MEASUREMENTS / REVIEW: Topping D90SE DAC - Part II: Resolution

In this Part II of the review of the Topping D90SE DAC, let's focus on the performance that this DAC can achieve. As you can see in the picture above, I've got the machine on the test bench ready to capture some data. Given the anticipated performance as described in Part I last time, the E1DA Cosmos ADC is used for the measurements being presented today.

As you can see on the LCD screen, it's playing DSD512 (22.6MHz) material at the time of this picture. For a more complete examination of the device's performance, let's see if we can capture both PCM and DSD performance.

This device does have a broad number of features, so let's try to hit the key (and also some not-so-key) features to better understand how well it works and if there are nuances that the audiophile should be aware of. As readers here I'm sure are aware of over the years, objective testing can often show anomalies that listening tests easily miss due to the limitations of human hearing and cognitive ability. Sure, one could argue that inaudible limitations are not important. I'd like to think that in principle, audiophiles have an interest in achieving both technical perfection as well as immaculate subjective performance. 

Before I begin with the results, there are a couple of things I noticed at the start of testing:

1. Using USB input, when there's no content being sent to the DAC, the D90SE seems to go into a "zero-detect auto-mute" that disconnects at the level of the hardware outputs. An engineer friend mentioned this to me in an E-mail recently. Playing digital silence over the S/PDIF inputs does not do this interestingly when I paired the DAC with my old Logitech Squeezebox Touch.

I'm not sure why the zero-detect auto-mute has been implemented for the USB input, or what benefits this might bring. Anyone know? I've see this in some portable headphone DACs over the years which I assumed may be for power-saving reasons.

2. The reason (1) was notable has to do with the pairing of the D90SE with the Cosmos ADC. For some reason, if I use the 48kHz sampling rate on the Cosmos ADC, there's an unusual 6.6kHz tone in the noise floor when the auto-mute is off. So if I'm not playing anything or I send the DAC digital silence for a bit through the USB input, I won't see this anomaly because auto-mute disconnects at the level of the hardware output. However, with S/PDIF input, this noise will show up continuously:

Once I switched to 96kHz, the stray tone went away. As such, the tests I'll be running here, unless specified, will be with the E1DA Cosmos ADC set at 96kHz. Interestingly, 44.1kHz was not an issue and there was no stray 6.63kHz noise found with other DACs like the Topping D10s, D10 Balanced, or the RME ADI-2 Pro FS (R Black Edition) using the Cosmos ADC at 48kHz!

I raised this with Ivan (E1DA). He dug around and was able to show a lower but similar pattern with his ES9038Q2M DACs (remember that the D90SE uses the older big brother, the ES9038 Pro). The hypothesis is that this is related to the master clock (MCLK) of the DAC (Ivan saw this with 25MHz clock on the ES9038Q2M). Maybe there's some ultrasonic content up there that's creating aliasing. If anyone has a high resolution oscilloscope (my lo-res Rigol oscilloscope did not show anything), I'd be curious if you see anything in the MHz frequencies with the D90SE. It's an interesting technical idiosyncrasy to be mindful of, so I'll keep an eye out for this with other devices down the road using the Cosmos. Just to be clear, there is no issue here around use of this device or the listening experience.

A. RightMark Comparisons (PCM)

As usual, let's start with RightMark at 16/44.1. Since most of our digital music is still at this resolution, it's important to make sure there are no concerns. As discussed previously, we'll use Filter 5 "Fast Roll-off Linear" as my preferred setting for both listening and measurements. Also, unless specified otherwise, I will use the default "4V" setting rather than the higher performance "5V" (we'll of course talk about that later):

Lots of data above! To the far left is the Topping D90SE result from the USB in from my DIY Raspberry Pi "Touch" streamer, Pi 3 B+ board. We then have a couple of readings using the Logitech Squeezebox Touch S/PDIF TosLink and Coax into the D90SE. Notice I've also included the RCA output measured through a 6' RCA interconnect and RCA-to-XLR adaptor (Raspberry Pi source).

On the right are a few Topping D10 comparison devices measured recently. For fun, I've included the Sony SCD-CE775 SACD/CD player bought 20 years ago measured awhile back. Although the ADC used to measure the Sony is not the Cosmos, at the 16/44.1 level, I've seen that so long as the ADC is capable of good hi-res performance, the results are reasonably comparable.

