advertisement


Noise Shaping

So given that existing recordings were not filtered properly, a post processing stage of a 18kHz 3dB point, 4th order or more low pass should reduce these artifacts?

Ironically pre-oversampling era PCM1610/1630 recordings behave rather nicely in the upper treble, from an aliasing PoV.

Half-band recordings, on the other hand, should be replayed with an oversampling filter as I described. I don't think there is any DAC or CD player manufacturer doing this. They are all still set on printing "frequency response 20Hz-20kHz +/-0.1dB" in their blurb.

But we are deviating ...


Earlier question: why no cascaded noise shaping?

Well, each noise shaper is built on assumptions about audibility of the resulting noise/distortion. If you mix noise shapers, you start mixing assumptions, and the outcome may well be a mess.
If there is any chance that a noise shaping process is living downstream the signal path, then better apply TPDF at your own stage.
 
Thanks Werner, to me that's new and interesting information about the why. In terms of the what, it seems I'm on the right track with a transition band from 19-22kHz (and I will now try 18-22kHz)!
 
Giggles. IME any such recording system is bought at great cost, unpacked, hooked up, the "engineer" spends best part of a day swearing at some box called a 'word clock' and tries to figure out why Cubase is exactly one hour out of SMPTE sync with the f***ing ADAT and when the right combination if swear words are found to rectify the situation it is never spoken of again. I'd be utterly amazed if you could find many/any studio sound engineers who understand this stuff down to the bits, bytes, noise shaping, filters etc level. Basically if most of what you are monitoring sticks to the tape you are done!

Alas, yes. I fear the Philips/Sony people did totally misjudge the ability of some in the 'music biz' to use boxes without any clue or care wrt what they were *actually* doing. Hence all the recordings that are clipped or have other problems like badly scaled levels causing patterns in the sample distribution and lousing up the results.

All made worse, I guess, by the tendency to bring in 'gurus' who can 'make it sell' ... by fouling it up. Loudness, anyone? :-/
 
And yet, the recipe is simple, provided we drop that old fetish of needing flatness to beyond 20kHz.

Just somewhere in the chain, ideally at the ADC side, start rolling off at 18kHz, and reach zero, or at least a suitably low level

Agreed. In part I guess this is why my old trick of using a decent analogue filter after reconstruction that removed cut away the stuff about about 18k sounded better quite often.

I wonder if one root of the problem was/is that many engineers who have even learned how the sinc shape is derived from IT realise that to be formally 'correct' it has to cover *all* the samples in the recording and have perfect resolution. So any real filter has to be an approximation. Hence the wisdom of keeping the danger zone clean of anything that might cause a problem.

As on some earlier occasions, I regret that you can't now easily buy those old Toko filters. They were quite good analog designs. All killed off, I guess by digital filtering.
 
At any time in the last 30 years, there have always been the odd CD player out there with a ~18 kHz low pass. These have usually been said to sound "more analogue" but "slightly dull".
I think few reviewers can hear anything like 18 kHz, so this is more likely down to the absence of dome tweeter break up hash triggered by artifacts
 
So, if one has the choice (e.g. HQPlayer) how does one choose which noise shaping to use? (Apart from just trying each one and forming a preference.) I.e. are there good reasons to prefer one approach rather than another?

- Richard.
 
So, if one has the choice (e.g. HQPlayer) how does one choose which noise shaping to use? (Apart from just trying each one and forming a preference.) I.e. are there good reasons to prefer one approach rather than another?

- Richard.

TBH I'm not clear what question you really have in mind. The choice of a reconstruction filter is a different matter to the choice of Noise Shaping. Although a given filter may well employ shaping.

For the end-user the simplest approach is to just 'play out the samples' if your DAC can accept them. If you're worried about 'ringing' or 'artifacts near the top of the band' then apply a filter. However if you just want to slope off the top off the band or similar, the filter shape is probably the main issue.
 
For the end-user the simplest approach is to just 'play out the samples' if your DAC can accept them. If you're worried about 'ringing' or 'artifacts near the top of the band' then apply a filter. However if you just want to slope off the top off the band or similar, the filter shape is probably the main issue.
I thought HQ Player was all about upsampling to DSD in software which will require noise shaping
 
I thought HQ Player was all about upsampling to DSD in software which will require noise shaping

Ah! OK, afraid I had no idea of that. :)

Afraid I'm now unclear what the advantage would be of that specific system. Many DACs upsample, possibly to lower bit depths. So far as I can tell, it is then mainly a question of how well that specific system is implimented.

Personally, though, I regard DSD as fundamentally flawed. In part because 1-bit is effectively impossible to fully dither. Also because whenever I tried to model it I found all kinds of idler problems, cyclic issues, etc. This was based on the DSD modulator - i.e. ADC / upsampling - designs they published. This is in addition to the well know problems of a risk of lockup, etc. And the almost impossible task of finding analytical solutions to determine if/when a high order system is free of such things or not.

