Using Victor's Oscillator as reference (1kHz, ultra-low distortion analogue sine generator). A quick peek at the APx555B Analogue & DAC Signal Generators @1kHz. And on playing music "for fun"!

Note the 2-pin power connector on the lower left.

As you've seen on the blog over the last year, I've been using the E1DA Cosmos ADC from Ivan Khlyupin for a number of measurements recently (early report discussed here). I remain quite impressed by the performance of this little device with some caveats (lower input impedance, individual unit idiosyncrasies including what "grade" device one has).

While the ADC comes pre-calibrated by E1DA and unless you really want to spend some time playing with it, one should probably just leave it alone. Nonetheless, it's pretty clear that this is the kind of device made by and for audio hardware geeks! ;-) One has the opportunity to use the Cosmos_Tweak Windows program to fine-tune the calibration by changing independent ADC 2nd and 3rd harmonic compensation amounts, SDM integrator setting, and digital filter for the ESS Tech ES9822 chip.

So then, for this post, let's spend some time in the area of calibration and see if we can tweak the ADC using the device above - Victor's Oscillator (US$57) - an ultra-low distortion, 1kHz analogue signal generator which comes pre-assembled and tested from Riga, Latvia. As you can see in the image, there are 2 single-ended RCA outputs, with volume control ranging from 300mVrms (-8.25dBu) to 2.7Vrms (+11.0dBu). Note that a 10kHz version (and other custom versions) can be purchased as well but for typical audio purposes, 1kHz is in line with human hearing sensitivity.

As you might expect, with the Pandemic still very much affecting global business, supply chains and shipping, it took awhile for this little box to arrive from Latvia:

It took the post man about 1 month between shipping and arrival at my doorstep here in Western Canada. Still not too bad but a bit longer than the estimate.

Here's a peek at the bottom of the board:

Based on the TI/BB OPA1656 opamp which you can see mid-board.

Over the years, Victor Mickevičs, the designer of this little board has updated the performance and specs. The one I have here is Version 5. Be mindful of this because earlier revisions (here's a circuit diagram of an older version back in 2017) had high output impedance, I've seen up to 600Ω. This high impedance would clearly be a poor match for the Cosmos ADC. For a 2.7V input setting, input impedance is 1kΩ balanced (~500Ω unbalanced). Thankfully, with the current Victor's Oscillator version 5, the output impedance is down to 100Ω which is a more reasonable match although absolute distortion will still be a little compromised. So long as the compromise isn't likely to intrude into the limits of human audibility, we should be alright.

Okay then, here's the Victor's Oscillator running on my test bench connected up to a ~35V/25mA power source and feeding the E1DA Cosmos ADC:

For best performance, the Victor's Oscillator should be housed in a proper enclosure which I'll do later in the year when I have some more time. Even without a shielded enclosure, performance is still excellent. Notice that I'm powering the board with an inexpensive variable DC power supply on the left (I'm using something like this 60V/5A bench power supply, <US$120).

As mentioned above, the board comes pre-assembled and tested to have extremely low distortions - 2nd and 3rd harmonics typically down at -155dB and below! This is excellent and even though impedance match isn't great, this will still be better than what's needed for human hearing. [The specs page for the Oscillator lists harmonics rise by +15dB into a 600Ω load which is still -140dB - this would be the estimate using the Cosmos ADC unbalanced at 2.7V setting.]

Okay then, let's do a little bit of optimization with Cosmos_Tweak. The input voltage setting on the Cosmos ADC is set to 2.7V. We'll use a simple XLR-to-RCA adaptor as shown in the picture above and generic 3' RCA cable.

