MEASUREMENTS: SMSL iDEA USB DAC / Headphone Amp (ESS Sabre ES9018Q2C + XMOS XU208)

In 2015, ESS revealed its latest generation of DACs at the CES show. One of the interesting new products revealed was the SABRE9018Q2C, a tiny 5mm x 5mm SoC claiming very high audio quality out of an integrated DAC/headphone amp operating with low power along with the ability to decode a wide range of sample rates.

I was contacted by ZugZugOrc on Computer Audiophile to have a look at a recently released USB DAC he wanted to buy for some measurements based on the above chipset. So we made an arrangement where he sent the package to me to have a look at and subsequently after I'm done, I would send it off to him to use :-). Thanks Zug for the opportunity!

Here's the device - the SMSL iDEA (can be found on Amazon for ~US$86 currently), a mini USB powered "stick" based on the ES9018Q2C "audiophile-quality" mobile DAC capable of accepting up to 32/768 and DSD512!

SMSL stands for  "ShuangMuSanLin" Electronics Company based in Shenzhen, China. I was sent the black one, but the device is also available in red.

As you can see, the box has the "Hi-Res Audio" logo on it which means it has been certified through the Japan Audio Society. It comes with 3 USB adaptor cables to connect the device to standard USB-A, microUSB (OTG cable for a typical Android phone or tablet), and USB-C devices.

As for the DAC itself, it feels sturdy enough in a metal enclosure, and tiny measuring 6mm x 60 mm x 16mm. On one side is a 3.5mm headphone jack and the other side a female microUSB. On the top is a low profile volume switch which controls the volume independently of the OS volume setting, so when I first plug it in, it defaults to a lowish amplitude setting. You have to push the volume + button a few times to get it to 100% (I didn't check how many clicks - probably at least 10) and also make sure the computer / streamer is set at 100% to truly get the highest output. I assume the volume control is done digitally. I like this, I can send bitperfect PCM and DSD to this DAC, and just use the hardware buttons instead of fiddling with software controls.

This is a fairly low-power DAC that can be used with mobile devices like cell phones and tablets provided enough power is available. Apart from the low power DAC/amp chipset, the other half of the equation for energy efficiency is the USB microcontroller chip. This device uses the XMOS XU208 (datasheet here), a new "second generation" Xcore200 device (I believe the specific device here is the 8-core, 128KB, ~500MIPS version, 1000MIPS with dual issue). It is this device that allows communication at 32/768 and DSD512 with the host USB2.0.

Before I get to measurements, remember that previously, I had reviewed and measured a number of products using ESS DACs. The 2013 Oppo BDP-105 used the ES9018, in 2014 I had a look/listen to the AudioQuest Dragonfly 1.2 based on the ES9023, PonoPlayer in 2015 with the ES9018K2M,  Light Harmonic Geek Out V2 also in 2015 is based on the ES9018AQ2M, and most recently I had the preview of the Oppo Sonica DAC and its ES9038Pro.

Each one of these ESS Technology DACs share the "HyperStream" architecture. If you look around, although ESS Technology tends to hold full datasheets close to their chest, we see that these DACs typically incorporate an oversampling anti-imaging reconstruction filter with programmable parameters, jitter reduction (by an onboard asynchronous sample rate converter), and then fed into the noise-shaped sigma-delta converter. For those interested, check out Martin Mallison's Rocky Mountain 2011 presentation (and more in the company's 2007 SABRE technical paper):

Given this pedigree and what I have seen previously, I certainly have high hopes for the objective output based on the latest and greatest "mobile reference DAC".

As usual, the measurements will be based on my standard setup as follows:

Device (eg. laptop, SB Touch, phone, HTPC) --> SMSL iDEA --> generic 6' phono-to-RCA cable --> Focusrite Forte --> 6' USB cable --> Win 10 measurement computer

Part I: Digital Oscilloscope, Digital Filter, Impulse Response

Here's the 1kHz 0dBFS 16/44 square wave output on the digital oscilloscope with the hardware volume at 100%:

Nice channel balance. Peak 2.73V corresponds to ~2Vrms. No clipping confirmed with 0dBFS sine wave (not shown). The configuration of the Gibbs phenomenon at the edges of the square waves suggest a symmetrical linear phase digital filter.