With good modern DACs, whether it's the D90SE or D10s or D10 Balanced, differences objectively at 16/44.1 are definitely not large (or significant) by any means! Modern high-res machines reproduce 16/44.1 with ease.

Let's have a peek at the graphs:

A few differences here and there between devices: the D10s' frequency response shows "ripples" due to the apodizing steep filter, the slightly higher noise level of the Sony SACD player (60Hz and 2nd harmonic at 120Hz in particular) can be seen along with IMD+N differences. Yeah, modern DACs have advanced in fidelity beyond CD players of yesteryear (even those with hi-res capability like this SACD model). Good to see that the Topping D90SE RCA output is free of the 60Hz hum.

Okay then, let's go hi-res and 24/96:

So for this hi-res round, we have the same series of Topping D90SE measurements including the RCA output. On the right we have again the Topping D10s, D10 Balanced, and let's put in the RME ADI-2 Pro FS R Black Edition which is based on the AKM AK4493 DAC chip. The RME has been set to +19dBu output (the +24dBu 0dBFS setting is just a little higher than what the Cosmos ADC XLR input could handle).

As you can see, we're achieving fantastically high levels of resolution here with each of these devices capable of >120dB of measured dynamic range and distortion levels of ridiculously small % figures!

Graphs for comparison:

Notice that the Topping D10s picked up a wee bit of 60Hz mains hum in this test (but way down at -138dB). D90SE remains very clean even with the RCA output.

Stereo crosstalk obviously is worse with the single ended output but still good around -90dB measured across the inexpensive 6' RCA-to-XLR interconnect/adaptor. Crosstalk can vary a bit with the cable used, I have another set of cables that achieves better than -100dB but I haven't bothered to remeasure; we saw this effect back in 2013.

24/192 and 24/384:

For completeness, I captured some numbers for 24/192 and even 24/384:

Yeah, again fantastic results. No real difference whether USB, Coaxial, or TosLink; whether from the Raspberry Pi or Logitech Squeezebox Touch (as usual, with a good DAC, bit-perfect streamers don't make much difference).

Due to some graphing bugs with RightMark, I can't show the comparison graphs properly like above. However, we can have a peek at the frequency response and noise floor at 192kHz (the 384kHz PCM graphs can be seen further below when we look at DSD playback comparisons) with nice right-left balance and clean noise floor (double checked that it's not just due to zero-detect auto-mute):

B. THD+N and 5V mode

Alright then. Let's have a look at good ol' THD+N aka "SINAD". I've already shown this as a preview with the Cosmos ADC article. Let's use the highest resolution "mono" mode of the Cosmos ADC, and the D90SE DAC's balanced output at 4V setting:

Impressive. Better than -120dB THD+N (120dB SINAD) along with better-than THD -130dB! Very nice right-left channel equivalence even at this high level of performance.

As you're recall, the D90SE also has a setting to push the output up to 5V. This will allow the DAC to achieve even lower noise levels, and in principle, we should see better THD+N:

While the THD value dropped a little for me (right at the absolute margins of the Cosmos ADC's limit for this input level in mono mode!), ultimately the extra voltage did produce a net improvement in lower noise level and hence breaking the THD+N -122dB level with the Cosmos ADC. In order to really improve the accuracy beyond this, a hardware notch filter (for example, E1DA's upcoming Analog Processing Unit perhaps) would be needed to better evaluate the residual distortion without that strong fundamental feeding the ADC and potentially self-generated distortion in the test equipment.

Not to be forgotten, here's the RCA output, 4V setting:

To achieve lowest noise measurement with RCA, I used the Surface laptop running off batteries so as not to contaminate the ADC capture with even miniscule 60Hz mains hum. Note that I'm just using a standard RCA interconnect and RCA-to-XLR adaptor so there could be small losses there.

With the higher noise level of RCA output, the Cosmos ADC "mono" mode was not necessary. Impressive results again, we're looking at around -118dB THD+N with RCA, again the best I've ever seen among DACs tested here. I did not bother testing the 5V mode with RCA output; as I've said before, this DAC's performance level really is aimed at balanced systems.

