Hi @jussi_laako,

I’ve recently ordered a Holo Audio Spring R2R DAC which appears to have a similar discrete resistor ladder design to your DSC1. It duplicates the ladder once for linear control ( not sure how this works) and again to give a balanced output.

I intend to use it nearly exclusively in NOS mode, which I understand avoids all the usual conversions and filters but am thinking of using HQP to send everything (Redbook, Tidal MQA (hopefully unfolded in Roon), Internet Radio) to it upsampled to DSD 512.

My question is whether that makes sense, or does having a NOS DAC remove all of the benefit of upsampling in HQP to DSD 512 ? I thought it might still have some benefit in shifting noise, but thought I would check with you.



I have the same dac coming soon as well and I was planning to do the same with HQP.

I imagine using HQP to upsample still has benefits. The benefit in using the Spring in NOS mode is to avoid its internal upsampling engine.

NOS DAC’s are generally good ones for doing upsampling, because they focus on only one thing - doing digital to analog conversion, without any digital processing. For HQP type of use cases, many DACs become more or less NOS DAC’s when fed with highest possible PCM rate or DSD.

So whenever there’s a NOS DAC that accepts particularly high PCM or DSD input rate, those should be good companion for upsampling with HQPlayer (or something else). Good thing is that you don’t pay for DSP features you don’t need or use.

If you play PCM sources from typical source rates like RedBook 44.1k, 88.2/96k or 176.4/192 you need oversampling/upsampling for proper performance. And that’s precisely what HQPlayer is all about. NOS DACs need that to perform well, but they don’t event try to enforce you any certain way to do it, but instead let you pick up the way you prefer and instead focus on doing just the D/A conversion, nothing else.

Does operating in NOS mode, especially 44.1k resulted in more out of band sampling noise? Does this noise have an impact on amplifiers and speakers? Many brought this DAC, including me are quite convinced that subjectively, playing back in NOS mode ‘sounds’ better than on OS mode. I believe OS mode is technically superior when it comes to SNR but I can’t find the correlation here.

Yes, both that and notable broad high-frequency roll-off, starting at about 10 kHz and reaching -3 dB at 22.05 kHz.

It depends on how much the analog reconstruction filter following the conversion section manages to cut out the high frequency distortion. Oversampling is used to help the analog filter do it’s job on reconstructing a smooth original wave instead of stair-stepped one.

Here is example output spectrum of 0 - 20 kHz frequency sweep of one NOS R-2R DAC running at 44.1 kHz:

And same DAC and same source data with upsampling to 384 kHz:

1 Like

Thanks Jussi, that’s very informative. I guessed the over-sampling to 384kHz graph with the ‘Spikes’ represent the image frequency of the original sample? Can these be removed?

What frequencies was the sweep running?

Here is example output spectrum of 0 - 20 kHz frequency sweep …

If that’s the true frequency content I don’t understand why resampling changes the spectrum so radically.

Sorry the sweep is exactly 0 - 22.05 kHz so full bandwidth of the sampling rate.

There are always images at multiples of the sampling rate. Higher the rate, higher those images go in frequency and further apart from each other. Sampling rate is like a mirror, there are positive frequencies above it, and negative frequencies (inverse spectrum) below it.

So in the first picture there are images around every multiple of 44.1 kHz, and since content has 0 - 22.05 kHz bandwidth the images are completely adjacent due to no oversampling. To make the spectrum clean, analog filter would need to be able to cut out everything precisely above 22.05 kHz but nothing below. This is practically impossible task for analog filter, but not for digital filter. So the first images are really strong.

In the second picture you can see images around every multiple of 384 kHz, and since the content still has 0 - 22.05 kHz bandwidth, there’s plenty of empty space between the content and first image and the first image is already -50 dB down and much higher up in frequency.

From both pictures, we can see that with this given analog filter the images would be practically removed by having upsampling to about 3.2 MHz sampling rate. But now with the sampling rate limited to 384 kHz there are still some left, albeit total power of the images being much much smaller than without upsampling.

Now it would be already feasible to design a higher order analog filter that would already significantly lower the remaining images around 384 kHz multiples.

You can also see that upsampling significantly reduces the HF roll-off in the 20 kHz band.