There she is... Linear phase filter as predicted. It maintains the positive polarity of the original impulse signal.

And here's the "Digital Filter Composite" as used in most of my recent DAC measurements based on discussions with Juergen Reis.

Overall the graph shows a "sharp" roll-off filter with good suppression of imaging above Nyquist. There's about -80dB of suppression between the ~20kHz and the ~40kHz imaging cluster. If you look at the Oppo BDP-105 measurements, you'll see that comparatively, the suppression was stronger with the Oppo - not an audible difference obviously! Also not surprising is that intersample overloading happens with the 0dBFS signal as I have found with almost all DACs these days.

Using a 20-ohm resistor at the output with a 1kHz sine wave playing at a reasonable volume resulted in a measured/calculated output impedance of approximately 0.5-ohms.

I got an inexpensive USB tester to look at current draw from the USB device during use:

With the music paused, the device is drawing 110mA at 5V. As you can see, I have some 16-ohm impedance JVC HA-FXC51B IEM headphones (sound great for the price!) connected to it. At 100% volume playback (extremely loud) while playing Depeche Mode's "Stripped" (might as well use real music off the album Singles 81-98), the current increased to ~210mA peak. It's important to keep these numbers in mind because if you're going to attach this to a phone or tablet, the device will need to be able to supply this amount of current for it to work. As you'll see below, I was able to use this with my Samsung Galaxy Note 5 phone.

For comparison, the Light Harmonic Geek Out V2 (measured here) sucks up 450mA idle, up to 500mA playing back in 100mW (into 16-ohm) mode, and up to 700mA in the 1000mW mode. The AudioEngine D3 (measured here) uses 160mA idle, with 280mA peaks using the above 16-ohm IEM.

Part II: RightMark Tests

Okay, time to run a few comparisons then... Unless noted otherwise, the measurements were obtained with my Surface 3 computer running Windows 10 loaded with the SMSL ASIO driver as found here (2017-04-11 version used).

16/44.1:

Okay, check out the line-up. We have the iDEA, PonoPlayer, Oppo BDP-105, TEAC UD-501, and finally the venerable Squeezebox/Logitech Transporter for comparison.

Honestly guys, 16/44 is a piece of cake for any modern DAC capable of hi-res. I feel I have to measure this because it is important given the amount of 16/44 content out there... But finding an anomaly is typically rare and usually "by design" such as the PonoPlayer. The most overt anomaly in the numbers is of course the PonoPlayer and its frequency response resulting from the digital filter setting used by Ayre emphasizing a very short impulse response (but at the expense of ideal frequency domain performance).

As usual, here are a few graphs from the data:

As you can see, the PonoPlayer shows its usual high-frequency roll-off and the slightly higher IMD+N can be seen.

24/96:
Going into hi-res, this is where we separate the men from the boys...

Not bad at all! This little SMSL DAC is able to demonstrate a very low noise floor on par with much more expensive devices.

Here are the composite graphs:

Overall very nice showing from the iDEA! You can still see a bit of an edge for the much more expensive Oppo BDP-105, TEAC UD-501, and Transporter devices in the IMD+N sweep for example. But man, these are impressive results for such a tiny and inexpensive device!

24/192:
Quickly, the 24/192 tests... No Logitech Transporter for the obvious reason that the DAC maxes out at 24/96.

Again,the SMSL iDEA achieves very respectable results comparable to much more expensive devices.

Part III: DSD64 / DSD128

I don't measure DSD performance often but over the years it's good to have a peek to just make sure things look and sound right... Here are the results using the 24/192 PCM signal converted to DSD64 and DSD128 using Weiss Saracon.