C. Stepped Frequency & Amplitude Tests

Using the stepped sine tool in Room EQ Wizard, we can have a look at the DAC performance across frequencies, and across output levels. As a start, let's make sure there are no anomalies across the 20Hz to 20kHz spectrum at different output levels - 0, -3, -6, and -12dBFS:

Looks great. For reference, the cursor is placed at -110dB and 1kHz so you can see the actual values of the harmonics in the legend. You can see the brown tracing is the noise floor stretching to the right. Many of the higher order harmonics lie below the noise floor. Even at -12dBFS, the THD+N is still at around -110dB across the spectrum.

Likewise, we can look at the harmonic distortions correlated with varying generator levels. Here it is for both left and right channels. Default 4V output level at 0dBFS is 4.22Vrms on my unit or 14.7dBu, cursor placed at 0dBu (0.77Vrms) for reference:

The harmonics are all buried below the noise floor until around -50dBFS (the -35dBu level). Notice that the THD+N plot is almost a straight line which doesn't "saturate" at higher voltage all the way to 4.22Vrms at 0dBFS. This bodes well for the higher 5V mode suggesting that the THD+N should continue to improve which we will look at shortly.

Before considering the 5V setting though, here's the same test as above with the DAC running at 192kHz (rather than 96kHz):

It's generally similar to the graph above with good THD+N but more "chaos" in the harmonic levels. The interesting thing is that when samplerate goes >96kHz, there appears to be less 3rd harmonic control with a rise in levels from -10dBFS up, thus increasing THD a bit; notice the little THD bump from +5dBu up. We can see this on the FFT (I held ADC stable at 96kHz for all measurements):

I've mixed it up a bit with both 44.1/48kHz-based -6dBFS 1kHz signals. The ESS DAC/ADCs have the ability to tweak THD compensation (as we have seen with the E1DA Cosmos ADC) so I'm wondering if there may even be optimization potential here to improve the higher samplerate (quad rate 192kHz and above) DAC performance. I know, this is a small thing, for the most perfectionistic of audiophiles of course. ;-)

Here then are the THD(+N) vs. Generator Level results at the 5V setting. To maintain the same input level setting for the Cosmos ADC as the 4V graphs about, I compromised and only pushed the generator level to -0.5dBFS so as not to overload the ADC. As such, the DAC can go a bit further than what I show here but I trust this is good enough.

Simply excellent performance. Nice right and left channel consistency at this higher setting as well.

Using the generator level data, we can look at linearity:

Very impressive. Down to -120dBFS output level, linearity deviates by <0.1dB! Yeah, that's basically "perfect". The graph above is at the 5V setting, but it looks just as nice at 4V and also the same for both right and left channels.

D. Multitone Tests

As a last section for these resolution measurements, let's do a few more multitone tests. (Already the RightMark battery above incorporates intermodulation distortion which requires dual tones.)

A few years back, I started using a triple-tone test for amplifier measurements in the AMOAR Score. Here it is with the D90SE:

Very nice. Again, great right-left channel equivalence. Distortion-free range can be seen down to better than -127dB. This DAC can be easily used as a signal generator for amplifier testing, but that would be overkill!

"Not good enough", you say?

Well, as discussed recently, we can use the "1/10 Decade Multitone 32" with this DAC.

For analysis, I'll use my standard hi-res FFT settings of 1M-points, with 8-sample average to clean up the noise floor. Let's examine from 15Hz to 30kHz to cover everything audible and a little more. For best resolution, we'll use the E1DA Cosmos ADC "mono" mode.

That looks very nice. We can use the level cursor to mark the maximum for any anomaly which is either representative of tonal distortion products (eg. harmonic distortion, intermodulation), or noise. In this case we're looking at that -132dB spike around 150Hz. 22-bits range, very nice!

Here's the same but with the RCA output (2V):

Not as clean as the XLR output above but still nice with better-than 115dB distortion-free range - better than 19-bits.

For comparison, we can look at the RME ADI-2 Pro FS (2018) at +19dBu (6.9Vrms) output level, XLR:

Indeed this older AK4490-based RME (not the newest ADI-2 Pro FS R Black Edition with AK4493) is not as distortion-free as the D90SE, but still excellent. Maximum distortion products around -120dB or about 20-bits of distortion-free range.

