or Connect
Head-Fi.org › Forums › Equipment Forums › Dedicated Source Components › iFi Audio Nano iDSD discussion + impression
New Posts  All Forums:Forum Nav:

iFi Audio Nano iDSD discussion + impression - Page 47

post #691 of 1720

And, PCM tests done. Equal wow.

 

120khz, 144khz, 168khz and 184khz heterodynes into the 32bit 384khz PCM test file,

120khz, 144khz, 168khz and 184khz heterodynes out of the iFi iDSD on the spec analyzer.

 

Obviously none of these would be possible if the iDSD was doing 24/192, as it's below 192's 96khz nyquist limit.

Absolutely no way it could be doing anything but 384khz, especially with the 184khz clean tone returning.

 

That also means that the iFi iDSD has at least a 0-184khz frequency response. Jeez.

I could use this thing to transmit radio if I wanted to.

 

Also "since I was in there",  looks like the "minimum phase" filter shifts the DSD noise ramp up a bit further from the audible band without really affecting audible band frequency response. So yes, "minimum phase" is a nice place to be, filterwise.

 

Impressed. REALLY impressed.

 

Do you guys at iFi have a back room filled with actual bona-fide wizards? Because the word "unbelievable" is what lead me to term it that when I was a skeptic a few hours ago. Now I can only use it in the colloquial sense. 


Edited by michilumin - 6/29/14 at 4:30pm
post #692 of 1720
Quote:
Originally Posted by michilumin View Post
 

And, PCM tests done. Equal wow.

 

120khz, 144khz, 168khz and 184khz heterodynes into the 32bit 384khz PCM test file,

120khz, 144khz, 168khz and 184khz heterodynes out of the iFi iDSD on the spec analyzer.

 

Obviously none of these would be possible if the iDSD was doing 24/192, as it's below 192's 96khz nyquist limit.

Absolutely no way it could be doing anything but 384khz, especially with the 184khz clean tone returning.

 

That also means that the iFi iDSD has at least a 0-184khz frequency response. Jeez.

I could use this thing to transmit radio if I wanted to.

 

Also "since I was in there",  looks like the "minimum phase" filter shifts the DSD noise ramp up a bit further from the audible band without really affecting audible band frequency response. So yes, "minimum phase" is a nice place to be, filterwise.

 

Impressed. REALLY impressed.

 

Do you guys at iFi have a back room filled with actual bona-fide wizards? Because the word "unbelievable" is what lead me to term it that when I was a skeptic a few hours ago. Now I can only use it in the colloquial sense. 

Welcome to the iFi fan-boy club with the rest of us!

post #693 of 1720
Quote:
Originally Posted by KmanChu View Post
 

Welcome to the iFi fan-boy club with the rest of us!

 

I am one, now, absolutely. Now if I can only find a place in the US to preorder. (Seriously.) Spent the latter part of today shooting out emails trying to see if anyone will be carrying part of the first allocation.

post #694 of 1720

I also have the Loki but I find that my Nano iDSD sounding more like analog tape, like my old Barclay Crocker open reel tapes I used to have. I miss that sound and the iDSD just brings that back for the most part. I think that IFI is on the right path with products that would be interesting for a  wide range of users and tastes.  It may be possible for IFI be a key player in getting people more interested in DSD recordings? The small high end audio store in Boulder on 30th street had three iDSD nano's for sale when I went there for a CAS meeting. 

post #695 of 1720

Sorry if I was a bit harsh, I am glad you have discovered for yourself that they do not lie about their products, and that you enjoy their products. Enjoy and Happy Listening :)

post #696 of 1720
Quote:
Originally Posted by michilumin View Post

And, PCM tests done. Equal wow.
That also means that the iFi iDSD has at least a 0-184khz frequency response. Jeez.
I could use this thing to transmit radio if I wanted to.

I think your approach is quite ridiculous.(edit: I don't think that anymore, I guess you are right) The range humans can hear is 20-20k Hz, so I would consider it quite normal to filter anything above 20k Hz regardless of sampling rate supported. There are reasons to use higher sampling rate, but widening transmitted range is the last one on the list. Most important thing that you gain when sample rate is higher, is that you can use analogue filter of lower order, i.e. it doesn't have to filter everything sharp after 22kHz but it can fall off gently to let's say 56kHz or more (edit: and because of that you will have in the output frequencies higher than 22kHz, and that means I just proven my first statement wrong). If you look at DSD1793 white paper there are plots of filter responses. You will also find there that DSD mode is implemented as analogue FIR filter. You will also find no statement about that chip supports max DSD64, but instead you will find N x fS, where they say N = 64 is default application, but they don't say you can't use bigger N. It's like with processors, they say 2.8GHz, but you can over-clock that to 4GHz if you like to. But if you do so you have to guarantee stable power supply and proper working temperature, which could be costly thing to do. I presume you didn't have much to do with electronic engineering, and thus you say things that create cloud of misunderstanding.

