Well friends, it's time to publish some measurements of my new RME ADI-2 Pro FS R Black Edition ("FS R BE") unit as DAC. For background, refer to the Preview from August and a couple weeks back, I had published results discussing the ADC function.
As you might be aware, since 2018, I've been using the older RME ADI-2 Pro FS ("Pro FS") for ADC and DAC duties in my measurements. It has performed like a champ and in fact, given that the ADC function is almost identical between the two units, there's certainly no reason to retire the "Pro FS" from measurement duties.
What we know is that as a DAC, the "FS R BE" should perform better than the "Pro FS" due to the fact that the internal AKM chip has been upgraded to the AK4493 from the previous AK4490. On paper, this represents a +5dB improvement in SNR and +1dB improvement in THD+N so we're not looking at Earth-shattering differences in specs numbers. Let's see if we're able to measure these differences on the test bench and identify any other changes to the DAC's performance along the way...
To start, let me show you what the equipment test set-up looks like:
For USB digital source, as you can see, I'm using a DIY Raspberry Pi 3B+ "Touch" sitting on top of the "FS R BE". I'm running Volumio on it loaded with the test signals in the USB stick poking out on the left and it's connected to the RME by a 6' length of generic USB cable. Notice that I've also taken out my trusted Logitech Squeezebox Touch. I suspect that a few of you might still be using TosLink and coaxial S/PDIF - the Touch (connected to Logitech Media Server via WiFi) can provide S/PDIF input to the "FS R BE".
For consistency, I will use the "Pro FS" as the ADC measurement device for much of these measurements. However, there will be times where I'll do some loopback of the "FS R BE" itself which I will point out. Unless indicated otherwise, measurements will be of the XLR analogue output from the "FS R BE" sent to the "Pro FS" by 6' lengths of Monoprice Premier cables. Some generic USB then will connect the "Pro FS" to my Intel NUC for data processing.
As with the previous ADC measurements, "FS R BE" firmware is USB 238/DSP 98, and "Pro FS" version 213/98.
Part I: Digital Filters and "Microscopic" Properties
I typically start by looking at some of the "microscopic" properties of the DACs. Usually, I'll hook the DAC up to my oscilloscope and have a peek at square waves and output voltage levels. For the "FS R BE", let's not worry about this because it's the same as the "Pro FS". Furthermore we already know the reference output balanced levels which are +4/+13/+19/+24dBu (corresponding to 1.23/3.5/6.9/12.3Vrms). See here if you want to check out pictures of square waves off the "Pro FS".
One area of interest over the last decade has been selectable digital filters. For the "FS R BE", we have the choice of 6 filters (1 more than the "Pro FS"). Here they are with pictures of the impulse responses (note that I tried to casually center these rather than being reflective of relative impulse latencies) along with the Juergen Reis-inspired "Digital Filter Composite" graphs I've been using for a number of years to examine filter frequency-domain characteristics and to look for overload behaviour:
The first 4 filters are relatively "orthodox" settings. Basically we're looking at minimum vs. linear phase settings with either sharp roll-off or short-length filters with resulting slow roll-off but more imaging/"aliasing" artifacts beyond the Nyquist frequency. In the graphs above, I used 44.1kHz signals so Nyquist would be at 22.05kHz. Remember that filtering is primarily of value for 44.1/48kHz signals; with 88.2/96+kHz samplerate, unfiltered artifacts would be at least an octave beyond human hearing.
The DFC graphs look even better than the older "Pro FS". No hint of overloading at all. This means that a few dBs of overhead have been added. Nice.
Now the last 2 filters are different "by design":
"NOS" setting is our "Non-OverSampling", unfiltered option which allows artifacts beyond Nyquist to pass as you can see in the DFC graph (NOS discussed previously). And for fun, we can have a peek at a 16/44.1 1kHz sine wave through this filter:
Oh those are pretty stair-stepped "digital" waveforms, right? ;-)
The "SD LD" filter stands for "Short Delay Low Dispersion" and we've seen this AKM option in other DACs, most recently the Topping DX3 Pro (V2). It looks like an intermediate phase filter of sorts discussed in the RME manual on page 83. It's steep enough to attenuate most ultrasonic distortions, provides low latency for realtime monitoring work, and maintains relatively linear phase characteristics within the audible spectrum. While there's no intersample overload behaviour between the 0dBFS and -4dBFS white noise signals on the DFC graph, note that the ultrasonic noise floor is a little higher than digital silence; not that this is an audible issue of course since we're looking at below -120dBFS.
Digital filter options are an interesting feature but IMO unless the filter is highly anomalous (like NOS or the highly relaxed PonoPlayer/Ayre filter), there's really not too much to say or hear. I'm happy with the linear phase "Sharp" filter for music enjoyment which I keep as default for both DAC and ADC. (Or I might fool around with the Ms. Goldilocks filter for fun! :-)
Another "microscopic" test that's nice to show as a reflection of the resolution for a DAC is the 1kHz -90.31dBFS undithered 16-bit tone:
We can easily see the defined digital levels of the undithered lowest 16th bit and the DAC is able to better reproduce the intended -90.3dBFS 1kHz sine waveform at 24-bits.
