Excluding hi-res and HD, why use a DAC with processing capabilities greater than 16 / 44.1?

[audiocat]

Hi Folks

 

We are seeing excellent digital - analogue - conversion. It's been done using 24bit or Delta output. Redbook doesn't need 24 bits but these dacs chips, power supply and pcb layouts are excellent. So Red book playback reaps the benefits and is along for the ride. These new dacs make mp3 sound good too.

[Miska]

Miska I think is looking at things from a time domain point of view and saying filter ringing should be audible as smearing of transients. Fokus I think is looking at things from a frequency domain point of view and saying for people like me who can hear to perhaps 15-16kHz, ringing should not be audible.

 

There are two aspects, one is audibility of the ringing as such, and another thing is it's implications as transient smear. The smear length of typical 1 ms is equivalent of one cycle of 1 kHz sine wave, while the ringing frequency itself is around the Nyquist frequency. It rings for many cycles - the overall smear time is length of the entire ringing period.

 

Like in a video, you may not notice content of a single frame, but if there's "motion blur" (content of previous frames leaking to next one) over 25 consequent frames you will certainly notice it. If you think that a frame coming out now still would have 1% of content from one second ago mixed in. If you have a stationary image, you don't notice anything, but if you have fast movement, it becomes apparent.

[JohnSwenson]

I throw up my hands. I'm completely out of my league when it comes to the science being discussed here.

 

Here's what I'd like to know. There seems to be a lot of excitement about FPGA's. Are they a more powerful platform from which to build an upsampling algorithm than a Core i7 in a PC or Mac?

 

A processor runs a program, a set of sequential instructions to perform a task. An FPGA (field programmable gate array) is a bunch of unconnectes gates (a "gate" performs a simple logical task, AND, OR invert etc). The user configures these gates how he wants them to be connected. They can be connected to perform arithmetic (add subtract multiply etc) comparisons, make decisions etc. One advantage over a processor is lots of things can be happening at the same time. For example the gates can be configured to be doing ten additions simultanneously, a processor would have to do 10 separate add operations, the FPGA can do then at the same time.

 

A trade off is that the processor is usually running faster than the FPGA, so it might be possible to do it in the same amount of time, depending on the clock rates etc. BUT the FPGA can be doing other things such as control decisions, format conversions etc at the same time. A processor would generally need separate threads for each and have interrupts and such to switch between control streams. It's an interesting engineering decision as to which is best for a particular problem.

 

Sometimes it's both, I have one DAC with both an FPGA and a processor, each doing what they are particularly good at.

 

For example reading a file of an NTFS disk is easy with a processor, there are code libraries you can just include and be done with it. That would be incredibly difficult on an FPGA. OTOH inputting a bunch of I2S streams, mixing them together, performing format conversion and sending out to a DAC chip, is not easy with a processor, but a piece of cake for an FPGA.

 

DSP is one of those tasks that can be done reasonably well with either.

 

 

 

John S.

[JohnSwenson]

For fun, I calculated that it would require roughly an Altera Stratix V FPGA to perform those calculations and that about 2000€/piece, so more that 6x as much compared to Core i7 4770K.

 

I did some calculation for a 512X upsampling filter and found an Altera FPGA that could do it for $300, either your filter is a LOT more complicated or Altera has cheaper FPGAs!

 

John S.

[Miska]

I did some calculation for a 512X upsampling filter and found an Altera FPGA that could do it for $300, either your filter is a LOT more complicated or Altera has cheaper FPGAs!

 

Well, my filters are quite complex and use very high precision, but another part that takes about the same amount of resources as the filter are the fairly complex delta-sigma modulators.

 

Core i7-4770K is 314€ here at the moment. And since I anyway have it, I don't need to buy it just for doing upsampling, it can do web browsing too and I can simultaneously use it to write this kind of posting here...

[Miska]

For example the gates can be configured to be doing ten additions simultanneously, a processor would have to do 10 separate add operations, the FPGA can do then at the same time.

