Saturday 13 August 2022

REVIEW: Sabaj A20d 2022 Version DAC (ESS ES9038PRO) [Part I] - Hardware, Filters, and Jitter.

[Disclosure: Most of the products I review have been purchased personally or loaned from friends. On occasion, a manufacturer or seller will contact me about a device which I might have an interest in. This Sabaj DAC is one such device which was sent to me from the company for an honest, independent review. As I have done over the years, my review process focuses mostly on objective performance with some subjective listening opinions. Objective data rather than pure subjectivity I believe will help substantially mitigate bias.]

Over the last few years, we have seen an impressive proliferation of high resolution DACs at very reasonable price points coming from audio companies in China. In the world of consumer electronics, this should surprise nobody given that much of our technological products hail from that part of the world. Everything from Apple phones, to computer motherboards, to high-tech multinationals like AMD and Intel having manufacturing and assembly lines there.

On the test bench, we have the Sabaj A20d 2022 Version (currently ~US$420). Make sure to take note of the "2022 Version" because there is also the original A20d "2021 version" which clearly looks very different, does not feature headphone outputs, but is also based on the flagship ES9038PRO chip.

At the heart of this DAC, the ESS ES9038PRO is aided by the ES9311 "ultra-low noise" linear regulator to provide reference voltage to the DAC. Headphone amplifier is based on the TI TPA6120A2.

As I have been doing in the last year (like with the Topping D90SE, and S.M.S.L. DO100 previously) I think it's good taking my time with multi-part reviews to examine the various facets of these modern devices. Unlike back in the "old days" of digital where CD players only handled 16/44.1 content, modern DACs provide a whole host of samplerates, PCM + DSD, filtering options, potential noise and jitter variation, headphone output and input options typically USB and S/PDIF variants; even with these tests, as an audiophile who prioritizes sound quality, I haven't even seriously examined the Bluetooth audio options available in these devices over the years.

I. A look at the hardware

In the image above, we see the components included in the box - the DAC unit, brief manual/pamphlet, USB-A to USB-C cable, Bluetooth antenna, power cable, and remote control (dual AAA not included). Everything came well packaged.

The power cable is a standard IEC plug which means the SMPS is integrated inside, making for a convenient package without external wallwart or power brick. The specs page indicates that power supply filtering is built in (so there should be no need to supplement with audiophile tweaks, not even power cables, right? ;-).

If you look at the front panel in the topmost picture, note both 4.4mm and 1/4" headphone outputs to the right. The 4.4mm headphone output is not balanced, rather, it's provided as a convenience for the user. Specs for the headphone stage look quite good - almost 0Ω impedance, with output power up to 1W into 32Ω and 2W into 16Ω, and a claimed THD+N down to -120dB (output level and load unspecified).

As you can see, we have a good sized high resolution 2" color display screen (LCD), large and bright enough to be visible from my usual seating position about 9-feet away in my sound room to easily see the volume level. I like the thoughtful, colorful arrangement of information available at a glance including volume level when in preamp mode with variable control (0 up to 99), digital input selected, whether headphone output is turned on (little headphone icon), whether PCM or DSD, and samplerate (or as in the picture above, Bluetooth LDAC variable bit-rate format). I hope other companies have a good look at this logical and informative layout.

Within the menu controls, you can change the brightness (5 levels). Also, you can set the "Dimmer" time which is how many seconds the display is shown (choice of always turned on, or auto-off up to 60 seconds). I like this auto-off feature but I wish it automatically turned on each time something changes like say the samplerate or switching between DSD/PCM. Also, I would have liked an intermediate setting where the screen stays on all the time but actually dimming to lowest brightness instead of turning off completely. Notice that there is no dedicated power LED so with the screen off, the DAC could still be on or it could actually be in off/sleep mode and we would not know the difference. Not a big issue IMO.

Other menu settings include the choice of 7 PCM filters, 4 DSD low-pass options (47/50/60/70kHz), the DPLL setting to fine-tuning jitter tolerance (level 5 default), and there's a switch for "SPDIF Mode" for "Normal" or "Processor" - the latter setting routes the S/PDIF interface data through the XMOS processor which allows MQA processing. Of note, there's also a headphone gain setting of LOW or HIGH (+8dB).