To iFi: Can you confirm that chip is running with N = 128 and N = 256, so that michilumin is kept happy?
Edited by koolas - 6/30/14 at 2:46pm
post #697 of 1720
Quote:
Originally Posted by ClieOS View Post

The latest firmware can be found here: http://ifi-audio.com/ifi-xmos-firmware/

I'm testing this firmware second day now, and must say I'm impressed. First, it did work perfectly out of USB 3.0 socket on my laptop, now I'm running in the office Foobar2k and I tested foo_output_dsd in native ASIO DSD256 mode using SDM Type D (FP32) PCM to DSD conversion algorithm. Must say sound quality is really great, but I think PCM playback has also been improved, because before I would stat that PCM converted to DSD sounds better than PCM played as PCM, and now I have hard time figuring out which one tastes better to me.

Good job iFi! rolleyes.gif

BTW. Now I can switch between DSD and PCM without device hanging-up.
post #698 of 1720
Quote:
Originally Posted by koolas View Post


I think your approach is quite ridiculous. The range humans can hear is 20-20k Hz, so I would consider it quite normal to filter anything above 20k Hz regardless of sampling rate supported. There are reasons to use higher sampling rate, but widening transmitted range is the last one on the list. Most important thing that you gain when sample rate is higher, is that you can use analogue filter of lower order, i.e. it doesn't have to filter everything sharp after 22kHz but it can fall off gently to let's say 56kHz or more. If you look at DSD1793 white paper there are plots of filter responses. You will also find there that DSD mode is implemented as analogue FIR filter, if you know what that is. You will also find no statement about that chip supports max DSD64, but instead you will find N x fS, where they say N = 64 is default application, but they don't say you can't use bigger N. It's like with processors, they say 2.8GHz, but you can over-clock that to 4GHz if you like to. But if you do so you have to guarantee stable power supply and proper working temperature, which could be costly thing to do. I presume you didn't have much to do with electronic engineering, and thus you say things that create cloud of misunderstanding.

To iFi: Can you confirm that chip is running with N = 128 and N = 256, so that michilumin is kept happy?

I disagree that michilumin's response is ridiculous - quite to the contrary. http://en.wikipedia.org/wiki/Heterodyne What he did use was a viable way of testing if nano iDSD actually does what iFi is claiming. The mere technique he used indicates that he has had "more than some" to do with electronics. Although the way he went first about claiming nano can not do as advertised was a bit harsh, I appreciate that after testing /proving for himself he admitted he was wrong - kudos for that. It certainly is a confirmation/second opinion from a third party - which can only benefit all concerned.

 

I would disagree with any audio system that is limiting frequency response to 20 kHz - particularly so steep as redbook CD with  a default "nothing beyond 22.05 kHz". It is one of the reasons why DSD, even DSD64, sounds more natural than PCM - it never needs to filter that steep as PCM. Each time DSD sampling goes up, so does the attainable frequency response. It is a tradeoff how far up in the frequency response one wants to go vs the outband HF noise inherent to the DSD. That is why practically any DSD DAC ever made use user selectable analog output filtering - so that each and everyone can choose what suits his/hers equipment/ears best. For example, Stax amplifiers driving a single pair of HPs can start producing noise with DSD64 recordings if driven to close to 0 dB - because ultrasonic noise from either DSD or SACD ( only different forms of DSD64 ) can exceed the capabilities of Stax amps in the ultrasonic range. That holds true for SRM1MK2 driving a single pair of Lambdas; lower powered Stax models may well begin noising at lower level - and driving two or more pairs of HPs in parallel will definitely produce trouble. If this occurs, in case there is user selectable filtering for DSD, a steeper filter has to be selected, as amplifier does not support out of audible band noise well enough. The whole point in DSD512 ( provided in micro iDSD) is in the fact one could (almost?) get rid of the output filtering, which always limits the quality of pulse response, while keeping the HF noise of DSD low enough. I am really interested in measured frequency response of micro iDSD with DSD512 - because it could in theory and practice reach "essentially flat", say - 3dB at approx 200 kHz . Depends on the filtering ( if any ) selected.

 

Similar frequency response considerations for the DXD - theorethically, up to approx 350 kHz. Again, tradeoff for gentler filtering than necessary for redbook CD can be aplied - and I would be interested in measurements for the choices iFi made. For example, Korg DSD recorders, when recording PCM 192/24, are flat ( ADC + DAC) to just above 80 kHz - and not to 96 kHz theorethical limit - and then steep filtering. By 96 kHz, the signal is next to nonexistant. DSD64, although attenuated, goes well past 100 kHz .

