Part II: Topping DX9 "15th Anniversary" Limited Edition DAC & Headphone Amplifier - PCM Filters, Frequency Response & DAC Resolution. (And Trump's ongoing oblivious narcissistic tariff buffoonery.) [Firmware 1.23 Update]

Continuing with the discussion in Part I, let's proceed with Part II of our look at the Topping DX9 "15th Anniversary" DAC as we wade into some of the technical capabilities and performance of this device.

As I discussed the other week, my belief is that for relatively "simple" devices like DACs where what we want is as-accurate-as-possible conversion of the audio data into the analog voltages, measurements are really all we need to characterize the accuracy, the transparency of the device using modern tests and instruments that are capable of capturing details beyond the ability of human hearing. The capability for any device to achieve transparency of course does not guarantee that we'll be able to enjoy the sound or the music which are determined by individual subjective psychological factors.

With that, let's jump into some PCM measurements of this Topping flagship product.

I. Digital Filters

Over the years of coverage here on the blog, I think we've understood quite a bit about digital filters (see here for some basics) and I hope you've also experienced these yourself.

Looking at the impulse responses, what we have in the Topping DX9 is the typical selection of 6 options available in the AKM AK449X DAC chips:

As you can see, they're organized basically into Linear Phase, Minimum Phase, and a couple of what I suppose we can call "atypical" (but useful) settings.

The default setting in the DX9 is F-3, the minimum phase, steep/sharp roll-off variant which is a fine setting. Looking at these time-domain impulses might be eye-catching especially for those who fear seeing pre-echoing, but human hearing isn't particularly sensitive to subtle phase changes of such small magnitudes.

Here are the frequency responses if we overlay each setting, focusing around the 22.05kHz "Nyquist" limit for a 44.1kHz sampled signal:

As you can see, the sharp filters F-1 and F-3 have the same general steep attenuation so they overlay, so too the slower F-2 and F-4 filters. What is different between these are the phase characteristics:

Notice the non-linear price we pay for the minimum phase settings compared to the linear and NOS options. The intermediate phase setting (F-6) is an interesting compromise which is both sharp with attenuation around Nyquist, as well as minimum pre-echo in the time domain, while the phase shift above 18kHz would be inaudible. In daily listening, I might be tempted to use this one as the "Goldilocks" filter similar to what was discussed years ago.

F-5 is the Non-OverSampling (NOS) setting where the filtering has been turned off and waveforms would look like stair-steps; apparently some audiophiles like listening to digital like this. See here for more details.

Over the last decade, we've seen the idea digital filters hyped up to entice buyers (eg. Meridian) or some companies use more snake oilish attempts to embed filter changes in the format (ie. MQA, QRONO) as if worthwhile. Well, at least for the Topping DX9 DAC, as far as I can tell, there's no MQA nonsense which I consider a positive. 🙂

Since the linear phase sharp roll off filter is the standard orthodox filter with flat phase response and good anti-imaging suppression, for my measurements, unless specified, assume F-1 is the filter I used.

II. PCM Performance (USB)

Next up, let's look at distortion. For these tests, I'll be using the E1DA Cosmos ADC "stack" like this:

MiniPC → USB → Topping DX9 → E1DA Cosmos Scaler → E1DA Cosmos ADCiso (Grade A) → Topping HS02 isolator → MiniPC

Notice that this is a simple setup which in the minds of some audiophiles might not be seen as optimal since I'm just using a MiniPC computer connected to the Topping by generic USB cable directly. For the astute audiophile, I trust that at this point in history, we realize that the computer/streamer makes little actual difference as bit-perfect USB sources. The Topping HS02 isolator blocks ground loops in set-ups like this as previously discussed.

I suspect for most users, it's this USB performance that we'd be using mostly. Also, unless specified otherwise, I'll be using the Line Outs (XLR or RCA) rather than the Pre Outs since I believe most of us would be using this as a standard DAC in a multi-component system.