There's a large multifunction push-to-click knob that allows you to adjust volume (preamp mode) and change settings on the menu. I'm not going to go into this in detail - suffice it to say the menu looks nice on the screen and easy to navigate; a couple of pictures below showing the "PCM filter" setting for example:


You can decide if you like the look of the unit with its double front rackmount-like-chassis-handle esthetic. I certainly appreciate that Sabaj got away from the basic black box style with this machine. The silver colored, angular metal box which measures approximately 8" wide x 8" deep x 2.25" tall including the handles (~1.5kg/3.3lbs) looks nice and will add a certain "vibe" to your system. The large black articulated push-to-click knob visually adds contrast to the front and works well for settings and selections. Push it in and hold for a couple seconds to put the machine into off/sleep mode. In day-to-day use, I'd probably use the remote more in my sound room.

Speaking of the remote, here's a picture of it:



Notice the instructions on the battery compartment. Hold down button "C" when first used to select USB DAC.

The remote is plastic in construction and quite typical of DACs at this price point. There's an on/off button, mute, toggle input, and volume control. The Fn button can be programmed in the menu system to switch to Bluetooth input or toggle between line out / headphone out. Maybe Sabaj could consider mapping other functions like cycling screen brightness.

And here is a closer look at the back end of the device:

(Nice black ribbed side panels to add a bit more visual contrast.)

Digital inputs include USB-C (note that it cannot do direct USB-C to USB-C, it's fine with USB-C to USB-A and probably OK with OTG adaptor) and both Coaxial and TosLink S/PDIF. USB can go up to 32/768 and native DSD512. This inability to connect USB-C to USB-C is not unique; I noticed this with the S.M.S.L. DO100 DAC as well. S/PDIF inputs can go up to the typical 24/192 and DoP DSD64.

The Bluetooth 5.0 input is capable of standard SBC, AAC (for Apple products in particular), aptX, aptX-HD, and LDAC (best quality, highest bitrate on Android phones). I connected to this with my smartphone and found the signal strength excellent within much of my house (one of these days I'll have to measure quality of these Bluetooth CODECs beyond SBC and what signal degradation with distance looks like!).

It features both single-ended RCA and higher-resolution balanced XLR outputs.

The USB interface chip is the more powerful 16-core XMOS xCore XU216 which is needed for full MQA "decoding" as well as "rendering" (as opposed to the common 8-core XMOS).

Using the Kill-A-Watt meter, I see that when turned on and playing through RCA/XLR, it's only sipping 5W through Bluetooth or USB inputs. When in sleep mode (turned off with remote controller or hold control knob down a few seconds), it drops down to <1W. Always good to see low-power utilization. The box barely gets warm after an hour of use.

II. Oscilloscope & Digital Filters

Let's start with the oscilloscope and a look at sine and square waveforms. The DAC is set to fixed 100% output level here (preamp mode off):


The Sabaj A20d 2022 Version demonstrates a peak output level through the RCA of 2.35Vrms unbalanced. XLR output was measured at 4.77Vrms balanced.

Notice the excellent channel balance of the right and left outputs of the sine wave. I turned the volume down to 100mV and registered <0.1dB difference in channel balance. The square waves look excellent as well - very symmetrically clean. The "Unlimited" square wave shows that the device is set to a linear filter when capturing these results ("Fast Linear" setting) with symmetrical pre and post-ringing. The "Bandlimited" tracing is simply what happens when appropriate low-pass filtering has been applied - as it should - to 44.1kHz samplerate material.

[Let the square wave graph be a reminder to us that when people talk about "pre-ringing" or "post-ringing" with filters, these are not of concern when our music has been properly filtered, as you would find with high quality digital audio! This is why years ago, I suggested that we not worry about the "pre-ringing" of linear filters even if unfiltered material such as impulse responses show it. Stick with linear phase setting when simply playing music from a DAC if you want highest accuracy with zero phase shift.]

Moving along to the 7 digital filter settings, here they are with a close look at the frequency response overlaid around the 22.05kHz Nyquist frequency (44.1kHz sampling) to see the filter roll-off:


We can see that the "fast" minimum and linear filters achieve -10dB at 22.05kHz, deeply attenuating by 24kHz. IMO, this is fine. Compare this to "ideal" sharp filters such as the Chord DACs which uses much higher order (tens of thousands of taps) processing to basically "brickwall" around 22kHz.

Then there are the "slow" filters that allow some imaging out to 28kHz or so, and various early roll-off, "apodizing" filters that cut off earlier just over 20kHz with deep attenuation by 22kHz.