I did ask some of the Philips people at the time about this and they accepted my results but said they couldn't explain further. This was years ago, though.

More generally, upsampling isn't magical. It can be handy for the designer and provide a good approach to processing. But it can't improve the SNR beyond what the source contains, etc. So *in principle* no better than a simple NOS followed by an excellent analogue filter. May be much easier to do in mass production, etc, though!

So in reality any questions end up being wrt careful implimentation and not making mistakes that unintentionally degrade the results.

Need to know more about a case to go much further.
 
(about HQPlayer)

Afraid I'm now unclear what the advantage would be of that specific system.

Control of filter and sigma-delta modulator quality. Most DAC chips cut corners with progressively higher sample rate. HQPlayer preserves numerical quality all the way.

But of course it offers enough knobs and buttons for the audiophool phraternity to mess it all up again ;-)
 
Control of filter and sigma-delta modulator quality. Most DAC chips cut corners with progressively higher sample rate. HQPlayer preserves numerical quality all the way.

But of course it offers enough knobs and buttons for the audiophool phraternity to mess it all up again ;-)

Thanks. I didn't know anything about it until now. Prompted by the above I did a quick search. It seems to be available for Linux, but so far as I saw, it's proprietary and closed-source. Is that correct?

It occurred to me that it may be possible to use it to do test conversions by piping or diverting its output to a file. That way, let people really examine what the effects / changes / etc are.

FWIW I did write a simple upsampler a few years ago.

http://www.audiomisc.co.uk/software/ARMiniX/Upsampling.html

But that was *just* for RISC OS to let people experiment with basic changes to a FIR reconstuction filter. I was prompted because the RO port to the Pandaboard only gave access to high sample rates. (HAL limitation so far as I could tell.) Hence any 44.1k/48k material had to be upsampled before going to the HAL. The snag being that the players people where using could only do clumsy linear interpolations. So those programs aren't really what's wanted here.

As per another current topic thread (MQA bad for Music?), I think it may help people if there is a program that lets then generate <high rate>/24bit -> <same high rate>/16bit Noise Shaped versions of files for comparison purposes. So unless you or someone else can do a better job before I manage, I'll have a go when I can.

It occured to me, though, that HQPlayer may serve the purpose very flexibly. But I'd then wonder about the details of what it was doing, and we'd need to grab the output to find out answers to questions about how much FLACing would reduce the size of the required files/streams this had been done to.
 
Control of filter and sigma-delta modulator quality. Most DAC chips cut corners with progressively higher sample rate. HQPlayer preserves numerical quality all the way.

But of course it offers enough knobs and buttons for the audiophool phraternity to mess it all up again ;-)
Does this mean that for say dsd 256 it upsamples to 24/11.2 mHz (with a sharp filter at 5.6 Mhz) before the modulator, whereas an ordinary dac chip probably upsamples "properly" to 384khz or so and then applies zero order or first order hold to get to the modulator rate?
 
Thanks. I didn't know anything about it until now. Prompted by the above I did a quick search. It seems to be available for Linux, but so far as I saw, it's proprietary and closed-source. Is that correct?

There can be problems with Linux, depending on your DAC etc. With my current DAC and Ubuntu version I can only get DSD128 using DoP. I think direct DSD to 512 needs at least kernel 4.9 and possibly a firmware update (when Ubuntu 17.04 comes out I'll find out). Yes HQPlayer is proprietary, and expensive IMHO, and the GUI could be improved. But with Win 10 DSD512 is worth it - again IMHO.

- Richard.
 
TBH personally I have no real interest in using DSD or LPCM rates above 192k as I can't detect any difference they might give. So the limits you outline would be irrelevant for me as an end-user. My interest would just be intellectual curiosity.

Is the limit an ALSA one, or somewhere else in the chain?

I don't currently have any USB DACs that go above 192k so can't even test if an plain alsa 'play' would play out DSD as 'pretend LPCM' - which is what I understand DoP to do.

I *did* ages ago have a USB DAC/ADC to test that was meant to do DSD128 using DoP but it seemed to have problems even with plain LPCM. In the end, I sent it back and didn't report on it as I think it was faulty. For DoP the behaviour seemed to be that its use of the 'flagging' bits were in a different order than specified in the documentation I read. But by then the device was showing other problems...

Doing over/up/resampling, etc, isn't inherently difficult provided you have some filter values, etc, in mind. So an open source player doing this and tweakable by the user should be possible. Has no-one done anything like this for Linux?
 
TBH personally I have no real interest in using DSD or LPCM rates above 192k as I can't detect any difference they might give.
As high as 192?
In the case of HQ player as I understand it is more of a software version of (most of a) delta sigma dac rather than being about playing hirez files as such.
Even if it did make sense I do wonder whether it would be better using some sort of multi bit like most delta sigma dacs, but then you'd have to find a dac to accept the input.
 


advertisement


Back
Top