 Exactly how ESS DAC's process DSD has been kept a bit of a mystery by ESS, and certain players in the industry (Mytek and Benchmark come to mind) get the nature of conversion, especially the nature of the volume control, egregiously wrong. For a BAD example of how the ESS chips implement DSD volume control, Benchmark, who otherwise is so excellent, gives us what sounds like more like a wild guess than reality. from the Benchmark website: "The net result is that the 6-bit converters have the near-perfect linearity of a 1-bit converter while achieving an 18 dB reduction in noise (due to the 64:1 parallel structure). This improvement delivers a 6-bit sigma-delta modulator that has an 18 dB noise advantage over a classic 1-bit sigma-delta converter (such as that used in DSD). The array of 1-bit converters also allows native DSD conversion with digital volume control. This combination of features is very unusual, but the ES9018 provides a unique solution to the DSD volume control problem. Normally, it is very difficult to implement a digital volume control (or any other form of digital processing) in a 1-bit DSD system, but with an array of 1-bit converters, we can set the volume by controlling how many DSD converters are turned on." ( you can click the black text to read for yourself on Benchmark's site). NOPE. Maybe if we wanted a mere 64 level volume control this could work, at the large expense of resolution and linearity. The digital volume control of ESS chipsets is a real 32 bit solution. 6 bits? No, that will not cut it. For 10 years I have searched for the answer to this question. Unfortunately, I have a hard time seeing things right in front of my eyes, tending to pay more attention to the periphery. So, I began downloading every datasheet for every chip ever made by ESS to see if I could find any clue. And then, as if I had been skipping over the answer since 2014, there is was. The function diagram that had alluded me for so long. You can see it at the top of page, and I will repeat it again right here. I posted to Head-fi in the HQPlayer forum this very signal path, along with an explanation how this architecture is actually nothing special compared to what we already know about DSD-wide, and how DSD is processed for digital volume control via Cirrus, AKM chips as well as others. The following post I made on head-fi. I hope you find it enlightening.......... ESS again has been tight-lipped on their method of DSD conversion, but I DID happen to find a diagram from one of their datasheets that answers a LOT of questions.
This confirms what you may have heard explained over the years by some really high-end, well respected engineers is not really correct, or at least they were obfuscating. The truth seems to be that the DSD path is not that different from what we have come to know about things like Sony's DSD-Wide, or how Cirrus, AKM and others covert DSD with DSP/Volume Control. The Digital Signal Path shown shows what I have suspected all along. DSD first encounters a FIR low pass filter, which creates a multi-bit signal. The signal is not 'decimated', and remains at the same rate as DSD input. (Decimation can mean more than one thing to me; in this case I say not decimated because no samples are removed, even if there is some redundancy.) The filter output can be just a few bits wide, but IF the volume control is used to attenuate, the output of the FIR filter is multiplied by a 32bit gain control, creating an even wider sample. Due to bandwidth restraints, we are probably not in a DSD unary code; we will be in binary. Next the binary code DSD multibit intermediate is sent to the IIR filter that is user selectable at either 47khz, 50khz, 60khz, 70khz for further noise control before it is sent to the noise shaper. ( Not shown in diagram is the sample rate converter and Phase Lock-Loop for jitter reduction). The noise shaper works as a high order multi-bit Delta Sigma Modulator. Considering each individual DAC (some ESS chips have up to 8 individual channels) has 64 unary/thermometer elements per channel, the modulator probably operates at 6 bits (binary coded). The 6 bit binary output of the Delta Sigma modulator at whatever oversampled speed is used (x128 or x256) is send to a logic system that converts the 6 binary bits into a 64 level UNARY code. ( note, that in unary code, 64 elements would actually mean 65 levels because all elements 'off' is zero. Likewise, 63 elements would be required for 64 (6-bit) levels). Furthermore, ESS uses here what they call the 'revolver' technique, which is effectively no different than the highly lauded DCS 'Ring' DSP. Levels are 'scrambled' since unary code is essentially multiple 1-bit signals that when added together equals the correct amplitude. It doesn't matter to which element the logic 'sends' the data. The output elements are all equal, and this creates exceptional linearity, avoiding element mismatch. Now HERE is where we get conversion that is similar to the Signalyst DSC1 (DSC2 DSC2.5 etc). DSC1/2/2.5 uses 32 equally weighted output elements. Unlike the ESS which receives multi-bit delta sigma with 6 binary bit / 64 individual levels, the DSC1 receives 1 binary bit DSD. It uses shift registers to create 32 1-bit streams. All the same 1-bit stream, but each stream is offset by 1 clock cycle that, along with the output elements, form a FIR filter, which in this case, is a type of very simple moving average filter, a CIC filter. (If you ever see the output of a CIC filter on chart, you will see the null points that give it its comb filter name) Yes, the DSC 1 design has similarities to the ESS in that both use unary coded 1-bit conversion for the DAC. The ESS however is a true multi-bit signal. The DSC1, along with many other brands who do something almost exactly the same, is really just a 1-bit signal as far as actual information is concerned. But the method of conversion is very, very similar. The 1-bit way is less complex, and there are those who prefer as little complexity and as little DSP as possible. I am not here to say which is better, or if one party is right and the other is wrong. From my PERSONAL perspective, I would use the DSC2 with HQplayer any day. Thankfully I am lucky enough to have one of the very rare in the wild DSC2 converters.
0 Comments
Leave a Reply. |
RSS Feed