Here are my Cosmos ADC's default parameters and what the 1kHz THD(+N) FFT (128k-points, 32/96 setting on ADC) looks like with the Victor's Oscillator before any fine tuning:

And with both channels connected, a reading in summed Mono mode:

Certainly very good as is and I think this is already fine for measurements. We can see a small amount of 60Hz mains hum but this is down at -130dB. Both the 2nd and 3rd harmonics are already below the 60Hz hum. Notice that my right ADC channel has slightly better noise and THD reading than the left with the default parameters. Remember, we're looking at remarkably low signal levels here using a reasonably priced hobbyist device; idiosyncrasies are to be expected.

Now then, in realtime, we can play with the parameters to eke out the best combination of settings to find a balance between how much 2nd and 3rd harmonics are present while optimizing the THD+N which includes the noise level.

The ES9822 ADC chip actually is a 4-"unit" ADC with ADC1+3 bonded for the left channel and ADC2+4 for the right. Presumably, internally it's very similar to the actual 4-channel part, the ES9842 PRO. A few tips:

1.  Play with the ADC_INT_SEL realizing the paired settings will affect the same output. This parameter changes the integrator options for the sigma-delta modulator and the effect can be relatively unpredictable. The default is 11 for all channels. I believe for most devices, keeping it at 11 is best although a friend was able to achieve better results with another value (I think 3 is a common one). Feel free to try different combinations. For me, deviating away from 11 either increased the noise floor slightly or even very significantly!

2.  Once settled on the ADC_INT_SEL, methodically optimize the 2nd and 3rd harmonic settings. There will be some interactions between them. I find the results somewhat more predictable than ADC_INT_SEL on my unit here.  For example, I see a range of values that would work quite well and on either side of the range, overshooting too high and too low increases harmonic distortion. As you can see below, my final calibration numbers for the 3rd harmonic went from -18/-18 default to -14/-15 optimized while I pushed the 2nd harmonic compensation value a little lower. Pushing the number too low will not necessarily improve the result.

3. Feel free to play with the FILTER setting as well for each of the 4 internal ADCs (again, 1+3 = left channel, 2+4 = right channel). For my unit, the default was "Linear Phase Apo". For consistency, I like to keep everything at "Linear Phase Fast" since that's what I prefer with most DAC playback (unless I want to get more fancy). Here too with the ADC, this is a predictable filter with generally flat frequency response out to Nyquist (apodizing will filter frequencies out a bit earlier) and maintain flat phase response unlike minimum phase settings.

And here are my final calibration and results:

Notice that the right channel still has a lower noise level, but the harmonic distortion amounts are now equivalent with improved left channel performance.

A THD+N of -119dB using single-ended RCA output with the unshielded Victor's Oscillator sitting on my work bench, powered by an inexpensive power supply I think is pretty impressive. The highest level 3rd harmonic is still down at around -135dB which is certainly way beyond "good enough" for measuring hi-res DACs, well past any human auditory limits I have ever seen, and dynamic range for sound rooms here on planet Earth.

Notice if you're buying one of these E1DA ADCs, they're offered as different "grades" depending on the performance level. This is determined by the SNR in MONO mode, measured A-weighted.

Grade A: 128-129dB(A)

Grade B: 126-128dB(A)

Grade C: 125-126dB(A)

Let's have a look at this for my pre-production unit. Disconnecting any cables attached to the ADC, here's the "silence" FFT (1.7V input sensitivity, 32/96, 128k-points, 8 average):

As you can see, we have a measurement of 127.9dB(A) which is just shy of a formal "Grade A". In stereo mode, the channels are about -3dB, ranging from 124.5 to 125.5dB(A) right and left respectively. Good enough for decent measurements, I think. ;-)

Before leaving this, let's do one more experiment...

Ivan of E1DA tells me that the Cosmos ADC performs best and is calibrated at 27°C. He monitors the temperature when calibrating with his "thermostat":

For me in the winter in the basement, the ambient temperature is down around 19.5-20.5°C. Let's see the effect of temperature as I warm up the ADC towards that optimal 27°C. This is a job for a hair dryer! ;-)

Indeed, I can measure an improvement in THD, especially lower 3rd harmonic as temperature goes up. I did not bother opening up the Cosmos ADC but I estimated that as it increased above 25°C, there was an observable improvement:

If the hair dryer warms up the device too much, noise starts creeping up. Anyhow, while there is a difference depending on how warm your ADC is, I'm not seeing a massive range and nobody's going to hear a difference between -131.6dB and -133.5dB THD I think.

