Five months after posting my initial "preview" of ECDesigns’ PowerDAC-R I would like to offer some concluding comments on my experience with this unique product.
I will also include additional information that was provided by ECDesigns in answer to some of my questions. Some of this information may already have been mentioned in the long ECDesigns thread on this forum, but it will be convenient to have it available here for future reference. Even if one does not grasp all the technical details, these explanations will surely convey the degree to which the PowerDAC is unique and how the solutions implemented contribute to one objective: obtaining the highest possible fidelity/accuracy in the digital to analog conversion process.
As ECDesigns is currently working on more powerful versions of the PowerDAC that are meant to be used directly with speakers, the comments below may also be helpful to understand what can be expected of these new versions.
Since this initial write-up, I have been able to compare the PowerDAC-R to the following DACs: Audiomat Tempo 2.9, Denafrips Terminator, Rockna Wavelight, Aqua La Voce S3. I wish I could have added the Mola-Mola Tambaqi to that list, as I have been curious to listen to it. These comparisons were conducted with other "audiophiles" in their homes. We did not always agree on everything, but it is fair to say that the PowerDAC generated interest and left most impressed. DAC comparisons are not easy, as the sound quality is so dependent on the rest of the system. My takeaway from all these tests is simple: these different DACs have more in common with each other than they do with the PowerDAC. There are several reasons, I believe, for the unique sound quality of the PowerDAC-R.
Source immunity of the PowerDAC
In my previous article (PowerDAC-R "preview") I attempted to explain how source immunity was achieved by ECDesigns. Since then, some further technical explanations were provided in the ECDesigns thread. For future reference, here is what John Brown explained:
The PowerDAC-R -only- collects the data from the incoming signal and places it in RAM (Random Access Memory). Next, the incoming sample rate is measured with a software-based frequency meter. So now the data is available, and the playback rate is determined.
Low jitter master clock (Vectron, Pletronics, -85 ... -90dB phase noise @ 10Hz) is used to generate a clean latch signal for the Fractal D/A converter. So we are -not- re-clocking a jittery clock signal and creating said radio transmitter. We are generating a new, clean clock from scratch that will be used to latch the data (we stored in RAM) into the Fractal D/A converter.
The Fractal D/A converter does not work with I2S or other serial data interface and requires no clocks to clock in serial data because there is no serial data. All problems related to high frequency clocks and high frequency serial data are eliminated this way.
All bits are presented to the Power D/A converter simultaneously (parallel). Then we wait until all bits are stable and switching noise is completely gone (because the bits no longer change). Now one single latch pulse (derived from the low jitter master clock) will write the new sample to the D/A converter output at the moment there is minimum electrical interference. This helps to minimize trigger uncertainty and provides a clean, low jitter latch signal.
In other words, we only have to load new data and latch this data once every sample. The highest frequency we need (data and latch) equals the sample rate (44.1 ... 192 KHz). With conventional DACs we need much higher clock and data rates, typically up to 24.576 MHz. These higher data and clock rates produce much more switching noise and because there is no radio silence during latching, jitter at the D/A conversion circuit will be rather high regardless of master clock phase noise (jitter).
The bandwidth we need for getting data into the PowerDAC-R D/A converter is therefore only 192 KHz. This helps to keep most of the source noise that still has to enter through Toslink out of the D/A converter output signal.
So what have we achieved with this?
1) The single, low frequency latch signal is -completely- independent from the source and because we use professional clocks, phase noise and related jitter is very low (data sheet specifies -85 ... -90 dB phase noise @ 10Hz offset). And that is pretty good.
Because the latch frequency is much lower than this master clock frequency, the impact of this phase noise is further reduced, this is not the case with standard DACs that need the native master clock frequency for serial data clocking.
2) The large bandwidth noise from the source (bandwidth up to 1 GHz is required to get the data through) is band limited by the optical Toslink interface to approx. 25MHz. This minimizes the noise injection into the D/A converter. Because we can't fully block all noise, marginal source dependency remains but, in most cases, this is so little that it's a non-issue
Owners of the PowerDAC-R were able to find out for themselves whether source immunity was indeed a reality. Opinions diverged.
I continue to use the PowerDAC-R with a very basic "non-optimized" source and could not be happier. Sometime after receiving the PowerDAC, I subscribed to Qobuz. The PowerDAC offered the same quality whether streaming my local files or albums on Qobuz.
When listening to the PowerDAC-R in others' systems, I always connected their high-end USB sources using ECDesign's own UT96 USB to optical converter. This small converter is a very basic model that costs only 150€: it does not contain fancy clocks, power regulators, isolation, etc... This is clearly not a piece of equipment one would normally find in a "high-end" digital chain. Yet, during these listening sessions, it never seemed to get in the way. How could that be? I can only conclude that the PowerDAC does correct for the high-level jitter associated with a Toslink connection while benefitting from the Toslink's perfect galvanic isolation from the source.
Fractal DAC architecture
The PowerDAC is based on an R2R ladder, but with ECD's own extended "Fractal" solution. I asked John Brown to explain this, and he was kind enough to provide the following illustrations.
The Fractal D/A converter is completely different from existing D/A converters as it chops up the most significant bits (the most critical one's for obtaining low bit errors) into tiny fragments. This is what the fractal circuit does. The fragmenting reduces bit errors and glitches. The downside is that we have to use more bits that represent the same bit depth (circuit gets a bit more complicated).