Finally, above you see the same test results with the inexpensive Topping D10 Balanced (XLR) and D10s (RCA). Notice the tendency of these DACs to show a wider "skirt" and sidebands in the lower frequencies, both DACs based on the ES9038Q2M chip. All-in-all, these are still very good results with 120dB and 115dB distortion-free range respectively.

As I mentioned before, one buys a high-resolution DAC like the Topping D90SE for the balanced output performance. Notice that the RCA output of the D90SE is still better overall than the D10s with cleaner low-frequency performance.

E. DSD Performance

Alright, let's finally examine DSD performance using the SoX-DSD signals as discussed last time. These will be XLR 4V results.

I spent quite a bit of time trying to understand the effect of the 4 IIR "Modes" available in the settings menu and as far as I can tell, the effects are minuscule, not even showing much difference at DSD64 up to 192kHz bandwidth. As such, I'll just leave it as Mode 1 for these tests.

As published the other day using the RME ADI-2 Pro FS R Black Edition, we can examine DSD test results with RightMark's 32/384 test signal encoded by SoX-DSD (SDM-8 setting). Here's the summary with the D90SE:

On the left we can see the DSD64/128/256/512 data. On the right, we have the PCM 24/384 performance for comparison. Right off the bat, we can see that DSD performance lags PCM by quite a substantial margin even if we ignore the "Noise level" numbers which tend to be inaccurate in the 384kHz RightMark test.

If we look at the frequency response:

Note: the pink "DSD512 M4" tracing is with IIR Mode 4 - I was wondering if there was any noticeable change. Basically no effect seen at DSD512, not surprising.

We can see interestingly that DSD64 and DSD128 show a different curve than DSD256 and DSD512. The fact that DSD256, DSD512, and PCM converge is interesting and perhaps speaks to the limits of the ES9038 Pro's "Hyperstream II" architecture when fed these high sampling rates. This is quite different than what we've seen with the AKM-based RME ADI-2 Pro or Topping DX3 Pro DACs where the higher DSD256+ rates will allow a bit more frequency extension.

The Topping D90SE's DSD64 and DSD128 frequency responses actually look quite a bit like the RME DAC with "DSD Direct" which I guess is good although it might be nice to have the option of a stronger filter like the 50kHz setting on the AKM DAC (as per RME's options).

Let's have a peek at the other graphs:

Here's an example of why the "Noise Level" graph is often inaccurate with RightMark at higher samplerates like 384kHz. Notice we can't see much of a difference in "Noise Level" among the DSD settings, but when you align the "Dynamic Range" graph with the -60dBFS 1kHz tone, we actually see that DSD64 and DSD128 have lower noise performance (better dynamic range result as per the summary chart) than with DSD256 and DSD512.

While I think a dynamic range of better than 110dB with DSD64 and DSD128 is great, the variability in the noise floor between the different DSD rates does seem a bit disappointing in the context of the class-leading PCM performance. As I showed with the AKM DAC test, the noise floor/dynamic range with those DACs tend to be more consistent with the DSD rate and tend to perform at least the same if not better with DSD256 and DSD512 compared to what I'm seeing here.

To double check that this fluctuating noise floor finding isn't just due to the 384kHz sampling rate on the ADC or RightMark issues, let's repeat using the 96kHz test signal converted with SoX-DSD:

Again, it looks like the "Noise level" and "Dynamic range" results still bounce around and we see that the DSD64 and DSD128 have lower noise performance than higher DSD rates.

To triple (!) check on that noise level graph, here's the "Dynamic Range" graph with the -60dBFS 1kHz tone using the converted 96kHz signal:

Yeah, looks like noise level with the D90SE is dependent on the DSD rate: PCM << DSD64 < DSD128 < DSD512 < DSD256.

Beyond RightMark, in DSD, let's have a look at a few other signals such as the 1kHz sine wave (-0.1dBFS to prevent clipping in the PCM-to-DSD conversion):

This is consistent with the variable noise floor shown above. The DSD playback THD values are still good with better than -120dB but once you add the noise level, the THD+N takes a big hit, worsening to around -100dB at DSD256.

For completeness, the 24-bit J-Test is fine in DSD, here's DSD64 and it was fine in DSD128 and DSD256 (but noise floor variation also easily seen consistent with above):

Finally, here's the "1/10 Decade Multitone 32" in DSD:

Yeah, I know I'm being ridiculously picky here with the "-116dB" value since the stray tone is up at 23kHz! The intent for this test is to be somewhat "extreme". ;-) Notice that within the audible range, there are no noise or distortions with better than 120dB (20-bit) range.

