Whether to Upsample? Yes or No?

[Hari Iyer]

After installing my Shangling ET-3 now around a month ago, i have been trying to up-sample the native sampling rates which for CD format is 44.1KHz to x2 and x4 sampling rates. I.e. either DSD64 or PCM192Khz and in-between. Also i have tried to bypass the sampling rate conversion (SRC) in the ET-3 and use the native sample rates that is original in the format. My subjective observations are based on my setup (amplifier, speaker, cables and room) which could be completely different than others and with my listening skills / limitations.

Initially i liked the upsampling conversion of 44.1Khz to 192Khz and even DSD64. The DSD64 sounded much rounded and with a slightly rolled-off highs which was smooth and to my liking. In comparison the PCM192Khz sounded more energetic and with more resolution. Off-late after comparing with the SRC bypassed, i find the harmonics and transients to be more accurate and natural if i don't fiddle with the sampling rate.

I tried to understand why this was so - and discovered - the upsampling method adds additional information to the original information by a process called interpolation which was not present in the original recording. This would mean coloration to me as you are fiddling with the original sample to derive at the new sample. This could be the reason why DSD64 sounds more soft, rolled-off and smooth and PCM192KHz sounds a little fatigued to my ears.

The above is my own personal observation and verdict. Will like to hear what others are observing if they fiddle with the original sampling rate using SRC.

Thanks for looking.

 

[Yelamanchili manohar]

After installing my Shangling ET-3 now around a month ago, i have been trying to up-sample the native sampling rates which for CD format is 44.1KHz to x2 and x4 sampling rates. I.e. either DSD64 or PCM192Khz and in-between. Also i have tried to bypass the sampling rate conversion (SRC) in the ET-3 and use the native sample rates that is original in the format. My subjective observations are based on my setup (amplifier, speaker, cables and room) which could be completely different than others and with my listening skills / limitations.

Initially i liked the upsampling conversion of 44.1Khz to 192Khz and even DSD64. The DSD64 sounded much rounded and with a slightly rolled-off highs which was smooth and to my liking. In comparison the PCM192Khz sounded more energetic and with more resolution. Off-late after comparing with the SRC bypassed, i find the harmonics and transients to be more accurate and natural if i don't fiddle with the sampling rate.

I tried to understand why this was so - and discovered - the upsampling method adds additional information to the original information by a process called interpolation which was not present in the original recording. This would mean coloration to me as you are fiddling with the original sample to derive at the new sample. This could be the reason why DSD64 sounds more soft, rolled-off and smooth and PCM192KHz sounds a little fatigued to my ears.

The above is my own personal observation and verdict. Will like to hear what others are observing if they fiddle with the original sampling rate using SRC.

Thanks for looking.

Hello sir

I think it depeneds on where the Dac is most comfortable at. Most dacs have a sweet spot in terms of sample rate, as well as input used. Iam upsampling Tidal to either 176.4 or 192 khz and find this smoother, and more detailed than native sample rate. I think this might be the sweet spot for my dac.

But in most cases, upsampling requires sophisticated algorithms and lots of processing power to do correctly, like HQ player for example. Even the older Sony HAP Z1-ES was quite popular for its sophisticated internal upsampling. Thus, it also depends on the quality of the upsampling, as I think there is more than just inter polation to it. But then we are talking about serious computing power, which might also get noisy.

 

[Hari Iyer]

Hello sir

I think it depeneds on where the Dac is most comfortable at. Most dacs have a sweet spot in terms of sample rate, as well as input used. Iam upsampling Tidal to either 176.4 or 192 khz and find this smoother, and more detailed than native sample rate. I think this might be the sweet spot for my dac.

But in most cases, upsampling requires sophisticated algorithms and lots of processing power to do correctly, like HQ player for example. Even the older Sony HAP Z1-ES was quite popular for its sophisticated internal upsampling. Thus, it also depends on the quality of the upsampling, as I think there is more than just inter polation to it. But then we are talking about serious computing power, which might also get noisy.

I think most DACs do not do any upsampling but only sync the cclocking speed with what is fed to it's input. Is the single bit DSD format any better SQ wise to the native PCM formats?. In DSD a single bit will have many samples and hence more interpolation to the original signal. Unless the original signal is recorded in native DSD, whether an upsampled PCM 44.1khz to say DSD256 add any real resolution to the signal or are they coloration due to derived via algorithms etc. I always find doing an SRC to DSD64 - I loose some HF and transient detail compared to native PCM, though the DSD64 sounds more rounded and smooth. I am unable to figure out which is accurate - bypassed SRC or upsampled.