Basically it looks good. There is minimal difference between the DSD and PCM measurements. Noise floor from 20-20kHz very comparable as per the numerical data indicating DSD playback about just as noise-free as PCM. Notice the slightly higher overall noise level with DSD64 which is a real finding and consistent with the fact that noise shaping is used aggressively with 1-bit quantization. We can see rising noise >20kHz on the graphs for both DSD64 and DSD128; cleaner with 24/192 PCM as expected.

Part IV: Jitter

One would expect jitter to be very well controlled:

Yes. Beautiful 16 and 24-bit Dunn J-Test results from this little DAC! This is what I expect to see with the ESS Sabre DACs. As you can appreciate, explicit claims about "femtoclocks" appear unnecessary for jitter control these days.

Part V:  Comparison between different source devices

Since this is such a tiny DAC which uses low power allowing its use with portable devices, I thought it would be fun to compare the output among the following...

Notice the power LED glowing purplish with the Surface Pro rather than the usual blue. This color tells you that it's working in "ultra" high sample rate mode - PCM 352kHz+. It glows red for DSD playback.

As you can see, I have the little DAC connected from top left going clockwise to: Microsoft Surface 3 Windows 10 using ASIO driver, Squeezebox Touch with EDO plugin, HTPC running Windows 10 Creators Update with native WASAPI USB Audio Class 2 driver, and finally my Samsung Note 5 phone running Android 7.0 "Nougat".

You can find the details of the HTPC (Skylake CPU) build here to check out the hardware in the big black box. Since that post however, I've added a nVidia GTX 1080 graphics card in there!

I had some trouble with the Logitech Squeezebox Touch. Initially the output from the DAC was a noisy buzz but with a few reboots and using EDO Kernel #12, it eventually worked. I noticed a similar issue with Android playing bitperfect through USB Audio Player Pro on the Note 5. I needed to make sure to start the app first, then plug in the device through the microUSB OTG cable in order for it to be recognized and initiated properly.

Despite issues with Linux/Android-based machines, I had no problems at all with the Windows ASIO driver, native WASAPI driver from Windows 10 Creators Update or Mac OS X over a USB-C MacBook.

Here are the measurements at 24/96:

As you can see, both numerically and in graphs, there are simply no differences!

And what about jitter?

Anyone see a difference of significance?!

Heck, to make it even more nasty, how about we run the nVidia GTX 1080 graphics board and Intel i5-6500 CPU full tilt at 100% with FurMark torture test and Prime 95 in the HTPC box?

You likely can't see the details in the image, but FurMark is running at ~200fps, GPU utilization at 100%, GPU temperature 78°C, internal fans going strongly, and I've plugged the SMSL iDEA DAC into a rear motherboard USB 3.0 port closest to the video card for exposure to any electrical/RF noise.

No difference! Not even increase in noise floor... And no anomalies noted at all in the IMD+N overlay graph.

How about jitter?

Nothing different. Surprised? You should be if you follow the typical "audiophile wisdom" expressed at the usual places by the usual people. Sure, I can show computer-generated noise affecting audio output in some situations (for example slightly more noise with this old internal PCI audio card with high load), but it doesn't appear that computer load affects the USB interface to this little DAC. Perhaps the little DAC's internal power regulation circuitry is very much capable of cleaning up whatever noise is there - all without needing to buy any filters or tweaks (which in the tweako-audiophile world likely would cost more than the SMSL DAC itself!).

Part VI: Conclusions

As you can see from some of the pictures above of the different systems I hooked the DAC up to, I had connected my AKG Q701 headphones to check for obvious changes in sound between the different setups. I also subjectively had a listen at times with my trusty Sony MDR-V6, Klipsh R6i IEM, and Sennheiser HD800 headphones and even downstairs with the main system using phono-to-RCA cables.