Yet again, we see that the DSD256 noise level is higher than at DSD64. DSD64 of course has that noise-shaped noise floor increase beyond 20kHz.

Conclusions from Part II

So as we end this second part of the review / measurements of the Topping D90SE, here's what we can say with confidence:

1. PCM 16/44.1 (CD- resolution) digital playback is simply no problem for high-resolution DACs these days. Over the years as DAC performance increased, we can show tiny improvements in the measured accuracy of 16-bit playback with yet smaller distortion numbers, and slightly higher dynamic range numbers. The ideal dynamic range of 16-bit playback would be 16-bits * 6.02 + 1.76 = 98.08dB.  Improvements in 16-bit performance among DACs have not been significant for years now. [The 16/44.1 blind test from a few years ago might be of interest if you haven't seen it.]

(Notice that over the years, the RightMark calculations with 20Hz-20kHz bandwidth can overshoot 98.08dB dynamic range a little bit when I measure the 16/44.1 test running the ADC at 24/44.1. The 98.8dB number we see with the D90SE, and also with the inexpensive D10s for example is the highest I have ever seen and I suspect as "perfect" a RightMark number as we'll see.)

2. Hi-res PCM 24-bit playback is fantastic with this DAC. We're seeing results here that surpass all my other devices across the board from noise level to harmonic and intermodulation distortions. Likewise, RCA output performance is excellent. As I suggested previously in Part I, I would not use this DAC in a single-ended system - the resolution we're looking at here simply demands that one pairs this with a fully balanced system from DAC to amplifier to extract the potential resolution.

3. On the picky side, I did notice that once you go to 176.4/192kHz (quad) sampling rate and above, the 3rd harmonic increased more than at the lower sampling rates when output level is above -10dBFS. I wonder if there's some tweaking potential here. I heard that ESS DACs are capable of applying multiple THD compensation parameters depending on DAC settings so perhaps a different load-out could be applied at and above quad samplerates.

The other day, I noticed that the newest RME ADI-2 DAC FS with the ES9028Q2M chip has the option to adjust 2nd and 3rd harmonic THD compensation (see manual page 28, E1DA's DACs and ADC can do this too). Maybe this could be something Topping could provide for end users who might want to tweak their DAC further assuming they have instruments to aid with the calibration process.

Normally I would not mention a small item like this, but since this is the "best" measuring DAC currently available, everything matters, right?

4. The 5V mode does improve noise level and I think the harmonic distortion amount is about the same. This is to be expected as it's likely that the ES9038 Pro chip reference voltage has been boosted in this mode. Dynamic range can increase by up to 1.9dB going from 4V to 5V, and so long as the noise level doesn't increase as much, then it's worth it. In my test here using the E1DA Cosmos ADC in mono mode, without the benefit of hardware notch filtering, I'm seeing an improvement of -0.6dB in THD+N; this is surely an under-reporting.

While it's awesome that Topping has included the 5V mode to squeeze out yet another bit of extreme performance, it's not like I'm in any way dissatisfied with the 4V setting! In fact, I'd be happy in day-to-day listening to keep the DAC at 4V XLR out. I see it a bit like "overclocking" a computer that's already way faster than one needs. ;-)

5. The DSD playback IMO is a bit of a disappointment considering the PCM results if DSD is important for you. I do have another ES9038 Pro DAC at home, the Oppo UDP-205, so I quickly had a look at the noise floor using the same 1kHz DSD64 and DSD256 files:

As you can see, the Oppo, while not posting a great THD+N on account of the high THD score, showed the same increase in the noise floor at DSD256 as the Topping D90SE. Perhaps this noise level difference is an inherent limitation of the ES9038 Pro chip?

AKM DACs like the RME ADI-2 Pro FS R Black Edition typically look very good maintaining low noise level with DSD playback. While I have not had the chance to try one out, it might not be unreasonable to consider that the Topping D90 predecessor with its AKM AK4499 DAC might be a significantly better DSD option. It's also less expensive especially when you don't bother with the MQA scam "feature". Similarly, the dual AK4497 Topping D70s looks very interesting. Yeah, you lose a few THD+N/SINAD points in PCM playback, but seriously, I'm not sure any of us will hear the difference anyways. I would certainly consider these alternatives if I love DSD or regularly use software like HQPlayer to upsample/transcode to DSD. (AKM-based DACs, especially these higher models can be in short supply these days due to AKM production issues after their 2020 fire.)

