Filter after filter, does it hurt?

[jussi_laako (Jussi Laako (Signalyst))]

@Marian

That would be true if HQ Player was free.

You likely use some software for playback anyway. Otherwise you wouldn’t be at the Roon forum.

@Marian

you could implement the delta-sigma modulator (i.e. up-sampling and bit reduction) in software and then have the DAC do just the D/A part.

Yes, this is what HQPlayer is about.

@Marian

I know there are ADC chips that allow you to bypass the decimation part and get the direct converter output (at say 256fs/6 bits), but I don’t know if that’s also true for DACs.

There are quite many DACs that allow this. With or without DAC chips.

@Marian

If the differences are not audible, the argument is moot. Although as an engineer, I have to admit it’s more intellectually satisfying

I know there are chips that accept DSD directly, but I was thinking whether there are DACs that accept multi-bit. 1-bit has inherent flaws and should in general be avoided, but that’s a different discussion.

No flaws that couldn’t be fixed. And care to demonstrate what are those flaws?

By the way, ESS Sabre is also “1-bit” in that sense .

[Marian (I stand with Ukraine)]

@jussi_laako

You likely use some software for playback anyway

I use Roon, but it does more than just ‘playback’.

@jussi_laako

this is what HQPlayer is about.

But it only supports 1-bit SDM, not multi-bit.

@jussi_laako

There are quite many DACs that allow this.

I’d appreciate some links.

@jussi_laako

By the way, ESS Sabre is also “1-bit”

I don’t know what it uses internally, the spec sheet doesn’t seem to tell, but I know some AKM chips use multi-bit.

@jussi_laako

And care to demonstrate what are those flaws

• Cannot encode zero.
• Tons of quantization noise. It’s all outside audible range, but needs a good amount of filtering.
• Signal-modulated quantization noise, no matter how much dither is added.
• Inability to dither properly (e.g. triangular). Only a tiny amount can be added without saturating quantizer.
• Reduced signal level (full scale at most -6dBFS after noise filtering)

The only advantage I can think of is a very simple D/A conversion (basically a comparator).

[jussi_laako (Jussi Laako (Signalyst))]

@Marian

I use Roon, but it does more than just ‘playback’.

Yeah, so does HQPlayer. But I think you got the point…

@Marian

But it only supports 1-bit SDM, not multi-bit.

It supports both…

@Marian

I’d appreciate some links.

TEAC UD-501, UD-503/NT-503, RME ADI-2 in DSD Direct mode, T+A DAC8 DSD / HA 200 / SDV 3100 HV, Holo Audio Spring / May, iFi DACs…

@Marian

• Cannot encode zero.
• Tons of quantization noise. It’s all outside audible range, but needs a good amount of filtering.
• Signal-modulated quantization noise, no matter how much dither is added.
• Inability to dither properly (e.g. triangular). Only a tiny amount can be added without saturating quantizer.
• Reduced signal level (full scale at most -6dBFS after noise filtering)

Not true any more than a multi-level SDM in these areas.

For example totally uncorrelated random noise output left-over from a good DSD modulator being 40 dB lower level than first of the fully correlated images with PCM input from the same DAC chip. End result being that you get best technical performance running these in DSD mode.

SDM D/A converter is a filter in itself. No matter how many levels it has. My open hardware design SDM D/A is 33-level converter and filter at the same time.

@Marian

The only advantage I can think of is a very simple A/D conversion (basically a comparator).

By the way you are likely listening to a lot of PCM recordings using decimated “DSD”, because lot of pro “PCM” ADCs use TI’s PCM1804/PCM4202/PCM4204 chips which are 1-bit ADC with on-chip PCM conversion.

[Marian (I stand with Ukraine)]

Thank you for the links. I was thinking of DAC chips that accept multi-bit input for the D/A stage (at 64fs or higher), not DSD.

@jussi_laako

It supports both…

I got the trial version, and it seems there are two output choices, PCM and DSD. For PCM, I don’t see anything above 16fs or the ability to select bit count. Am I missing something?

@jussi_laako

totally uncorrelated random noise output left-over from a good DSD modulator

It can easily be demonstrated that there is no such thing.

@jussi_laako

you are likely listening to a lot of PCM recordings using decimated “DSD”

I know… You get what you can find. I own a DA-3000 recorder, just for fun. But at least I don’t own DSD I have better use for my storage. Ideally, DSD should only be used as a D/A stage, if at all.

