In this blog installment, let's look at the "ultra high-end" of sound. Of course, I'm not talking about the audiophile "high-end" marketing term which is meaningless (beyond just another phrase for "expensive"). Rather, let's look at the frequency high-end, especially all the stuff our DACs can produce in the ultrasonic range!
If you've ever wondered, ultrasound devices used in medical imaging typically function at around 2MHz on the low end up to about 15MHz. However, the term "ultrasound" simply refers to wave
compression and rarefaction outside of the hearing range which by convention are those above 20kHz or so.
Remember that over the last few months, I have been using the RME ADI-2 Pro FS ADC as a measurement device. This device is capable of operating at 768kHz although up to now I have rarely used it at the highest rate. So why not push the system up to that limit and examine the spectrum up to the full 384kHz bandwidth to see some ultrasound components? While we're on this topic, let's also see if we can examine the frequencies beyond using an oscilloscope.
I. A look at the Digital Filter Composite (DFC) at 768kHz sampling rate...
II. Filtering the 384kHz samplerate signal...
We can do the same process using the SMSL iDEA:
What we can say about the iDEA is that the frequency response is rolling off already quite a bit by 192kHz. I didn't check if this is because they're using a very weak digital filter. HQPlayer is obviously effective as you can see. Again, whether anyone cares is another matter :-).
III. MHz range ultrasonic noise?!
[As an aside, to those who might have a Rigol DS1000-series scope, there is a Python-based project whereby higher quality FFTs can be derived (software on Github). I'll give it a look when I have more time.]
|On screen: A 100kHz signal off the SMSL iDEA...|
Hmmm, ugly. It looks like the high amplitude, high frequency signal is too much for the little DAC/headphone amp to handle; note the grossly elevated noise level across the spectrum and "skirting" of the primary signal. Furthermore, there's also a strong image of the primary tone out at 234kHz. Knowing this, let's cross-correlate this with the oscilloscope FFT to make sure the two peaks can be seen easily there as well:
Yup. We see the 2 frequencies. Each horizontal division is 200kHz and we can also see the 427kHz image a little further out.
Can this anomaly be fixed with software upsampling like HQPlayer, JRiver, etc? "Yes we can!" - as in Part II above, assuming you think adding a high quality digital filter is worth it for 384kHz samplerate material, it can be done with good results...
In comparison, when we put the same 0dBFS 150kHz (24/384) test signal through the Oppo UDP-205 with standard "linear phase fast" filter setting, we don't see any issues with the noise floor nor is that 234kHz image present. In fact, it looks excellent! No need for HQPlayer here:
We see only a single 150kHz primary peak as expected on the Rigol FFT. There's possibly something out at ~1.3MHz however but clearly at much lower level than the primary signal.
Moving along, let's push this even further to levels that the RME ADC cannot capture. How about a -3dBFS 300kHz 24/768 sine signal? Here's the SMSL iDEA:
We see imaging of the 300kHz primary signal out at 468kHz. This suggests that the iDEA doesn't use a digital filter with 768kHz audio data, or if implemented, a weak filter.
And how about the Oppo UDP-205?
Looks good with just the single 300kHz tone. Nothing terrible all the way to 2.4MHz... Perhaps a bit of noise around 1.2-1.4MHz, but again this does not appear to be of high amplitude.
For "fun", here's a wideband 24/384 white noise signal using both DACs:
Despite the low resolution of the FFT, as expected based on Part II, we see that there's more high frequency roll-off with the iDEA. While the resolution obviously isn't good enough to catch small anomalies in the noise floor, there does not appear to be strong noise all the way up to 2.4MHz. If anything, with the Oppo, we continue to see a "little something" around 1.3MHz.
Before I dismantled the set-up, I thought I'd just throw out one last test. Here's the Oppo using a track from "Archimago's Insane and Unreal Esoteric Tests" collection - a 24/384 signal with 140 & 170kHz -6dBFS sine waves:
Even with these high amplitude and very high frequency signals measured at the analogue output, I see nothing all that scary using the high-resolution RME ADC, nor am I seeing anything of concern with the oscilloscope scanning out to 1.5MHz.
IV. A quick look at the TEAC UD-501 (TI/BB PCM1795 DAC)...For years, since 2013, I used the TEAC UD-501 as my reference DAC. Although I don't use it as much these days, I figure it would be good to put the TI/Burr Brown PCM1795 DAC chip to the test as well using the 768kHz DFC and my oscilloscope.
While there's some intersample overload with that wideband white noise with 0dBFS peaks, this does not look bad although we have some -70dBFS distortion out beyond 330kHz. The DAC chip implements 8X over-sampling of the 44.1kHz signal to 352.8kHz and what we're seeing are images out at 332.8+333.8kHz and 371.8+372.8kHz.
One "nice" feature with the TEAC UD-501 is that we can turn off the digital filter from the DAC menu. This makes the DAC into a "Non-OverSampling" (NOS) device. Here's what the DFC looks like with the TEAC in NOS mode:
Remember this distorted "cascade" when you listen to devices that are NOS at 44.1/48kHz without appropriate upsampling applied! In fact, using the TEAC, it's a much cleaner looking pattern than one would see with old NOS parts like the multibit Philips TDA1543.