Over time, I have been moving away from the old RightMark 6.4.5 Pro software because of various bugs and stability issues. The software hasn't been updated since 2018 and while it still works in 2025 on Windows 11, there are a few issues that affect graphing of 192kHz results. Also, when paired with the E1DA ADC, make sure to play with the volume setting of the ADC again to 0% for Stereo Mode after starting the program; I suspect RightMark resets the value when it starts up. If you don't do this, the stereo crosstalk measure will only capture one channel and calculate half the actual value.

Bugs and stability issues notwithstanding, RightMark is still a good, quick, screening measurement. Here it is for the Topping DX9 USB input, XLR balanced output using 16/24-bits and also the most popular 44.1/48/96/192kHz sample rates:

For comparison, on the right side, we can see the 24/96 balanced output data for the Sabaj A20d (2022) (ES9038PRO), the SMSL DO100 (2x ES9038Q2M), and Topping D10 Balanced (single ES9038Q2M). 

And here are graphs of the Topping DX9 data at 24/96, XLR out, probably the most common hi-res bit/sample rate used for music production:

Nice concordance between the left and right channels suggesting that the construction, parts quality, are up to high standards for a modern high-resolution DAC.

I found what might be an anomaly at 192kHz sampling rate. For some reason, it looks like the DAC is restricting the full frequency response. Here's what I mean:

With filter set at F-1 (linear steep), the frequency extension of 192kHz is only slightly above 96kHz sampling rate, whereas it looks like at 384kHz, the DAC shows an appropriately full bandwidth. Hmmm, I wonder why that is and whether Topping should have a look at the firmware to check if there's a bug in how the 192kHz data is being handled. (In my detailed testing over the years, it's not unusual to find bugs that companies are able to fix with firmware releases as far back as the Oppo Sonica DAC in 2017.)

[Firmware 1.23 has fixed the bug - see addendum.]

As you can see in the RightMark results, this is clearly a high-resolution DAC with those kinds of numbers. Noise and distortion basically at ideal values at CD-resolution 16/44.1 and achieving expected improvements with 24-bit hi-res data.

Let's switch over to REW for a detailed plot of the typical 1kHz THD+N, for both XLR and RCA outs:

RCA 1kHz THD+N -119dB and XLR -121dB at 0dBFS, as measured connected directly to a MiniPC by USB are certainly excellent numbers. I suspect if I added another USB isolator between computer and DAC or was more careful with the cables (I used long 12' XLR cables, 6' RCA), it's possible I could get those numbers even better. Friends, when you're measuring THD+N at -121dB and want to get down to -122dB or better, this is when fancy interconnects might make a difference.😉 Totally academic of course as whatever gains would be all well below human audibility!

Next we can look at the stepped THD vs. Level graphs; let's examine this with the XLR left and right outputs:

Interesting little wrinkle around -25 to -30dBFS. Since distortion is so low, what we're seeing is a fluctuation in the noise level; THD+N still around -90dB where this occurs. Again, academic since this would not be audible with <0.005% anomaly. Otherwise, the THD+N v. Level is very linear.

Speaking of linearity; here's the zoomed-in output level linearity:

Less than 0.5dB deviation until below -125dB. This could just as well be the limitation of the ADC itself rather than the Topping DX9 DAC.

Since audiophiles generally appreciate that harmonic distortion isn't highly audible unless severe, let's have a look at intermodulation distortion. To mix it up a bit from typical measurements (like SMPTE or CCIF), here's the Linkwitz IMD tone which is a 1:1 sine wave at 1kHz and 5.5kHz which the late Siegfried Linkwitz often used in his measurements like this:

So using that IMD tone vs. output level, this is what it looks like with the XLR balanced line out:

Nothing scary. A little technical bump in Linkwitz IMD between -25 to -60dBFS, all below -80dB (<0.01% distortion).

Let's have a look at the "hi-res" 1kHz THD+N vs. Level graph captured at 0.1dB increments to examine the nature of the harmonic distortion pattern:

Compared with a device like the original SMSL DO100 which uses ES9038Q2M chips, notice that the Topping DX9 with its AK4499EQ does not show the regular sinusoidal distortion pattern.