 

My rant "a couple of posts ago" regarding DSD512 has to be viewed in light what it means on the recording side and consequences it will have on the support required for playback.  It means piles and piles of money in order to provide true DSD512 recording with all the benefits it offers - the first being extended frequency response well past 20 kHz.

 

Nothing against iFi providing it in  micro iDSD. Priciples apply equally to the nano iDSD - with the limitation it is 2 times slower and less capable in everything - but both are incredible good value.


Edited by analogsurviver - 6/30/14 at 6:30am
post #699 of 1720

Hi All,

 

Disclaimer: Before we proceed, we would like to first apologise. We cannot give out the full details/answers to all of your technical questions relating to our designs. As you know, AMR/iF’s technical proficiency is quite idiosyncratic and that’s what makes iFi different from other audio manufactures. Notwithstanding, we always try to disclose as much as we can.

 

We would first would like to thank Michi Lumin for taking time on her Sunday, going to the Lab and spending time (and time is money) to perform all the tests and to go public with the results. As is our perennial position, measurements are not the end all and be all. It pays to be circumspect (even though they support iFi’s contention). Always listen and judge for yourself that is all the proof you need.

 

Michi, you seem to have some nice test tones or test tone generator for DSD, what do you use?

 

We were prepared to run a set of tests today and post the results, but this seems redundant now. Less work is always appreciated around here. There remain one or two questions to be addressed, though.

 

Please understand that we will not divulge details how things work on a technical level, we spend enough time figuring things out for ourselves. What we can give out has been done over on the micro iDSD Crowd-Design thread – including chipset implementation (read: uprated engine) and XMOS programming (read: uprated gearbox) as both are needed, so don't just focus on the former.

 

But the question remains, do we “overclock” the DSD1793 and why the DSD1793?

Addressing the latter first, one of the reasons we can in fact execute DSD512 and double speed DXD (on the iDSD micro) and DSD256 and DXD on the nano is precisely because of that old dusty chip. It has a sound quality that others have described as “analogue”, which is PRECISELY what we at AMR/iFi strive for.

 

Our ultimate goal is sound that has all the good bits of digital (low noise, wide bandwidth etc.) but with the sonic signature of “analogue”. There is enough “digital” sounding gear out there, no need to add even more.

 

 

This brings us to question number 2, do we “overclock”? In the sense of running higher clock-speed than stated under “maximum ratings” allowed for the DSD1793, the answer is a categorical “not on your nelly”, we do not overclock; we do not run the chip faster than it can go; we do not risk overheating; dropped bits; system crashes and all that stuff you get with “overclocking”.

 

You wont find this in any iFi product.

 

However, it is true that we run clock speeds other than those suggested by BB (now TI) and we use modes and settings other than those suggested by BB/TI. The modes and operation settings and limits are of course not publicly documented by TI/BB. One needs to refer to some confidential documentation regarding the internal silicon design of the chip (able to converse with the actual chip designers themselves, not the marketing guys who drafted the datasheets helped). While this was of much assistance, it still took us quite a while and it needs considerable custom-coding on the XMOS side as well, to produce data that the DSD1793 can accept correctly and decode at those speeds.

 

 

How does this compare to the most obvious alternative solution, the ESS platform?

There are many off-the-shelf USB solutions and connecting it to an ESS DAC Chip that figures out by itself what format and sample rate it receives and configures itself suitably makes live very easy. Come to the office at 11, break for lunch at 1, come back at 4, down the pub for happy hour at 5:30 and we still designed a world class DAC today.

 

So is it a lot simpler? Yes, but we are now face with limitations (tech limitations, sound limitations) built into ESS’s silicon. We were not willing to accept these limitations. We want to deliver a Supercar DAC (read: no speed limits).

 

 

Do we have a back room filled with actual bona-fide wizards?

Well, there is this Harry character who works here and who claimed he went to Hogwarts after his 7th Guinness, but we don’t take him very seriously on that count. He does have a strange soldering scar on his forehead though (a soldering iron is not a magical wand!).

 

No, there is no magic to this, just blinking hard graft, attention to detail and meticulous and extended testing of parts we work with, plus recognition of the wisdom of recently sainted Bob Pease who said “Datasheets never include the part you really need to know”.

 

From the iFi tech team

 


Edited by iFi audio - 6/30/14 at 8:26am
post #700 of 1720

biggrin.gif
post #701 of 1720

 


Edited by Mr Creosote - 6/30/14 at 12:22pm
post #702 of 1720
Thread Starter 
Quote:
Originally Posted by Mr Creosote View Post
 

...

 

Err... what?

post #703 of 1720

iFi team, what do you think of AKM's new high end chip, AK4495SEQ? I really hate the digital sound signature of Sabre chips so I can see you guys are very good at choosing the right chips for your products. AKM says this is their best chip, have you tried it or are you thinking of using it for any future product?

post #704 of 1720
Just testing DSD playback under Linux, and apparently it just works (with latest iFi firmware) with HQPlayer.
post #705 of 1720

"Datasheets never include the parts you really need to know."

