Can someone explain the CLANS noise shaper?

Wow, love the new update so far. DSD 256 has never been more stable on my system. I can see that my processing speed quadrupled as well for upsampling to my Chord DAVE. Love it, love it, love it.

Anyway, can someone explain what the CLANS noise shaper does? I am using the 7th order variant of this.

Thanks!

Bump for this. Anyone know what the CLANS noise shaper does? So far, it doesn’t seem to make any difference in processing speed. And I haven’t heard a huge difference in sound.

2 Likes

One for @brian.

CLANS stands for “Closed Loop Analysis of Noise Shapers”.

Inside the Sigma-Delta modulator–which converts multi-bit PCM to a 1-bit DSD signal–there is a component called a “noise shaper”. This is comprised mainly of a filter which pushes quantization noise inherent to 1-bit systems out of the audible range. Every system which produces a DSD stream (whether it be in software or in the mastering environment) must do some form of noise shaping.

“Traditionally” designed noise shapers are essentially built to spec using roughly the same filter design concepts as are used for analog/IIR filters (speaker crossovers, etc). The “Order 5” and “Order 7” filters fall in this category.

CLANS uses an iterative optimization technique called dynamic programming to improve on the technical specs of a traditionally designed filter. At filter design time, the computer chugs for a while by making tiny iterative changes to the filter designs and tracking improvement or compromise in certain technical characteristics. The filter that pops out at the end is the most optimized one that it could come up with.

The CLANS filters don’t use any more or less CPU at runtime than the other options because the filter topology (7th order loop filter) is the same either way.

How much of a sonic difference you perceive will depend a lot on how your DAC is architected. Small differences in noise shaper design are most noticeable on discrete DSD DACs like the HOLO Spring, T+A DAC-8 DSD, DSC-1 since these DACs perform no further processing of the signal in the digital domain. DACs with their own signal processing–like the Dave–may obscure some of the differences depending on exactly what is going on inside.

22 Likes

Wow–thanks for the very thorough explanation. I don’t hear any difference on the DAVE, but I’ll keep playing with it.

Hi @brian,

Is there any difference between 5th vs 7th order noise shaping when doing DSD64? My understanding is DSD64 limitation is noise starts to rise above 22.5kHz this can go as high to -70dB and above the noise floor. Does switching to 7th order noise shaping pushes the noise even further away?

Higher order filters increase the dynamic range in the audible frequency space at the expense of a “louder” noise floor at smaller inaudible frequencies.

So they are, in a sense, pushing the noise further “down” in the audible frequencies, but the noise floor rises more steeply, since it’s a higher order filter.

The best way to push the “loud” part of the noise floor out to higher frequencies is to choose a higher DSD sample rate. DSD64 is fairly tightly constrained.

I think the SNR improvement for DSD64 between the extremes (5th order non-CLANS -> 7th order CLANS) is about 20dB.

Of course, the DSD modulator isn’t the SNR bottleneck. Your ears, the room, and the amplifier-speaker system all have less dynamic range than either modulator choice…

9 Likes

Thanks @brian, that’s brilliant!

@brian

Do you know if the USB DAC in Oppo’s 205 implementation (ES9038PRO) performs any of their own signal processing?

I’ve tried all of the DSD combinations and sometimes I think one might sound better than another but then again…the differences are vanishingly small compared to a convolution filter and/or a few room treatments.

Regardless, upsampled DSD256 into USB sounds wonderful.

Happy holidays to you and your team, awesome work!

Larry

Yeah, it does. It’s an S-D architecture dac, so everything is getting upsampled/modulated.

1 Like

This has been my experience also and, I believe, a number of other users. Room treatments and/or convolving Room EQ is the best “bang for buck” SQ improvement for many systems.

1 Like

@brian

I don’t wish to belabor the point but do you consider upsampling in Roon to DSD256 worth the PC processing power (electricity and dissipated heat, fan noise) when a Sigma-Delta DAC is going to upsample to DSD512 or PCM768 anyway?

I probably just answered my own question…but I’m curious what you recommend.

Thanks

All upsampling algorithms are not created equal. Nor are all sigma-delta modulators. You can spend more (CPU time) and access a more sophisticated algorithm. The idea behind software based upsampling is that the software can potentially do a better job, or optimize the process in certain deliberate ways (e.g. apodizing, min-phase, slow roll-off, etc).

I have zones in my life where I employ software based upsampling, and others where I do not. I know better than to tell a room full of audiophiles what is “worth it” or not :slight_smile:

7 Likes

@brian,

Understood, and fair enough, you’re a wise man.

Cheers

1 Like

@brian… is there a know different behaviour between the four possible filters?.. my NUC 8i3is able to a upsampl

Here’s a post Brian made about the filters in another thread:

Ok thanks, interesting post from Brian

What I was trying to ask, but too late and I had to sleep, was… I use the precise linear setting, upscale pcm up to 512 DSD (but that happen the same way with DSD up to 128 and a did decent dac,… why changing from 5th order to 7th order to 5th order clans and the to 7th order clans the sound increase in bass but mainly in mid high range with sone eccess of glare harshness? I was expecting to have a more musical detailed music but, especially the 7th clans doesn’t seems to me very musical, let say the opposite.
Is there something I’ve not understood?

2 Likes

What you hear is the final arbiter of what works best in your system. The order of a filter is the number of samples used to generate the output. Higher order filters can be steeper than lower order, meaning more HF content. That can be good in some systems, not so good in others, hence the choice.

This article on filter design sets out the various trade offs that have to be made.

2 Likes

ok, thank you :slight_smile:

1 Like