[analogmusic]

I have to say congratulations on achieving your dream of one million taps in 2016
 
I thought this was 5 to 10 years away...?

 

[AndrewOld]

   
I have already looked into that, and its possible that eliminating the ADC won't actually make a great deal of difference. The majority of the BOM cost is metalwork, FPGA, PCB and power supply. The ADC savings (analogue integrators and pulse array) is actually quite small and would probably be outweighed by the costs of producing two products.
 
I want as many people as possible to enjoy the benefits of M scaler tech, so we are trying our best to reduce costs.
 
Rob

Great! I hope that means that Davina will therefore be substantially cheaper than the Blu 2. The casework should be much less complex since there is no fancy hinged lid for the cd compartment, there is no cd mechanism or associated electronics, and if the ADC doesn't make a great deal of difference to the price then the Davina/M Scaler should be a good bit less expensive than the Blu 2. Here's hoping! All the best for the New Year.

 

[mmerrill99]

Congrats, Rob on achieving a transformational change in SQ with 512,000 taps & progressing it further to 1M taps
Loom forward to your further experiments in 2017
"That is the question I aim to answer with the Davina project - I will make 768 kHz recordings, decimate it down to 48 kHz, M scale it back up to 768 kHz, to create another file. So we will know for sure what these losses actually represent subjectively by comparing the two files."

Like you I am fascinated by auditory processing & psychoacoustics with respect to our audio reproduction systems.

It's fascinating to hear your experiences with refining timing & small signal accuracy in digital audio - both of which seem to be intertwined & of crucial importance to SQ.

One area you might find of interest & which is attempting to research the brain mechanisms involved in auditory processing is "Auditory Scene Analysis" or ASA. A good, recent (2014) overview of ASA is found here "The what, where and how of auditory-object perception"

Also lots of research on temporal coherence in ASA

 

[DaveRedRef-III]

Rob, have to say I really liked the presentation. You made a complicated subject accessible to lesser mortals. Thanks

 

[Crgreen]

So I can start saving, is it possible to provide a ball-park figure for Davina?

 

[Hubert H]

£7,995 seems to be Chord's current favourite figure

 

[Crgreen]

Yes, I saw that: the same price as the new CD transport, which would be rather disappointing. As has been pointed out, one would have thought it would be cheaper given that there's no costly CD mechanism and extra casework, the ADC doesn't seem to be make a big difference to cost, and apparently, Chord are trying to keep the cost down.

 

[Hubert H]

I wonder how much cheaper it could be though, if we remove the VAT and dealer mark-up (I think it's 40% for the big items) that leaves ~£4,750. If the removal of the mechanism and metalwork saves £500 (which I doubt) that gives an end price of £7,140.
 
Looking at the BLU2, I think I'll wait until Davina comes out although I may try to get a BLU on loan to test prior, I wonder why they didn't just put some extra inputs on the BLU? They did the same with the DSX1000, I'd have bought one if it had the BLU's dual BNC inputs - glad they didn't though, Dave is better

 
 
 

 

[Crgreen]

I suppose we'll have to wait and see how things develop. At the moment however, aside from the cost, getting a CD transport primarily for the m-scalar facility seems a bit of a sideways move.

 

[Hubert H]

Agreed, but so is getting an ADC for M-scalar to some extent. I've not used my BLU since I got a music PC this year and was holding on to see if an upgrade was available, it will be up for sale shortly.
 
Trouble is, we've bought into it, like it and are greedy for more...

 

[VerBla]

Reading all this about "M-scaler" and the BluMk2 could someone explain to me or point me towards an explanation by Rob about this M-scaler and what it does? Thanks! 

 

[Mython]

  Reading all this about "M-scaler" and the BluMk2 could someone explain to me or point me towards an explanation by Rob about this M-scaler and what it does? Thanks! 

 
Have you read the previous page?

 

[VerBla]

   
Have you read the previous page?

I did, but I kind of malfunctioned the first time. Just read through it again. So the M-Scaler basically means a WTA filter employed in a different way with 1M taps in contrary to the normal WTA filters used in the Mojo and Hugo which have a lower value taps?

 

[ecwl]

I did, but I kind of malfunctioned the first time. Just read through it again. So the M-Scaler basically means a WTA filter employed in a different way with 1M taps in contrary to the normal WTA filters used in the Mojo and Hugo which have a lower value taps?

There are many ways to think about digital filtering and taps. All are legitimate. But this is how I think about it. Most DACs need to upsample from 16/44 to 24/704 prior to playback. Imagine the original analog waveform during recording, the 16/44 are quantized signals from that recording. Most DACs cheat with short tap length filters. When they actually use their non-apodising filter or apodising filter to upsample to 24/704, because it's an approximation, they're a bit off from the original analog waveform. The frequency response/spectrum within the audible range is the same as the analog waveform. You would think that if the frequency response and spectrum are identical, it wouldn't matter. But according to Rob Watts, there is an audible difference because the waveforms are not identical. There are two ways to better reconstruct the original analog waveform in 24/704. One is to use a better digital filter like WTA so that with the same tap length, the upsampled waveform can converge on the original analog waveform better than other digital filters. Another way is to use longer tap length so that there is more calculations done so that the 24/704 waveform becomes closer and closer to the original analog waveforms.