And here are the Digital Filter Composite (DFC), Reis-like graphs, of the 7 filters - let me split them into the 4 standard fast/slow linear/minimum settings, then the last 3 settings that introduce an earlier roll-off and/or intermediate phase:



This looks like a typical assortment of ESS DAC filters. None of the filters afford up to +3dB overhead for intersample overload as demonstrated by distortion of the 0dBFS white noise signal. In music playback, this should not be an issue unless you play lots of compressed, loud recordings. Use volume normalization like ReplayGain or implement a little bit of overhead (like "headroom management" in Roon) and it'll be perfect. Otherwise, the filters perform very well and we can see evidence of low noise floor, wide dynamic range, good suppression of "aliasing", and low intermodulation in the 19+20kHz signal.

Absolute phase is maintained as you can see in the impulse response tracings.

As discussed in the past, filter settings are quite subtle unless they introduce multi-dB changes at the high end (for example, the -3dB slow roll-off with NOS into 20kHz should be audible especially with younger ears).

I am happy to stay with a typical "Linear Fast" setting which is what I'll use for the measurements ahead and in my listening impressions.

III. Quick peek at MQA

This DAC supports MQA so it might be of benefit to TIDAL HiFi+ users. Let's just have a look at a few of the filter impulse responses when "rendering" audio to 192kHz:


Compare these first few impulse responses to the Topping D90SE last year or the "full monty" from the Dragonfly Black. Yup, it's consistent with the "standard" MQA filter collection.

Here's your "blue LED" to confirm MQA "authenticated" sound as displayed on the screen of the Sabaj A20d 2022:


Notice that I'm using my old Squeezebox Touch TosLink optical output to feed the Sabaj ("OPT" input, S/PDIF setting in "Processor" mode). The digital audio is just 24/44.1 but through the "magic" of lossy encoding/decoding, we have 352.8kHz displayed on the DAC. The blue dot to indicate "Studio" (pseudo-)authenticated sound.

For completeness, here's the magenta MQA color when fed MQA data of uncertain provenance like the test signals I'm using to extract the impulse responses above:


IV. Jitter

Using that set-up above with the Logitech Squeezebox Touch as digital source for USB and S/PDIF (using Triode's EDO plugin for enhanced digital out), let's have a good look at jitter performance, starting with USB:

While audibly inconsequential, we see miniscule noise spurs in the 16-bit test below the jitter modulation signal. The 24-bit J-test results are literally "textbook", cannot get better than this even with the signal accelerated to 96kHz samplerate with a 24kHz primary signal outside of human hearing.

Let's do the same with the S/PDIF inputs - Coaxial and TosLink optical:

Simply beautiful. Even when pushed to 96kHz - something not typically done - we see the absence of jitter anomalies (this is done with a very detailed 1M-point FFT, ADC running at 192kHz).

To push the margins of jitter even further, as if 96kHz is not enough, let's take this to an extreme 192kHz with a primary signal at 48kHz (temporal accuracy becomes more important with higher frequencies, hence jitter effects are more evident):

Very nice! The sideband anomalies due to jitter are at most around -140dB from the peak of the primary signal. As extreme as this test already is, the result here is even better than the S.M.S.L. DO100 DAC (-130dB below primary signal).

I played with the DPLL menu setting a bit; supposedly lower values could improve jitter performance, and larger number improves compatibility. I did not see any significant difference with the Squeezebox Touch so just kept the setting at the default of 5 for the remainder of my tests.

Looking at the 192kHz test above, this is an example of how asynchronous USB tends to be better than S/PDIF for jitter performance (even if absolutely audibly inconsequential!). Also, for the S/PDIF interfaces, coaxial tends to be a little better than TosLink in most devices I've tested even though TosLink provides galvanic isolation from electrical noise.

There's nothing to worry about with jitter these days using modern DACs like these! As I noted years ago, we can listen to music with jitter added and recognize that it takes quite a bit of anomaly in order to be audible; far higher levels than any reasonably decent DAC I've come across over the years. Bottom line is to not worry about the "jitter boogeyman" unless demonstrated otherwise. In my measurements here, I just used generic USB, coaxial and TosLink cables and a decade-old Squeezebox Touch as source. With a good DAC like this, the source is not that important - a good DAC functions at the level of "Bits Are Bits".

There is however one idiosyncrasy I found with this DAC and S/PDIF jitter. In the menu settings you can change "SPDIF mode" from "Normal" to "Processor". My measurements above were all on "Normal" setting. Here's what happens when switched to "Processor":

Hmmm. As far as I can tell, the only purpose "Processor" mode provides is MQA decoding. It looks like whatever it's doing has added quite a bit of jitter-like sidebands. Some kind of unintended consequence as the processor tries to sniff out an MQA signal? Maybe it's applying MQA decoding to the lower bits even though this is not encoded?

