MEASUREMENTS / REVIEW: Topping D90SE DAC - Part I: The Hardware, Filters, MQA, and Jitter Performance

Today, let's start with what I'm aiming to be a 3-part review/discussion of the Topping D90SE DAC. I figure I'll take my time on this since things are busy around here and I hope not to purchase more hi-fi DACs going forward, so I might as well savor the moment. ;-) As a "more objective" audiophile who has a vision of aiming for "high fidelity" / "transparency", there is a clear target and end-point to what's needed from hardware performance especially with DACs.

Readers of this blog know that I've had a number of Topping devices reviewed here over the years. In fact, for any single brand of DACs, I think I've reviewed more Topping gear here than other brands. This, I believe, is a reflection of a brand that provides multiple products at price points with features that actually speak to me as a consumer interested in value which includes features, and price. I have certainly not been disappointed by overall quality to this point.

At a current retail price of around US$900, this is not an inexpensive model. As per most of my reviews, I bought this through usual retail channels so what I'm reporting on here is not any potentially specially-selected unit sent from the manufacturer.

Feature-wise, this is a DAC only but accepts multiple inputs - USB, HDMI-style I2S, S/PDIF Coax, S/PDIF TosLink, and wireless Bluetooth 5.0 (SBC, AAC, aptX [LL,HD], LDAC). While it has both unbalanced RCA and balanced XLR outputs, there is no headphone amp. For that, Topping recommends grabbing their top model the Topping A90 (~US$500).

What makes this DAC interesting in mid/late 2021 is that this is reputed to be the highest fidelity converter in the world (available to consumers) at this point in history, Topping's "flagship" device. 

Let's take a look...

A. The physical unit & the DAC technology

The package comes in a simple but attractive basically all-black box with the Topping brand name in metallic print and some usual side stickers to indicate serial number and product color.

The DAC itself is just a little less than 1.1kg. Dimensions are approximately 22cm wide x 15.5cm deep x 4.5cm tall. This is about the same width as the RME ADI-2 DAC (similar price) but deeper and not as tall. As you can see, there is a small Bluetooth antenna, a pack-in USB A-B cable, and stock IEC power cable in the box. Plus there's the user manual (with already a number of Audio Precision graphs, similar to ASR review) and warranty pamphlet with info about other Topping products. The warranty says: "free from malfunctions and defects in both materials and workmanship for one year".

The front of the unit has 3 buttons - POWER/SEL, and a couple for up/down volume or for moving the cursor when in the options menu. The set-up options menu can be accessed by holding down the SEL button while flipping the power switch on at the back. Here's what the menu looks like:

Notice item 12 - there is a way to set the XLR output to either 0dBFS 4V or 5V mode. An engineer friend who has this DAC noted that the 5V mode likely increases the reference voltage to the ESS Tech ES9038Pro chip (indeed increasing the 3.3V to 4.06V based on my friend's measurement) which ideally can improve the signal-to-noise ratio by up to 1.9dB. Distortion (like THD) should be about the same while noise floor improves, thus bettering the THD+N.

Other menu items include filter settings for PCM and DSD, turning on/off "preamp" function (ie. fixed 0dBFS output without volume control when off), Bluetooth ON/OFF, as well as the IIS settings discussed below, etc. The DSD filter choices appear to be the default IIR attenuation settings in the ESS datasheet (page 49) with your choice of -3dB at 47/50/60/70kHz.

The rear of the machine consists of the IEC power plug and switch to the right. The middle segment consists of the various digital inputs. There's the AES/EBU XLR connector and S/PDIF coaxial and TosLink derivatives for up to 24/192 and DSD64 via DoP input. We have a Bluetooth antenna connection (capped in red). USB for the typical family of bit depths and samplerates - up to 32/768 and DSD512 (DSD256 DoP). Yeah, MQA is available for USB and the S/PDIF family of coax / TosLink / AES inputs.

