I’m not so hard on him. He told the truth as he saw it. At least he’s trying to be consistent, unlike some who use different tests, different equipment for each test and/or don’t seem to understand how to use their measurement equipment. Being consistent has its own drawbacks, and perhaps there’s a better solution, but I don’t think measurements in reviews should always offer to do custom measurements for every company whose devices are reviewed.
PS Audio offers a full refund if you don’t like it (at least in the States). So you can try it in your own system at little risk. I think everyone should do this anyway. For that matter I think people should at least try the DS with a good preamp.
This is tangential, and I apologise, but I’ve been wondering about this for years. What’s the main idea behind ASRC, and what are the key differences between it and what you do in your DACs? I’m pretty sure that VCXO is included in the answer to the second part there. If ASRC doesn’t involve bending the local clock to match the average rate of the incoming data, does that mean they end up with a kind of quantisation error in the time domain?
The simple idea is that to match two clocks whose rates aren’t consistent or aren’t related by ratios of small integers you digitally simulate a D/A using the input clock and resample it using the output clock (digitally simulating an A/D.) Obviously, there are going to be some interpolation issues, especially as the relative phase of the two clocks keep changing sample to sample…
Stereophile has been notoriously stubborn about how it does measurements. I’m an owner of Thiel speakers, which were designed to be time and phase coherent at a listening distance of at least 8 feet (and optimally 9 or 10 feet). When JA would measure Thiels at 50", they would not perform optimally, but the speakers would still get criticized and mis-characterized in the review based on JA’s inappropriate measurements.
If JA wants to use a standard set of measurements even when a piece of equipment is designed to different standards, then it would be appropriate to discuss the reasons for any deviations in the measurement results. That doesn’t seem too much to ask of a competent reviewer.
Also in June’s Stereophile on page 141 there is an article about Darren Myers authored by Julie mullins. A must read for PS Audio admirers of him and his quest for designing affordable high quality audio equipment.
For more chatty version this is a fairly detailed description from a person who worked on the CS8420 ASRC chip amongst others.
Beware, it’s a forum thread, but one dedicated to this explanation only.
My June issue of Stereophile arrived today. It’s interesting that the MK2 didn’t make the cover, nor is it in the first few pages of the index. I don’t know if there’s a correlation to devices they like and photos in those two sections. I haven’t thought about that before. I wonder if a manufacturer can pay their way onto the cover or the index.
JA stated in the article “for logistical reasons, I performed the measurements of the DirectStream MK 2 earlier in the review period than I usually do.” He goes on to say that he usually waits until the end of auditioning to ensure that he isn’t unduly influenced by his knowledge of the measurements. Even though he didn’t want to be influenced by his measurements there is the possibility that he was.
Where is “the reviewers are corrupt and Hi-Fi magazines are unethical and conflicted by advertisements and don’t ever print any negative reviews or comments” crowd?
Had an Antipodes K50 Server-Player and a Bricasti M1-MDx DAC in the house this weekend, along with a number of expensive silver audiophile cables, e.g., Nordost USB, as well as my usual batch of slightly better than Amazon Basics cables. Ran various A-B tests with myself, my neighbour and one son.
Maybe it was the quality of the equipment, but no one succeeded in identify the fancy USB, AES or SPDIF cables.
We could not identify the Antipodes K50 over my ASRock 4x4 running Euphpony Stylus either. But all agreed that the K50 is a very nice massive piece of kit with lots of capability and choice of outputs.
Sorry to say, the Bricasti M1-MDx just won out over the PSA DS MK2. Admittedly the PSA is not so expensive. But it is newer. Generally, the Bricasti managed the same quality of soundstage with a clearer presentation. It was mainly in the mid-range that the Bricasti sound was clearer. This was with the Bricasti set on its base zero filter. Hopefully once we get past the teething problems with the MK2 pops and whitenoise and we get Ted’s updates for music, I can invite the neighbour back for a DAC rematch.
With reservations early on, and a review revision by Stereophile following a PSA DSD FW update addressing concerns identified by Stereophile IIRC. My advice, don’t put too much faith in reviews, at best use them as a pointer for gear you may wish to consider.
That thread on ASRC was brilliant, thanks for sharing.
@tedsmith now that I think I understand how ASRC works to some degree, could you please expand on your comment about JA’s jitter measurement? In particular…
“A test that is designed to only measure the jitter generated by the DAC proper”
is that sending an ultra-low jitter digital sine wave to the DAC and measuring the frequency smearing in the analog output?
“ASRC encodes any incoming jitter from the digital inputs irretrievably into the audio before the DAC proper and hence incoming jitter isn’t picked up by the measurement”
This one confuses me, but is it my lack of understanding or could you have worded it better? Because if jitter is embedded in the ASRC output data (right value at the wrong time) then wouldn’t we see the frequency smearing in the output anyway?
Per the thread joma linked ASRC has the effect of LPF-ing the input clock jitter (below 3Hz in that specific example) so could the lack of being “picked up by the measurement” actually demonstrate that ASRC’s jitter rejection is what they claim it to be?
As for the sound quality and preference, that same thread describes the need for a huge lookup table of convolution coefficients – over a million rows worth given 2^20 interpolation factor, and in their instance each row contains only 64 coefficients. Is it possible that the ASRC sound deficiency is due to them using only a small number of taps as the trade-off for having to store so many of them? I think that in your implementation with synchronous SRC and a lot more computational resources you need orders of magnitude fewer rows of coefficients but also use orders of magnitude greater number of taps?
It’s really fascinating to consider how the clocking strategy you built for the DS could be applied to other kinds of DA circuits. I understand your argument for maximum linearity with high frequency 1-bit operation but even so, the general problem of jitter rejection from unsynchronised sources is so neatly solved with your VCXO and input signal sampling system. I wonder if modern silicon production could make that feasible as an off-the-shelf solution to compete with PLLs and ASRCs, though the price of the crystal itself might put it in a very different cost category.
