MEASUREMENTS: Roland Mobile UA-M10 DAC (PCM, 1-bit, 4-channel, Balanced Out)

Roland UA-M10. Note colorful level LED, 2 volume buttons, 2x3.5mm outputs; the one on the right labeled "A", and the one to the left where I have a cable connected "B".

In late 2019 when I visited my friend one afternoon, he showed me an interesting DAC he owns. It's the Roland Mobile UA-M10. This device has been on the market for awhile (since around late 2015), it's also a little more expensive than most small-form-factor, USB-powered DACs out there these days at around US$200-250.

Physically, it's a small, light, aluminum box with two volume buttons up top and a bright LED indicator that dances with the music when playing. The bottom surface has a couple of rubber strips across the length of the box that protects from scratching whatever surface you're putting it on - nice touch. Like many other USB DACs these days, it's completely powered by the micro-USB 5V input. The heart of this box is the AKM AK4414EQ DAC which interestingly is a 4-channel DAC (see datasheet). The device makes use of this by having both a headphone out phono plug on one end plus a line out phono output on the other, each of which could be playing independent content. The driver has the ability to tell the DAC which mode it should operate in.

Phono output B and micro-USB digital interface. 1MORE Quad Driver headphones attached to output A.

Have a look at the device settings:

Note that I am running the latest firmware 2.02 and the Windows 10 driver is 2.0.0.

Notice that the settings panel is clearly quite different from your typical DAC. As you can see, it supports ASIO in Windows with the usual buffer size for control of latency vs. reliability. You can set the samplerate and have Windows match that with the tick box at the bottom. To the right you see the setting for "Output Mode" - there are 4 settings and these are described in the larger window on the left side.

For the most part, 3 of the 4 modes keep this box in stereo 2-channels output with either variable headphone output or fixed-level line out setting which can be adjusted by that "Attenuate level (line out)" setting. Notice that there is a 4-channel setting which is cool and you can imagine all kinds of things you might want to do with this. Maybe you're on a plane and one person wants to watch a movie on channels 3/4 while the other listens to music on channels 1/2. Pro audio software might be able to use the channels for monitoring/mixing outputs. Maybe it can act as a DSP crossover for 2-way speakers? While the attenuator has 4 fixed values (each step -6dB/1-bit attenuation) for the line out, there are +/- volume buttons on the top of the device for controlling headphone output and it's nice that the LED will respond telling you the volume level when the buttons are pressed.

One negative about this extra complexity is that the UA-M10 is not a plug-and-play device if you hook it up to a Linux box. For example, I tried a few ways to get this DAC to play nice with my Raspberry Pi 3 B+ running Volumio; I managed to get it running once but was unable to replicate the success in general.

That "1-bit" D/A conversion mode (red arrow) allows the computer to play DSD64 audio directly and resamples all PCM into DSD for playback to the DAC. I presume this resampling is done inside the DAC as I don't see an increase in CPU utilization - the product has something they call S1LKi "silky" DSP (ESC2 DSP chip inside?). Remember that these AKM DACs are internally delta-sigma converters anyway, therefore regardless of this "1-bit" setting, it's still a bitstream DAC; the only difference is whether the conversion from PCM to 1-bit is happening transparently inside the DAC or through Roland's S1LKi algorithm.

I see that a great post with objective data on this DAC has already been published by Raoul Trifan back in late 2016 on Head-Fi (as you may recall, Raoul wrote an article for this blog measuring the +5V output from USB ports back in 2018 - thanks man!). I'm using the same drivers and firmware as him so what he described there applies to this post as well. Have a look at his results for more around headphone power, opamps, and the relatively high output impedance. Also, the comments about 4-channel mode not being accessible to ASIO is true (4-channel playback through WASAPI and DirectSound only), so make sure to read that informative post for more details.

Since I'm not a big headphone guy, I'm glad that Raoul has already discussed much of that :-). My interest today is more to document a few other items like objective measured sound quality just to "round out" the capabilities of this DAC if you own one or might still be interested in one of these.

For reference, here is the measurement chain used:

ASUS i5 Windows 10 laptop --> generic USB-A to micro-USB cable --> Roland UA-M10 DAC --> 6' phono-to-stereo RCA --> RME ADI-2 Pro FS ADC --> generic USB --> measurement Microsoft Surface Pro 3 laptop

Remember that over the years, I've used different ADCs for measurement, so the results obtained today will be with the RME running on battery power for best resolution.

For simplicity, the majority of the results will be done with 0dBFS 2-channel "A:phones, B:line out" mode through output B. We'll compare the difference "1-bit D/A Conversion" makes, as well as a quick look at the 4-channel and balanced output modes!

