What are user’s preferences on the option of Roon’s upsampling to their NDS or ND555 network player?
Do you play the format as is?
Do you upsampled all PCM format to the max support format, namely 24/192?
Do you upsample/convert to DSD format (DSD64 for NDS, DSD128 for ND555)?
Is there any advantage of the offboard upsampling?
I play as it is. No up sampling.
I did try up sampling but did not like it
I found that there was a loss of transparency with Roon’s upsampling. However, this was several versions ago and there are reports that Roon have significantly improved their engine in recent versions.
Running a Titan with 64GB of Apacer Industrial Wide-Temp RAM, I find that my NDX 2 + XPS-DR reveal even the subtle differences between having volume leveling on vs off. Full DSP? It’s still fairly obvious to me.
I would imagine that upsampling (we assume this means upscaling rather than oversampling) has a positive impact on lower end DACs as reconstruction of lower resolution audio requires a lot more heavy lifting than with higher resolution audio. While the higher amount of data going through seems like it generates more work, it requires far less interpolation and extrapolation to reconstruct a convincing analogue signal.
The better the DAC is, the less positive impact this would make until you get to a point where the DAC does it better job of reconstruction without the data being “pre digested” in some way.
This is why the audible difference between 16/44.1 and 24/192 (or even 32/768) on a Walkman or iPhone is absolutely huge but relatively tiny on high end hifi.
I’d not discourage anyone experimenting. But conversion to DSD is probably not ideal for any DAC that unpacks DSD into a PCM bitstream first, which all Naim DACs do. It’s adding two steps of potentially lossy conversion.
I play as is… given the Naim and Chord DACs oversample anyway… there is no performance advantage of doing it separately.
Upsampling is not recommended as it introduces mathematical errors, oversampling is usually preferable to upsampling as depending on implementation such as zero value sample insertion, no interpolation errors are introduced into the sample stream. Naim and Chord for example use zero sample value insertion in their over sampling methods.
But as always in hifi better doesn’t mean preferable, so one has a choice…
Oversampling is always an even finite multiple of the original sample rate… like 48 to 96 or 192 … or 44.1 to 88.2 or 176.4…. the typically inferior upsampling is not… like 44.1 to 96.
I’ve tried upsampling with both Roon+HQPlayer and Audirvana and prefer the more natural sounding bitperfect variety on both the nDAC and the Holo Spring3/KTE ladder-DAC (even at 705.6/768kHz oversampling). But I’d say it depends on your system/preference or even mood, for me upsampling doesnt kill the music but (very) slightly overdo details.
Roon can do power of 2 conversion to the max rate that your streamer supports. So 48 to 384 and 44.1 to 352.8.
Yep oversampling is usually much preferable compared to upsampling, if your DAC doesn’t natively over sample… but that’s quite a lot power one is using to transmit that data to the DAC where it will be able to likely perform that function more efficiently and therefore create less electrical noise.
Agree. I recall that the HQPlayer developer said something to that effect, but more on a positive note.
On my Linn NG KDSM streamer, I find that upsampling to DSD 256 (7th order clans) sounds better at HF contents.
I was watching a review of a Holo Cyan 2 R2R DAC, which offers a NOS approach, therefore any change to format has to be performed off-board pre-DAC.
This is something I had not set in Roon, just using the DSP conversion for formats not supported by the NDS, so 24/353.8, DSD128, DSD256 etc (of which I only have 20 titles in these formats), so downsampling/downscaling rather than any upsampling/upscaling, plus the ‘unfolding’ of local MQA upto 24/96, of course.
So wondered if any NDS/ND555 users pre-processed any local library material.
Never thought to try that with Roon - will give it a go!
@badger1, I would be interested to hear your feedback on the Roon DSP upsampling, I’ve tried HQPlayer, but it is limited to 384/24 via async USB B input, and I find that HQPlayer is only stable below and at 192/24 with good SQ results.
Quick test - it’s a subtly different presentation at first listen but will listen more and report next couple of days. My Mac M3 laptop has no issue with the conversion.
Hi Sean
If I understand right, the Organik Dac Linn designed has a delta-sigma modulator that upsamples the audio data to 1 bit pulse-width modulation stream at 98.304 MHz which in turn is fed to a discrete switched-resistor network which performs the conversion to analogue.