Realize that all these measurements are without any 1kHz notching. Much of the distortions we're seeing here are the results of the ADC being fed the high level 1kHz tone. On pro measurement tools like the Audio Precision APx555B, notching and quantification of the residual distortion is incorporated in the sine analysis mode. The way forward then is to use a filter like the Cosmos APU which hopefully will be forthcoming not long from now. With that, the ability to measure even lower THD will be possible.

[Another interesting product for the Cosmos ADC is the Cosmos Scaler which Ivan has been discussing recently. This will act as a pre-amp front end for the ADC, increasing the input impedance to 20kΩ balanced, and even incorporating auto-ranging capabilities. Assuming minimal effect on noise level and good price, this looks like a great tool!] 

While this post is about fine-tuning and obsessing over small differences, let's make sure to maintain the perspective of being "reasonable objectivists" as well! Already I believe I'm measuring way below human hearing thresholds. Does it really matter if the measurement gear can produce results down to around -120dB THD+N as a hobbyist (from RCA, not even balanced!)? No, of course not. Objective performance as it pertains to audio hits the end of its practical utility when nobody's going to hear a difference due to the limits of perceptual ability.

Multiple lines of evidence have already pointed to the awareness that distortions can be quite high before we notice anomalies. Whether it's listening tests done on this blog on THD audibility (using actual music) or studies like this one from a number of years back reported on Axiom Audio (with "distortion" frequency tones played over music), human hearing isn't as amazing as some "Golden Ears" might want to suggest.

In comparison to objective limitations, adequacy based on subjective opinions are more difficult to define, if not impossible due to the boundless limits of imagination and other psychological characteristics as humans.

Anyhow, for those who have one, have fun tweaking the Cosmos ADC. The Victor's Oscillator is an inexpensive, ultra low distortion 1kHz reference source that can get the job done if you want to go the extra mile!

[For completeness, you'll also find other ultra-low distortion signal generators out there. For example, here's the JanasCard design by Vojtěch Janásek. Looks like the latest version of this also uses the OPA1656 op-amp.]

Addendum: May 2022
For more measurements of the Victor's Oscillator with 1kHz notch filter for improved accuracy, see here.

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

Here's how the Victor's Oscillator measured with the RME ADI-2 Pro FS which I still use regularly on my test bench for its many convenient features and rock-solid driver performance. Furthermore, the RME has an excellent analogue input impedance of 18kΩ balanced and 9kΩ RCA.

Input sensitivity is set at +13dBu (3.5V) for these measurements with the Victor's Oscilloscope output at 2.7V:

Lovely right-left channel equivalence. Notice that my RME has slightly better left channel results.

Using the same Victor's Oscillator reference source, we can see the relative differences and limits between the RME and Cosmos ADC. Again, a clean 1kHz notch would certainly be beneficial to obtain a more precise harmonic residue reading.

While the RME ADC (without a notch filter) may not be able to grab results down to numbers like -120dB THD+N, realistically, this level of performance is still excellent and IMO able to measure beyond the needs for human hearing.

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

As a bit of a follow-up to the post back in March 2020 showing the innards of the Audio Precision APx555 B-Series, some readers wondered about the difference between the analogue sine generator and the DAC output from that machine.

I asked my friend if he could do a 1kHz loop back for us to have a peek. In the context of this post with the Victor's Oscillator, let's see the generator results from the AP:

As you can see, the DAC (ie. digital generator) is not as "clean" and we can see the harmonics (primarily 2nd and 3rd) up around -115 to -120dB. Best 1kHz THD+N is around -113dB for the DAC and a phenomenal -123dB for the analogue "high performance sine generator".