[jussi_laako (Jussi Laako (Signalyst))]

@Marian

I was thinking of DAC chips that accept multi-bit input for the D/A stage (at 64fs or higher), not DSD.

I’m not horribly interested in DAC chips in first place. I find discrete implementations more interesting.

@Marian

I got the trial version, and it seems there are two output choices, PCM and DSD.

PCM (8-32 bit) and SDM (1-8 bit). DSD is just an obscure marketing term, so I prefer not to use it.

@Marian

For PCM, I don’t see anything above 16fs or the ability to select bit count. Am I missing something?

You have output rate choices up to 1.536 MHz and “DAC Bits” configuration option let’s you choose anything from 8 to 32 bits.

@Marian

I own a DA-3000 recorder, just for fun

I have two RME ADI-2 Pro’s, so I can record up to 768/32 PCM and DSD256 stereo (with HQPlayer Pro).

And one Merging Hapi, so up to 384/24 and DSD256 at 8 channels.

@Marian

But at least I don’t own DSD I have better use for my storage.

DSD is more efficient than hires PCM. I have quite a bit of DSD recordings, up to DSD256. You can get a lot of stuff for example from NativeDSD.

[Marian (I stand with Ukraine)]

@jussi_laako

DSD is just an obscure marketing term

Totally agree! In that sprit, maybe you should only use PCM (not SDM), since PCM uses SDM for bit reduction anyway.

@jussi_laako

DSD is more efficient than hires PCM.

That is impossible, considering how much noise (i.e. useless information) is encoded in DSD. Besides, I’m not interested in HD, just well mastered CD.

BTW, you should read this AES paper on DSD, it’s really interesting. It shows, among other things, that 8-bit PCM at 4fs, while half the rate of DSD64, outperforms it in every aspect.

AES E-Library » Why 1-Bit Sigma-Delta Conversion is Unsuitable for High-Quality Applications

[jussi_laako (Jussi Laako (Signalyst))]

@Marian

BTW, you should read this AES paper on DSD, it’s really interesting. It shows, among other things, that 8-bit PCM at 4fs, while half the rate of DSD64, outperforms it in every aspect.

I’ve said, and demonstrated, many times before that it is total ■■■■■■■■■ It is 20 year old paper demonstrating that it’s author doesn’t know how to design proper a delta-sigma modulator. Only thing it shows that badly designed modulator will perform badly. Competent modern designs bunk it completely.

[Bert_Dijkstra]

Oops…I got lost in my own topic.
However, I would like to come back to my question.
First of all thank you Jussi for your answer about the filter settings.
I hope you were able to sleep well.

Unfortunately I am still a novice especially when it comes to certain terms.
In the proposal for PCM I chose the suggested sample rate (sample rate> auto, rate limit> 768000).

@jussi_laako

Just in case, select a sharp linear-phase filter.

PCM filters are explained in the help document. The word sharp is rarely used. Perhaps other words are used for this word, such as: sharp/extreme cut-off and attenuation. I’m thinking of poly-sinc-ext/2/3 or xtr/-short-lp.
Are the Gauss and sinc S/M/Mx/L filters also part of the sharp linear phase filters? Do you have any advice for any of these filters to start with? I now use LSN15 for Dither.

I don’t quite understand your advice for DSD material.

@jussi_laako

If you send DSD there, select 50 kHz DSD filter.

Is this a choice in the HQPlayerd configuration page? Do you automatically choose this (50 kHz) when you choose a filter 1x rate or 2x rate and higher?

@jussi_laako

I’d recommend using ASDM5EC modulator at DSD256 rate.

I know where to find the ASDM5EC modulator and I assume that the DSD rate is determined by the music file. Currently a file of DSF DSD256 is playing. In Roon it shows this:

And the Topping shows this: DSD 11.28MHz

This is what my Config.page currently looks like:

In this topic I ask questions to Jussi, but anyone who has experience with this is welcome for advice.

Thanks!

[jussi_laako (Jussi Laako (Signalyst))]

@Bert_Dijkstra

PCM filters are explained in the help document. The word sharp is rarely used.

I thought you were asking about DAC filter selection… You can choose your preferred HQPlayer filter, poly-sinc-ext2 is a good starting point.

@Bert_Dijkstra

Is this a choice in the HQPlayerd configuration page?

No, I meant the ESS Sabre filter in the DAC.

@Bert_Dijkstra

This is what my Config.page currently looks like

My recommendation is to choose “SDM” as “Output mode” instead of “Auto”.