Needless to say, I would recommend leaving the S/PDIF Mode to "Normal" if you have no need for MQA. (DSD64 playback through S/PDIF via DoP does not require "Processor" mode.)

V. Summary of Part I

As we end off this first part of the review, here are some general comments about the Sabaj A20d 2022 DAC so far:

1. This DAC certainly has a unique look. The silver color and angled design could add a nice accent to your sound system. Obviously, appearance of something is bound by subjective judgment. I think it looks great and if you have a bunch of black audio boxes, this silver unit will stand out.

2. Nice color LCD screen, a good size to see volume setting, and other useful data shown. There is a "Dimming" setting which I typically will leave off (ie. screen on all the time). Currently what this setting does is to turn off the screen after a set number of seconds of inactivity. Maybe it would be good to set a brightness value for dimming (eg. lowest brightness intensity) instead of just turn off the screen completely since there's no power LED to let the user know it's still on.

An "auto-sleep" function might also be nice. For example, if there's no audio being played for 60 minutes, it would be nice for the machine to turn off by itself to save power.

I'm focusing on this stuff because as you can imagine, DAC performance is becoming so good these days I think companies will need to focus more on "creature comforts", displays, power saving, other feature and avoid usability nuances! (I'm happy to report that there's no clicking/popping sound with first audio playback using a Linux streamer - noticed with the S.M.S.L. DO100, which hopefully the company can address.)

3. We see a selection of 7 ESS ES9038PRO PCM digital filters. Different manufacturers will use different names for them, and Sabaj has elected to use easy-to-understand terminology here thankfully. As an example, look at the unnecessarily complex acronyms Mytek uses for their Manhattan II filter settings.

Based on my listening (and our blind test results), filter phase effects are subtle assuming no significant frequency response changes and of the options available, I prefer a standard linear setting not just in testing but also in personal listening. Be mindful of potential intersample overload with loud, compressed music hitting beyond 0dBFS true peak. Use of volume normalization like ReplayGain or "Headroom management" (-3dB) in Roon will certainly sort things out.

4. MQA is available... The 16-core XMOS will fully decode an MQA signal. We can see the standard impulse responses used in "rendering" to a higher resolution like 192kHz.

The S/PDIF "Processor" mode which allows MQA decoding resulted in some anomalies in the J-test suggesting either processing being done affecting the lowest bits of the signal or actual worsening of jitter itself. Engineers at Sabaj might want to double check.

Keep the "SPDIF mode" setting as "Normal" if you have no need for MQA through S/PDIF.

As usual, you know how I feel about MQA. ;-)

5. Jitter performance is excellent (apart from the S/PDIF "Processor" mode comment in [4]).

The jitter performance of this DAC is exemplary. The DPLL setting might be helpful for compatibility if you connect an extremely jittery device to the DAC. Relaxing DPLL to >5 might help in those cases.

Size comparison with the RME ADI-2 Pro FS R Black Edition, and S.M.S.L. DO100. The Sabaj is about 3" deeper than the RME.

So far, this DAC looks great and the performance is certainly very promising! And yes, it does sound excellent as a high-fidelity DAC. But let me not get ahead of myself quite yet - all in good time. ;-)

Next week, we delve into some detailed analysis of the measurements of resolution and distortion from the line-level RCA and XLR outputs in Part II.

8 comments:

  1. Hi Archi

    Impressive. Looking forward to the next parts of your review.

    It is a pity, that they ignored the intersample overhead.

    Juergen

    ReplyDelete
    Replies
    1. Thanks Juergen,
      Yeah, going forward, I think it's good for these increasingly high dynamic range DACs to give a dB here and there to improve the overhead. With objective awareness, I think at some point even objective audiophiles will be very much satisfied with the numbers we're seeing and move on ;-).

      For me, benchmark numbers these days of around -120dB THD+N and dynamic range around say 125dB are so impressive that I don't think manufacturers or consumers need desire more. I think it would be more impressive now to achieve "THD+N -120dB with excellent linear sharp filter and +3dB intersample overhead" than say "THD+N -125dB" only.

      As you know, few bother measuring to check that overhead. I think it was back in ~2014 when Benchmark released the DAC2 that they spoke of +3.5dB overhead peaks. Now that we're many years beyond, achieving distortion of such low magnitude, I think it's time to widen the net and not worry too much about just THD+N/SINAD.

      All the best to you and yours!