An R2R DAC with 16 bits resolution can output 2ˆ16 = 65536 different analogue levels. The bits represent the following values:
bit0 = 32768
bit1 = 16384
bit2 = 8192
bit3 = 4096
bit4 = 2048
bit5 = 1024
bit6 = 512
bit7 = 256
bit8 = 128
bit9 = 64
bit10 = 32
bit11 = 16
bit12 = 8
bit13 = 4
bit14 = 2
bit15 = 1
If a DAC outputs 2 Volts, the MSB ("Most Significant Bit" - bit0) would represent 1 Volt. 1 bit error represents 2 / 2ˆ16 = 30 microvolts. So in order to maintain 16 bit accuracy, the MSB voltage would need to be accurate down to at least 30 microvolts and that's very difficult to achieve with any practical circuit. The lower bits bit6 ... bit15 represent much smaller fragments of the output signal and are therefore less critical as an error here has less impact on the output signal.
The Fractal converter fixes this problem by chopping up the problematic MSBs (bit0 ... bit4) into smaller fragments. The remaining bits can be converted using a R2R ladder.
The fractal converter in the PowerDAC-R has 18 bits resolution that represents 262144 levels (auditory system absolute maximum resolution). So we start off with larger numbers:
fbit1 = 32768
fbit2 = 32768
fbit3 = 32768
fbit4 = 32768
fbit5 = 32768
fbit6 = 32768
fbit7 = 32768
The remaining bits are converted using R2R ladder:
bit3 = 16384
bit4 = 8192
bit5 = 4096
bit6 = 2048
bit7 = 1024
bit8 = 512
bit9 = 256
bit10 = 128
bit11 = 64
bit12 = 32
bit13 = 16
bit14 = 8
bit15 = 4
bit16 = 2
bit17 = 1
We now have a 22 bit fractal converter with 18 bit resolution. The advantage is that the fractal bits have less weight compared to bit0 and bit1 of a comparable R2R ladder DAC (bit0 would represent 131072). This translates to reduced bit errors and lower glitch (glitch energy reduced from 32768 to 8192 or factor 4). Another advantage is that we can now obtain 8 times lower output impedance with the same bit switches and resistors as the fractal resistor value is 8 times higher compared to a R2R ladder converter. With same error introduced by a MOSFET switch (internal resistance RDSon) it will have 8 times less impact on bit accuracy.
Volume control and signal path
I mentioned in my "preview" how the volume control implemented in the PowerDAC was so unique. I also asked John Brown for some explanations of this mechanism.
In the PowerDAC-R the 3dB attenuation steps are obtained by selecting 2 different D/A converter supply voltages, 3V1 and 4V4.
Bit shifting aligns the same data with less sensitive (fractal) bits in the D/A converter. Each shift offers -6dB attenuation. Both power supply and bit shift attenuations can be combined as follows:
0db, 4V4 supply voltage, no shift (vol.9).
-3dB, 3V1 supply voltage, no shift (vol.8).
-6dB, 4V4 supply voltage, 1 shift (vol.7).
-9dB, 3V1 supply voltage, 1 shift (vol.6).
-12dB, 4V4 supply voltage, 2 shifts (vol.5).
-15dB, 3V1 supply voltage, 2 shifts (vol.4).
-18dB, 4V4 supply voltage, 3 shifts (vol.3).
-21dB, 3V1 supply voltage, 3 shifts (vol.2).
-24dB, 4V4 supply voltage, 4 shifts (vol.1).
-27dB, 3V1 supply voltage, 4 shifts (vol.0).
Within the limited bit-shift range, no audible degrading occurs and no extra components are added to the signal path. The PowerDAC output impedance remains constant regardless of volume setting. With any kind of analogue volume control, there will always be degrading (noise, distortion) and the in / output impedance (and related sound) changes with each setting.
In my “preview” 5 months ago I mentioned testing the volume control by comparing the sound with and without a preamplifier. I have settled on using the PowerDAC-R directly connected to my power amplifiers and no longer use my “reference” preamplifier (The Truth). I was able to test a direct connection to tube power amplifiers in two other systems and compare it with the use of a preamplifier. It was very interesting to hear the change in sound. The results are sometimes surprising, as a preamplifier can compensate for some of the shortcomings of an amplifier! In one case I did prefer the “direct connection” but in the other it was harder to decide. The PowerDAC-R is limited to 10 volume steps (total gain of 30db). In some cases, that may be too sensitive to offer sufficiently fine control over a given power amplifier and speakers.
The PowerDAC may not offer a perfectly transparent signal path - no piece of equipment ever will - but it may very well come closer to that ideal than any other audio system has to date. This minimal signal path is composed of very few components. The analog signal is directly produced from an electrical power source (linear, regulated) going through MOFSET switches and resistors. Nothing more.
With speakers, the powerDAC-R's performance is still very dependent on the rest of the analog system: cables, amplification, speakers. However, the powerDAC-R used with high quality headphones directly connected to its RCA outputs can already give us today a good idea of what to expect with the future S model and speakers. Of course, the result will depend on the quality of the headphones used - each headphones introducing some level of distortion. Surprising results may be obtained with some headphones who’s low-level of distortion were not so obvious with conventional headphone amplifiers.
The forthcoming S models will contain a larger number of Fractal bits, extra volume steps with a much higher maximum volume level and lower output impedance. The higher number of Fractal bits will provide even higher accuracy, and the ability to drive speakers directly from the PowerDAC's outputs will eliminate the need for traditional amplification. Exciting times ahead!