So yes, the simple answer is Blu has longer tap length than DAVE which has longer tap length than Hugo/Mojo.

There are several consequences to WTA and longer tap length. Because the 24/704 is closer to the original analog waveform, you end up with much lower higher frequency noise because you can imagine when we say the poorer upsampling being further off from the analog waveforms, the difference is very small and when you translate that into the frequency domain, you're essentially getting less antialiasing artifacts and less high frequency noise.

Another consequence that I still don't fully understand is that transient timing and timbral accuracy improves with longer tap length. It's just what Watts and many DAVE owners hear. Watts calls this improved timing accuracy. I still struggle conceptually why this happens. I can imagine the amplitude of the analog waveform from longer tap length being more accurate except our ears don't hear amplitude, we hear frequencies. So why better timing? Not sure.

 

[Rob Watts]

 

I did, but I kind of malfunctioned the first time. Just read through it again. So the M-Scaler basically means a WTA filter employed in a different way with 1M taps in contrary to the normal WTA filters used in the Mojo and Hugo which have a lower value taps?

There are many ways to think about digital filtering and taps. All are legitimate. But this is how I think about it. Most DACs need to upsample from 16/44 to 24/704 prior to playback. Imagine the original analog waveform during recording, the 16/44 are quantized signals from that recording. Most DACs cheat with short tap length filters. When they actually use their non-apodising filter or apodising filter to upsample to 24/704, because it's an approximation, they're a bit off from the original analog waveform. The frequency response/spectrum within the audible range is the same as the analog waveform. You would think that if the frequency response and spectrum are identical, it wouldn't matter. But according to Rob Watts, there is an audible difference because the waveforms are not identical. There are two ways to better reconstruct the original analog waveform in 24/704. One is to use a better digital filter like WTA so that with the same tap length, the upsampled waveform can converge on the original analog waveform better than other digital filters. Another way is to use longer tap length so that there is more calculations done so that the 24/704 waveform becomes closer and closer to the original analog waveforms.

So yes, the simple answer is Blu has longer tap length than DAVE which has longer tap length than Hugo/Mojo.

There are several consequences to WTA and longer tap length. Because the 24/704 is closer to the original analog waveform, you end up with much lower higher frequency noise because you can imagine when we say the poorer upsampling being further off from the analog waveforms, the difference is very small and when you translate that into the frequency domain, you're essentially getting less antialiasing artifacts and less high frequency noise.

Another consequence that I still don't fully understand is that transient timing and timbral accuracy improves with longer tap length. It's just what Watts and many DAVE owners hear. Watts calls this improved timing accuracy. I still struggle conceptually why this happens. I can imagine the amplitude of the analog waveform from longer tap length being more accurate except our ears don't hear amplitude, we hear frequencies. So why better timing? Not sure.

That's a very good precis of the situation.
 
But to answer your question why transient timing improves with tap length I can illustrate the two extremes; a tap length of 1 tap (a NOS filter followed by analogue filtering) and an infinite tap length filter. Here is a couple of slides from my Mojo presentation:
 


 
Now this is just a simple illustration and it shows the two extremes one tap filter giving worst case 100uS or so timing error; and an infinite tap length filter reconstructing the transient perfectly. Somewhere in-between we will get acceptable levels of timing errors - but the only way we can test for this is to build long tap length filters, and change the parameters - tap length, oversampling rate, and algorithm - then keep listening - and that's what I have been doing for the past 20 years.
 
But what is really cool is that with Davina we can actually know for certain what the subjective losses are with 1M tap length filters. Ideally, the most powerful listening test is one where you can hear no difference from say 768>decimate 48>interpolate 768, and Davina will tell me how much difference we get in absolute terms. I will be publishing files - the original, and the decimated/interpolated version. In case the bandwidth limiting changes the sound (it will probably make it sound better) then I actually have three files - original, bandwidth limited, bandwidth limited/decimated/interpolated.
 
Sampling theory has nothing to say about bandwidth limiting - only that at FS/2 and greater it must be 0 output. I have already designed 300 db bandwidth limiting filters, so it will be very curious to see how these actually sound.
 
In 2013 a paper was filed in a physics journal, and this talks about the Fourier uncertainty, and the importance of timing. On this website, they have some samples where the signal has identical frequency response, but timing information has been destroyed, so try playing some of these tracks:
 
http://phys.org/news/2013-02-human-fourier-uncertainty-principle.html
 
The original paper is very interesting to read, but is not easy to follow. Now Fourier uncertainty is the timing problem characterized mathematically. Now I have always felt that we needed to minimize Fourier uncertainty by making sure the windowing function was greater than 1 second (this requirement was from listening to other problems) - and guess what - we only get windowing functions of greater than 1s with 1M 16FS filters.
 
Rob