Something I noticed was that the S/PDIF optical TosLink connector can be a little small for certain cables and required quite a bit of pushing to get a cable in there securely. This was not an issue with an old Acoustic Research TosLink cable but tight for some glass cables I own. I double checked and others have noticed this as well with their units. No big deal but if you're using TosLink, make sure it's all the way in and secure otherwise you might hear dropouts and errors.

Notice there's also the HDMI interface here for I2S input labeled as IIS-LVDS (I2S - Low Voltage Differential Signaling). Maybe at some point I will have a source that can send a signal to this, but I fail to see a need. This port can accept up to 32/768 and DSD1024 native (still DSD256 DoP).

Since there is no single standard for the I2S port, it's good that the Topping can configure for compatibility between different "families" of devices with differing pin assignments. Here are the relevant settings from the manual:

Note: Item 11 - Polarity setting is around these other options because it only refers to IIS output polarity as relayed to me by a friend. Apparently this doesn't do anything for the other PCM/DSD inputs.

I guess you'll just have to look at your source device pin-outs to check polarities and which pin is used for DSD signaling. Here's an interesting resource if you want to look up various HDMI pin assignments.

We have both balanced XLR and unbalanced RCA to the left of the back panel. The DAC can be set to either one or both outputs active. Let me just say right now, unless there are features with this device you really want/need, if you're only planning to use the RCA outputs, consider something like the Topping DX3 Pro, maybe the Topping E30, or the SMSL M300 Mk II even (which still has balanced out but much cheaper) if one wants to keep with the chi-fi brands. The intended resolution this device produces IMO is really only of benefit if the rest of your analogue chain is capable of extremely low noise, almost certainly requiring truly balanced preamp and amp(s) to justify the performance gain.

Slightly curved and "ribbed" sides. Aluminum construction all around.

The remote controller is the standard plastic "Topping RC-15A" which is identical to ones supplied with some of their other DACs. For example, the Topping DX3 Pro uses this controller as well. The central round button selects whether RCA / XLR or both outputs active. Left and right selects which digital input. +/- for volume.

Of the lower round buttons, the "FIR" button selects one of 7 digital reconstruction filter options. The "AUTO" button puts the DAC in "auto standby" mode which will put it to sleep after 1 minute of not seeing connection on the selected input and wakes up when it does. The bottom right button is for brightness and there is a selection of Lo/Med/Hi brightness with an "A" setting that is of medium brightness then automatically goes blank if no display changes in 30 seconds, waking again when changes like volume adjustments are made.

At this price point, I would have certainly liked to see a more robust remote controller (metal), preferably customized for this DAC without unmapped buttons (like that headphone button).

Before going into measurements, I think this is a good time to talk about what makes this DAC special in terms of high fidelity performance. At its heart, the D90SE is built on the ESS Tech ES9038Pro DAC chip which actually has been out since 2016 with a potential for mono, 2-channel, and 8-channel conversion output modes. Back in 2017, we considered the Oppo Sonica DAC, and in 2018 looked at the Oppo UDP-205 - both also based on this chip.

Beyond just channels, the ES9038Pro can also function in either current or voltage modes with the current-mode preferred for lowest distortion. And in order to get the cleanest, most linear performance when transforming the current into analogue voltage output, high-quality current-voltage converters (I/VC) have to be implemented. As you can see in the Topping D90SE literature, they've incorporated independent 8-channel I/V conversions. No cutting corners with voltage mode operation, 2-channel mode, or even combining channels before the I/V stage in the D90SE design:

The Topping D90SE 8-channel I/V conversion using TI OPA1612A "SoundPlus" opamps. 

As a 2-channel stereo device then, 4 of these OPA1612 I/VC outputs are combined in parallel for each balanced channel. A friend who had a look at the circuit observed that the outputs are connected via 200Ω resistors for a combined output impedance of 50Ω per phase, 100Ω balanced output impedance. While I have not tried it myself, it's suspected that there's enough current here for the XLR outputs to act as a pretty decent balanced headphone amp - with the appropriate cables and adequately sensitive higher impedance headphones of course!

