[Whazzzup]

nice work, thanks

 

[JaZZ]

  Yes the ADC anti aliasing filters have a tap length requirement, and the requirements are "fairly" simple - linear phase, aliasing set to a level that is no longer audible. But if you set that to say -144 dB (24 bit accuracy), you need about 2K taps or more - which is much greater than normally employed. ADC chips do have significant aliasing, and normally the chip designer says its OK as the errors are at say 21 kHz. What they fail to realize is that that is audible, as it will change the time domain linearity, which is audible.
 
Now the mathematics is pretty clear - a DAC interpolation filter needs an infinite tap length, infinite oversampling FIR filter to perfectly reconstruct the original bandwidth limited signal. So I will keep pushing the tap length, and oversampling rate until one can no longer hear the difference. For the ADC filter, the maths simply states that at >= FS/2 the level is zero, i.e. there is no signal at 22.05 kHz and greater for 44.1 kHz. The fun will start when I actually listen to these filters, and find out what stop band attenuation is really needed. But even if 200 dB was good enough, then that would not need enormous tap lengths.
 
Originally Posted by Rob Watts 
 
The 164,000 taps and SQ is a long story, something I will perhaps talk about closer to launch.

 
I take it you're talking about 44.1 kHz. And I must say the 26,368 taps implementation in the Hugo sounds great. How about higher sample rates, say 96 or 192 kHz? Is the filter steepness the same as with 44.1 kHz? Moreover: Can it be expressed in dB/oct.? And finally: Would 164,000 taps (or more) definitely make Hi-Res recordings obsolete?

 

[Rob Watts]

   
I take it you're talking about 44.1 kHz. And I must say the 26,368 taps implementation in the Hugo sounds great. How about higher sample rates, say 96 or 192 kHz? Is the filter steepness the same as with 44.1 kHz? Moreover: Can it be expressed in dB/oct.? And finally: Would 164,000 taps (or more) definitely make Hi-Res recordings obsolete?

No, dB/oct is for IIR filters, FIR filters have much faster attenuation per Hz, and its not a simple number.
 
Clearly, more taps = better timing accuracy which will vastly reduce the gap from high rez recordings. But if the master recording was 96kHz, and they used a poor sample rate converter (SRC) to get 44.1 kHz then no amount of increased taps will solve the SRC distortion and noise. So I guess my answer is that it depends upon the recording. Still anybody that has a Hugo will know - 16 bit 44.1 is capable of exceptional SQ and musicality.
 
That said, decimation (if done correctly) with extremely long tap length WTA should be completely transparent - I should know the answers to this with the pro audio ADC currently under development after the listening tests planned for it.
 
Rob

 

[Sam Edwards]

  I should know the answers to this with the pro audio ADC currently under development after the listening tests planned for it.
 
Rob

So I was right!
-se

 

[JaZZ]

  No, dB/oct is for IIR filters, FIR filters have much faster attenuation per Hz, and its not a simple number.
 
Clearly, more taps = better timing accuracy which will vastly reduce the gap from high rez recordings. But if the master recording was 96kHz, and they used a poor sample rate converter (SRC) to get 44.1 kHz then no amount of increased taps will solve the SRC distortion and noise. So I guess my answer is that it depends upon the recording. Still anybody that has a Hugo will know - 16 bit 44.1 is capable of exceptional SQ and musicality.
 
That said, decimation (if done correctly) with extremely long tap length WTA should be completely transparent - I should know the answers to this with the pro audio ADC currently under development after the listening tests planned for it.

 
Since 26,368 taps is still quite abstract to me, I was trying to get a picture of the filter characteristic from a passive crossover network perspective. I imagine it reaches –144 dB well within an octave (maybe 400 dB/oct. as a rough estimation). So the Hugo's amplitude response has a drop-off of 0.000... dB at 20 kHz?
 
While I'm aware of the importance of a filter sharpness and steepness close to infinity to cover all aspects of the Nyquist theorem with a sampling rate of 44.1 kHz (due to its close proximity to the audio band) – how does it look with a sampling rate of 96 kHz? Does the filter still have to be as sharp and steep as possible for optimal sound quality? And at which frequency is it with the Hugo, then?
 
I'm still amazed how good it sounds with all recordings. This despite the fact that – before digitization – all of them (at least the non-hi-res) had to go through an anti-aliasing filter certainly not of the Hugo's caliber. Hence the damage is already done, so to speak.

 

[Dobrescu George]

not necessarly. 
 
to make an analogy, most video data is streamed at 24 fps. there are very powerfull and resource hungry algorythms that can over-process every image, at full resolution to make everything go smooth at 60 fps, or even 120 fps. even for anime, which were not made to be seen at such high framerates. it is amazing. so for audio, something similar can be done. interpolation should work, even if the anti aliasing filter destroyed lots of data in the ADC stage. It is harder to wirte the interpolation algorythm for audio, on the other hand, as the data does not miss seconds of data, but much finer parts of a second, which are harder to determine, than it is to combine two photos in a video.

 

[JaZZ]

I don't know much about video, but interpolation for «natural motion» only works so much – depending on the scene and the implementation. And in analogy to this that's where the Hugo excels. In fact interpolation of digital audio data is common technique, it's called upsampling or oversampling and helps with low-pass filtering to prevent aliasing. Other than video interpolation it is even (theoretically) able to reconstruct the waveform to a degree that it's undiscernable from the analogue original.
 