As discussed here, it looks like this sinusoidal pattern no longer shows up with the latest ES9039Q2M designs which I suspect means the latest generation ESS HyperStream IV DACs do not have this. Interesting pattern that jumps out in measurements but questionable whether the human ear could detect such objective findings.

III. Multi-Tone Tests

Beyond the single and dual tones, let's have a look at some multi-tone results. First, here's my Triple-Tone Total Distortion + Noise (TD+N) measurement which is what I often use as a score for devices that shows off more complex harmonic and intermodulation products:

Of course it's 2025 not 2024 :-).

A value of -117dB TD+N for XLR output is excellent. I'm even more impressed by that RCA TD+N value of -116dB, one of the lowest noise RCA outputs I've seen on DACs. For context, the best amplifiers I've come across might be able to achieve TD+N of -100dB to -105dB at high volumes on this test.

Another thing to note not just in the FFT above but in the other measurements as well is the excellent right-left channel balance from this DAC, typically showing <0.05dB difference. 

For an even more complex signal then, here's the 1/10 Decade Multitone 32:

Again, should have been 2025.

Notice the very precise overlap of left and right signals, plus the peak level detected above the average in pink, captured over a few minutes for the 32-sample average. The highest noise peak is around 180Hz at -126dB; this implies around a 21-bit distortion-free range. Beautifully clean Multi-Tone 32!

IV. USB Jitter

Before closing off these measurements of USB input to the Topping DX9, let's have a peek at the Dunn J-Test, both 16-bit and 24-bit signals:

As per the jitter demo discussions, I really don't believe jitter is audible even though we can detect its presence quite easily in the form of side bands and "skirts" under the 11kHz or 12kHz tones of the Dunn J-Test variants. I think it's quite silly in the 2020's to still hear certain people talk about jitter as if they are "easily" able to hear its presence! Absolute rubbish, and uninsightful; be careful of such reviewers who are typically using jitter as justification for you to spend money on products unnecessarily (stuff like this).

For the jitter measurements above, we're looking deep into the noise floor with E1DA Cosmos ADC operating in mono mode. For the 16-bit signal we can see the LSB (16th bit) 229Hz square wave modulation tone and all jitter/noise products are below that. And for the 24-bit J-Test, all distortion products (jitter or noise) at worst around -140dB below the 12kHz signal (which itself is actually -6dBFS in level).

Bonus: To stress the system further, let's accelerated the J-Test up to 96kHz with the primary tone now at 24kHz:

As you can see, running the USB 2.0 interface to 2x the usual 24-bit J-Test speed induces more jitter sidebands; both small amounts of fine jitter very close to the primary signal as well as broader sidebands. This is of course a kind of stress-test since the 24kHz is already ultrasonic and we're still looking at sidebands at least -130dB down.

Topping DX9 - on the test bench...

V. Summary, Part II

So, what have we learned in these measurements of the Topping DX9 "15th Anniversary" DAC tested with USB in? Well, not only does it look great, sound subjectively excellent, but objectively we demonstrably have a high performance digital-to-analog converter under the hood (which we can peek into of course thanks to that glass top! 🙂).

The selection of 6 digital filters are certainly fine. Personally, I'd stick to F-1 linear phase steep filter, or maybe F-6 intermediate phase filter which provides a nice compromise of minimal pre-ringing for low-DR music with clipped samples while showing insignificant nonlinear phase in the audible frequencies. The F-5 NOS setting is great for those who prefer to defeat the filter and do your own software upsampling with products like Signalyst HQPlayer or maybe players like Roon.

