Filburt

The AD1896 is a sample rate converter. It is there to convert sample rate, and an incidental effect of this process is it can reclock the signal and thereby provide jitter rejection. Although, I've seen it suggested that this is a poor use of an ASRC because the jitter will be integrated into a sampling error post-SRC. I'm not sure if there is a verdict on how bad this problem is in real-world application. Aside from that, I'm not exactly sure what you're trying to say. It sounds like you are trying to talk about ratios (e.g. integer/non-integer) but mixing the oversampling operation of the AD1955's filter with the resampling operation of the AD1896. Both use interpolation, filtering, and so forth but the AD1896's function is distinct from that of the AD1955. Anyway, it seemed to me that the main issue in the DIYHifi discussion was over whether CrystalLock was an independent, synchronous reclocking scheme that did not rely upon the use of an ASRC.

OK, I think I understand the confusion now. What the AD1955 has is an oversampling filter, which in very simplified terms has the net effect of pushing imaging products away from fs/2 (22.050KHz in the case of CD Audio), so you don't need a steep filter there which is both costly as well as is difficult to do well. You could do a sinc comp at the output to combat the rolloff, but oversampling I think is the better solution. Anyway, not sure why he's referring to the oversampling filter as an "upsampling" filter, as people usually associate that with SRC, but I guess it's inconsequential. I don't know what you mean by SCR, though; not familiar with that term, however the filter in the 1955 does use interpolation, and I guess it's "synchronous" and "integer" in the sense that a. it isn't a sample rate converter (so it's "synchronous" with the original Fs) and b. the oversampling frequency is an integer multiple of the sample rate. Anyway, aside from that, what I had asked about above (and what Alan answered) pertained to whether the ASRC is actually disabled in CrystalLock mode.

The AD1896 is the ASRC. It is not a D/A converter. I don't understand what you're trying to say with the integer/non-integer multiple unless what you mean is the ratio between original SR and the resampled SR is not an integer, however it still doesn't make much sense since the AD1955 doesn't resample. There was a thread on DIYHifi.org a while back about the AD1896 and CrystalLock mode - DIYHiFi.org; View topic - Lavry DA10 DAC & CrystalLock?. I haven't seen much discussion about the matter since then; kind of wonder what came of it all.

Ah, OK. I didn't see that in the manual. Thanks for pointing that out So, are there any salient differences you're noticing between this and the DA-10 apart from USB input?

Want to re-emphasize, I have no idea what chip it uses; mostly I just wondered some others' thoughts on the matter of differences from the DA-10 spec. I can't remember if I ever tried feeding the AD1955 eval board a 200KHz sample rate or not; maybe it can accept it. Perhaps Mr. Lavry will at some point say what it uses. One thing that does leave me rather skeptical about the proposition is I guess I would have expected it to have been part of the announcement.

Wasn't impressed by the DA-10; maybe this is an improvement. Anyway, total conjecture time... I see no mention of CrystalLock, and it specifies 200KHz sample rate max. It also mentions digital volume control, but says it's "analog"; not sure if this means there's some sort of digitally-controlled resistor ladder attenuator system or if it's just marketing-speak for something on-chip with the d/a. Anyway, the other two features might suggest: a. switch to just using an ASRC or something involving one, and b. the 200KHz sample rate may mean change in D/A converter, since the AD1955 (used in the DA-10) at least doesn't spec 200KHz. If so, this could suggest a switch to the ESS Sabre (or Sabre32), as 200KHz is its max spec'd sample rate, and it contains an ASRC-based jitter rejection component. I think the Sabre also has digital volume control although I don't recall off-hand. If it's using a Sabre, I wonder if it's using the voltage outputs or the current outputs.

-45dB for the Music. Haven't tried the tones yet.

I love the "true to our design methods...differential input stage" LOL. Of course it's a differential input stage, it's an op-amp! Making something discrete doesn't automatically make it better. It takes substantial effort to do better than a good IC, and just dumping something in because it's discrete is asinine.

If you like Celebrator you might like the Dogfish Head Palo Santo Marron.

I'm sorry but I don't understand what you're trying to say here.

Digital volume control pushes the data into lower bits in the dac, with peaks falling below the MSB. This is generally not the best way to get maximum performance out of a d/a converter, as it will lower your DNR and SNR, and the upper bits are generally the most linear operating region of the converter. This is especially true in R-2R dacs, but seems to be true for at least current output delta-sigma dacs as well, as they still use bit ladders and are still subject to many of the same physical limitations, in addition to some other constraints (such as the impact on noise shaping both in filtering and in ladder assignment [e.g. 'scrambling' and such]). There really isn't much use in 32 bits for playback if you haven't the dynamic performance to support it. Well, other than I guess you can push bits down into noisy lower bits instead of outright truncation, but I'm not sure there is much practical difference to truncating the LSB of a 24 bit track versus pushing it into lower bits of a 32 bit dac. However, 32 bit depth is useful for storage when doing many stages of processing, so I guess it can be useful in a professional setting when you don't want to have to downconvert to listen to the result. Other than that, it seems the Sabre32 has better dynamic performance than the original Sabre, so it is still an improvement even if one of the new features is essentially superfluous.

Can you be more specific about the basis upon which you conclude this? Thanks

Rachmaninoff - Piano Concertos Nos. 2 & 3 performed by Stephen Hough and the Dallas Symphony Orchestra

I don't mean to say distortion means nothing. What I mean is that THD isn't very helpful other than as like a preliminary "something is wrong" kind of measurement. Harmonic amplitudes compared against psychoacoustic modeling is a lot more helpful in terms of discriminating between 'good' and 'bad' performance. Measurements are tremendously useful; I used them in my work on my DAC over the summer. It's just evaluating overall audio performance appears to (at least to me) require a finer measurement than what THD offers . One thing to watch out for in DACs, for example, is you can take some standard uni and duotone THD measurements and think yourself distortionless, then you run a complex multitone config to try to sim actual music and watch your harmonic distribution outright stink on top of having some aharmonics or overall indications of DIM problems. Anyway, like I said, I'm just trying to help out. Good luck on your project

I'm not trying to bag on you; I was just trying to help since we've both apparently tried to engage in projects having to do with measuring real-world audio performance via measurements. I've had some success doing so, so I tried to pass on some of what I'd learned through experience to help in this endeavour as I'm interested in the project . I probably should have been a little more specific, though. What I meant by looking at the harmonic amplitudes is analysing the overall distribution against psychoacoustic models so you can get a good idea of what the significance is in terms of overall audio performance. I've tried the other thing you're suggesting before. It does work; it's just harder to quantify the results enough to make qualitative predictions. However, if you're persistent, and do a lot of tests to figure out what the audible significance is of each component of the result (or can figure it out from some psy models), then it can be helpful.

Try a multitone generation with dynamic amplitude as the reference if you can, as this would best approximate something music-like. With something like weighted averaging you can probably get a decent metric of what the relative distortion is. I actually prefer running an FFT and looking at the harmonic amplitudes, but it gets pretty difficult once you do something like what I just suggested above. If I'm stuck with static amplitudes and only uni or duotones then FFT is pretty much the only thing useful, as THD will tell you almost nothing. There are theoretical and practical reasons why balanced could help but it's also plausible that it doesn't pan out to much practical advantage in the case of headphones due to systemic limitations. I've not tested it myself so I don't know >.<

Looks like it's a PCM1704-based platform with some sort of FPGA or off-the-shelf DSP to do filtering. Could be a Blackfin in which case the HDCD decoding may be done by the PMD200 firmware + Teac's stuff instead of having two ICs (e.g. DSP + PMD100) plus would make taking those high sample rates a lot easier. So, that at least is pretty solid. I imagine the power situation is OK too. So it's probably the output stage that is the problem. Knowing Esoteric they probably used something silly like NE5534s (or 5532s), or OPA604s for I/V and probably the follower for the active LPF. In essence, that's going to sound kind of sterile and crappy. If they used something like MKS or certain MKP caps for the LPF, particularly WIMA's (I haven't entirely figured out why this is yet but it was reproducible across a few circuits I tested), this may exacerbate the effect. Anyway, the long and the short of it is that it could plausibly be worth checking out, but you may end up having to invest in someone modifying it to some adequate amount for you. I imagine one of the DIY-for-hire guys could do it *shrug*.

I'm not sure what else can be said other than I'm not surprised.

1G has Wolfson WM8758 as noted earlier. I found I didn't really like the new Cirrus Logic chip's sound quality in the Classic. Don't know about the 2G yet, though. Will have to make a head-to-head comparison when I get the chance.

Nevermind. Apparently it uses the Cirrus Logic chip.

Does anyone know if the new Touch has the Wolfson or the Cirrus Logic CODEC?

That's mostly what it appears to be if you look at the feature specifications and overall description of the device. The only difference in the description is the 740c uses a Blackfin while I guess the Dacmagic uses a Freescale. I'm not sure that's really material, though, if they're using the same code with the same ops etc. I'd have to see what specific components it uses in the output stage to tell for sure, but it sounds like structurally it's the same but for the DSP, which means the rest may at least be modifiable to conform with or exceed the performance of the 740c. I'm not a huge v-out dac fan but the WM8740 is pretty decent; I'd take it over the Asahi-Kasei line and certainly the Crystal Semi and Cirrus Logic dacs.

The i/v is LT1363; I don't know if the buffer thereafter is as well or not as the label is underneath that resistor. The top chip in that row of three on the first picture is a line buffer for data, clock, etc. (it's a 74ACT244). The second chip down looks like a DF1704. The fourth I don't know which part specifically but it's some sort of input receiver, and looks like it's one of TI's.

LM317/337 are regulators

YGPM