It sounds excellent as expected, consistent with the objective results. I had a listen to Martin Roscoe's Nielsen: Complete Piano Music, the 2017 remastered Paul McCartney's Flowers In The Dirt (DR12), and selected tracks from Ansermet's The Royal Ballet: Gala Performances (DR12, 2009 remaster). No complaints about the DAC sound. "Unveiled", low noise floor, neutral sounding. As I have heard many times with ESS Sabre DACs, the high fidelity "clarity" of the sound is often described as "clinical" in nature by some listeners. Others might even describe some kind of "digital glare"... The way I hear it, there are some recordings that are just overly bright which I attribute to the album production and mastering rather than the DAC. There are ways to counteract this... For example, a device like the PonoPlayer is based on the Sabre DAC chip as well but when playing 44kHz audio, sibilance and "glare" would not be a problem because of the purposeful high frequency roll-off.

With an output impedance of ~0.5-ohms and good output voltage of ~2Vrms into a high impedance load, volumes achieved with the headphones I tried were certainly good enough for my use ranging from "loud" to "very loud" (remember, don't go deaf!). However, the ESS 9018Q2C's integrated amplifier section is rated at 49mW into a 32-ohm load, THD 0.1%. This means that if you have low impedance and low sensitivity headphones, there might not be enough current to drive these demanding loads. Within reason, this should not be a problem since this DAC is meant for mobile applications. A device like the Light Harmonic Geek Out V2 (previously measured) with 1000mW (into 16-ohms though) might satisfy those who want more gain and of course get a dedicated headphone amp if you need even more than that (for example to power planar magnetic 'phones)!

By the way, for Apple MacBook users, I can confirm that the DAC works well connected with the USB-C adaptor. It shows up in the "Audio MIDI" panel without fuss and allows sample rates to 32/768kHz.

So, what have we learned?
1. Objective quality is excellent with this device. It measures very much like the Oppo BDP-105 in fact! That's actually quite amazing and I was certainly impressed when I first saw the results I was obtaining; I had already had a listen and suspected the results would be excellent.

2. Again, the digital source did not matter to the analogue output from this DAC whether it was a desktop type computer, Squeezebox Touch, laptop or cellphone. No evidence of noise issues even comparing the output from a cell phone with that from a desktop computer running a nVidia GTX 1080 graphics card at 100% load!

3. Excellent jitter suppression from the ESS DAC is again shown. Not that jitter has been a significantly audible issue for years with decent modern equipment...

4. If you're going to use this device with a portable source - cell phone, tablet, etc... double check compatibility given the 100-200mA current draw. I can confirm that it works with my Samsung Note 5 cell phone (see item 5 next). I was also able to get it to work with an old first-gen iPad using the 30-pin Camera Connect Kit! Unfortunately I didn't have the Lightning CCK to try with my wife's iPhone 6 or iPad Air 2.

5. The one main concern I have - issues with Linux-based devices including Android. I suspect there's something not quite right with the "plug-and-play" UAC2 drivers interacting with this device. What happens is that when I first plug the device into a Linux/Android machine (like the Pi "Touch", Note 5, Squeezebox Touch), the power goes on and the device goes into some kind of inappropriate starting state. When I try to play something, the output is an annoying buzzing noise with the audio being played at a very low volume. What I have to do is start the app (eg. USB Audio Player Pro on the Note 5) or configure the music program (eg. for the Pi "Touch" with piCorePlayer, tell Squeezelite to use "USB audio" and Output setting to "front:CARD=v12,DEV=0" in my case then save and reset), then unplug and reconnect the DAC. It works fine afterwards.

Of course the above issue is annoying and should not be like this. SMSL, please have a look into this! Might need to update the firmware for improved Linux compatibility? I have sent an E-mail to SMSL and will update if I hear back...

As I close off, I think this is another example of how as technology continues to mature, we're seeing higher and higher audio quality out of less and less expensive devices. Something like this SMSL iDEA - powered off USB even from a cell phone, in such a small form factor, easily measuring within striking distance of much more expensive DACs would not have been possible a few short years ago! And with the flexibility to play back at PCM 768kHz and DSD512 sample rates as well. All for less than US$100!