 

So, I've learned a valuable lesson on this.  And - folks, please, I need to implore. This isn't about it "measuring well vs sounding well".  I just, okay, well, let me back up a bit.

 

I do indeed work with the Super Audio Center, making SACDs and, specifically, helping engineer Sonoma. That will beget a boo and a hiss from some, and an "Oh, OK" from others. I understand - we're trying to make it better and trying to get it to do the higher DSD rates. Really. That's my "stake" in the game.

 

However, if that could be called a "night job", my "day job" is working with Software Defined Radio and high frequency digital signals.

 

And then, paired with my DSD alter ego, I'm kind of always thinking about how this fits in with audio. So, I do apologize that that's what's flying through my mind most of the time.

 

So to answer AMRs question: I have a whoooole lot of scientific acquisition ADCs and signal generating DAC setups in the lab, to test into the GHz range. I've got DSD sweeps at least for DSD128 up to 150khz, and the ability to generate tones and heterodynes up to 500khz for up to PCM 768khz. When you get into the realm of sampling and reproducing RF, this kinda stuff is needed. Most of this is my own software "test harness", some of it comes from RF engineering, some of it from audio.  I'm currently in a "chipset mess" myself, dealing with two high frequency ADCs that aren't doing what they're supposed to. So yeah, that's where my head's been at.

 

Please don't kill me. I'm not just talking about measurements.

 

I had been familiar with the DSD1700s in the Meitner DACs, and watched TI/BB's development into the DSD1791/DSD1793 chipset (same thing, different flavor; SW vs HW control), and later on into the DSD1792/1794 chips, and most recently, the PCM1795.  (Which, I will admit, has good 'gozinta' specs, but not so great 'gozouta' specs. And sounds kind of dull.)

 

Funny thing, out of those setups, the old, dusty DSD1700 in the Meitners always sounded ... a little bit better to me. According to the spec sheet, they shouldn't.  In sound, they just did. So trust me, I am no stranger to the disparity between measurements and sound.

 

But my concern wasn't "is it good, or is it bad" - for most people, that's all they should have to worry about.  For me I really wanted to know if "this new Nano thing" really could "exhibit" DSD128 and DSD256, and, it turns out, indeed-it-do. 

 

To determine that, "true or false", not qualitative - a "benchy, measure-ey"  test was needed. Again, a "litmus test", Y or N, not a qualitative test. Needless to say that "while I was in there", I noticed that the Nano performed... impressively. Please understand: Performing a heterodyne test near Nyquist frequency is a legitimate way to tell what sample rate a device is operating at, at least "positively". It's not a "measurement", it's a binary check. If there's not a filter in the way, it'll tell you.  If you put a 180kHz heterodyne into a system running at 24/192kHz PCM, you'll get either nothing, or a real mess, out. However if you do get the same clear signal you put in, out of it, then, that's proof that it is running at that rate.  So I'm not suggesting the "quality" of such a device should be measured, this was more of... exploratory surgery.

 

And now that I think of it, after all this mess - there's more oddity in the mix: The old DSD1700s can do DSD128.They're not supposed to. The old DSD1700s can almost do whatever you tell them to. Sorta. I mean the specs sure don't say they can - they were made in a time when if you said "DSD128" you'd get a blank stare. Or if you even said "PCM 24/192" you'd maybe get a tilted-head glance. So, really, AMR basing these DACs on the "slight" evolution of those chips, the DSD1793, makes a whole lot of sense now.. 

 

Basically, I had my eyes glued to the damned TI/BB spec sheet.  And was doing what a lot of engineers tend to do after being mired in projects after a while: not thinking outside of the parameters. What AMR said about needing to know what went on inside the silicon, and even knowing some of the chip developers, all makes a whole lot of sense to me now.

 

I really feel that my approach here on head-fi was wrong. I meant to be incredulous but not outright denying, and I took that too far.  Trust me, I feel bad. I had to take a few days off listening to my Nano because of what transpired here.  And i'll probably go back to read-only mode after this.

 

But it was based on a cynicism that I've gathered up dealing so often lately with "promised functionality" not being "delivered functionality". And I am sorry that that colored this. And I need to put it in check.  When I heard "AMR runs the DSD1791/1793 in a secret, unsupported manner", my first reaction was "Oh come on."

 

But, after testing it, and finding out that, indeed, it does, I gotta say, thanks for doing engineering "the right way" vs the "confined" way. I always preached that, and this week I found myself caught up in it.

 

Not good. But maybe there's a reason for everything that transpires. Think outside the box, and all that. AMR is certainly doing that.

New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Dedicated Source Components
Head-Fi.org › Forums › Equipment Forums › Dedicated Source Components › iFi Audio Nano iDSD discussion + impression