Filter after filter, does it hurt?

Recently I ran a Topping D90SE trial. I also use HQPlayerd/NAA.
Now I’m wondering about the filter settings. On the Topping you always have to choose a filter setting with both SDM and PCM. With HQPlayer you always have to choose a filter with DSD files, while with PCM you can choose for none. Filter after filter, does that hurt?

No it doesn’t.

If you send PCM there, use 705.6/768k output rate. This bypasses the digital filter on ESS Sabre. Just in case, select a sharp linear-phase filter.

If you send DSD there, select 50 kHz DSD filter. I’d recommend using ASDM5EC modulator at DSD256 rate. Alternatively you can use ASDM7 at DSD1024.

1 Like

The point of spending money on an good DAC is not to worry about what you send to it. D90SE is one of the best DACs out there, so I see no reason to bypass anything that’s going on in there. Just select internal filter 1 or 6.

The point is that even on best DACs out there, the digital filters and modulators are still resource constrained implementations.

But if measurements are stellar, does it really matter how constrained its resources are?

Depends which measurements you are talking about… If you are talking about ASR SINAD rating, I just giggle a bit. :smiley:
(as they don’t measure reconstruction accuracy at all, instead they use analyzer’s anti-alias filter to complete the reconstruction)

Yes, ASR, and it’s not only SINAD. All measurements are stellar. If reconstruction was not accurate, you’d see it in THD, or the 32 tone test, or the linearity test. I believe that the AP analyzer is more than capable of taking accurate measurements.

No those are not valid for measuring it. Those are only valid for the purpose they are created, which is not measuring reconstruction accuracy.

Partially yes, if you know how to use it and measure the right thing. Otherwise it is just an expensive device for fooling yourself.

I also believe Amir knows how to use it.

Of course you are free to believe what ever you want. That belief doesn’t mean it would absolute truth about anything.

I forgot to mention that I also understand those measurements. Just in case it was not clear that I consider my beliefs to be educated.

Then you should also understand what they are fit for and what they are not fit for. For analyzing digital filter performance you look at things like pass-band ripple, shape of the roll-off, stop-band attenuation, oversampling factor performed by that filter, image band levels, etc.

For analyzing modulator performance you also need to use complex dynamic signals rather than steady state signals.

The white noise spectrum shows the shape of the filters. If you want details about each filter, you can look at the spec sheet. For example:

PCM Filter Characteristics (Fast Roll Off, Linear Phase)
Pass band Ripple and Bandwidth 0.002dB 0.453 x fs Hz
–3dB 0.484 x fs Hz
Stop band < –120dB 0.55 x fs Hz
Group Delay 35 / fs s

Yes, that is one of the examples, and remember that the on-chip digital filter works only up to 352.8/384k rate (8x filter). Quite pathetic figures, really. Those specs tell that there’s notable amount of pass-band ripple. And 120 dB stop-band attenuation equals to about 20-bit reconstruction accuracy, but doesn’t yet tell anything about image levels of the digital filter output rate. In addition it tells that the filter is leaky, since 120 dB is only achieve beyond 0.5 x fs.

People here talk about filters with a million taps or so, and here we are talking about on-chip filter that can have max 64 taps…

These figures can be improved over massively by bypassing the on-chip filter. Unfortunately ESS Sabre doesn’t offer direct access to the actual D/A conversion stage, so it’s performance will be always somewhat limited by it’s constrained on-chip DSP.

Not sure what’s pathetic about that. It’s also the rate before the delta-sigma modulator, which up-samples more.

That is more than enough. Even if you live in the most quiet room in the most quiet neighborhood, you are using less than 100dB dynamic range. Not to mention the best analog chain after the DAC barely matches that.

Just curious, how many taps does HQ Player use in the best filter?

And one more thing: I believe 44.1/16 done right is all we humans need.

Fascinating discussion (on both sides) that illustrates why I recently bought a turntable.


@jussi_laako, what dac chip would you recommend?

I’ll take a wild guess that he’s going to recommend HQ Player no matter the DAC, plus a super-computer to run it, in case you have more than one zone.

Because it limits the reconstruction accuracy because it leaves images around multiples of the digital filter output rate. For example I run full digital filters to 11.3M - 49.2M rates before handing over to the advanced delta-sigma modulator.

It is not, if you want to reconstruct sources accurately. And yes, no trouble measuring the difference also from power amp output. You can certainly hear errors at -120 dB levels.

Remember that SNR figures are for 20 kHz bandwidth. But hearing masking bandwidth for frequency peaks are much more narrow! For example you can easily hear signals 40 dB below 20 kHz bandwidth noise levels.

And dynamic range is something totally different.

Depends on your filter and conversion factor. But you can get up to about 260 million.
Ripple figures can be below 0.000000005 dB and stop-band attenuation >300 dB.

But already something like 1000 taps instead of just 64 would be great improvement.

It just happens to be extraordinarily challenging to reconstruct accurately, because the chosen limits are so close to audio band. Hires is much easier.

1 Like

If possible, no DAC chip at all, but instead a discrete implementation.

But if you need to have, one that allow direct (bit-perfect) pass-through to the D/A conversion stage. Like AKM, TI or Cirrus Logic.

1 Like