Here’s the NOS mode output in narrow band:

And here’s the same with upsampling to 384k:

From the upsampled plot you can also see how the DAC’s THD increases as function of frequency (the increasing slope above 22.05 kHz).


Because the upsampling digital filter removes image frequencies between the original 22.05 kHz Nyquist frequency and the new 384 kHz Nyquist frequency minus the negative frequencies -> 384 - 22.05 = 361.95 kHz. So for perfect reconstruction of RedBook you could try to design an analog filter that has 96 dB attenuation by 361.95 kHz. This means available transition bandwidth of 361.95 - 22.05 = 339.5 kHz instead. For 20 kHz bandwidth that would be 384 - 20 - 20 = 344 kHz. Compared to 44.1 - 20 - 20 = 4.1 kHz for the original NOS.

1 Like

Wow! Thanks Jussi, so the purpose of over-sampling is to shift the first image (44.1k) as far as possible from the 0-22.05k bandwidth, in this case it is shifted to 384k. At this frequency, it is much easier to filter off?

My understanding of analog vs digital filter is the later can provide an accurate and sharp cut-off but at the expense of severe ‘ringing’ (impulse response). Would it be possible to do without digital filter but over-sample the first image as far as possible then do a gentle analog filter? This is similar to what DSD high sampling allows gentle analog filter without resort to the use of digital filter.

Filters ring – that is what they do. It is a product of bandwidth limiting the signal. Pre ringing occurs in linear phase FIR digital filters – because the filter is run both forward and reverse to maintain phase linearity, whereas analog filters can be run only forward, hence post ringing. The same can be true of IIR digital filters – no pre ringing at the expense of phase linearity.


Agreed, but doing a digital filtering so close to the corner frequency of 22.05k, together with pre and post ringing of a sharp digital filter, there’s a lot ‘energy’ needs to be dissipated across the spectrum, since 22.05k is at the upper audio band, this energy will have some undesirable effects in the audio band.

By over-sampling, there by shifting the sample image to very high frequency then apply a gentle analog filter with minimal post ringing is more desirable.


Given that RedBook has Nyquist frequency (22.05 kHz) so close to the 20 kHz audio band edge it unfortunately requires very steep digital filter for proper reconstruction. There are various different kinds of digital filters with various different levels of aggressiveness/steepness, so if you run the filter in software you have quite a bit of freedom to choose which one performs best for you…

With RedBook the issue is even more on the production side, because input frequencies higher than 22.05 kHz alias (fold) down to the audio band. So even more aggressive filters are needed there. For playback side, implications of these filters can be replaced/modified by using “apodizing” upsampling filters.

Since DSD is “non-Nyquist” sampling system where bandwidth it not defined by Nyquist frequency, things are different. There the analog filter requirements are defined only by the shaped noise at ultrasonic frequencies. Since the noise slope is fairly gentle, it is possible to deal with this using pure analog filters. This goes to the same category as class-D amps.


Yes, with FIR filters you can have linear phase (50/50 pre/post-ringing), minimum phase with only post-ringing, anything between the two (named “asymFIR” in HQPlayer) or even maximum phase filter with only pre-ringing (nobody wants that).

Practically IIR filter can be converted to FIR filter too.


Thank you @jussi_laako for your continuing involvement on this forum and the crisp explanations and graphs set out in this thread. It is absolutely refreshing to read high level explanation from a person with deep understanding of the underlying principles shorn of all marketing hype.

I have bookmarked your post showing the narrow band (audio band) effects of upsampling to 384 kHz for later reference. It is the clearest explanation/demonstration of the benefits of upsampling that I have seen.


“Yes, with FIR filters you can have linear phase (50/50 pre/post-ringing), minimum phase with only post-ringing, anything between the two (named “asymFIR” in HQPlayer) or even maximum phase filter with only pre-ringing (nobody wants that).”

I agree with Andy. I have missed any earlier explanation like this quote. Very handy to know.


Thank you Jussi!

Thanks for this response and the response to MusicEar’s question. I misread the frequency scale in the original spectrum plots and that threw me off (I see now they are fractions of MHz) so this makes a lot of sense to me now. My eyes just aren’t what they used to be…