MUSINGS: The hunt for -150dB THD with the Victor's 1kHz Oscillator; A look at Paul K.'s Multitone Loopback Analyzer; And the pinnacle of audio streamer pricing? The Wadax Atlantis Reference Server. (And REW IMD vs. Level Step!)

I like a good challenge ;-). A few weeks ago, in my post on the E1DA Cosmos APU, in the comments, MG discussed achieving THD -155dB on the Victor's Oscillator! In my original post, I showed a picture of -141dB. Subsequently, I then fiddled around with the set-up and achieved around -148dB.

Well, on one of my lazy afternoons, I thought I might as well give this another try. The aim - at least THD of -150dB. So I packed up some parts I needed and moved it into the quietest room (electrically and acoustically) in my house. Bits and pieces laid out on my ottoman where I place drinks on late night listening sessions. ;-)

Clearly this would have to be run off battery power to remove any mains hum. I went with my trusty RME ADI-2 Pro FS ADC, powered by a 12V/8300mAh TalentCell lithium battery pack for convenience in case I needed to fool around with input voltage settings. The E1DA Cosmos APU was powered with a 5V battery pack.

Of the components above, the only one that's new is the DC-to-DC Step Up Converter (US$10) which is able to take in 4.5-28V and convert up/down to 5-35V; notice on the display if you zoom in that I pushed it a little higher to 35.5V using the 12V input from the 3000mAh lithium battery. The converter is based on the XLSEMI XL6009E1 Buck-Boost Converter; a convenient little voltage regulator for DIY projects (maximum 4A current, recommended to put a heatsink on the chip if >2A, the Victor's Oscillator only needs a paltry 30mA or so).

DC-DC Converter.

A little trial and error was needed to find the optimal output voltage to feed the Victor's Oscillator (which is how I settled on 35.5V). And then some more trial-and-error to figure out the best output level for best THD. It ended up being 1.90Vrms:

I found that above 1.90Vrms, the 3rd harmonic started to creep up.

Here's what the final optimal THD / THD+N FFT looked like:

Beautiful! -150dB THD achieved, with -154dB or lower individual H2-H9 levels. -121dB THD+N with the RCA unbalanced output is phenomenal. No 60Hz hum since we're fully battery powered in a room away from running electrical appliances and such.

Notice that I used my standard FFT settings: 128k FFT, Blackman-Harris 7, 8 average. ADC running at 96kHz. No need pushing to 1M/2M-point FFT. ;-)

So, here again is the previous comparison I made with the Audio Precision APx555 B-Series:

Notice that the APx555 looks like it had a small 50/60Hz low-level hum far left, just below -140dB.

And another look at the new measurement with -150dB THD, linear scale for comparison with above:

After the test, I hooked the Victor's Oscillator to my main sound system for a listen. How does it sound? Well boys and girls, this is one "smooth" and "natural" 1kHz tone. Pure tone of the gods, friends. Having experienced this -150dB THD level of audiophilic nirvana, I can check one item off the bucket list. ;-)

Anyhow... I suspect with some further work like putting the unit in a metal enclosure, maybe shortening (currently 2' RCA-phono cable) and improving the cable from the oscillator to the Cosmos APU, I might be able to get the noise down even further. Maybe running off some alkaline batteries (MG suggested that Victor tests with 4 x 9V batteries in series) instead of the lithium battery + DC-DC converter system will get even closer to the -155dB THD level Victor speaks of!

No urgency of course. Unlikely there will ever be an actual audio DAC achieving -150dB THD at 1kHz with RCA output around 2Vrms much less! Furthermore, we do not "need" -150dB THD in consumer audio playback regardless of how much of a "high fidelity" fanatic we might want to be. No actual audio system (amplifier + transducers) in the foreseeable future would be able to play audio with such fidelity. 

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

Not content with excellent, freely-available tools like DeltaWave (comparator for input signals), Distort (adding distortions to music/signals), and even Earful (hearing test app), Paul K. has expanded his tools to include Multitone Analyzer (MA) to make home measurements even easier using an all-in-one signal-generator / analyzer app.

The use of this program is quite straight forward. You point out the output/input loopback device(s), it creates a signal (based on selection from the various "Test Signal" dropdown list), plays it to the output while recorded on the input, then analyzes the loopback data to determine accuracy including characteristics like Total Distortion + Noise (TD+N), Signal-to-Noise Ratio (SNR), Effective Number Of Bits (ENOB), IMD amount, delay through the system (latency), harmonic levels, etc. Measurable values depending on which kind of test selected.