When it comes to PCM performance, the objective results are excellent with low noise, low distortion. The only anomaly I came across was at 192kHz (and 176.4kHz) showing an unusual pattern in the frequency response which seemed to prematurely roll off. Years ago I measured the Topping DX3 (AK4493 chip) with the same set of digital filters, and I did not see such an issue with 192/176.4kHz. I'm guessing there's some kind of firmware bug. Higher sample rates like 384kHz and 768kHz look fine. Needless to say, subjective reviewers would not notice such an issue since we're looking at a roll-off up around 50kHz. Topping is looking into this and I'll update if/when I hear from the company.

With the usual 1kHz pure sine wave signal, the XLR line outs from this DAC were capable of achieving THD+N -121dB (SINAD 121dB) on my test bench quite easily. At this stage of my audio reviewing journey, whether it's actually -121dB as I measured or -124dB (see Topping advertised measurements) is immaterial for listening pleasure. 🙂 The way I see it, any DAC that can push THD+N below -120dB is worthy of engineering respect and better than -100dB is all I'd confidently feel I "need" for music enjoyment through a modern hi-fi system (assuming other tests like frequency response, crosstalk, etc. looks good)!

Measurement sweeps like the 1kHz THD vs. Level show excellent linearity with <0.5dB deviation all the way down below -125dBFS. Distortion pattern looks benign with odd-order harmonics (we are looking at very low distortion amounts!) predominating from -45dBFS and higher. More complex measurements like the IMD also is low and correlates with the general THD pattern, and as you can see with the Triple-Tone TD+N and 1/10 Decade Multitone 32, they're looking very good as well; as good as the best DACs I've measured in the last decade.

As for the USB jitter measurements, it's also really good (better than the $200+k dCS Varèse at least 😉). What I've generally found over the years is that if one is looking for the absolute best look Dunn J-Test, the higher end ESS DACs (like ES9038Pro devices or presumably modern ES9039Pro which I have not tested) tend to show pristine results with remarkably clean graphs. There are exceptions to this; for example the RME ADI-2 Pro FS R Black Edition with AKM AK4493 was remarkably "perfect". With the Topping DX9, we can see tiny low-level jitter/noise on the 24-bit J-Test. Absolutely nothing to be concerned about as listeners of course; even if not perfectionistically pristine!

Next time in Part III we'll proceed with a few more measurements focused on DSD, the S/PDIF inputs, and headphone output to cover a few of the other features.

As usual, let's end with some music!

That's Tracy Chapman with "Fast Car" (from Tracy Chapman, DR14 original CD release, 1988). The 35th Anniversary 180g LP was just released the other day for those LP collectors. This video was captured at Wembley Stadium in 1988. Amazing story with Tracy going from street busking in Massachusetts to playing in front of 90,000 at the Nelson Mandela 70th Birthday Tribute Concert.

In 2023, Luke Combs re-recorded "Fast Car" (Gettin' Old , DR6 stereo, DR13 multichannel/Atmos) which won the CMA Single and Song of the Year. IMO the original recording with Tracy's raw emotions expressed based on experiences in her upbringing, is indisputably superior. 

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

In other news of the day, I see the American Clown Playground Bully Show continues to inflict financial pain this week with further trade war escalation. Let's have a quick look...

Some stock market recovery after the 90-day tariff pause announcement. Other indices like the small cap Russell 2000 taking an even worse beating than the S&P500 above (-24%), a measure of just how poorly the policies are going to affect American companies without large international exposure.

The bond market continues to show unhappiness with dropping prices and increasing interest rate as per the 10-year US Treasury Yield:

Also see here. Maybe it's time for Congress to come to
the rescue to provide stability if they have the balls to respond.
The world already knows Trump's character as capricious.

The Administration plays dumb. Notice that this important marker of interest rate was heading down through the first part of 2025 already, and now Trump has upset the apple cart. He's gonna have to capitulate sooner than later I suspect. Otherwise, get ready for bond vigilantism and the "Moron Premium" slapped on the country.

The Chinese are unlikely to bend the knee to a bully like this as that would signal dishonor in the face of what the world sees in Trump and those who would enable him. Likewise, Trump's threats are severely negatively affecting the Conservative Party's chances in our federal election in Canada in a couple weeks.