If you want a look inside the Topping D90SE, check out these photos. Great shots. Notice the power supply is a switching device. I hope this is a good reminder for all those folks who somehow have a "thing" for expensive third party linear power supplies. In fact some of the highest performing gear in the audiophile world are actually based on switching devices. So long as the designer knows what he/she's doing and of course verify with measurements that there's no noise/distortion. As consumers, I think it's best not to hang on to simplistic beliefs in what kind of power supply is "better".

Speaking of power, the specs note 8.5W in use. It does get slightly warm over a few hours. Not as warm as the RME ADI-2 Pro FS models (12W).

Let's start looking at the measured performance...

B. Measurements: Oscilloscope, digital filters, and MQA

As usual, let's just start by putting the output onto an oscilloscope to observe the overall levels and morphology of waveforms.

As you can see this looks great. Precise overlay of sine and bandlimited square waves as a demonstration of excellent channel balance. This is the RCA output with Vrms measured at 2.1V using default 4V (XLR) settings. Balanced XLR output measured 4.22Vrms. When turned to 5V mode, I read 5.22Vrms from the XLR.

While the square wave is bandlimited, typically one can still see hints of whether the filter used is linear or minimum phase. In the default setting, one can see that in fact a minimum phase filter is used (as witnessed by the asymmetrical square wave transition "ring").

Indeed, the manual tells us that the default setting is "PCM FIR / Mode 3" or "Fast roll off minimum".

Here's a look at all the filters offered in the form of my "Digital Filter Composite" (a variant of the "Reis Test") frequency-domain graphs along with the impulse response to examine the time-domain impulse waveform. For convenience let's split it up into the first 4 settings and then the last 3:

For the most part, the filters are able to absorb about +2dB intersample peaks so on the whole the 0dBFS white noise signal typically doesn't clip. The only exception interestingly was Filter 3 which also happens to be the default!

Here are the remaining filters:

Looking at the ESS ES9038Pro datasheet (see pages 51-54), these appear to be the defaults available on-chip. Since I generally prefer linear phase filters, minimal intersample overs, and steep roll-offs with no imaging artifacts, Filters 1, 5, 6 look like they're the best for these characteristics. The steep roll-off "hybrid" setting 7 also looks good if you play a lot of dynamic compressed music with clipping and want to reduce potential pre-ringing but prefer not to go with a minimum phase setting (we discussed this and intermediate phase settings years ago).

From my measurements (and listening), I'll choose Filter 5; a standard "fast roll-off linear" for much of the testing with flat frequency characteristics out to 20kHz. But wait... There's more.

I don't think I've seen folks measure DACs to show the slight differences when they change filter settings. So while I'll leave most of the "fidelity testing" to Part II, here's a quick preview using RightMark Audio Analyzer with the E1DA Cosmos ADC in stereo mode running 24/44.1 tests for each of the filter settings:

Notice that the differences are minimal for noise, dynamic range, and distortions but the frequency response (20kHz upper end roll-off) is significantly different. Even within these numbers, notice that Filter 5 interestingly gives us the best results. (Filter 3 is similar but minimum phase which is usually not my preference.)

We can look at detailed frequency response graphs (again let's split this into Filters 1-4 and then 5-7 for clarity):

We can better appreciate the roll-off characteristics and interestingly the "rippling" in the audible frequencies with the "apodizing" Filter 1 (similar to the Topping D10s). We can easily see the "slow-roll" character of Filters 2 and 4. The "fast-roll" Filters 3 and 5 have the best flat frequency extension. Filter 6 indeed gives us a clean "brick-wall" at 20kHz and the unusual looking Filter 7 looks fine but doesn't quite give us a full 20kHz extension. So overall, again, Filter 5 is my preferred and I would consider Filter 6 if I want a literal 20kHz "brick wall".

Small technical nuances - don't lose sleep over this.

Let's talk about MQA...

I'm definitely no fan of MQA ;-). I believe it's simply worthless these days so I hope Topping didn't pay much money to implement it into the DAC. We know that MQA also implements its own filtering mechanisms. Since I'm going into this review to capture quite a bit of detail on the D90SE, let's have another peek at MQA in 2021.