To start, let me show you this picture with a few settings I'm using for reference:

Notice that I'm using ASIO4All in the settings with options menu pointing output to the Topping D10 Balanced, input is the E1DA Cosmos ADC. Notice the red arrow pointing to the fact that I maximized my buffer size to 2048. I'm using an older Intel NUC i5 computer and sometimes the system speed might not be fast enough so I have seen errors during the playback+recording process; doing this avoided problems for me. You can also increase process priority, and make sure you're not running other apps in the background like a web browser. As usual, I typically set audio devices as 24 or 32-bit, 96kHz in Windows settings.

I pointed an orange arrow to the "L+R" box because I was confused for awhile why the program kept crashing. Turned out I had set this box to "L" only. Even if measuring just 1 channel, keep this "L+R".

The green/aqua arrow is pointing to the "Harmonics" setting which defaulted to "10" but I brought down to "9" to be consistent with most measurements done. No reason why not to use 10 of course other than convention.

Finally the purple and gray arrows are pointing at the "Multitone" min/max frequencies which defaults to 20Hz/20kHz. This is usually fine but since I was about to make a frequency response measurement for 96kHz and I wanted the full spectrum, let's look down into 10Hz and all the way up to 48kHz bandwidth. This program uses a multitone to derive frequency response so set these numbers to the full range if you want wider than 20/20k default.

Here's what the frequency response looks like using a detailed "Multitone 1000" - first with the "Spectrum" tab then with the derived "Freq Response" tab:

I'm using a linear phase, sharp digital filter firmware for the D10 Balanced; not the default minimum phase.

The detailed frequency response looks great - as expected.

Using complex multitones, we can "stress test" DACs to make sure very complex signals are clean. Often we see 32-tone tests and I've used the "1/10 decade Multitone 32". Well, here's what the "Multitone 200" looks like with MA and the Topping D10 Balanced:

From top to bottom, we have a range greater than 100dB distortion-free. We can see the TD+N value of -103.5dB in this example for an overall measurement. Notice BTW that I had set the measurement range to 10Hz - 24kHz (you can see this represented in the white color). Typically we set this to 20-20k so just be mindful when comparing measurements to make sure it's apples-to-apples; in some cases like this the result would not be much different for TD+N whether 10-24k or 20-20k since the difference is just very low level noise.

Intermodulation Distortion (IMD) is good to look at to see how the system performs beyond a single-tone THD+N/SINAD, easily done in MA - here's the SMPTE 60/7k 4:1:

Notice also the opportunity to do level sweep measurements - here's an IMD level sweep using DIN 250/8k 4:1, -80dBFS to 0dBFS, by 1dB increments:

Nice to see the TD+N plotted along side. We can look for any unusual "humps" in the graph to see whether there are egregious DAC output anomalies at various output levels.

Finally, let's just end off with the 1kHz THD+N graphs and compare with what I usually do here on the blog (Room EQ Wizard's RTA function). Note that this is without the aid of the E1DA Cosmos APU:

I showed a couple of MA measurements above - 8-average and 1-average with the 0dBFS 1kHz tone. Notice that the T(H)D+N from the "8 averages" graph (-121.6dB) is significantly better than the single-capture no average one (-116.3dB). This is because MA uses time domain "coherent averaging" where a repeated tone will average to the true level but uncorrelated noise will average down to zero. For example, we can see that the THD (low level but consistent harmonics) remained similar (<1dB difference) while the noise level gets averaged down resulting in the SNR improving significantly. In fact, any measurement that includes noise level like the TD+N likewise will show a greater change using coherent averaging.

Here's my typical measurement using REW, 8 average but without coherent averaging turned on (as per the highlighted option):

As you can see, allowing for small inter-test deviation, the REW result is basically the same as the "1 average" MA one.

It's good to keep in mind that the coherent averaging function is another one of these parameters we need to pay attention to (and one of many potential reasons "Why you can't trust audio measurements", right? ;-). For those of us who run measurements, this effect should be obvious and accounted for when publishing data... Be mindful that companies could publish results that seem to show extremely good THD+N performance using this technique without mentioning to consumers; I have not seen this thus far.

There are other tests provided by Multitone Analyzer like the J-Test. Here's an example of the 24-bit 48kHz J-Test using 1M-point FFT, linear scale, focusing into my usual 5k-18kHz region where sidebands typically congregate around the 12kHz tone:

Jitter test for the Topping D10 Balanced (linear sharp digital filter firmware). Notice how low jitter is with sidebands down at -150dB. As usual, jitter is not an issue at all with even inexpensive modern USB DACs.