There is a multistage FPGA enabled process of upsampling and noise shaping for PCM files. There appears to be minimal dsp applied to DSD files as they are converted to a pulse-width modulation stream at 98.304 theoretically retaining the timing of the original 1 bit file.
So by getting streaming software like Roon to offload some of the conversion to 1 bit it is likely to change the sound produced due to a) different implementation of conversion algorithms and b) a change to whatever noise shaping is going on in the NG.
Whether that change is better I guess is a matter of taste. Linn must have put a lot of work into the preparation and conversion of PCM files to 1 bit and tune it to their preferred sound signature. I found the change (FPGA v Roon upsampling) to be super subtle - not really sure I could pick it out in a blind lineup. Perhaps a slight softening of some transients. I spend an age a while back comparing DSD recording to PCM recording on my Tascam 3000DA and I eventually could pick which is which so I can hear some of the differences. But I can’t be bothered to spend hours testing this for the NG! I could spin a coin and probably be happy! Good news with Roon is that it is really easy to switch over and on my hardware setup super stable.
My other takeaway Sean is that a Mac M3/M4 sitting on my lap connected via 5GHz wifi (one of my unifi APs is about 4m away in the next room and is hard wired into a different switch) is more than enough server and network oomph to get the file from my NAS and create a perfect bit stream to the NG.
Combine this with the extra network isolation Linn did with the selekt and NG casing/isolation designs, and some common sense networking (dedicated switch for hifi, decoupling switches with utp cable, sensible networking settings etc) and I am getting way better results than my old KDSM streamer. And crucially these factors (server, switch, cabling) seem to have negligible impact on linn’s NG noise shaping now so long as they are setup well in the first place imho. That wasn’t the case before. I got some great help from members here on good simple network setup that was worth its weight in gold all now implemented. Thanks to all who helped.
So I abandoned any messing around with Melco servers, network cables, switches, ferrite chokes as a total waste of my time. All they did was get the old KDSM closer to how the new KDSM now sounds, but imperfectly. Others may have fun playing with these things in their systems and rooms to try and fine tune/override their streamer’s noise shaping but for me in a Linn NG world with my system, it’s not worth any effort whatsoever.
Offloading some of the organic DACs conversion workload to a modern hi powered PC/Mac is relatively easy now with SSD/Multi Core processors/GPU (apple silicon or Nvidia). Personally I prefer not to second guess Linns dsp but if I was unhappy with it at any stage I could easily experiment with different software for conversion and offloading. Did you play with any of the different options and what did you hear?
Good to chat on this as always. Jon.
As an addendum, all I said applies to Linn DAC architecture. My understanding of Naim DAC architecture is that it is fundamentally different. DSD data is converted and upsampled to 32 bit 705.6 kHz PCM. Similarly pcm is over sampled. The PCM is fed into a Burr Brown DAC. Not sure what benefit in this case there is in upsampling before passing data to the streamer but again I assume there would be a change of some kind to the noise shape.
I’ve been playing around with the Roon DSP (aka Roon Muse) on and off for some time due to curiosity, I use Roon to upsample the stream to DSD 256 (filter linear, precise phase, SDM 7th order clans). I find that the Roon DSP results in some subtle improvements in timber definition with some tracks with HF content, but less transient attacks and softening the bass.
So in summary, the Linn streamers are still good with the Roon or the stock Linn App with no upsampling or oversampling. HQPlayer DSP is another thing, it can and is capable of producing very high SQ, however it is a bit clunky, and we can discuss more, perhaps in another topic (if anyone is interested).
Yes i just heard some similar things. I played ‘chihiro’ by Billie Eilish and bass line was significantly more blurred and less detailed with the roon upsampling. I wanted to switch it off asap. I hadn’t noticed that particularly in the tracks I played earlier.
Happy to discuss more elsewhere as this is a thread for nds. I suspect there is limited gain trying to second guess Linn’s noise and sound profiling and sound testing given its implemented with some meaty fpga coding. Lesser dac and streamer combinations may have different mileage?
Try oversampling instead… it can sound a lot better than upsampling. Upsampling can cause sample data stream errors through interpolation, oversampling doesn’t.
Oversampling is always power of 2 to the original sample rate… upsampling can be any ratio… hence the errors.
That is why Naim internally use over sampling rather than up sampling for example, but you can force it to up sample on the digital out, which never sounds the best.
Oversampling in the Linn DAC context kind of makes no sense? But will try it now.