After all these years, the MQA "origami" encoding scheme is still under wraps, the detailed mechanics of it presumably only known to those who have signed industry NDAs. To some degree, the fact that not more is known or the "first fold" decoding reverse-engineered with freely available software probably speaks to the lack of interest after 6 years or so since introduction. Nonetheless, like before (starting with the review of the AudioQuest Dragonfly Black in 2017), we can have a look at a few of the filters implemented when the DAC detects MQA's control stream and "renders" the output thanks to the work of Måns Rullgård.

Display during playback. Note the input --> output settings, sampling rate, type of data (usually "DSD" or "PCM", "OFS" here for MQA), and dB attenuation.

You can see that when I send it a "decoded" 24/96 MQA file to "render" (the "final unfold"), in this case targeted at 192kHz, the D90SE displays "OFS" on the top right as opposed to PCM or DSD. I guess that's supposed to mean "Original Frequency Sampling" (or just technically "Original fs", original sampling frequency). Sorry to say, I don't see a blue LED on this device so I'm not as relieved that my MQA content is truly "authenticated" by the artist in the studio. ;-(

If MQA is playing a 96kHz stream with no further "rendering" or "unfolding", this is the filter it uses on the Topping D90SE:

Now if MQA is told that the original content is 192kHz, as far as I can tell there is no actual further "unfolding", rather all they're doing is using filters with very poor anti-imaging properties ("aliasing", "leaky") to just fill in ultrasonic content. Let's have a peek at the first 6 of these filter impulse responses:

Compare these with the AudioQuest DragonFly Black "MQA full monty" from 2017 and we see that the morphology of these filters is about the same. As usual, why MQA chose these settings, and whether there is any rhyme or reason remains an enigma. Likewise, it's an enigma as to what kind of "neuroscience" any of this is based on, or how this is supposed to "de-blur" or have improved time-domain properties!

Maybe one day, reporters/audio writers interviewing Bob Stuart might want to demand some explanation. Or maybe it's just better for everyone's sake to let MQA die and sweep away this "audiophile" embarrassment into the annals of history.

Here's a -4dBFS white noise in 24/96 (48kHz bandwidth), played back with built-in DAC Filter 5, then put through the MQA 96kHz filter, and "rendered" with Filters 2 and 4 to an "OFS" of 192kHz:

Note: This was captured with the RME ADI-2 Pro FS hence the rising ultrasonic noise to 192kHz (24/384 sampling).

Behold the effect of MQA on the frequency domain for what is supposed to just be a "flat" frequency response white noise which is very well demonstrated by the D90SE's usual Filter 5 playback.

Hmmm, tell me, how is MQA in any way "better"? Notice the slow roll-off with the 96kHz filter instead of the intended flat frequency response out to 48kHz. As for the "rendered" output using #2 and #4 filters as examples, notice how they basically fill out the non-existent >48kHz content with imaging artefacts up to 90-95kHz, again, presumably calling this the last stage of "unfolding". Notice that MQA #2 accentuates non-existent 50kHz "content" while MQA #4 attenuates 50kHz slightly.

I'd love for MQA to show actual DXD (24/352.8) content vs. the same music encoded to MQA 24/44.1, then decoded through this meat grinder. I suspect they're too afraid to show us anything more than just some grossly averaged FFTs over the years.

Let's see if I can make it even clearer that this is not just "noise" (random or irregular fluctuations) that's being filled up by MQA when it's upsampling (to 192kHz). Consider this (take note Jim Austin, as per discussions here):

As you can see, I've created a 96kHz file with content starting at 17.5kHz rolling-off by 45kHz. Obviously music should not have high levels out to 40+kHz, but you never know and I'm trying to be illustrative here. The shape is intentional so we can identify the pattern across the frequencies. With the D90SE Filter 5, we see a very accurate reproduction of the intended waveform. The moment you send 96kHz content for MQA playback, the filter used results in the top right tracing. It's not a strong or clean filter as it obviously allows a bit of "imaging" beyond 48kHz, at least it's not horrendous and we see about -50dB attenuation of the peak around 57kHz.

