Upsampling is exciting and significant across many digital disciplines, not just music, such as images, videos, and video games (for example, Nvidia DLSS uses AI to generate more frames).
The idea of upsampling is grounded in mathematics. You have a certain number of points and use a function to calculate additional points. As this can work well in images, videos, and gaming, does it work well in music?
Over the past years, I have used many upsampling software and, lately, the built-in one in Audirvana. To my surprise, as upsampling is hard to do without coloring the sound, Audirvan upsampling filters (PCM and DSD) are very good.
This post is to drive a conversation on this subject. Please share your thoughts. And if you believe music upsampling is of benefit, how and why?
Based on your feedback, I will share additional insights I learned with you. Please consider the harmonics built into the music. I do not mean harmonic distortion but the harmonics, which are a part of each music note when recorded, and what happens to harmonics when we upsample. I am keen to dig deeper into this.
It’s all good to try stuff, and I do, but my conclusion is to keep playback as close to original recording as I can. That way, I learn to trust particular labels/recording engineers etc. for my future purchases.
It’s all a matter of taste.
By-the-way, I think I have the best taste of anyone I know.
If your DAC provides an un-fettered 1-bit PDM (DSD) signal path to a simple low-pass filter output circuit, up-sampling/modulating all PCM to DSDxxx will present a more natural ‘analog like’ signal, due to the nature of 1-bit PDM (DSD) encoding…
Up-sampling does not create any new harmonic information, and nothing beyond the Nyquist Fs of the original PCMxxx encoding (unless using something like RDOT fluency logic interpolation used in TEAC and Esoteric DACs) [see below] … However the increase in bit-depth from 16 bits to 24 or 32 bits reveals low-level information that is hidden by the noise of the 16 bit playback that is actually encoded in the original ADC. Another advantage of up-sampling to higher PCM sample-rates and modulation to DSDxxx, is the increase in sample-rate presents a digital signal more compatible with the output circuit of the DAC… PCM-centric DACs employ Sigma Delta Modulators to up-sample the PCM signal to a higher sample-rate, compatible with its output circuit design (typically multi-bit).
RDOT, which applies an analogous interpolation method using fluency logic, is a technology that was developed to enable reproduction and playback of frequencies higher than 20 kHz that are lost by 44.1kHz/48kHz digital signals. Based on the information read, analogous data is generated between the waveform samples, which also results in data above 20 kHz.
I don’t use upsampling. I tried upsampling to higher rate PCM and DSD but then the music lost certain qualities which are important for me. Maybe I try it again with a new M3 CPU Mac next year.
The upconversion function uses RDOT-NEO (Refined Digital Output Technology NEO) to smoothly augment PCM digital audio signals and upconvert their sampling frequencies 2, 4, or 8 times (384kHz maximum). RDOT, which applies an analogous interpolation method using fluency logic, is a technology that was developed to enable reproduction and playback of frequencies higher than 20 kHz that are lost by 44.1kHz/48kHz digital signals. Based on the information read, analogous data is generated between the waveform samples, which also results in data above 20 kHz.
It’s using an ESS DAC, which has it’s own SDM up-conversion and is a PCM-centric, multi-bit DAC… so DSD would not be it’s forte anyway… it doesn’t provide an unfettered 1-bit PDM (DSD) signal path to a simple low-pass filtered output circuit…
ESS ES9038Q2M DAC: Industry’s highest performing 32-bit mobile DAC with unprecedented dynamic range and ultra-low distortion.
If you deliver a high-quality up-sampled PCM DXD 384 or 768kHz signal derived from PCM 16/44.1 kHz or higher 24 or 32 bit PCM signal to this DAC, it will work less hard and potentially sound better, because it will be working closer the SDM sample-rate which reduces the load on the SDM DSP…
Yes, on paper you are absolutely right. But it doesn’t work out with my old MBP . The results were not pleasing and I preferred upsampling disabled. With PCM 7xxkHz I got distorsions and DSD sounded too soft for my taste. Maybe the culprit was the old MBP and not Audirvana.
Well, DSD will never sound right on that DAC… Choosing the proper filter settings in SoX or r8Brain in concert with a complimentary filter choice in the ESS DAC, is important… It is not as simple as most would like to think… the up-sampling algorithms in Audirvana are tools… the better one knows how to use them the better results one will get out of them…
Not wrong, Phillip to feel special, especially in music.
You put it correctly when you mentioned the reference is the original recording. Ideally, all systems should sound the same by producing precisely what is in the original design. Unfortunately, this is not the case. Equipments and the environment in which they operate tend to influence the sound a lot. Unfortunately, some people do not mind that. Audiophiles do.
Absolutely. I spend days fiddling with my system to get… the best from my equipment. I love it.
And then sit listening for a few weeks and have to move the speakers for cleaning… and off I go again. Magnepan speakers are notorious cor placement-issues.
I am a little late to this posting party but seeing as I have just spent 17 hours trying to come to a landing on this very topic to soothe my FOMO. My conclusion is it all comes down to the individuals preference and his equipment
My methodology was to select my favourite well recorded tracks and then played each a) bit perfect b) up sampled power of two to 705 or 768 kHz (Or 192 kHz to the KEF’s) and c) up sampled to DSD 256 or 128 depending on the endpoint capability.
My playback chain - M3 Max MacBook Pro, 128gb ram, 4 TB internal SSD with library. Studio 2.6.3 on macOS 14.2.
Two playback scenarios MBP via USB to Chord Hugo TT2 and Focal Utopia 2022 headphones. Or MBP via wireless UPnP to KEF LS60’s
After intense listening, my conclusion was that each playback scenario sounded great to me with only very minor differences in SQ. I found that upsampling PCM to 705 or 768 kHz to the Hugo TT2 gave an impression of greater separation between instruments and vocals and enabled me to focus more on artistry in the track than the native resolution PCM, but I repeat it was a minor improvement. Upsampling to DSD256 via the Hugo sounded warmer with a very slight loss in my perception of individual instrumentation
With all three playback scenarios the use of computer resources was significantly different. Processing of the files to DSD 256 would send CPU usage up to 20% for 15 seconds whilst processing and eat a few GB of Ram. With my MBP specifications this is absolutely no issue and playback was faultless however for people with old macs with core i5 processors and only 8GB ram I would have a concern on reliability during playback.
So with my ears and my preference for neutral, incisive sound and large soundstages I prefer upsampling PCM with both my Chord DAC and KEF speakers. I then did a mini test on my wife and she preferred the warmer sound of upsampling to DSD.
Conclusion from me therefore is that everyone perception is different and they should go with the sound they enjoy the most whilst taking into consideration your playback chain and its capabilities (particularly with wireless UPnP) to ensure a faultless experience. I.e don’t upsample to DSD256 and above if you get no perceived benefit and are not confident in your computer capabilities or network infrastructure.
I should also state Audirvana Studio produced stunning SQ and faultless playback throughout my 17 hour marathon, yet immensely enjoyable, testing today.
Choosing the proper filter settings in SoX or r8Brain in concert with a complimentary filter choice in the ESS DAC, is important… It is not as simple as most would like to think… the up-sampling algorithms in Audirvana are tools… the better one knows how to use them the better results one will get out of them…