If we look at just the 19 & 20kHz sine waves in the oscilloscope with filtering turned off (NOS), while the resolution isn't good enough to see individual frequencies (like the separation between 19 and 20 kHz), we again will see the cascade of ultrasonic distortions:
Isn't that "pretty"? Not for an audiophile who desires the cleanest, highest fidelity signal! :-)
The maximum samplerate for the TEAC is 384kHz. At this high samplerate, there is no digital reconstruction filter applied and one would expect therefore that something like my 140 + 170kHz signal (@ 384kHz as per Part III) will show imaging distortions as well:
Yup. Notice that even with the relatively low resolution of the Rigol FFT, we can see the images out to the 524+554kHz and 598+628kHz pairs. Otherwise, not much else of concern out to 1.5MHz.
V. Ultrasonics and DSD...I suspect in the back of your mind, you must be thinking - "What about DSD!?" After all, what kind of article talking about ultrasonic frequencies for the audiophile neglects DSD, right?
That is indeed an important question which we must not avoid regardless of one's opinion on the value of having music encoded as DSD. The simple fact is that with a 1-bit digitization scheme, quantization error for each sample is large, +/-0.5 for complex signals. This translates to a large amount of noise in the signal which needs to be addressed if we're aiming for clean, high-fidelity playback.
Over the years we've seen this noise manifest every time I measure DSD output, most noticeable with DSD64 and DSD128. In order to improve the noise floor within the audible frequencies, aggressive high order noise-shaping is used to shift the noise to the ultrasonic range and DACs will typically then implement an analogue filter to roll-off high levels from there.
For comparison then, here are the "DFC" graphs of DSD64/2.8MHz, 128/5.6MHz, 256/11.3MHz, and 512/22.6MHz from the Oppo UDP-205 using HQPlayer 3.25 to perform the PCM-to-SDM realtime conversion with an i7-3770K CPU + nVidia GTX 1080 GPU using "poly-sinc-xtr-2s" resampling (the "poly-sinc-xtr" setting is extremely CPU/GPU hungry especially at DSD512). I chose to use the "DSD7" 7th order SDM setting as a demonstration of the kind of aggressive noise shaping often used to improve dynamic range below 20kHz.
There are a couple of unexpected noise spikes in the "Digital silence" plot at DSD64. Not sure where that's coming from. Otherwise, nothing unexpected here.
Notice the ultrasonic noise pattern - how quickly and how high it "takes off" beyond 25kHz with DSD64 especially!
As you can see, noise is not as severe with higher samplerate DSD and is similar to PCM by the time we hit DSD512. Remember what I've said many times before... The higher the DSD samplerate, the more the objective characteristics look like PCM; in the case of these DFC diagrams, the true noise floor is below the ADC's ultrasonic modulator noise level by the time we hit DSD512. If you like the sound of DSD, it's quite possible that the noise is experienced as euphonic/preferable, hence the high noise DSD64/128 patterns could sound "better", with "more air", appear "more three-dimensional", "more analogue", etc. with some sound systems.
VI. In Conclusion...With the <US$100 SMSL iDEA, the >US$1000 Oppo UDP-205, both using ESS DACs, and adding the TEAC UD-501 from 2013 with its TI/Burr Brown DAC in the mix, I think we can say based on the data from the RME ADC with 384kHz bandwidth that there's really not much in the "high end" at all during PCM playback when using appropriate digital filters. Sure, likely there's ultrasonic stuff below the limits of the RME's AKM AK5574 ADC modulator noise floor, but we'd be looking for signals below about -85dBFS at frequencies significantly above 100kHz!
Furthermore, using the oscilloscope's FFT function, though the results are of much lower resolution, we are not seeing large amounts of ultrasonic noise in the MHz range.
In contrast, we can certainly see significant amounts of ultrasonic content when we turn off the digital filtering (ie. NOS DACs) and with DSD playback, especially DSD64 and DSD128. So, for audiophiles, if the extremely high ultrasonic imaging found with NOS playback and persistent DSD64 ultrasonic noise are not considered generally objectionable, that's in fact evidence that ultrasonic content is not a big deal and do not severely affect sound quality.
By the way, remember that the SMSL iDEA is a tiny, USB-powered DAC. In these tests, I'm connecting the DAC to my Intel i7-3770K desktop CPU currently overclocked to 4.0GHz. Inside that machine is also the nVidia GTX1080 GPU, 16GB of DDR3 RAM, an SSD and Firecuda hard drive, connected to the home ethernet network with no special power conditioning. Notice that despite what should be a "noisy" computer system, that neither the 768kHz "DFC" nor the oscilloscope readings show terrible noise.
As discussed in late 2018, I'm far from impressed by audiophiles who seem to believe that there's all kinds of noise, jitter and whatnot affecting the sound output from otherwise competent DACs connected to computers through USB. Anxiety over things that appear to have no basis in reality amounts to perpetuation of unsubstantiated fear, uncertainty, and doubt. I hope magazines and on-line sources seek to find balance and evidence when these kinds of ideas are presented.