However, I wasn't referring to digital data, but to the indispensable low-pass filtering of the analogue signal befor conversion to digital data. I don't really know how it's done, but imagine it to be far from perfection in the sense of the hugo's «reconstruction filter». That's why I wonder how the Hugo can nevertheless make so much of a difference compared to most other DACs, since the signal is already contaminated with a «bad»filter. Of course more recent recordings may use higher sampling rates from the start, and downconversion in the digital domain would do less harm than the former scenario.

 

[Dobrescu George]

  I don't know much about video, but interpolation for «natural motion» only works so much – depending on the scene and the implementation. And in analogy to this that's where the Hugo excels. In fact interpolation of digital audio data is common technique, it's called upsampling or oversampling and helps with low-pass filtering to prevent aliasing. Other than video interpolation it is even (theoretically) able to reconstruct the waveform to a degree that it's undiscernable from the analogue original.
 
However, I wasn't referring to digital data, but to the indispensable low-pass filtering of the analogue signal befor conversion to digital data. I don't really know how it's done, but imagine it to be far from perfection in the sense of the hugo's «reconstruction filter». That's why I wonder how the Hugo can nevertheless make so much of a difference compared to most other DACs, since the signal is already contaminated with a «bad»filter. Of course more recent recordings may use higher sampling rates from the start, and downconversion in the digital domain would do less harm than the former scenario.

you are totally right! of course, in video it is simple to construct motion and frames based on another frames. 
 
i listened to some old iron maiden, and the sq was great, but it lacked some of the quality of newer recordings, so i think that you are right. hugo can do just so much it does to what has already been badly recorded.
 
let's hopa that a ADC from rob comes out, and many bands will use it!

 

[MT0227]

   Still anybody that has a Hugo will know - 16 bit 44.1 is capable of exceptional SQ and musicality.
 
Rob

 
I will certainly attest to this.  My DMB studio library is 16/44 FLAC and it sounds fantastic.  There's no substitute for excellent mastering and production.

 

[Jalo]

Very interesting discussion. Since the tap length is what Hugo's claim to fame, or at least one of the important one. I hope Rob can enlighten us the non dsp folks what is tap length, we understand that it is something the longer the better. Is it also having something to do with the timing domain? Is it kind of like sampling rate but happened after the adc stage? Why is it better as it get longer? How does more taps help with better sound? Please see if you can use lay language to explain. Thanks.

 

[JaZZ]

I'm not really an expert, but an important background is the Nyquist(-Shannon) theorem. For a bandwidth of 22 kHz you need a digital sampling rate of at least 44 kHz to get a proper reconstruction of the original analogue signal, and at the same time it needs a sharp, steep (low-pass) filter at the upper end to prevent aliasing. The theorem is actually only valid for a filter with infinite sharpness and steepness, so in reality the 44.1 sampling rate is hardly capable of reproducing the original signal. Except for the case where you get relatively close to the filter ideal.
 
Filters, especially sharp filters, create inadequacies in the time domain: crippled transients, «ringing» near and on the filter frequency. The theoretically ideal filter, though, would concentrate the whole transient corruption at the filter frequency and there ring infinitely – without any harm to the music, since it's outside the audio band (~22 kHz). At least that's how I understand it.
 
I get it that the number of taps represents the complexity and sharp-/steepness of the low-pass filter. So the Hugo's low-pass filter – also called «reconstruction filter», since it's part of the signal reconstruction according to the Nyquist(-Shannon) theorem – is close enough to the ideal to result in a remarkable improvement of the sound compared to most conventional digital/analogue converters.
 
I hope Rob will chime in and correct some possible inadequacies of my interpretation.

 

[Dobrescu George]

@JaZZ the read was truly hard to understand. i think that if somebody asked what these taps are, it will be much harder for them to understand after reading that.
 
now that it comes to it, i think that i have the wrong definition of what the taps in hugo mean.

 

[JaZZ]

Actually it's quite simple.

It's like in the speaker world (which I'm more familiar with): The sharper/steeper the filter, the more complex. A first-order filter consist of just a serial capacitor (high-pass) or inductor (low-pass). It makes for a drop-off of 6 dB/oct. A second-order filter consist of a serial capacitor plus a parallel inductor or vice-versa. It makes for a drop-off of 12 dB/oct. A third-order filter consists of two serial capacitors plus a parallel inductor between the two (high-pass) or two serial inductors plus a parallel capacitor. This results in a drop-off of 18 dB/oct. A fourth-order filter consists of two serial capacitors plus two parallel inductors after each of them or vice-versa. It results in a (final) drop-off steepness of 24 dB/oct.
 
I would think that the above could be seen as something like an equivalent to 1-4 taps.

 

[MT0227]

   i think that i have the wrong definition of what the taps in hugo mean.

 
For me, they're what I do to my desk with a pen; but can't hear because I'm listening to music via my Hugo.

 

[Whazzzup]

 
 i think that i have the wrong definition of what the taps in hugo mean.

For me, they're what I do to my desk with a pen; but can't hear because I'm listening to music via my Hugo.

Brilliant.
Jazz thanks for that anyways, it was kinda like a dream world where I'm nodding my head but have no clue why. The closest to understanding what I don't understand ever.
does make the Monteverdi choir renditions of some Bach im listening too on my hugo sound fantastic I guess.