With current technology capability, we find that the best sound quality requires four distinct steps.
- Server. To take a music file or chaotic internet stream and stream it over Ethernet to the Player. This ideally involves very high power, but high power typically introduces high levels of noise.
- Player. To take the streamed asynchronous feed from the Server and send a digital audio signal over asynchronous USB to a re-clocker. This involves medium power with some noise penalty.
- Re-clocker. To take the asynchronous USB feed and turn it into a precisely clocked synchronous feed (I2S, AES3 or S/PDIF) to the DAC stage. This requires low power and so typically generates very low noise.
- DAC. To convert the digital signal to an analog signal. This is best to be physically separated from the first 3 stages, mainly due to noise considerations. It is a good idea to mix as little digital with the analog as possible.
Note that all of the first three steps involve some buffering and re-clocking. The precision of step 3 is important because the output signal is synchronous, but the sound quality is still impacted by the quality of the previous steps.
We sometimes see someone write that an asynchronous input is best because timing (clock) becomes irrelevant. This is utter nonsense. The DAC needs accurately timed data and an asynchronous input is relatively poor at that, by definition. In fact, in both the case of digital audio over Ethernet and over Asynchronous USB, the streaming protocols used apply a fair degree of timing discipline. The principal benefit of an asynchronous input is that you can control the arrival rate of packets to avoid the buffer at the receiving end running out or over-flowing, which would cause drop-outs. As you progress through the steps, the timing gets better, the size of the buffer gets smaller and the signal gets more synchronous.
Each of the first three steps listed above requires a different set of trade-offs to do its task optimally. A single piece of hardware can do all three steps, but cannot do all of them well. With current technology we still need these three steps of buffering and re-clocking to get excellent sound quality.
The big question is WHERE each of these steps SHOULD be located. With a CD player, all steps were more or less synchronous – enabling low power and low noise – but required error correction when a bit could not be read in real time. The level of error correction was much higher with CD players than anyone cared to admit.
With computer audio, separate DACs re-emerged as a product, but DACs had to grow an asynchronous input in order to deal with the low-quality signal coming from standard computers. Early on, firms like Genesis, Audio Alchemy etc., produced re-clockers, and in DCS’s case, they added a re-clocker/up-sampler into their stack.
But later, DACs added USB inputs, therefore incorporating both step 3 and step 4, and so heavy-duty re-clocking moved inside the DAC. Later, many DACs added Ethernet inputs, therefore incorporating steps 2, 3 and 4. But placing computer steps inside a DAC is a very poor architectural decision, and was done because so many audiophiles persisted in using low quality computers as their music server.
Ideally, each step should be completely isolated from the other, but if you combine any into a single box, then the K50 is the best architectural decision, because it includes all of the computer, and purely digital, steps together. And it keeps the computer steps away from the analog step.
To be clear, we are saying that the DAC should only have synchronous inputs, to get the computing steps out of the DAC. To receive high bit-rate files this means you would use I2S. There are several benefits to using I2S, including that it is what DAC chips need to be fed with anyway. I2S is specifically designed to get the clock data to the other end of the transmission as cleanly as possible.
Our view, that the re-clocker stage should be separated from the DAC, is not as controversial as you might think. For a start, most DACs have only very basic re-clocking chips and circuitry. At the higher end of the market, we are seeing a growing trend for high-quality re-clockers (step 3) to be separated from the DAC, such as with the top-end models from MSB, Playback Designs, Chord and Auralic.
Our point is that arguing whether Ethernet or USB or I2S is better than the other is irrelevant. All of those transmission methods are used in a computer audio solution, either inside the boxes or between the boxes, depending on the architecture. The argument is not about the transmission method, but about the architecture – what should be separated from what, and if you combine stages for simplicity, which stages can most safely be combined.
The difference between using the K50 USB output and using one of the Digital Outputs is not about the outputs, it is about the architecture. Using the K50’s Digital Outputs takes the signal one step further than the USB output. Whether a DAC’s best input is Ethernet, USB or AES3 depends on what the music server is that you feed it with. Given the very high quality of the K50’s re-clocker, we recommend the Digital Outputs over USB.
Regardless of the architectural arguments to remove the re-clocker from the DAC, I can confidently say that very few DACs have a re-clocker that comes anywhere near the quality of the re-clocker we now use in the K50.
For more information see our Technology page