 

For example Core i7-4770K can do 16 parallel calculations (like square roots for example) with 64-bit double precision floating point values running at constant 3.5 GHz speed and peak at 3.9 GHz. It can also calculate for example both sine and cosine simultaneously for the same value with single instruction (at 80-bit precision). More complex operations like sin, cos and sqrt at double precision FP take a lot of space on FPGA.

 

There are also various operations where you need to complete previous calculation because it's result is needed in the next calculation.

[Miska]

Well, my filters are quite complex and use very high precision

 

John,

 

I'm not sure about yours, but mine can go from 44.1k to 24.576 MHz, so it doesn't need to be simple 512x ratio. I always use one fixed frequency clock so the filters need to adapt to what ever is the input rate. Could be 64 kHz too, there are lot of pro-audio interfaces that can record at 32/64/128 kHz.

[sandyk]

John,

 

I'm not sure about yours, but mine can go from 44.1k to 24.576 MHz, so it doesn't need to be simple 512x ratio. I always use one fixed frequency clock so the filters need to adapt to what ever is the input rate. Could be 64 kHz too, there are lot of pro-audio interfaces that can record at 32/64/128 kHz.

Hi Miska

What fixed clock frequency do you use, and what brand/type (including PPM and phase noise specs) do you find gives you the best results ?

 

Regards

Alex

[Superdad]

Well, my filters are quite complex and use very high precision, but another part that takes about the same amount of resources as the filter are the fairly complex delta-sigma modulators.

 

Miska: I am glad that you and John are finally in the same virtual "room" together.

 

As you know, John and I (well, John's expertise, my marketing ideas and sponsorship) have been hatching plans for a sophisticated yet elegant DAC-chip-less DAC for a while now. And while some parts of Ted Smith's design for PSA are similar, John's ideas go further and avoid some limitations in several areas (can't go into those publicly).

 

One aspect of Ted's design which I'd like you--as a respected designer of SD modulators capable of running into multiple formats--to comment on is his choice to SD to a "one-bit" bi-level as opposed to doing a discrete x-wire/x+1 level multi-bit (I put it that way so others reading won't think I'm talking about PCM).

I think I know part of why Ted did it that way (though I should not presume without asking him), in that it makes it easy to then use video opamps fed into an LPF--to in effect have no output stage--and to not have a ton of parts.

 

But assuming one can (John can) feed a 22-28MHz 6-wire/7-level DSMed signal into a discrete summing stage>LPF for direct output, which way do you think--from a modulator design perspective--would be sonically preferable?: 5.6MHz "1-bit" or a higher rate multi-bit? (And yes, I know you would ultimately prefer a 9+wire/10+level discrete setup, but at some point cost becomes an issue for us; already considering a $250 FPGA for the SRC.)

 

In answering, feel free to assume your own DSM algorithms in each case. Not that we could approach the quality of your DSM (unless you license?). And maybe we can assume similar DSM orders for both cases: 7th-order for 1-bit "DSD"? 7th-9th order for multi-bit? We would like to experiment with lower order. Maybe mid-twenty megahertz rates will make lower order more acceptable?

(I may have some facts and concepts mixed up a little bit, so feel free to correct.)

 

Thanks and regards,

 

Alex Crespi

[Fokus]

Fokus I think is looking at things from a frequency domain point of view and saying for people like me who can hear to perhaps 15-16kHz, ringing should not be audible.

 

I am always speaking of and thinking in both domains, as they are the same.

 

Take a linear phase filter with pre-ringing, once more.

 

If the filter cut-off is in the audible range then the pre-ringing is audible as a separate artefact.

 

If you move the filter cut-off frequency upwards then this audible artefact disappears.

 

If someone can provide me with sample music(*) files of two >20kHz filters that are identical, except in their amount of pre-ringing and that can demonstrate an audible difference to a majority of non-naive listeners then I would welcome that. I have been looking and trying for ten years, with listener panels, but so far ... nothing. So I lost all interest, I cook my filters pragmatically (just as Miska, sacrificing some above 18kHz to get to a shorter filter kernel, and always ensuring sufficient alias suppression - yes, I only care about the ADC side of things), and I carry on, concentrating on things that really matter.

