Filburt

Ah, OK. Scrambling is a pretty cool function. I guess in sort of general principle it's another form of noise shaping, as it basically appears to be about randomising distribution of the data across the MSBs in the dac (which is composed of several 5 bit ladders). Thereafter, you'd want a tightly controlled filter (interpolation, LPF, noise shaping, etc.), which I guess is the major point in which the SHARC comes in, though maybe it controls the scrambling as well. Good filters take a lot of processing power, hence the external DSP. Dusty - Yes, I think you got my point about reconstruction. However, I think perhaps I wasn't clear about what I think of digital processing. I'm not against it as some sort of blanket rule; in fact, I think signal processing is crucial to good performance. The concern in this case was that the use of an ASRC in this manner could undermine the performance of the digital filter, which is why I brought it up.

I think the most interesting use of DSP in this context is with the AD1955 where it seems it can use something like the SHARC to externally control it. Anagram utilises this, AFAIK. It apparently improves the function of the 'scrambling' function as well as the digital filter (which is required to really take advantage of scrambling). grawk - I don't think I even for a moment suggested that it wasn't of paramount importance that people enjoy their gear and that it sounds good. Discussing design on head-case doesn't interfere with that general premise. However, I do think good design and good sound are pretty closely related, and that it's worthwhile to investigate what makes for good design, plus it's interesting to have a discussion with other head-casers about it

I don't see any mention whatsoever about matching in the APL specs, and it's not exactly like they hold back on mentioning just about everything they possibly can about their work. I'm not even sure how you could effectively match something like an AK4397, actually. luvdunhill - I'll have to read over it and try to make sure I understand what they're doing before really commenting on it.

Interpolation isn't the same as something like a feedforward from the higher resolution master. Secondly, the interpolation is of course dependent upon sampling precision, and the problem is that both limitations in the interpolator as well as in the signal (e.g. jitter) end up creating substantial error. After all that, much of the error will probably sit right in the passband and whatever is out of it is greeted by lower precision in the filter. I'm not saying interpolation of any sort if an absolutely poor idea in all situations, but I think it's better handled at the filter and in much more tightly controlled environment with an eye towards improving filter performance rather than giving pretty numbers for the spec sheet.

Aside from possible problems of expectation bias (e.g. did you blind test it?), I do wonder which order you tested it in. I've been trying to refine my testing procedures lately and I'm finding, as two compared things get closer to one another in character, I find the second one has a higher probability of sounding better, irrespective of the actual order in which the two things are played back. I'm pretty sure the reason, at this point, is that I'm already primed to some degree to pay attention to certain aspects of the music by the time I've finished the first sample, and so the second sample ends up sounding better by virtue of my having a greater degree of acuity thanks to already having heard the sample. Aside from the issue of upsampling and of limitations in the output stage (e.g. I'm not a fan of switched capacitor filters), I do wonder about the wisdom of paralleling dacs together that probably aren't matched. Instead of getting error cancellation, you could just end up with error intermodulation.

How, specifically? You're already going to get interpolation, noise shaping, filtering, etc. out of the internal oversampling filter on the converter. Delta-sigma converters like the AK4397 already have an abundance of this as they don't function without it. Upsampling just gives the filter less headroom; it just pushes the passband edge up unnecessarily and you can end up with garbage in the pass or transition band that would have otherwise been past the stopband edge. On top of that, it takes more processing power, so it can't use as many taps (which just lowers the resolution of the filter).

Upsampling isn't going to bring forth that which was not there to begin with, plus you get to trade jitter for AM distortion in the process. The 4397 doesn't even have the dynamic performance for 24 bits; I don't see much point in feeding it 32. Plus you're stuck with that switched capacitor output instead of being able to do a high performance i/v + buffer. So, in some sense I suppose, there is a bit of irony to swapping a PCM1796 (that's what the 3910 has stock) for AK4396s or AK4397s as an 'upgrade' when it is the higher end part, especially when considering the output options. I'm not trying to be a total buzzkill on this. Rather, I am just a bit perplexed by some of the design decisions and this seems like an appropriate place to discuss it

What '32 bit mod'? The AK4397? All that means is it'll take a 32 bit word input. Other than that, it's pretty much just a run of the mill AK SCF output dac.

I don't know what the operation is like. Might be controlled by a microcontroller, or otherwise just by pulling some pins high/low.

I don't know if it is for sure, as I haven't looked a schematic for a "current conveyor", but it sounds like it probably is since Pars referenced Hawksford (who calls his own I/V "current steering"). It probably can just route the output of the modulator (or whatever is at the tail end) to either a switched capacitor filter internally (for voltage output) or straight out to external i/v.

Current steering seems to be regarded as the best. I don't know if this is because of the stuff Gilbert was talking about with overloading the input stage, if it's due to simply discrete offering better performance parameters (I haven't seen as much discussion about discrete op-amp versus current steering), or because of something else like better slew characteristics. I haven't yet tried colin's i/v though I will probably some time this summer.