This can be represented on a THD+N vs. Generator level graph:

Fantastic right & left channel precision.

My friend noted that although not as high resolution, the DAC output is well tuned to have stability across a large range of output levels so it's predictable as a reference. Also, because the analogue sine generator has a wider "skirt" at the base of the fundamental frequency, the DAC output is better for "phase noise" measurements. The other obvious benefit with the DAC generator is the flexibility to create more complex and non-sine signals.

The implication of the DAC's curve flattening out by around 0dBu is that THD at best is around that -112 to -113dB level.

Notice the analogue sine generator's best THD+N performance occurs at various output levels depending on the "autoranging" circuitry on the input side. For example, +16dBu (or around 5Vrms balanced) is a good level for the APx555-B to measure at - a "sweet spot" if you will. This would be a voltage level that guys who measure with the AP might commonly want to target to achieve cleanest results.

So, overall, yeah, I'm guessing that music playback on the AP's DAC should sound good (THD+N -113dB certainly very respectable), but only if you can get past the fan noise. Note that these days something like the <$US150 Topping D10 Balanced DAC will provide a similar or better THD+N performance. In fact, I'd like to thank John Yang at Topping for providing me with linear phase filter firmwares for the D10B and D10s so I can use them as signal generators for testing with flat phase response.

Big thanks to my friend for the data with his APx555-B. Maybe another time if he opens up the machine again, he can poke around and let us know which DAC chip the AP uses. :-)

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

To end, I cannot help but point everyone to this interesting forum post on Steve Hoffman Music Forums: "Do you still play fun (NOT audiophile) music on your high end stereo?"

What a "fun" question ;-).

Of course a question like this can only be answered totally subjectively with no right/wrong judgment. Nevertheless, in my heart, I certainly wish that everyone is able to answer "Yes!" to such a question as audiophiles!

For me, my "first love" was not the audio hardware. Rather, achieving "hi-fi" playback developed secondarily as a means for the intent of playing back the music I enjoyed starting in my teenage years which happens to be mostly '80s rock and pop. To this day, playback in my sound room will regularly revisit those "old" tunes including the performances of Belinda Carlisle, Rick Astley, Donna Summer, Samantha Fox, Whitney Houston, Michael Jackson, 38 Special, Icehouse, Alison Moyet, Duran Duran, Richard Marx, Tears for Fears, OMD, UB40, George Michael, Rod Stewart, Madonna, Alphaville, Phil Collins, Pet Shop Boys, Gloria Estefan, Cheap Trick, Aerosmith, Depeche Mode, Steve Winwood, Corey Hart and the like in no special order...

Of course times have changed and these days the list of "guilty pleasures" include many younger names. Mix this up with some world music and Asian pop (my daughter likes BTS and my wife enjoys tunes from various Korean and Chinese dramas as well). In my home, the majority of music played at any one time would not be considered typical "audiophile music" (which I guess can be generalized as classical, light jazz, female vocals, or from music labels catered to supposedly high fidelity recordings/packaging). 

To me, the distinction is not whether a piece of music is 'audiophile' or 'not audiophile'. Rather, it's whether the music was well recorded and sounds good to me - almost wholly subjective value judgement. The only times I feel more compelled to listen to "audiophile music" I guess are those situations where I'm "critically listening" for the purpose of writing reviews. Honestly, these days, I generally don't find "critical listening" (as opposed to "listening for enjoyment") a particularly fun or meaningful activity so will try to keep it to a minimum for when-needed situations.

Usually, for me it's not fun to just sit there and listen with eyes closed in isolation - much better to be with loved ones, friends, and even better, tapping feet and singing along as appropriate.

News this week: Apparently Pope Francis likes his classical and tango music. So here's the question - did he leave this store with a CD box or were those LPs in hand? Is the Holy See on the side of digital or analogue? ;-)

Stay safe and make sure to enjoy the music! Obviously.