You can even customize your own test signals (see the documentation). Great stuff!

Again, amazing work, Paul! Much appreciate all the tools you're providing for audiophiles, in the process demystifying hardware performance, and I think over time growing more comfortable with the importance of measurements. Correlating the objective performance and subjective experience of one's own listening I think will go a long way in helping dispel fear, uncertainty, and doubt; psychological elements that IMO advertising departments and self-promoters play with, as much as creating hype.

Paul: Would love to see a way to turn off the coherent averaging (just to clean up the look of the noise floor) and also would be nice to get FlexASIO working as an alternate to ASIO4All if possible!

As for REW, how about adding SMPTE/DIN IMD to the Stepped Sine/Step level function, not just single tone, John? ;-)

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

I probably missed a few other expensive "audiophile computers / servers" here.

To end, let's talk about what I think has a good chance of being at or near the pinnacle of price as a class of products. As an Audiophile Computer Server, the Wadax Atlantis Reference Music Server (see recent TAS review here) hailing from Spain carries one heck of a ballsy price tag!

At a hernia-inducing weight of 106lbs, asking price starting at US$59,000 (up to US$76.5k with the optical cable and module for their DAC) for a relatively small 2TB M.2 SSD storage system, these people want you to spend twice as much as the previously discussed Taiko SGM Extreme! While the Taiko at least looks pretty good to me with its clean design, I find the Wadax to be rather ugly. Obviously this is an opinion, and I guess the outlines and angles in the industrial design could go well with Wilson speakers if you like that kind of look.

The Wadax Server is relatively inexpensive compared to the Wadax Atlantis Reference DAC, which costs an astronomical US$145,000 (touted to have "5500 discrete parts" - is that supposed to be good for sound quality?). I'd love to know what the THD+N/SINAD is of that baby. ;-)

Anyhow, like with other expensive Audiophile Computer Servers, here's their "attribution theory" in the marketing to justify having done something useful for the (utilitarian) purpose of enhancing sound quality (emphasis mine):

"But from an audiophile perspective, there are a number of issues with current bit-perfect protocols. In theory, a buffered input at the receiver (DAC) end of the transfer prevents cable-induced distortion, but this overlooks two important factors. In any buffered system, the system itself has a transfer function that is constant and thus effectively invisible. But more importantly, digital transfer actually occurs in the analogue domain! The USB encoder has to convert the digital data to voltage square-waves in order to transmit it down the cable, reconverting to the digital domain at the far end. The problem is that producing perfect voltage square-waves is far from simple, even with a purpose built, high-end USB encoder.

Once again – in theory – this shouldn’t matter, as the buffered input clocks and reconstitutes the data. The problem is that this process effectively ‘counts’ data blocks rather than assessing their shape, so that distortion and displacement in the waveform, induced in the analogue domain, remains incorporated in the reconverted digital signal, invisible to the digital encoder. That doesn’t matter in many real-world computing cases, where the raw data is essentially simple binary. But in the case of audio recordings, that data is incredibly time, amplitude and phase sensitive, with small errors rapidly eroding the integrity of the whole.

Once we researched the problem and realised its implications, a whole series of related issues fell into focus. In particular, it explained why different USB cables sound so different – and why different lengths of the same cable also sound different."

As usual, there's nothing in this marketing material to suggest the expensive box does anything useful nor is there evidence that sound even changes based on the above rationale. There's an unusual section in the text talking about "counting data blocks rather than assessing their shape". At the end of the day, digital transmission is basically the accurate waveform representation of "bits" and "counting" of those waveforms to reconstruct the data. So long as the squarish waveforms are good enough to represent the data with wide enough tolerance from errors, what does it matter their exact "shape"? Digital audio data transmission is no different from "real-world... simple binary". Furthermore, if there are indeed "small errors rapidly eroding the integrity of the whole", we'd be experiencing errors in the USB transmission and the errors will be obviously non-bit-perfect with audible degradation (as per this demo years ago).

Interestingly, they're making a big deal with the "three rotary knobs that allow users to adjust/compensate for errors in the rise-time and amplitude of the sent signal as well as the spacing on the return channel (counteracting echoes and reflections)". They call this the "Digital Waveform Control" and touts "more than 200 parts mounted on 12 proprietary circuit boards".