The moment you tell MQA that the source "original fs" is 192kHz, then it will switch to one of those "rendering" filters above. Look at what Filters 2 and 4 did! Simply gross "mirror imaging" showing up in the octave between 48-96kHz. This is not simply some kind of random noise being added that's supposed to look/sound like the original recording! In a true 192kHz recording, even if 48-96kHz were just noise, it would never be a mirror image of the stuff from 0-48kHz, right!? Yet this is what MQA does regardless if it's being implemented on an entry level Dragonfly DAC or this state-of-the-art device.

How can anyone with any hi-fi / accuracy / objective pretensions be pleased by this kind of effect? I accept if a person subjectively claims this "sounds better" to them since any kind of distortion can "sound better", but I question such persons' devotion to good fidelity! Would it not be better to just turn off MQA and let the normal D90SE filter handle the digital reconstruction than introduce all this junk!? (Seriously Stereophile, I don't believe you should be OK with this or worse, act as shills for this nonsense - how many years have we been at this charade?)

Hopefully this is the last time I need to drag up measurements of MQA. Simply disgusting what MQA has been claiming, and barely do we see any pushback from the "mainstream" press (or some "objectivists" who know better). I think MQA is a bit of a joke these days amongst audiophiles and I suspect articles that tout the wonders of the "blue identification dot" are simply there for much-needed click-bait. Don't click on that link! ;-)

By the way, the MQA rendering algorithm is kinda sneaky. You can't just throw a stream of impulses at it and expect the system to show you the goods. You need to embed the impulse within some low level signals since there seems to be a form of detection at work. I don't know if this is just a necessary part of the algorithm or done purposefully as a way of concealment. Let's just say MQA conceals a lot of what they're doing. Considering how consistent they have been, I suspect "concealment" and "secrecy" are simply part of the corporate mission statement.

C. Measurements: Jitter

Usually, I will leave measurements of jitter a bit later after looking at DAC resolution, but I figure I might as well get this over with. Here are the 16 and 24-bit J-Tests with USB, S/PDIF Coaxial, and S/PDIF TosLink inputs:

As you can see, the graphs are of both the 16/44.1 and 24/48 J-Tests captured from the USB fed from a "Raspberry Pi 3 B+ Touch", and both S/PDIF tracings are from the TosLink and Coax outputs of the Logitech Squeezebox Touch (which I have used as a standard of sorts for many of my S/PDIF jitter tests over the years). The E1DA Cosmos ADC is used here for the very clean noise floor and very low jitter with minimal "skirting".

Like I said, it's good to "get this over with" because the performance is simply excellent across the inputs! As usual, don't freak out about jitter. Alas, I don't have any HDMI / I2S source to see what the J-Test would look like using that. To be honest, given how great this looks already, I don't believe the I2S interface would improve things given the expense. It would not surprise me if the I2S interface had worse performance and would not be the first time that something expensive in the audio world actually offered lower performance if this be the case.

I've seen some comments about optical interface drop outs with some devices and ESS DACs. Apparently this has to do with excessively tight jitter rejection parameters so I'll keep an eye out to see if there are any issues with the D90SE or firmware updates needed. I did not have any issues with the SB Touch's TosLink.

One last thing, let's have a look at the hi-res full 24-bit J-Test spectrum from 10Hz to 20kHz (256k-point FFT):

That's pretty darn lovely, down to the very low level 24th-bit 250Hz jitter modulation tone embedded in the J-Test signal easily captured! Wow.

As you can see, there's absolutely no need for expensive streamer devices. The little Raspberry Pi 3 B+ "Touch" with just an inexpensive power supply delivered the data "perfectly"; no worries about noise, no jitter at all.