I. Oscilloscope, Impulse Response, and Digital Filter

Through the digital oscilloscope, here are the standard 1kHz 0dBFS sine wave and bandwidth limited -3dBFS square wave playback (16/44):

That looks pretty good. The two channels are well balanced. 0dBFS sine wave is clean without clipping at the peaks. Likewise, good looking bandwidth-limited square waves. Now, let's click on the "1-bit D/A Conversion" option and see what happens:

Well well well... As you can already see, 1-bit conversion adds a significant amount of noise to the tracings. Notice also that accentuated "post-ringing" with the square waveform suggesting that with standard PCM playback, the DAC is using a linear phase filter whereas when you turn on "1-bit", it's now a minimum phase filter implemented. Despite these differences, turning on 1-bit mode does not significantly change output volume which is important for A/B listening.

We can confirm the type of digital filtering used by having a peek at the impulse responses:

Notice that they both maintain absolute phase, and both are relatively steep filters (characteristic of the long 'ringing'). The difference is that standard PCM is a linear phase filter compared to the 1-bit setting being minimum phase. As expected, unless a strong low-pass filter is applied, DSD conversion results in more ultrasonic noise which is exactly what we're seeing here in the traces.

With regards to filtering, this is what the "Digital Filter Composite" (inspired by the "Reis Test") looks like with standard PCM (16/44.1) signals sent to the DAC and the corresponding 1-bit setting as well:

As we can see, both are steep filters with the PCM (non 1-bit) setting a bit steeper and cleaner overall. In the frequency domain we can easily now see the amount of ultrasonic noise rising beyond 22kHz when "1-bit" setting is activated. Both filters show a little bit of intersample overloading with the 0dBFS signal that is cleaned up with the -4dBFS signal. Notice that the 1-bit setting results in a couple of tonal peaks at 38kHz and an octave above around 77-78kHz when playing digital silence.

This is an early taste of what the 1-bit conversion is doing. For the next sections, let's focus on standard direct PCM playback and come back to the "1-bit A/D Conversion" setting in Section III.

One last thing I checked was the headphone output impedance on Output A. Using my setup, I got a result that was higher than what Raoul measured; he found 10Ω while I got a result of 15Ω using a 1kHz signal and 51Ω resistive load.

II. RightMark Audio Analyzer (straight PCM)

As you can see above, the 1-bit mode results in quite a bit of noise coming out of the device. I don't consider this setting optimal because these frequencies are simply not part of the audio signal being sent to the device and represent a form of distortion. As such, let's start with just straight PCM to get a taste of the DAC's resolution abilities.

16/44:

Remember, while everyone was all excited about high-resolution audio especially a few years ago and we have more content these days, good ol' CD-resolution 16/44.1 is still the most important samplerate/bitrate for all DACs. And indeed, across the board these days with modern DACs, there is little differentiating devices. Assuming one controls volume and potential issues like impedance mismatch, for the most part, DACs will sound the same if they measure similarly. Looking at the graphs above, we see some outliers. For example we know that the PonoPlayer was designed with a very weak minimum phase filter so it rolls off earlier than other DACs which could be audible. Also, we know that the AudioQuest Dragonfly Cobalt particularly has more distortion "baked in" as you can see; not ideal but this would not be terribly audible.

The Roland performs well and we'll talk about the subjective sound quality below.

24/96:

Since I found the Dragonfly Cobalt to be overall inferior to the Red, I took that one out and added a comparison with the Topping D10 DAC (measured here) as another one of these "USB box" type devices (and quite inexpensive at <US$100).

While the Roland's distortion level is good compared to something like the PonoPlayer, it does tend to be a bit noisier with about 17-bit dynamic range when fed a 24-bit signal. As expected, the Oppo UDP-205 (RCA out) with its ESS9038Pro DAC is capable of 20-bits performance and remains my high-fidelity reference.

24/192:

Alas, RightMark has a bug with its graphs for crosstalk and distortion sweeps at 192kHz. Nonetheless we can still see that the Roland, like with 24/96, has a higher noise level compared to the other DACs including the PonoPlayer, but it does have good frequency extension.

III. With 1-bit Conversion

Now let's turn our attention to the Roland's touted "silky" S1LKi, "1-bit" DSP mode.

In the summary above, we see what happens to the numerical measurements when you turn the 1-bit conversion on. I trust for the audiophiles who have followed my blog over the years, it comes as no surprise that 1-bit/DSD conversion of PCM is not a lossless process. It will add some noise and distortion; inevitably we will see a reduction in objective fidelity of varying degrees.

While the conversion is doing a fine job with a "simple" 16/44.1 signal, let's look at the graphs for the 24/96 and 24/192 data to get a sense of how the conversion is treating hi-res test signals...

It does a pretty good job with 24/96 as you can see. Notice the rise in noise level just after 20kHz which tells us that this is basically conversion to DSD64. You'd need a DSD128 level of performance to keep the modulator noise down beyond 40kHz. Here are 192kHz graphs:

Again, notice the very significant rise in noise just after 20kHz. There's little low-pass filtering being applied and we can see a rather high -54dBFS noise level at over 80kHz (the true noise level might even be higher than this because the RME ADI-2 ADC is applying its own filtering).

Remember that these kinds of results, especially the noise level would be familiar to those of you who may have used DSD <--> PCM converters in the past like these.

IV. Jitter

Let me show you some rather "beautiful" high-resolution 16-bit and 24-bit J-Test graphs:

Very nice, eh? Tiny +/-4kHz sidebands on either side of the 12kHz fundamental on the 24-bit J-Test. These can be seen beneath the jitter modulation tone in the 16-bit test as well. Nothing of audible significance.

As I've said time and again, jitter is simply NOT a problem these days; even with a relatively inexpensive USB-powered device like this released around 5 years ago. Remember this when uninformed audiophiles continue to complain that "bits are not bits" and that timing/jitter can be a problem despite the simple fact that it is not an audible issue these days with any reputable DAC.

I know, people still want to sell us stuff like this. Uncontrolled sighted listening impressions where people claim to hear "obvious" or "clear" differences and attribute them to "jitter" should be safely ignored unless they can come up with some reasonable evidence.

V. 4-Channel & Balanced output modes!

Let's now look at a couple interesting features this DAC has over most others. First, this DAC is capable of 4-channel output which is very cool - I don't know of another USB-powered DAC with quad-channel capability. As I suggested above, maybe you can make use of this as a simple 2-way digital crossover, perhaps quad-channel surround playback, or if you just want dual independent stereo playback from one computer (eg. sharing music playback for a friend while watching a movie!).

I was curious if 4-channel mode sound quality suffered when playing 2 streams simultaneously. Let's test this out.

On Output A (channels 1/2), I'll have VLC play a movie (the excellent V For Vendetta, HEVC 1080P Blu-Ray rip, AC3 soundtrack), while I play the 24/96 RightMark test on Output B (channels 3/4) to my RME ADC! To keep it simple, both streams are set to 24/96 Direct Sound output, 100% volume for the test audio channel, ~70% movie volume which was quite loud for my headphones. The test signal channel at 100% would be "bit-perfect" without using ASIO or WASAPI.

Here's how the computer screen looked while I was running the test - notice that I'm using foobar for playback of the test signal and you can see the Roland Device Settings set to 4-channel mode:

RightMark 24/96 results:

As you can see, there is essentially no difference whether I used 2-channel mode or 4-channel with all channels playing! Note that this is a rather old i5 CPU ultrabook (2.6GHz dual core i5-3317U), doing software playback of the 1080P HEVC-encoded movie plus AC3 decoding. Despite all the extra multitasking CPU cycles needed, notice that there was really no significant noise added to the analogue output, and barely any difference with the IMD+N distortion sweep. At most, I suppose if we peek at the crosstalk graph, we can make out a consistent marginal difference with slightly higher 4-channel crosstalk.

Let this serve as a reminder that with bitperfect output, typically there is no noise added to a signal even when a CPU is working relatively hard (here's another test example). So long as the DAC is getting the bitperfect audio data without buffering issues, there is no sonic difference regardless of the software running (this includes playback software and OS).

The Roland, like the PonoPlayer, can output in balanced mode, I can use those same 3.5mm phono-to-XLR cables for Pono and measure if there is any benefit with this mode (no point testing 16/44.1 with balanced output, let's just look at 24/96 and 24/192):

Simply put: not worth the hassle to use this DAC in a balanced configuration. Notice that the only parameter that improved marginally was the stereo crosstalk; not that this would be noticeable with actual music. I suppose if you had very long wires, there could be benefit with noise rejection, but with typical lengths <10' at home, the results suggest one should just stick to single-ended mode. In contrast, the PonoPlayer actually did benefit from using balanced output with lower noise level (test results here). So there you have it, here's an example where just because a device is capable of sending out a balanced signal doesn't mean the resolution would be any better.

Needless to say, no difference with the ultrasonic noise level with "1-bit" mode using balanced out.

VI. Subjective sound quality...

You might be wondering, "How does this DAC sound?" :-).

Well, it sounds as you would expect from the measurements! It's quiet, has good frequency extension when attached to my main sound system and does a great job with dynamic transients. The main thing to be cautious about is the relatively high headphone output impedance of around 15Ω. For best quality, it's good to pair this DAC with some high impedance headphones of 100+Ω.

I had a listen to a few tracks using the 1MORE Quad Driver IEMs (32Ω only). Not bad still, and certainly the headphone amp is powerful enough to play uncomfortably loud without audible distortion. However, I can hear a bit of mid-range accentuation likely due to the poor impedance match. For example, Neil Young's Harvest album had him sounding a bit more hypernasal than usual with this 1MORE-Roland combo :-).

Let's talk about the DSD "1-bit A/D Conversion" mode. With this device, one can easily do an instantaneous A/B comparison between PCM and "1-bit" conversion playback because as shown above, output volume remains the same. Just tick the setting box and the device will switch immediately with a <1 second pause. As I had noted back in 2013, DSD resampling does subtly change the quality of the sound, making vocals more "present". Using headphones, I perceive a sense of "smoothness" ("S1LKi" DSP after all!) and at times the impression that the soundstage is "larger", more "expansive". I found it rather hit-and-miss whether I like that effect. For jazz and acoustic music (like say Musica Nuda), it could add a bit to the ambiance but with synthetic electronic or modern pop (I was listening to some Aphex Twins Drukqs), it seems to take away from the "bite" that I enjoy with the transients and I think it's better turned off for this kind of music.

Regardless, I would definitely classify the effect as subtle. It's the kind of thing where with an immediate switch one can detect the difference, but once you let your mind enjoy the music and not focus on that qualitative aspect of it, within 10 seconds I won't notice the difference. How much of this effect is due to the difference in filtering or the excess noise from the DSP conversion is hard to say without controlled listening tests. The "1-bit D/A Conversion" mode is an interesting feature, but nothing to get too excited about and from an objective perspective, turning it off does result in better resolution/fidelity playback.

VII. Conclusions

While obviously not a "new DAC on the block", the Roland Mobile UA-M10 is certainly an interesting device with more features than standard USB DACs. The dual headphone/line outputs along with the flexibility to assign functions like 4-channel and balanced out make this product stand out.

The "1-bit D/A Conversion" algorithm does change the sound and I'll leave it to you whether you like this kind of thing. Personally, I found the difference to be minor.

I didn't bother testing playback of DSD64 material with this DAC simply because I think we are far into the twilight of DSD as a music format and there's little native DSD material worth owning anyway (remember, many SACDs are just standard-resolution PCM-sourced music).

Remember that because of the unique features controlled by the device driver settings, this should be seen as a DAC aimed at Windows and Mac OS users. The form factor is actually very nice and would have made for a nice external DAC slapped behind a small Raspberry Pi enclosure. Alas, it's too bad I could not get it to work consistently with Linux.

Thanks to my friend for letting me borrow this rather unique USB DAC for a little while :-).

--------------------

Pssst... Want to hear about some snake oil? Check out this conversation between Darko and Jeff Dorgay (TONEAudio) if you dare. I think the conversation itself IS snake oil - hinting at products and ideas like the various snake oil classes I discussed previously. You're a better man than I if you were able to listen for more than 15-minutes without quickly skipping through those 80+ minutes spoken by men who apparently have no means to test reality and are IMO desperately in the business of selling you stuff. Darko seems to have a soft spot for USB cables, and yes, sighted listening is prone to bias.

Seriously, these days when we can easily measure device resolutions beyond 16/44.1, detect picoseconds of jitter, quantify inaudible content like the ultrasonic noise of DSD conversion, devices that sound obviously different would also make them measurably different as well.

Unless the sound of a device is truly "obviously different" and can be picked out easily, all sighted listening claims regardless of whether we're talking about cables, or speakers, or DACs, or digital streamers must be considered to be at best low-level "evidence" when making claims. That discussion between Darko and Dorgay is a typical example of what it sounds like when people "shoot the breeze" with little facts, plenty of assumptions, drawing from questionable anecdotes. One comes out of that experience ultimately learning little of anything either about the "big picture" or specifically about any particular device - dispensable chatter with no enduring truths expressed, or IMO even ephemeral entertainment value.

In closing... Gimme your blind test results: "Is high Harmonic Distortion in music audible?" :-). Two weeks left until end date of submissions (end of April 2020)!

Stay healthy and enjoy the music!