help". And no fallacies involved. I've seen it happen multiple times. Audiophiles are sure they know which amp is which, until they listen unsighted. Then they can't tell and those obvious differences suddenly aren't.
Doesn't mean the reviewer or some people can't hear the difference between this DAC and a $1000 Chinese one. But it means lots of readers of this publication, some of whom are very sure they can, can't.
...probably don't want help evaporating that abject mythology. That's not a mystery; it never was, and I appreciate your saving us even more time.
I recommend continuity: Reduce all your exploits in great sound to more of the same artificiality that makes everything sound equally forced and distilled. Go AB some chifi. Use one recording, take 15 second snippets, keep score, and always involve agreeable minds. Music is to be a miserable, smug competition. Then make ambiguous references to these vague fields full of your imagined projections of the shortfalls of others that you'll personally never suffer from because you're you. You advance fine audio by leaps and bounds by the day.
Your smugness and arrogant prove nothing about your supposed point, but lots about you.
At 44.1kHz, we think we can hear filter artifacts. Maybe. On certain program.
At 96kHz original program, the artifacts are gone. We hear no filtering issues, regardless of program.
Technique doesn't matter. "Splines" don't matter. All of this seems 100% related to magnitude response over time.
The $1M question becomes: On a 44.1 signal, is it less harmful to do a high quality up-SRC before filtering? Or avoid SRC and filter at native 44.1?
In our experience, and apparently in the experience of CH, Weiss, and others, using a pristine SRC to 192 or 384 (etc) is the better approach. SRC to 96 is probably just fine, as well.
Is any of this deeply unique engineering? Naw. Simple stuff, really.