Speaking of the PCM 768kHz and DSD512 sample speeds, if you look at the ESS information about the ES9018Q2C, they actually only mention that the DAC is capable of 384kHz and DSD256 (11.2MHz) in the product brief. I wonder if the higher sample rate is just undocumented in the publicly available information or if the XMOS USB microcontroller might be doing downsampling before feeding the DAC. I'll have to take a look at this another time with my oscilloscope. I'd be interested if anyone has insight into this. In any event it's not like there's actual content at 768kHz or DSD512/22.4MHz to really benefit from these ultra-high sample rates (upsampling in JRiver 22 to these high sample rates sounds great)!

Ahhh, portable audiophilia... With Rebecca Pidgeon, The Raven (24/88 Bob Katz remaster) audiophile standard of course... :-)

The other day, I was listening to the set-up above while doing some paper work. Very impressive sound indeed through the 300-ohm Sennheiser HD800. As it is, these 102dB/V headphones sound loud enough with excellent bass extension. Given the measurements, it would literally take headphones like these in a quiet room to appreciate the remarkable accuracy of this DAC! There is obviously no way I would be able to appreciate the nuances this DAC can provide walking around the neighbourhood with headphones on.

Given the state of the hype these days, I suppose some might wonder about the importance of MQA and obviously the fact that this device doesn't support it. Clearly, I'm not the guy to talk to since I don't believe MQA makes much difference other than as a partial lossy compression scheme, provides questionable benefits with no evidence of actual value beyond a few testimonies.

But remember that the XMOS Xcore200 microcontroller inside does have 8 threads, 128KB of SRAM, what looks like 4Mbits (512KB) flash memory (Winbond 25X40) and something along the lines of 500 to 1000 MIPS of processing power. This appears to be much more powerful than what's in the AudioQuest Dragonfly Red/Black series (Microchip PIC32MX). I wonder how much processing speed is needed to implement even a single MQA unfold to 24/88 or 24/96 and if this level of processing speed from XMOS is enough...

As I said a couple weeks ago, digital audio truly is democratizing high-fidelity output to the point now where judged purely on ability to reproduce high resolution sound, there's no need to look much further than a device like this. Of course that doesn't mean I'll be trading out a nice desktop DAC because I still want balanced XLR output, there's value in having other digital inputs like S/PDIF, and a nice desktop device like the TEAC UD-501 or Oppo Sonica DAC just looks nice on the audio "rack" :-). I certainly do not discount the importance of non-utilitarian factors.

Thanks again ZugZugOrc for the opportunity to take this little DAC out for a spin! Originally the plan was that I test it out then send it on to him in the UK... Well, considering just how well this device worked - amazing fidelity in such a small portable package -  I decided to just buy it off him and use it myself! It would also make a great gift at some point for potential audiophile friends :-).

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

Looking over the news feed this week, Munich High End 2017 has passed with a few announcements but it doesn't seem like there was much actually new. The What Hi-Fi? reports seem quite thorough showing pictures of the "latest and greatest" generation of gear.

Of the things on display, the Mag-Lev turntable certainly is eye-catching (began as a Kickstarter project). I wonder what kind of wow and flutter measurements it achieves as it seems to wobble a bit based on the videos I've seen of it in operation. Other measurements like noise level and distortion would be very interesting given what must be quite a strong magnetic drive unit. I wonder if the low-voltage moving magnet cartridge would pick up distortions and would a moving coil cartridge be even more susceptible as they tend to have even lower voltage outputs. Also, good to know there's a UPS system for potential power failures... Stability in operation also would be important for accidental knocks on the floating platter I imagine! Will be interesting when it's released and we read some reviews on this device in daily use.

Hope you're all enjoying the music as usual... Also remember, stay sane and rational :-).

ADDENDUM (June 3, 2017):
As requested, here's the RightMark frequency response curves across various low impedance/resistance loads: