Saturday, 5 August 2017

MEASUREMENTS: SMSL A6 as DAC (Part II)

SMSL A6 Integrated Amp - playing through USB 24/352.8kHz signal. Notice the matte/satin finish of the aluminum case. Probably need to wipe it down once awhile if dusty.
As I mentioned last week with the initial review of the SMSL A6 integrated amp, this is a versatile device. It features an AK4452 DAC first released in 2015 which though not "top of the line", does perform very well with a potential 115dB SNR and is part of the "Velvet Sound Architecture" range of "Premium" DACs. For those keeping track, currently, the flagship DAC from AKM is the AK4497, capable of 5-channels and rated at 128dB SNR; they call this series the "Verita" which competes with the best of the ESS DACs like the ES9038Pro previously discussed in the Oppo Sonica DAC measurements and review. By the way, another claim to fame of the AK4497 is that it has a "volume bypass" mode which allows the DSD data to bypass the internal delta-sigma modulator; a more direct processing path... As usual with new features in otherwise mature technologies, whether this is audibly better is probably up for debate.

Today, what I wanted to test out with the SMSL A6 are the RCA and headphone outputs from this device. As usual, along the way we'll have a peek at jitter results, the antialiasing filter being used, and check out the various inputs - USB compared to the S/PDIF variants of TosLink and coaxial.

To begin, let me list the usual set-up for most of the measurements to get that out of the way.

USB 2.0:
Microsoft Surface 3 laptop --> 6' USB cable --> SMSL A6 --> 6' generic RCA --> Focusrite Forte ADC --> 6' USB --> Measurement i5 laptop

S/PDIF:
Microsoft Surface 3 laptop --> 6' USB cable --> CM6631A USB-to-S/PDIF converter --> generic 6' TosLink/Coax cable --> SMSL A6 --> 6' generic RCA --> Focusrite Forte ADC --> 6' USB --> Measurement i5 laptop
As you can see, that's the typical set-up I've used over time. Having kept variables stable, I will as usual compare the output quality with other DAC's I've looked at.

I can confirm that the A6 works well for PCM audio using Windows 10 Creators Update's native USB Audio Class 2 WASAPI driver. Installing the ASIO driver from the SMSL download site (currently 2017-07-01 release) opens up native DSD playback (SMSL: You might want to change the ASIO driver name to something more specific than "ASIO for Generic USB Device"). I can also confirm that it works well with my Logitech Touch using Triode's Enhanced Digital Output (EDO) app and piCorePlayer with a Raspberry Pi 3 so I suspect it works well with Linux in general.

Part I: Digital Oscilloscope, Digital Filters, Impulse Response, Headphone Out

As usual, we begin in the microscopic domain. Firing up the oscilloscope, here is the 1kHz 0dBFS 16/44 square wave through the rear DAC RCA output:

Nice overlay of the 2 channels indicating very good stereo balance. The DAC output is "clean" in that it appears to be a direct path out, fixed at 100% level and will not change with the volume control knob. The DAC output is therefore meant to be paired with either a preamp for volume control or using digital volume control on your computer.

Peak voltage excluding the ringing overshoot is 3.35V or around 2.4V RMS with a 1kHz sine wave. Based on the nature of the ringing, we can see that SMSL is using a minimum phase filter in the device, likely with a steep roll-off. Over the last number of years, most DACs have transitioned to this minimum phase setting as shown recently with the Audioquest Dragonfly Black and Oppo Sonica DAC. Remember though that use of minimum phase filtering is common; every iPhone and iPad I've measured already use minimum phase filters!

So here's the impulse response for this SMSL:

Yup, minimum phase, sharp filtering as suspected. And using the overlay graph as suggested by Juergen Reis over the years, this is what we see in the frequency domain:

This looks very good. The antialiasing filter is strong, removing the ultrasonic noise very well. Intermodulation distortion from the 19 & 20kHz sine signal is very low. There is still a little overloading of the 0dBFS noise signal which is essentially gone with -4dBFS attenuation... This is rather typical of almost all DACs these days including those from ESS and TI/Burr-Brown in the hi-fi world.

I then turned my attention to the headphone output section. As I mentioned in the previous post, the headphone output is not the most powerful out there. Based on the published specs by SMSL, into 32-ohms it's capable of 330mW, 240mW into 150-ohms, and 122mW into 300-ohms. Fed into my Focusrite Forte ADC line level input (something like 100kohms impedance), the 0dBFS square wave looks good but with the power limitations, it will start clipping the peaks beyond a volume setting of 30 (out of 40).

More realistically for typically lower impedance headphone loads these days, here's a snapshot of the 1kHz square wave at volume setting 25/40 without clipping across a 32-ohm load:

Again we can see the DAC's minimum phase filtering and associated post-ringing. Very nice channel balance. For some context, with my 1MORE Quad Driver headphones rated at 32-ohms and 99dB/mW sensitivity, normal comfortable listening level would be around 10/40 for modern compressed pop and 18/40 for classical music; plenty of headroom to raise the volume.

As usual, I measured the voltage drop across a 20-ohm resistor using a 1kHz sine wave to estimate the output impedance. I'm getting a reading of 2-ohms off the headphone amp. Obviously it would be nice to see <1-ohm but this is fine with headphones 16-ohms and above. I checked the frequency response across the 20-ohm and indeed it was flat with no evidence of DC-blocking capacitors in the pathway.

Part II: RightMark Tests


16/44.1:
As usual, we start with "standard resolution" CD-variety samplerate and bitdepth. Always good to make sure the objective output is good considering that the vast majority of what we're likely listening to is still of this bitrate.


As you can see, to the left is the DAC output using the USB interface followed by S/PDIF interface inputs (Coaxial and TosLink). There's not much to say really. From the perspective of dynamic range and noise level, it's all about the same. Nonetheless, we can see differences in distortion and crosstalk especially evident in the graphs:

Basically the AudioQuest Dragonfly Black is not in the same league as the others in terms of fidelity with higher distortion and crosstalk. Remember though, I did say the Dragonfly Black nonetheless sounds fine and IMO this is a reflection of the limitation of human hearing; at some level, differences are just not going to be obvious unless you literally compare side-to-side.

24/96:
Turning to high resolution then, let's have a look at how the same devices compare:


As we get into the world of increased bitdepth and higher samplerate, clearly we see differences between DACs and even the different digital inputs start to separate out.

For the SMSL A6, S/PDIF TosLink and Coaxial inputs have a slightly different frequency response possibly a reflection of the variation in time base between the external asynchronous CM6631A USB to S/PDIF interface used compared to the direct USB input to the A6 going through its internal CM6632A chipset. There's also slightly higher distortion with the S/PDIF inputs compared to the USB interface. Notice the higher crosstalk and 60Hz mains hum with the A6 compared to the Oppo Sonica DAC and TEAC UD-501 DAC. As usual, we must put these relative differences in context given the price differential plus I think it's fair to say that neither a 60Hz hum at -120dB nor -90dB crosstalk would pose any impediment to musical enjoyment!

Like above with the 16/44 test signal, notice mostly the difference between the Dragonfly Black and the others.

24/192:
Finally, we look at the highest resolution that my ADC can capture - 192kHz.


Instead of the Dragonfly Black which cannot manage 192kHz, I have here the Pi3 with the HiFiBerry DAC+ Pro board for comparison. Unfortunately, the RightMark software has problems with plotting out 192kHz samplerate crosstalk and IMD+N sweep graphs.

Differences between the S/PDIF frequency response compared to USB input appears more pronounced. Also, numerically at least, we can see the HiFiBerry DAC+ Pro HAT board isn't up to the same kind of objective quality as the other DACs when it comes to THD and IMD distortion levels. Nonetheless, as a HAT daughterboard sitting over a Raspberry Pi 3 powered by an inexpensive Enokay 5V 2.5A switching power supply, the DAC+ Pro does achieve excellent low noise level.

That TosLink and Coaxial digital inputs do not measure as well as the USB interface is no surprise. This is not the first time we've seen this. Recently, measurements of the Sonica DAC show the same tendency suggesting that until proven otherwise, with modern DACs where high quality USB is emphasized, it is likely that the USB interface will perform objectively superior to the much older S/PDIF digital inputs.

I'm impressed that the A6 is capable of handling 24/192 over TosLink actually. Quite often, TosLink input is capped at 24/96 due to the fibre optic specs used having lower bandwidth. The fact that this is the case suggests a good quality optical-to-electrical transceiver. As a general rule of thumb, even though I did not hear any digital errors during testing, it's probably best to stay with 24/96 transmission over TosLink.

Remember that the A6 can handle 384kHz and all the way to 768kHz! Honestly, I doubt we'll ever have commercial material released in 768kHz (nor is there any need), but here's a quick peek at 384kHz using the 24/192 test signal upsampled to 384kHz with iZotope RX 5:


Using my ADC examining the analogue output signal, it does look like there is marginally higher noise when the DAC is running at the faster (384kHz) sampling rate. It's small in any event and not something one needs to be concerned with.

Part III: DSD

So, what about DSD performance you ask?


These results for DSD64 and DSD128 were obtained using the RightMark 24/192 test signal upsampled using Weiss Saracon. I think this is a fair assessment because it's very rare to find pure DSD material which hasn't at some point gone through a PCM phase for mixing or mastering.

As you can see, the SMSL A6 did a fine job with DSD. Not unexpectedly, there's quite a bit of noise above 25kHz with DSD64 and even with DSD128 (cyan), the noise floor rises earlier than the modulator noise from the ADC (that white increase in noise above 50kHz with PCM 24/192 input).

Part IV: Jitter

As I've expressed countless times in the past, I don't think jitter is a problem these days with any reputable DAC device; certainly not a device like this one with a modern chipset for both the asynchronous USB communications and converter!

Nonetheless, we can see that it's not absolutely perfect. We can see a pair of elevated sidebands close to the primary signal on the 16 and 24-bit tests. Also, a few spurious noise spikes around 7.4kHz and 12.8kHz. Not audible in any case.

How about with the other inputs?

Not bad at all really. With the 16-bit J-Test, we can see that jitter is least with USB, followed by TosLink, followed by coaxial. Interestingly, the 24-bit S/PDIF J-Test results look cleaner than the USB! All of this is of academic interest only and speaks to the idiosyncratic nature of different devices found only with detailed analysis using instrumentation.

Part V: Headphone output

As I mentioned above, headphone output impedance was around 2-ohms using a 1kHz sine signal. Also, I noticed waveform clipping above volume level 30/40 into a high impedance load like my Focusrite Forte ADC. For completeness, I listened with a few headphones including the AKG Q701, Sony MDR-V6, Sennheiser HD800 and 1MORE Quad Driver IEM. Thankfully, even with the AKG's which I've always found to be most difficult to drive I didn't need to push the volume beyond 25 before it became too loud for my listening preference. BTW, I can't believe AKG's claim of 105dB/mW for the Q701, it's more like 93dB/mW as listed in the Headphone Voltage Cheat Sheet.


Therefore, I figure it would be interesting to see how well the headphone output measures at 25/40 volume level. This is a bit louder than my usual with any of these headphones but below the clipping point across a 32-ohm load to check the resolution.


As you can see from the summary table looking at 16/44, 24/96, and 24/192 data comparing the headphone output volume at 25/40 across a 32-ohm load versus the direct RCA DAC output, there is some sacrifice in noise level and intermodulation distortion through the headphone amp. But it is still capable of higher than 16-bit resolution with a low noise floor! Not bad.

Composite graphs with 24/96 test signal:

The frequency response graph shows a slight bass roll-off of around -1.5dB at 20Hz through the headphone jack. There's also very slight accentuation of treble across the 32-ohm load. Otherwise, the other differences aren't particularly suggestive of audible change.

Remember that just like speakers, headphones are quite variable in their impedance, frequency response, phase characteristics, distortions, etc... across the audible frequencies resulting in much more sonic variability than just these measurements of the output from the headphone jack at a specific resistance load. Headphone-amplifier matching is obviously important.

Part VI: Conclusions

As I discussed last week, subjective listening tells me that the SMSL A6 is a nice sounding integrated amp for a very good price. This post now gives us more detail about the objective quality of the internal digital audio conversion.

SMSL has chosen to use the steep minimum phase filter setting of the AK4452 DAC chip. It does a great job of removing ultrasonic distortions. The channel balance is excellent. RightMark testing shows that the DAC portion can handle high-resolution audio input with a measured resolution of over 18-bits. I would stick with USB input as interface of choice with slightly better noise floor and lower general levels of distortion compared to the S/PDIF TosLink or coaxial inputs. I doubt it makes much difference in any event. Although my measurement ADC doesn't go beyond 192kHz, upsampling the test signal to 384kHz does suggest a slightly higher noise floor with such a high samplerate (just by 1dB, so it's no big deal). In the same way, accuracy of DSD64 and DSD128 output looks very good. I don't know anyone who "needs" 384/768kHz PCM or DSD256/512, but they're there in the feature set...

S/PDIF TosLink and Coaxial inputs do demonstrate more jitter at 16/44 with this device. As usual these days, I do not believe the slight jitter would be audible even though measurable.

The headphone output sounds good but keep in mind the 2-ohms or so output impedance which should still be fine even for low impedance 16-ohm headphones (are there any good headphones <16-ohm impedance?). Also, because I noticed some clipping behaviour over 30/40 headphone volume, I recommend using higher-sensitivity headphones (say 96dB/mW and up) to give you some headroom to pump up the volume.

Objectively, this device measures well in terms of the quality of DAC performance by interrogating the RCA and headphone outputs. The results are consistent with the subjective opinion formed last week and in fact, this device holds up very well against much more expensive DACs like the Oppo and TEAC put up for comparison (remember, this isn't even a fair comparison because the A6 is actually less expensive and is an integrated amp). Apart from its obvious role in a bookshelf type sound system, as I mentioned last time, given the size, I think this box would be great on a computer workstation fed by USB playing to passive speakers.

Having said all this, there is finally of course the matter of assessing the quality of the ICEpower/B&O 50ASX2 Class D amp. We'll look at this another time as I put together a suite of amplifier tests and gather more data to compare.

Until then, my dad wants his DAC/amp back... Surely this is a sign of a recommended product that has successfully captured the intent of the purchase :-).

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

With summer now in full swing it's time to take some time off again. I hope everyone is enjoying the music and have some time to unwind...

This is also the time of the year when I spend time on my other hobbies like photography. For those who like taking pictures, especially if you print a lot and hate the ridiculous cost of ink jet cartridges, have a look at the Epson Expression ET-2550. The 4-ink "ecotank" probably will last me at least the next 2 years for my 4"x6" and 8"x10" non-critical photo printing. The image quality is good, and although it's more expensive to start (~US$300 for the printer, Epson wants their ink income front-loaded), this will certainly save me money given how well it has performed over the last three months already! (The printer's wireless network connection convenience also now a "must have" for me.)

Remember to give the MQA vs. Standard Hi-Res Audio Blind Test a try and get me your results if you have a hi-res system capable of playing back 24-bit 176.4/192kHz files! I had a quick peek at the results so far to just make sure there were no burning issues and let's just say it's looking interesting :-). Survey closes September 8th.

7 comments:

  1. I am all for simplicity, and this is, of course, an interesting unit. For most applications, however, I would find the output power too limited. About the same money buys an equally versatile Yamaha receiver, but with a lot more power.
    As an exercise in ultimate simplicity for bedroom use, I recently bought a tiny (pocket book sized) British Ava Maestro 50 digital amplifier. It has just an optical or coaxial input, and some 2x25 watt output power. Its selling point for me was an auto on/off function. So we keep it in the wardrobe, with a Chromecast Audio as its only source. We don't have to get out of bed to turn it on ar off, and it is completely out of view for less visual clutter (why can't have all amplifiers an auto on/off switch?). The limitation is the output power. I use a pair of Wharfedale Diamond 9.0 speakers of 86dB sensitivity, and amplifier power is only just enough in our small bedroom. I tried it with my less sensitive Harbeth P3ESR desktop speakers, and the amplifier clearly lacks the power to drive those.

    ReplyDelete
  2. You correctly noted that this D/A converter uses a minimum phase filter. This type of filter has become popular because it seems to provide a "more natural" time-domain response. However, most people fail to realize that the phase errors introduced by a minimum phase filter are cumulative. If there is more than one minimum phase filter in the signal chain (from analog microphone to loudspeaker), the phase errors will accumulate. For this reason, no more than one minimum phase filter should be used in the signal chain (if even one). If the minimum phase filter "improves" the time domain response, it does so on the first pass. All other filters in the signal chain should be linear phase. Passing the output of the minimum phase filter through a linear phase filter will not create a prering (try it if you don't believe it). The linear phase filter will not change the time domain response of a band limited input signal. In contrast, each pass through a minimum phase filter will add additional time-domain errors. --- Food for thought

    ReplyDelete
  3. Hello, Archimago. Pls tell me about kind of samples for impulse response test. Usually i play waves which are generated from izotope or wavetones.com. I didn't have osc, so i just make record in hi-res domain. After that I look at the files in the audio editor.
    Nevertheless my results (single impulse and square wave) looking "as generated" - too clear, without any ringing. What i doing wrong? Please describe your procedure in more detail

    thank you!

    ReplyDelete
    Replies
    1. Hi Adova,
      Impulse response can be captured by playing an impulse WAV (16/44, you can just create this in an audio editor as a file with occasional 100% single samples separated by a second worth of silence) through your DAC and using the ADC to capture it at a higher samplerate (24/192). You'll see the ringing pattern show up. I typically use Adobe Audition to create a smooth interpolated curve out of the capture.

      Delete
    2. Yes, i do this way already. Usually i use wavetones.com as impulse generator. If i set width 1 sample only, it looks like "classic linear phase impulse" (with pre- and post-ringing). Is it right thing? I guess test impulse must be clean. Of cource, I can get the clean impulse from Wavetones, but it needed to increase sample width more than 1. I will be very grateful if you just send me your impulse file to my mail adovapravda@gmail.com.

      Delete
  4. Hi Archimago,
    Did you measure the actual headphone power output? I found a strange thing: from its specs which is published on the official website, SMSL states the headphone output is 330mW at 32-ohms, however, it raised to 545mW at 64-ohms. Normally, output power will decrease as the impedance goes up, how does this come out or is it just a typo? Thanks

    ReplyDelete
  5. Why are all the tests done at 44khz sampling rather than a higher sample rate? I am a complete noob and always wondered why.

    ReplyDelete