 

[prem]

I liked 44.1 oversampled to 88.2 or 176.4. IME, a direct multiple of 2 or 4 seems to work. Never liked it upsampled to 96 or 192 or DSD. I found it sounding unnatural at least in the dacs I owned or heard. Maybe in some dacs it works.

 

[muralimmreddy]

Some random analogy

Decade back I used to have then ground braking Sigma DP-1 camera, which is true full-frame sensor i.e.) there is sensor for each R, G, B for each pixel it only had 14 Mega pixel, where as others (Canon, Nikon etc) has insane MP. But they all has only one of RGB sensor per pixel, and then they do interpolation based on surrounding pixels. Sigma DP-1 pictures were real, true to life pictures taken were aesthetically pleasing. True, authentic, and quality over quantity,

I tried all sampling with lumin X1, they all sound processed may be for synthetic music its good

 

[Hari Iyer]

I liked 44.1 oversampled to 88.2 or 176.4. IME, a direct multiple of 2 or 4 seems to work. Never liked it upsampled to 96 or 192 or DSD. I found it sounding unnatural at least in the dacs I owned or heard. Maybe in some dacs it works.

Ok. I have never tried upsampling to 88.1 and 176.4 yet and not done an A/B vis-a-vis native sampling rate. Maybe another day. Thanks for the insights Prem.

 

[Hari Iyer]

Some random analogy

Decade back I used to have then ground braking Sigma DP-1 camera, which is true full-frame sensor i.e.) there is sensor for each R, G, B for each pixel it only had 14 Mega pixel, where as others (Canon, Nikon etc) has insane MP. But they all has only one of RGB sensor per pixel, and then they do interpolation based on surrounding pixels. Sigma DP-1 pictures were real, true to life pictures taken were aesthetically pleasing. True, authentic, and quality over quantity,

I tried all sampling with lumin X1, they all sound processed may be for synthetic music its good

Does interpolation and upsampling make the images surreal? I don't know if audio and video have the same yardstick for a comparison. But opinions could vary widely here too.

 

[Yelamanchili manohar]

I think most DACs do not do any upsampling but only sync the cclocking speed with what is fed to it's input. Is the single bit DSD format any better SQ wise to the native PCM formats?. In DSD a single bit will have many samples and hence more interpolation to the original signal. Unless the original signal is recorded in native DSD, whether an upsampled PCM 44.1khz to say DSD256 add any real resolution to the signal or are they coloration due to derived via algorithms etc. I always find doing an SRC to DSD64 - I loose some HF and transient detail compared to native PCM, though the DSD64 sounds more rounded and smooth. I am unable to figure out which is accurate - bypassed SRC or upsampled.

From my understanding, Delta sigma dacs down sample DSD to PCM before the stream enters the Dac chip for conversion. Only true 1-bit convertors can convert DSD to analogue. And since my dac is delta sigma, I never bothered upsampling to DSD though the streamer can do upsampling if required.

I havent tried it, but there are people using HQ player for upsampling to DSD, and it apparently sounds stunning, when used in a 1-bit Dac architecture for conversion. So I would either stick with original DSD content, of which I have a few. Or use something like HQ player to upsample to DSD if the Dac is worthy of the effort. Till then, as Prem rightly mentioned, Iam happy just upsampling tidal at 44.1 or 48 khz to 176.4 or 192 khz.

 

[Hari Iyer]

Thanks @Yelamanchili manohar for your insights. How then does a multibit converter work?

 

[Yelamanchili manohar]

Thanks @Yelamanchili manohar for your insights. How then does a multibit converter work?

Good morning sir

Sorry I dont know anything about Multi-bit dacs or R2R. Never had one .

 

[sachinchavan 15865]

As per my limited layman understanding:

Upsample is at best a smart guess by an interpolating algorithm about the frequency response between the two consequent sets of data separated by milliseconds (or is it microseconds, whatever). There’s no way to ensure whether it matches the exact missed data on the original analog signal which got lost during digitisation. We can only say whether the upsampled response is to our liking or not. Yes. It will sound smoother than non-upsampled signal, but how close is it to the original analog signal captured by the mic? There’s no way to say that.

So my Cambridge CXN V2 upsamples everything to 384 kbps. I have no control over it. I may say I like the resulting, unless I compare it with the analog recording, I cannot say how close the interpolation was to it. Also, I’d have to compare it with a bit perfect non-upsampled response with only analog filtering done by the DAC to say if the upsampling actually improved or worsened upon it, sonically.

 

[matbhuvi]

The reason why many don't prefer upsampling is because it alters the original bits. It is really hard to retain the original bits when you upsample.

I have the original Shiit Yggdrasil with Analog Devices AD5791 chip. It does an internal upsampling upto 8x without altering the original bits. I am big fan of its tonality. So much that i ensured to use a solid state preamp in the chain and didn't want anything altering the chain.