Conclusions from Part I

Okay guys and gals, let's end here for this first part of the Topping D90SE DAC review. Some general take-home points:

1. The Topping D90SE is much more utilitarian than luxury. While the metal build quality is good, it's not particularly big, heavy, or calls attention to itself. This is simply a DAC that aims at accuracy to the source data with a good complement of inputs including wireless Bluetooth. You might need to use a little bit of elbow grease to get that TosLink cable securely in!

- A suggestion for Topping if they update the firmware - I would love to see a "FIR Filter Lock" in the set-up menu that prevents easy switching of PCM/DSD filters with the "FIR" remote control button. In use, I think most listeners select a filter option and stick with it. No need to fool around with this and I don't like accidentally pressing the button and having the filter switch so easily without notice.

2. Topping designed this DAC to optimize the performance of the ESS ES9038Pro by running it in 8-channel current mode. They have implemented 8 independent I/V stages for each of the channels. 4 channels are then combined in parallel for each of the final 2 stereo channels.

3. My engineer friend commented on the fact that there are 3 switching power supplies inside. A larger MeanWell 100-240VAC to 15V 1.4A block, and then 2 separate smaller MornSun for +/-12V rails used internally. Again, audiophiles, no need to fear switching power supplies in very high performance devices these days!

4. The device has 7 built-in digital filter settings, likely just the default options built into the ES9038Pro chip. Most of the filters have a decent amount of intersample overload protection of ~2dB. I found Filter 3 (default) setting the least tolerant of intersample overs. I like linear phase, and I like flat response to 20kHz. Filter 5 IMO is the best technically - the standard "Fast Roll-off Linear".

5. Yeah, I paid money to buy an MQA device. I feel ashamed ;-(. However, I don't subscribe to Tidal (basically the only "outlet" for MQA music) and turned off MQA decoding in Roon, so I'm not feeding MQA any more money. The silliness needs to end. Consumer apathy (no $$$ for MQA stuff) along with strategic outrage and calling out the BS will probably achieve that goal.

6. It doesn't look like there's been any change in the way MQA's filters are implemented over the years. I did not bother checking out the noise shaping/dithering but even if something changed there, no biggie. MQA's filters are a step down from the normal DAC options even when just playing 96kHz material, and when upsampling to declared 192kHz+ "original sampling frequency" content, imaging artefacts are even worse.

Some MQA advocates might still say - "but you can't hear the ultrasonic distortions, so MQA still sounds good!" This is problematic because if ultrasonics have no effect, then why bother even having claimed 96+kHz material? By all means, go use a NOS DAC (no reconstruction filter) and fill up the ultrasonic frequencies with "stuff", and claim that every 44.1/48kHz track is actually 384kHz because you've magically "unfolded" high bandwidth content! But is this kind of playback "high fidelity", "accurate", and what "perfectionist audio" is about? I certainly hope not!

7. When it comes to jitter, the Topping D90SE performed fantastically through the USB, coax and TosLink inputs. We see basically "ideal" J-Test results as one would expect from "femtoclock" level performance. Definitely no need to buy any kind of add-on devices that's supposed to "reclock" or improve time-domain performance. Until proven otherwise, you're likely wasting money on this kind of stuff in general, if not potentially even worsen performance especially with a device like the D90SE!

Next time, let's delve into some more "fidelity" characteristics in Part II.

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

An audiophile engineering friend sent a note to suggest having a look at the Audio Check website. It looks like it has been updated with new test tones. Have fun with some of the self-tests like the THD test or check out your room acoustics with MATT. There's also a plethora of downloadable test signals, and you can even use the online signal generator. Wonderful stuff!

From the same signal processing PhD who came up with Audio Check, have a listen to Neural Symphony! This might help to mask tinnitus if you suffer from that. But even if you don't, it sounds quite amazing especially with headphones. Make sure to check out the other effects like The Sound of Thunder, Rain Noise, Night Blue or Cinematic East. Convincing synthetic nature sounds and even music. Brilliant.

Have a great week ahead! Stay safe. I trust these days we're generally vaccinated in the developed world and in a place of vigilance rather than fear - or worse - paranoia.