By showing the Oppo UDP-205 results, a machine released in Spring 2017, I think this is also a nice demonstration of how designs become optimized over time. Even though the internal DAC chip is the same (ESS ES9038 Pro), the performance of the Topping D90SE is clearly superior. By the way, those sidebands you see around 1kHz are power-related in the Oppo measurements. No such anomaly with the D90SE which is also using a switching power system.

Bottom line... With good filters and no jitter to speak of as discussed previously, ultra-low distortions throughout the audible frequencies, THD+N better than -120dB, more than adequately low crosstalk, linearity of the analogue output practically "perfect", no difficulty negotiating a complex multitone signal with 22-bit distortion-free range, what more would human ears need objectively from a high-fidelity digital to analogue converter?

Friends, as a "more objective" audiophile, I don't think I have any "right" to demand more on the PCM playback side. This is certainly the highest resolving device I've come across here. The idea that this is the highest-fidelity DAC on the market today obviously resonates with what I'm showing as well; Topping really squeezed out everything they could from the ES9038 Pro flagship! (Barring maybe a little more THD compensation optimization of course at higher samplerates. ;-)

Nonetheless, even though personally I rarely need DSD playback, I do have some reservations about DSD performance, likely inherent limitations in the ESS converter. This is perhaps an area that future ESS DAC chips can improve upon instead of wasting time and chip space on the nonsense MQA decoding stuff they've added of late!

Next time, for our final (Part III) installment, let's talk about listening impressions, let's also post up the AMPT track recorded from the Topping D90SE, and some concluding thoughts.

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

I updated to Windows 11 on my main workstation this past week. This is a free upgrade for licensed installations of Windows 10 and was relatively quick and painless. Compared to maybe 15 years ago, Windows these days is very stable so I don't see any urgency to upgrade for the time being unless you like living on the edge with your tech (like me ;-). There are not that many differences, and certainly no "killer" features as far as I can tell currently.

Okay audiophiles, before we see some question about "Does my DAC sound better in Windows 11"? The answer is of course "No". The device drivers have not changed; after all, "Bits Are Bits" with good quality hardware as you know. And any reputable high-fidelity DAC should be performing in a bit-perfect fashion.

Yeah, there are some look 'n' feel differences which overall I think works reasonably well. Microsoft seems to be gunning to dissuade those with older hardware from upgrading to Win 11 (and maybe in the process nudging folks to upgrade their hardware) - see CPUs supported, and the need for TMP 2.0 (should be OK for computers from 2016 onwards). If you do have older hardware, you can still try a Windows 11 installation by using the registry hack but it's uncertain what kind or support or updates one should expect in the days ahead.

BTW, Windows 10 will still be officially supported by Microsoft until October 14, 2025. So much for Windows 10 being the "last" version of Windows.

Anyhow, so far so good. There are still optimizations to be had especially if you're running AMD Ryzen CPUs which should arrive in the next month (I'm not noticing any issues on my AMD Ryzen 3900X computer).

And in other news, The Absolute Sound reviews a US$8000 glorified NAS with USB output (Fidata XS20U) "housed in a thick billet-aluminum case, enclosing a 3.2mm steel chassis that mounts the internal components", which serves up digital data with "exceptional tonal and timbral accuracy". Various supposed attribution theories of why this magical device is special - all neatly wrapped up in the long article - presented for your entertainment. Hilarious I suppose, if not a little sad that this is what "high end audio" is often about these days and the way they're trying to sell such products.

As a body of knowledgeable, technologically astute, evidence-based hobbyists, we can hopefully all look at these kinds of articles and demand more. I'm guessing this is why TAS no longer accepts comments in fear of obvious responses to questionable-value products like this and the wholly subjective mere testimonies they give.

Happy Canadian Thanksgiving this weekend, eh?

Stay safe and hope you're all enjoying the music. I'll make sure to give thanks for the amazing sound quality we can achieve these days without the need for massive budgets!