Again, is this complexity supposed to be impressive? I don't understand why they didn't just select the "best" rise-time and signal amplitude settings to begin with to reduce risk of errors as per formal USB (presumably USB 2.0) spec or their optical output. Are we to believe that reducing rise-time or dropping signal amplitudes (input and output) somehow can improve digital accuracy? This is weird and I'd love to see measurements of DAC output changing based on those adjustments! In digital, either we drop the signal below a safe threshold (thus creating potential for bit errors), or we don't; there's no need for some kind of variable knob to seek for an "optimal" setting. This kind of thinking is just fooling around with old-skool audiophiles who still engage in analogue-think when it comes to digital; which seems to be exactly what these guys want you to do (echoes of old John Swenson talk - I don't think he has ever produced any evidence over the years for his claims).

I'm quite sure digital audio end-users would not want to play around with knobs that could lead to errors or worsen jitter if set too high or too low! This is literally what's being implied by the controls. Are there any other manufacturers claiming such a thing?

As expected, the reviewer Robert Harley hears all kinds of things. Apparently "The optical interface seemed to provide an octave of deeper extension, with greater solidity and power below 40Hz." Wow, that's really quite an awful USB interface inside there to filter those <40Hz frequencies! Furthermore, of course Harley feels that there's benefit in those knobs since his (unspoken) job appears to be selling this nonsensical product to you [nothing new]:

"The Speed control subjectively seemed to affect the timing of the bass guitar and kick drum; when ideally set, the two instruments combined synergistically, almost as though they were one pulse driving the rhythm. Moving the control out of its optimum setting (which you set by ear) seems to make the bass lag, with a less precise sound and diluted propulsive drive. Of course, the adjustment to the waveshape has no objective effect on the timing of the bass, but that is the subjective impression rendered by the Speed control."

So fellow audiophiles, "In Robert's ears we trust!" or are we reading bovine excrement? Do we believe that there has been "Qualitative advance in sound quality that is unlike the improvements I’ve heard in a single product" or is this even more serpent lipid? I'll leave it to you... I'm just the simple audiophile who says "Bits Are Bits" unless Wadax can demonstrate otherwise to show us the benefits of their science.

While it's impossible to know with luxury products, I think there's a good chance that we're seeing the pinnacle of expensive Audiophile Computer Servers for this economic cycle for various reasons including gross lack of value as a product class. Nobody has a crystal ball (that works 100%) and it doesn't take too much for any company to put up a big MSRP sticker, make up stories of why it's worth it, and put up a fancy web site. But I think it will be hard to to top the shameless audacity of this Wadax Atlantis Reference Server computer.

BTW, it's thoughtful of Wadax to include a height-adjustable USB cable support when you need to install your "phat" and heavy USB cable. Furthermore, check out the girth of that patterned yellow/gold AKASA optical garden hose / anaconda Wadax uses to connect to the DAC - hilarious and IMO very ugly!

It's good to see that TAS also reviewed the Topping D90SE DAC just before this Wadax box. Talk about a juxtaposition of no non-sense value high-fidelity gear versus IMO frivolous expense intended to boost the ego of certain individuals who feel that this represents some kind of objectively meaningless "high end".

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

Alright dear audiophiles, hope you're all doing well as we enter June and on our way to summer 2022 (here in northern hemisphere of course)!

I was recently listening to local boy Michael Bublé's latest album Higher; good standard vocal pop as you'd expect. Feel free to de-clip this album from DR6 to DR10. The title track is a nice upbeat tune, and the duet with Willie Nelson on "Crazy", while predictable, is enjoyable. "Higher" DR value to start with for vocal albums like this would have been nice especially for those folks who shelled out for the 24-bit version! As suggested years ago, I think there should be "Advanced Resolution" versions to differentiate standard CD release from "hi-res".

I've also been catching up with Malaysian singer/rapper Namewee's (黄明志) discography lately. Really clever Chinese / South East Asian multi-lingual / dialect stuff there! Some of the songs have been controversial (like this, and this) but freedom-loving adults with a sense of humor should not have a problem with this I think.

Hope you're enjoying the music!

----------

Addendum - May 31, 2022

And POOF! Just like that, the Wizard of REW has done it again with another beta!

Room EQ Wizard 5.20.9 Beta with "IMD vs. Level" step measurements. Check out the various IMD options on the right with also the opportunity to run a custom setting (set in the Generator panel, your choice of frequencies, either 1:1 or 4:1 amplitude ratios).

With the multichannel Pro upgrade, here's what it looks like running the stepped DIN IMD vs. Level measurement in stereo to the E1DA Cosmos ADC:

And we can look at the saved distortion vs. level step graphs after the data capture:

Oh yeah! Easy simultaneous comparison of right & left channels from -130dB in this example. Will need to test a bit more and start utilizing this capability in upcoming reviews.

Wow John. Now that's fast software development! Thank you! :-)