      Delete
  2. Hi "Mago"
    Maurizio from Verona- Italy
    As Always a perfect analysis, with a lot of measurements.
    Congrats 👏🏻 MQA yet.!? Oh my God...🙏🏻
    But.I'ld Foward at your attention an Evaluation that isn't like Darko's Marketing Videos, but Is produced by Goldensound that with You and Amir are my 3 Audio Gurus..! https://youtu.be/Fa1y9JRip68
    Title Is Very exciting.. "Why you can't trust Audio Measurements"..
    Curious of your enlightened judgment 💡
    Buona Vita, Mago.🤔

    ReplyDelete
    Replies
    1. Greetings Maurizio,
      Nice hearing from you! Hope you're having a nice summer in Verona!

      Yeah, I saw that GoldenSound video. Made a comment on it also a few months back:
      http://archimago.blogspot.com/2022/05/demo-software-de-clipping-of.html

      I think it's important for me to remind audiophiles that I believe human preferences do not need to follow measurement parameters so whether one "trusts" measurements or not for purchasing decisions is one's choice. For me, philosophically, the audio hobby is about achieving "high fidelity playback". So this means I want to seek out objective measurements that can tell me if I'm listening to flat frequency response, low noise, low distortion and low temporal anomalies (like jitter, phase anomalies, etc.). I appreciate that this philosophical leaning is a personal choice just as much as someone else who says "I believe in my subjective preferences, know what I like, and I'll follow how I feel". This is why I think it was important to write about my philosophical leaning many years back when faced with an article in Inner Fidelity (after Tyll left):
      http://archimago.blogspot.com/2018/07/musings-zen-and-art-of-high-fidelity.html

      It's a feedback loop of sorts because then I seek out hi-fi gear, and in turn develop a sense of what that sounds like, I tune my ears to enjoy this rather than the boosted bass/treble of a V curve that I might have followed in my younger years!

      It's not hard to show what GoldenSound spoke of in the video (looks like he's just using the title as click-bait). Yes, we can all make errors in the measurements and often changing a parameter here and there like how many points FFT used, whether averaging is turned on, or line smoothing will impact the numbers. This is why we talk about the standards we set when doing measurements. Audio Precision is the industry gold standard; but since I'm doing this as a hobbyist and find it more fun to explore for myself, I'll provide my own disclosures/standards as well like this:
      http://archimago.blogspot.com/2019/11/the-measurement-of-amplifiers-rig-moar.html
      or what I'm doing with DSD recently:
      http://archimago.blogspot.com/2022/06/notes-on-dac-dsd-1-bit-pdm-measurements.html

      Those of you who don't want to spend US$30,000 on Audio Precision gear can follow along and try things out yourself! There's great freedom in not needing to be locked into specific commercial hardware or software. The evolution of what can be done on the measurement side like John Mulcahy's REW, or the work from Paul Kane, or the hardware from E1DA is progressing beautifully! ;-)

      Yeah, MQA is still floating around. I don't mind capturing a few of those impulse responses depending on how much detail I want to achieve with a product review just to see if there has been evolution in some of what MQA is doing under the hood. And yeah, sometimes we see that the MQA feature might actually have unintended effects like that "Processor" mode which users should be aware of.

      Regarding Darko, I quite enjoyed his recent video on room acoustics and the RT60:
      https://youtu.be/dp_OdILUEkA

      I think he did a good job on that! Maybe he can spend more time on DSP room correction in time as well. (Discussions about the room here: http://archimago.blogspot.com/2020/08/summer-musings-soundroom-speaker-layout.html)

      There's a lot of potential education we can all present for audiophiles after basically a generation of miseducation. Whether "more objective" or "more subjective" leaning in our philosophies, so long as we can present in ways that help hobbyists clearly differentiate facts from opinions towards achieving "better sound" (and define what that means!). Intelligent, thoughtful, tempered discussions would be characteristics of the important cultural shifts forward in the life of the hobby, I think!

      Delete
    2. BTW, APx555b will never show you such a low noise floor FFT in the 'Jitter' test, and actually for two reasons:
      a) it is too noisy audio analyzer
      b) its jitter is higher than DUTs
      You can check Amir's tests, no one has side-bands <-160db, even -140db there stands like "excellent" or "textbook" ))

      Delete
    3. Indeed E1DA,
      There are ways to improve on the AP results without that asking price. ;-)

      Delete
  3. Is there anyway to adjust the balance (left-right) using the remote control?

    ReplyDelete
    Replies
    1. Hi fgk,
      No, not at this point unfortunately - no balance control (although the Topping D90SE has added this feature recently). I agree, with a screen like that, just a simple balance control would be awesome even if not adjustable using the remote.

      Delete