I'm interested in showing up, but I won't need a hotel room there

Get the pico; it isn't really even a contest. The iMod is nothing more than a substitution of electrolytic (or sometimes I guess teflon or other film cap) for the tantalum caps. This can be a significant improvement, but you're still limited by the performance of the on-board codec which is, even by Wolfson's standards, fairly average from the standpoint of what you can achieve if your goal isn't portable-class power consumption and integration. I like my iPod Touch a lot, and the portable codecs from Wolfson are some of the best available, but they are not in the same class as Wolfson's high-end desktop offerings (e.g. WM8740, which is what is in the Pico). Secondly, the pico buffering schema is very good and uses an extremely low distortion chip. Plus, I have some doubts that the 404 headphone amp is in the same class as the Pico headamp that uses an optimised AD8397-based solution.

I use ATH-CK7s with my iPod Touch. It works rather well and provides good isolation for portable use, which sounds like what you're intending. I'm not outright thrilled by some of the mids emphasis (kind of a class AT thing to do) but I just recently got them and it does seem to be balancing out with some hysteresis in the drivers from initial use. The PK2 is also a good suggestion, especially if you don't want to go the canalphones/IEM route. I seem to be in the minority in that I think the PK2 sounds substantially better than the PK1. It's a great earphone overall, but the isolation may not be enough for you if you're in a noisy environment.

I think the RT is now under the Audiocap branding, with the RTX (still Polystyrene) being in the Multicap line. From what I understand, they're both excellent for audio applications.

You could try Multicaps (polystyrene dielectric).

PCM1704 is only +/- 1.2mA. If they aren't using much gain, they're using a relatively high value resistor, which would be even worse.

I sort of regard passive i/v as the common fallacy of simple is better that seems to float around in audiophilia. No more complicated than necessary (standard razor principle) might make sense, but simplicity for simplicity's sake seems unwise to me. As you note, these chips are designed to have a near-zero impedance on their output. You're right that op-amps can't provide that into the MHz range; not even close. The feedback factor drops into oblivion by that point. Although, an old revision of the AD8055 used to have about 70dB of OLG to 2MHz (see - http://www.jancorver.org/info/atv/baseband/files/AD8055.pdf). Current steering i/v seems to be the way to go. luvdunhill - I can barely read the charts on that article. The distortion doesn't look particularly low to me, though. They also probably have to gain the signal quite a bit to get line-level output, which just means you throw in distortion from the VAS (in this case, resulting in what seems to be a lot of odd-order products). Might be worth trying to install a current steering i/v and see how you feel about it.

Oops; I meant to say PCM1738, though it's essentially the same deal with the 1704; arguably an even bigger problem.

Jeez; I'd hope not. 1704 + Passive is bad news; that chip is not designed to be used like that and you can get a boatload of distortion by doing it. It'll probably still sound OK, but you're really squandering the potential of the 1704 by doing it. If you don't want to do an integrator w/ NFB (e.g. op-amp) you can do a current steering i/v which just steers the current into a resistor, so that way the converter sees near-zero impedance on its output but you're still relying on a resistor on the end for i/v. The AD844's design allows it to be used in a common base config as far as I understand. I've used the AD844 before, both in the standard manner and I think at least similar to how Pedja does it. Seems to work okay, but I've found I'm not a big fan of CFB op-amps for the most part. For I/V, I've had great luck with the THS4031, and liked the AD825 (though I get better results as a filter/buffer). AD829 and AD8021 seem like they might work well, too. My DAC, though, will hopefully soon have a current steering I/V in it.

Either seems like a reasonably interesting choice. It would have been nice had Krell used the PCM1704 but I guess this needed 192KHz spec'd support? I'm not sure. From what I understand, the 1704 actually runs at wider bandwidth and will apparently take up to 768KHz input, so I would think you could do 192KHz + 4X OS. Oh well. The other thing is, at least looking at the board pictures I've seen, I can't figure out how the Krell does I/V if it's not op-amps, which is confusing considering the really beefy discrete output stage (e.g. seems like output stage would be overkill for the I/V). If it does use op-amps, it might be worth looking into upgrading them with something higher performance depending on what's in there. I like using the THS4031/32 for I/V. The Marantz looks like it's a similar config (op-amp I/V, discrete output stage) but I'm not sure which is better between the converters being used (the NPC vs the BB), and it looks like the Krell output stage design might be more sophisticated.

They're supercaps; frequently called "double layer" etc. Nichicon lists a series that goes up to 4000F (http://www.nichicon.co.jp/english/products/pdf/e-jd.pdf), although maybe they have some higher capacity ones that they aren't listing. EPCOS apparently makes some 5000F ones. Digikey and Mouser carry Supercaps, though Digikey's selection seems to be better.

What? I have K271s and I don't find them difficult to drive at all.

AFAIK neither is true. 2vRMS is pretty common nowadays, and in the past you might see between 2vpp and 2.828 (1vRMS). 1vpp just makes it extremely difficult to get the full dynamic performance out of the converter.

A lot of the sound is probably due to the PCM1702. The 1702, along with the 63 and 1704, are tremendously nice sounding converters IMO. However, you can improve the performance of your DA-500 by changing the op-amps. It seems to have M5218 and NJM4570 op-amps; change the i/v converter to THS4032, and use maybe a single-to-dual adaptor with a couple of AD825s for the buffer. Add in some bypassing/decoupling if there isn't any. Try maybe 0.1uF across the rails and 100uF to ground, although there may be more optimal configurations. I think you'll enjoy the results