Inevitably, we must ask ourselves then, is there evidence here of anything to worry about when it comes to ultrasonic noise? Is there any ultrasonic noise that likely will be affecting the "sound" of our DACs? From what I've found and what I hear, I have no concerns about the SMSL iDEA, Oppo UDP-205, or TEAC UD-501.
As for software upsampling (eg. HQPlayer, Roon, JRiver...), yes, they work and can improve DAC output where the built-in digital filter may not be great. Software upsampling is also particularly useful for NOS DACs (as discussed before). Remember this the next time you hear an audiophile claim that their Audio Note "1x oversampling" DAC somehow sounds remarkable. I believe this says more about the individual's preferences and other sound system components than actual high fidelity performance from the DAC itself. Remember though that if you have a good DAC already with good filtering (like the Oppo with its "linear phase fast" setting), do not expect remarkable sonic revelations using software upsampling. At best, differences will be subtle.
Historically, there was a time when audiophiles and magazines hyped up the importance of ultrasonic frequency reproduction. Remember back in the late 1990's and early 2000's with the advent of SACD and DVD-A, the idea of "super tweeters" was in fashion? Back then, I remember reading magazine articles tout the ability of these ultrasonic transducers for improving soundstage and "air" (I discussed this back in 2017). Consider the possibility that these "super tweeters" could actually make sound quality worse if not implemented properly by adding yet more drivers to the speaker design.
These days, the need for "super tweeters" and claims that extended high-frequency response being beneficial are quite rare. In fact, if we look at the in-room frequency response in speaker measurements, typically there's roll-off into 20kHz ("house curves" with high frequency roll-off are often subjectively preferred). Likewise, headphones, including "hi-fi sounding" devices like the Sennheiser HD-800 typically have more than -10dB dips by 20kHz compared to the amplitude at 1kHz. Furthermore, little recorded music actually contain large amounts of >20kHz content (remember, studio microphones also have frequency limitations). This, combined with the fact that most of us adults likely at best can perceive frequencies up to 16-18kHz argues strongly against needing to pay much attention to ultrasonic content. As I noted above, if audiophiles aren't complaining about the ultrasonic distortion/noise of NOS DACs and DSD64, what's there to worry about? Therefore, technically, likely the only time we need to be concerned is with unstable amplifiers unable to handle excessive high frequency content.
To end off... These are probably some of the most insane/extreme measurements I've published in this blog. However, I'm sure I'm not the only person curious about the presence of ultrasonic frequencies. Remember, as per the post a couple weeks back, test tones like these are obviously not like "real music", but they certainly can at least give us some idea of the technical limits of equipment and whether the gear still performs in an expected fashion.
In any event, while I'm sure we will continue to discuss and measure using 384 and 768kHz sample rates, I hope this is the last time we need to discuss frequencies in the MHz range from audio DACs. :-)
Despite the numerous complaints against audiophile magazines/press and the false ideas they may perpetuate, I do appreciate John Atkinson's Stereophile in maintaining some objective perspective for how reviews should be presented incorporating both artistic sensibilities and scientific evaluation. This is far and away superior to many alternatives out there both in print and online (no need to name names - rather obvious). Congrats to Mr. Atkinson on the years of work and the legacy this will certainly leave.
It's good to hear that "major editorial changes are not envisioned". Hopefully this implies that Stereophile can maintain the balance between subjective and objective perspectives in their reviews. In my opinion, the objective measurements in the magazine are the portions that endure and make articles worthy of repeated reading over the years. The objective results speak of true performance. Stereophile has documented the advancement of technology over the decades, and helped educate hobbyists seeking to understand the pursuit of fidelity as practiced over the generations. This is the stuff that transcends ephemeral, subjectively idiosyncratic, toss-away fluff that fades like the fads of this world or even just the next news/product cycle.
At a time in history where the boundary between fact and fiction has been blurred sometimes beyond recognition, where it seems "my truth" has frequently overtaken "the truth" even in common sense, rather unequivocal matters, it would be rather unfortunate at least for North American audiophiles to lose the perspective Stereophile has brought to our discussions.
Hopefully, as Mr. Austin takes over the helm of the publication, he will not forget the science of his earlier years (talked about in the announcement article linked above). In the words of the inimitable physicist Richard P. Feynman:
"I would rather have questions that can't be answered than answers that can't be questioned."
"Religion is a culture of faith; science is a culture of doubt."I'm certainly not calling for the magazine to turn into some kind of serious scientific journal; rather, let's talk about fun hardware and great music while maintaining a decent level of reality-testing and willingness to examine unconventional beliefs using scientific principles.
Here's to hopefully a helpful, truthful, honest, intelligent, reasonable, educational while also creative, entertaining, and ultimately successful tenure for Mr. Austin. Here's to wishing that Stereophile remains a publication of significance in building the culture of the hobbyists it seeks to serve.
Until next time, I hope you're all enjoying the music!