Having said that, HQPlayer does bring out a big benefit. Jitter reduction. Unless you are using a really high end transport with ultra low jitter like Hifi rose 130 with 10mhz internal clock, it is really really hard to get away with jitter in digital transport. That jitter will impact the smoothness, separation of instruments...basically, you will find the music flowing easily. For people like us who can't invest in such hiqh quality transport, upsampling will help to reduce the time domain error caused by jitter. That is why people call the upsampled music sounds smooth. The catch is, you need to find the right filter that can maintain the timbre / tonality. If you can find one, then it is brilliant to upsample everything. If you can't then it will sound really bland and too smoothed out and artificial.

I use NS9 shaper and Sinc-Mx filter which i really liked with Yggdrasil. The other one are poly-sync-gauss-hires-lp with gauss 1 dither. HQPlayer filters are way better than cheapo upsampling which can cause an echo like notes in music. That is the downside of doing upsampling any single single chip based devices. Some of the filters would need good compute / gpu to do the upsampling which won't be there in those chips.

The key here is to focus on the vocals. If you can maintain the same vocal tonality before and after upsampling, then you have hit a jackpot.

 

[Hari Iyer]

Past 10 days I was listening to x2 and x4 upsampling of the 44.1khz. TBH, I did not like the upsampled version. If I upsample then I have to set the RME DAC as master clock and the transport becomes slave. When I keep the native sample rate then the RME DAC sets it's sampling rate as per the source sampling rate and the transport is master clock and RME is slave clock. I like the tonality of the native source as they sound to me more natural and resolved compared to the rounded artificial upsampled source. YMMV.

 

[matbhuvi]

Past 10 days I was listening to x2 and x4 upsampling of the 44.1khz. TBH, I did not like the upsampled version. If I upsample then I have to set the RME DAC as master clock and the transport becomes slave. When I keep the native sample rate then the RME DAC sets it's sampling rate as per the source sampling rate and the transport is master clock and RME is slave clock. I like the tonality of the native source as they sound to me more natural and resolved compared to the rounded artificial upsampled source. YMMV.

Interesting..both your DAC and transport support clock input?

 

[Analogous]

Past 10 days I was listening to x2 and x4 upsampling of the 44.1khz. TBH, I did not like the upsampled version. If I upsample then I have to set the RME DAC as master clock and the transport becomes slave. When I keep the native sample rate then the RME DAC sets it's sampling rate as per the source sampling rate and the transport is master clock and RME is slave clock. I like the tonality of the native source as they sound to me more natural and resolved compared to the rounded artificial upsampled source. YMMV.

@Hari Iyer from your observations above it’s possible you may belong to the NoS camp in terms of sound preferences.
Since you are doing comparisons, why not include a NoS DAC into the mix and decide? (If possible to borrow or audition). Remember to use same source and files.
Sometimes old tech is still good (Tubes, R2R)
Different from Chip and FPGA tech that upsamples. Which is better would be up to personal preferences.
As always there are many decent sounding DACs that use either tech, or both and a few of each topology that sound brilliant.

 

[ryanrulez4ever]

I use HQplayer to upsample Redbook content from tidal or qobuz and sometimes even DSD 128 /256 to DSD 512 to my T + A DAC 200, and it sounds beautiful, I regret not doing this earlier but I guess everything happens at a certain time for a reason

 

[amrutmhatre90]

The DAC should support NOS (Non-OverSampling). I'm using the Gustard R26 with HQPlayer for upsampling, similar to @ryanrulez4ever , and I really enjoy the sound. While the upsampling on the Shanling ET3 isn't on the same level as HQPlayer, it’s still a significant step up from most off-the-shelf options.

How a signal sounds on a DAC can vary, too. Some DACs perform best with Redbook content, and their performance with higher-res files may differ. Ultimately, everyone’s experience will depend on the specific DAC, the upsampling method used, and the source material.

 

[matbhuvi]

After playing with HQPlayer for couple of weeks, I decided to stick with Redbook for my Yggdrasil DAC. Yes the HQPlayer improved the instrument separation making the music flow so smoothly and you can listen for hrs without any fatigue. But, the ever so smothering of vocals kept on bothering me. For the vocals heavy genre that i listen to, especially for my DAC, I decided not to go with HQPlayer route.

 

[matbhuvi]

I went ahead and bite the bullet. Got RS130 streamer of Hifi Rose. The result is just dramatic. I am not longer missing HQPlayer. To give an analogy, i felt like moving up from 128kbps to lossless. You get the similar benefits when the streamer is good. RS130 has replaced iFi zen with linear psu