 

 

 

(* Or even just a killer test sound.)

[Miska]

If someone can provide me with sample music(*) files of two >20kHz filters that are identical, except in their amount of pre-ringing and that can demonstrate an audible difference to a majority of non-naive listeners then I would welcome that.

 

HQPlayer has number of filters where you can choose between linear and minimum phase versions. Both versions have exactly same frequency domain response. There is also one asymmetric variant that is between linear and minimum phase. With lot of content it works to the other direction too, since the filters are apodizing.

 

I use a minimum phase filter for most of my listening of pop-music and linear phase filter for classical music.

 

If we again look at the ringing source example:

 

And compare it to 4x minimum-phase upsampled version:

 

...most of the pre-ringing is gone.

 

(note that waveform is with dB vertical scale)

 

I'll see if I could make the test file I recorded downloadable. Those just tend to be big...

[manisandher]

FWIW, I've found the differences between the various filters in HQPlayer to be easily audible. For me, HQPlayer has been an indispensable tool in learning how various digital filters and noise-shaping schemes affect the sound. (Having a filterless NOS DAC capable of accepting the 24/768 rates output by HQPlayer has also been a big help.)

 

Mani.

[Jud]

FWIW, I've found the differences between the various filters in HQPlayer to be easily audible. For me, HQPlayer has been an indispensable tool in learning how various digital filters and noise-shaping schemes affect the sound. (Having a filterless NOS DAC capable of accepting the 24/768 rates output by HQPlayer has also been a big help.)

 

Mani.

 

And Mani, while I would certainly expect your DAC to be a big help, I can also say for everyone else's benefit that I was quite surprised by the differences I heard through a Dragonfly.

 

Again, I have some thoughts about slightly more rigorous experiments, particularly wrt transients.

[Paul R]

I am always speaking of and thinking in both domains, as they are the same...

 

Ummm no. Addressing just the partial quote above.

 

Mathematically an audio signal may be expressed (after suitable manipulation) in the frequency or time domains as equivalent expressions, but they are not the same thing in terms of human hearing. Misunderstanding this is what leads to so many arguments and issues.

 

In laymans terms, the time variation in a digital signal at conversion time to an analog signal, is in itself, a signal that a human can detect. It produces a second order artifact of the playback of the original signal. This is why some kinds of jitter can be heard. Even though the jitter itself presents a vanishingly small signal, it acts to modulate the playback and that modulation is audible. And also only really understandable when viewed in the time domain. (Same idea as FM transmissions by the way...)

 

In a different sense-the reason things are switched from the time domain to the frequency domain is simply because the math is easier in one domain than the other. Some of the properties and effects of the signal, filters, transmission medium, reproduction, conversion, and so on are also easier to understand in one domain or another. However, the folks that believe they can totally underdstand the entire picture from only one domain or another are sprouting rubbish.

 

You are welcome to hold an opposing opinion, as is anyone else. But the fact remains one needs to understand both domains to even begin to understand all the phenomena associated with audio reproduction. Failing that, one is always easily led astray from reality- either by hawks trying to sell $30,000/meter quantum enhanced speaker cable, or hawks trying to convince one our digital reproduction is already perfect and cannot be improved upon. Both ideas of course, being provably false to the satisfaction of most reasonable people.

[wisnon]

Okay- I am a little bit confused here. The Loki I am playing with here does not have any such issue. Perhaps because it only eats DSD64 and never switches rates? No switch, no pop. I will endeavor to get it to click or pop today- that is a show stopper issue for me. I have also not heard of that issue with the Bifrost. Seems odd that those two would put such a device out.

 

I know the LampizatOr will click an pop, but the designers are open about that and have some reasoning behind it. It is why such a fine sounding DAC is not on my radar.

 

The Lampizator just started using a new PCB for DSD...no more clicks and pops in any situation. Indeed, from early December with the USB flash update, clicks are greatly reduced in volume already...now totally gone.The Lampi DSD is the REAL direct stream....