I usually show this overlapping for the two channels so we can examine for any imbalance and DC offset - no problems here. Basically this tells us that the RME DAC is capable of resolutions beyond the basic "standard" 16-bits and has excellent channel balance even at such low levels.
Part II: RightMark Test Suite
Okay, so we know that the filters have been improved and we know that the "FS R BE" is obviously capable of high-resolution reproduction... The question then is "How much?" in comparison to the older "Pro FS" and compared with a reference (let's use my Oppo UDP-205).
While it rarely yields useful information, I do like to start with standard 16/44.1 measurements still simply because most of our music remains "standard resolution". Here's a summary table and the specific graphs for comparison (all measured with XLR output, RME devices set to +13dBu reference, linear phase "sharp" filters used throughout):
|Note for the frequency response graphs, I used the "sweep sine" measurement here. The "multitone" result can be skewed when presented with slight clock differences.|
The only notable difference in these charts is the IMD+N sweep where we can see that the "FS R BE" and Oppo perform better than the older "Pro FS".
Next, let's go into 24/96 Hi-Res territory:
As you can see, we have 24/192 and 24/384 numbers here for the "FS R BE". As noted above, I used the +13dBu (3.5Vrms) reference level for these RME measurements. Notice that I included a 24/96 result at the highest +24dBu (12.3Vrms) output level and was able to get slightly better measured dynamic range although likely this is hitting against the limits of the measurement ADC again.
Oppo UDP-205 and Topping DX3 Pro V2 results follow to the right in the summary table. I find the Topping DX3 Pro rather impressive for a little ~US$200 DAC!
Here are the graphs done with loopback 24/384 for the RME ADI-2 Pro FS R Black Edition, "Sharp" digital filter:
At 24/384, we have a flat frequency response with less than -0.5dB at 100kHz. Note that with loopback, RME has implemented a compensation filter on the DA side to keep the results essentially flat into 100kHz. This is a thoughtful nod to those who want to use the ADI-2 Pro for measurement purposes (described in the manual on page 79). Noise level is well controlled into 50kHz. Beyond the rising ultrasonic noise of sigma-delta conversion, there is no evidence of issues arising from the stock switching power supply used. Notice how precisely the two channels are balanced.
I would argue that when level-matched, if a supposedly "high fidelity" DAC sounds very different from a high performance machine like this RME or the Oppo, it would be a result of intentional distortion (like this ridiculous "euphonic" audiophile mod), noise/hum, software settings (eg. EQ, DSP applied), user error, or suboptimal system integration (eg. impedance mismatches). The idea that a hi-fi $10,000+ DAC somehow could inherently "sound better" than even a reasonable ~$200 DAC (like the Topping DX3 Pro) makes no sense these days apart from the above factors. IMO, price-performance correlations are meaningless now absent objective evidence.
Part III: Jitter
As I have said for years, if you really think jitter is audible, then stick with the asynchronous USB interface (asynchronous ethernet should work just as well). In general, TosLink is a more "jittery" interface, and depending on the DAC, commonly coaxial input is better than TosLink. [Try out the Jitter Simulation Demo.]
In the tests above on the "FS R BE", jitter is very low regardless of S/PDIF or USB input used. The obsessive audiophile will see that USB input is performing better with a narrower "base" of the fundamental frequency and also lower spurious noise (like ~17kHz with 24-bit S/PDIF).
If you're wondering, I did not use any fancy TosLink or coaxial cables of course. The coax is a homemade 3' length of Radio Shack cable with gold plated connectors I put together 20 years ago. The TosLink is some >20 year old 6' Acoustic Research branded cables with presumably plastic fiberoptic material.
Dare I say that "Bits Are Bits" with high quality modern devices these days? Of course I dare! ;-) Only low quality digital gear makes bits sound like anything else.
Part IV: A detailed look at THD+N
Using a -1dBFS sine wave at 24/96, 256k-point FFT, averaged over 4 reading, we see that the "FS R BE" with its upgraded AK4493 achieves a THD of -117 to -118dB and measured THD+N of around -113.5dB for both channels (the right channel was a little better even). This is within spitting distance of RME's specs reporting THD+N of -115dB at 0dBFS with my measurement perhaps also limited by the ADC. The "FS R BE" THD+N is about 4dB better than my older "Pro FS".
A THD+N of -113.5dB is in line with and even exceeds the "official" AK4493 spec sheet value of -113dB. If I push the "FS R BE" to +24dBu output level, I've seen the average THD+N better than -114dB.
A result of -100dB/0.001% THD+N is already phenomenally distortion-free for human consumption. Let's also not forget that the dB scale is logarithmic.
Part V: Level and Frequency Step Measurements
Using REW, we can now start looking at the detailed harmonic structure and linearity errors as we vary the generator level (level step).
As you can see, this is a very linear DAC from -120 to 0dBFS. I basically had to zoom in to show that tiny deviation in linearity above. And when we look at the harmonics across the levels even at the highest +24dBu output, it looks very clean with little "saturation" of THD+N approaching 0dBFS:
Part VI: DSD
Part VII: Headphone Output
|Note simple 1/4" to 3.5mm stereo adaptor then 3.5mm phono to RCA cable for headphone out measurements.|
Overall the graphs look great.