I don't know what's happening in the U.S. in terms of educational and moral standards, but internationally people are not stupid and plainly see the oblivious narcissism of the man and his administration. While countries might play nice, let's not mistaken etiquette for respect. Oblivious narcissists are by nature weak because of their inherent low self-esteem even if outwardly they appeal to the impression of strength initially; that "bubble" of confidence. In the face of the facts of what's happening, Trump will almost assuredly fold sooner than later (already he's afraid of rising iPhone prices) but I fear the seeds of the next recession have already been sown even if we see a nice rebound initially as he walks back on all these nonsensical tactics. It's hilarious listening to people like his press secretary Karoline Leavitt trying to paint lipstick on a pig as his "art of the deal". 😂 

[Ummm... This is not about "spiritual warfare". It's a battle between truth vs. lies, competence vs. charlatanism, wisdom vs. buffoonery, and I would have hoped that one who professes to have Christian insights/wisdom would know the difference! I don't think prayers like this are enough if one does not "walk in truth". What would Jesus do?]

As humans, we tell stories to each other to form "tribes" (including we the "audiophiles"!), and share intersubjective connections through these tales we tell each other. National identities get formed through such stories, religion is another one, and the value of fiat currencies is another - arguably the most powerful item of faith that connects all humans - money. Throughout history, financial dislocations also logically correlate with the fall of empires and dynasties.

Speaking of the value of money, while everything is playing out, the U.S. dollar has been taking a beating with the index below 100 now; all in the face of potential inflationary headwinds if the trade relationship especially with China doesn't get sorted soon:

So with currency getting weaker, bonds selling off (increasing interest rate), and the stock markets on a rollercoaster ride but still down substantially for the last month, restoration of confidence is needed, lest the risk of capital flight out of American assets persists.

As a nation with tens of trillions in debt (something like $36T, much of it thanks to Trump's first term), it's one thing to improve things like government efficiency so as to pay back the bills quicker (done wisely of course!). But obviously, it's not smart to upset global trade with your lenders without a clear strategy. Now the Trump administration is creating doubt as to whether America could ever pay back its bills by having a government willing to risk recession and trade wars by presidential edicts. Those who are less well off in society who voted for this MAGA Republicanism will be harmed the most. Yeah, the billionaires who lined up at Trump's inauguration will lose money on paper, but are they likely to lose their cars, jobs or homes?

Seriously, I don't think anybody believes the USA could ever pay back its bills completely, nor have "Full Faith and Credit" in the country regardless of what Congress or the POTUS claims. But I suspect the vast majority of us in this present generation are just hoping the music keeps playing so we get our work done, bills paid, and achieve a decent quality of life. At some point in the future, the game of musical chairs will stop. Probably most of us are just hoping in won't be in this or foreseeably, our kids' generation.

[While this blog isn't generally about political or financial matters, as discussed privately over the years with some readers, I know many of you guys have interest in and thoughts about this stuff. As audiophiles, we see all kinds of snake oilish characters and hear false claims all the time so it's not hard to call out such political actors either. 🙄 

Furthermore, the audiophile pursuit requires a certain level of net worth, a fair amount of expendable income, since these toys are luxury products and we're arguing about "First World problems", I know many of you here are educated professionals so this stuff matters. 

Here's an interesting podcast interview for those who enjoy thinking about macroeconomics and trends.]

Hope you're doing alright fellow audiophiles. Regardless, let's at least enjoy the music...

Addendum: April 21, 2025

Topping has updated the firmware now to fix that 176.4/192kHz frequency response anomaly with firmware 1.23.

Confirmation of frequency response fix:

Nice to see the company responded quickly - within 10 days of my E-mail with a new firmware to try. I assume the new firmware will up available shortly at the usual website page.

I think this is an example of why it's good to have independent objective reviews done. High-tech components these days are complex and there are potentials for bugs of all sorts to creep in that otherwise would not be audible. Obviously, we could not depend on purely subjective reviewers to pick up issues like this!