Filtering

[christoffe]

http://www.phasure.com/index.php?topic=3146.msg34305#msg34305

Hi Peter,

stupid/dumb question: there are custom filter settings marked with (NOS)
-especially for NOS1 users?

Joachim

[PeterSt]

Hi Joachim,

No, not especially for NOS1(a) owners. However, it is obvious that the NOS1(a) is the only D/A converter which is NOS/Filterless (not Oversampling and no Analogue Filter) with an input possibility of 32/768Khz (in effect 24/768). This means that :

a. The 705600 filters can not be used by any other D/A converter;
b. The DAC really doesn't touch the sound herself.

Ad a and b.
Which thus means that the NOS1(a) is the only DAC that can apply the 705000 filters optimally and that what I have measured over here will apply to all of you owning this DAC.

So yes it should be a bit special but in the end everybody can use the filtering and as how it was/is with the good old Arc Prediction it will overrule to some extent what's (filtered) in other D/A converters.

I hope it is clear a bit !
Peter

[dsm]

Hi Peter,

After seeing this thread, I guess I am wondering why there is a 705600 filter and a 705600 (NOS) filter. I see they have different settings but why are some labeled (NOS)?

Thanks,

David

[xp9433]

No, not especially for NOS1(a) owners. However, it is obvious that the NOS1(a) is the only D/A converter which is NOS/Filterless (not Oversampling and no Analogue Filter) with an input possibility of 32/768Khz (in effect 24/768).

Peter

Can you please explain a little further, for a technical digital novice like me, how the NOS1 differs from something like the budget iFi iDSD Micro that will accept 705.6 and 768 PCM input. The iFi doesn't upsample or down sample that input as far as I understand.

Thanks
Frank

[PeterSt]

After seeing this thread, I guess I am wondering why there is a 705600 filter and a 705600 (NOS) filter. I see they have different settings but why are some labeled (NOS)?

Hi there David,

Well, Doh ! I only see now that hese "NOS" filters should have been removed !!
Or at least that is what I think, because these were the mere rough trials (with measurement but not with an explicit THD target); possibly I told somewhere that the old ones were still present for comparison and that I left them in on purpose.
But I will remove them now, for the next version.

Thanks a lot for bringing this to my attention,
Peter

[PeterSt]

No, not especially for NOS1(a) owners. However, it is obvious that the NOS1(a) is the only D/A converter which is NOS/Filterless (not Oversampling and no Analogue Filter) with an input possibility of 32/768Khz (in effect 24/768).

Peter

Can you please explain a little further, for a technical digital novice like me, how the NOS1 differs from something like the budget iFi iDSD Micro that will accept 705.6 and 768 PCM input. The iFi doesn't upsample or down sample that input as far as I understand.

Thanks
Frank

Hey Frank,

First off, yes, I now recall from a few months ago that iFi was going to support 768 input. So apparently this is now the case ? OK, then my remark about "the only one" out of further context is not correct any more.

NOS = Not (or Non) Over Sampling.
Over Sampling = A higher internal (say in-chip) sampling speed than what the (music) material requires; a technical thing. This is what SDM (Sigma Delta Modulation) chips do. They oversample (at least) 64 times and next use an 1-bit scheme. What happens net with such a chip is that the internal sampling speed is 2.8MHz (64 x 44.1KHz) at a bit depth of 1. And/thus also : the original PCM stream (PCM assumed) was converted from e.g. 16 bits to 1 bit;
It has become "DSD" ...

When the digital filtering is applied how I adhere it, then the material (like 16/44.1) is Upsampled to a higher sampling rate so that samples can be injected which imply a better granulate audio stream (noise gets less because "step size" gets less).
Upsampling : The audio stream is made a higher sampling rate and while doing that bits are added;
Each 2x Upsampling step requires one additional bit (or otherwise things will be out of "balance"). Thus e.g. 16/44.1 upsampled to 88.2 needs to be 17/88.2 (the 17 is "at least").

Side note :
Ask me each other year what the difference is between Oversampling and Upsampling, and each other year you will receive a slightly different answer (better insight).

A genuine PCM converter (like the NOS1 with its R/2R ladder D/A converters) does not do anything internally to the sampling rate or bit depth. It receives a PCM audio word (for 16 bits this could be 0111 1001 0010 1101 for one channel) and it converts it to a voltage with mathematically follows from those bit values. For example, 1000 which is decimal 8 could imply (for easy thinking) 8mV (milli-volt). 1001 would be 9mV. 1010 would be 10mV. Or about 6dB more : 0100 is 4mV and 1000 is 8mV (twice as much = 6dB).

An SDM based converter "modulates". It assumes a given base voltage, and (in very rough terms) when it encounters a next 1 (in the 1 bit stream) it highers the voltage with "some" level. When a 0 is encountered it lowers the voltage with "some" level.
A bit can only carry the value of 0 and 1 and there is thus no means to tell the "1 bit DSD stream" that the level must stay the same. It always goes up and down (when it actually needs to stay the same) and it always creates a lot of noise in doing so, because the "going up" value is actually the largest voltage value possible for the D/A converter, and it could be 5V. However, the 5V "going up" happens in a very very short time and right after that it goes down again. Think like the 2.8MHz I mentioned and how this is 1/2,800,000 of a second and how this very brief "spike" of voltage is averaged with the previous "build up of level" from the previous samples (this is actually PWM - Pulse Width Mosulation). Or think like this : when we spit out 20 subsequent 1's or 20 times 5V, then the level is really 5V in the end. But when after a 3rd 1 a 0 follows, the level -say 20mV at that moment- drops again to 12mV-.
So something is modulated all the time.
That this SDM based system is also a system of feedback is something else. It is for the better, but for the worse at the same time (think feedback amplifiers and not everybody will like them).

Do you need to understand this ?
No. If you'll only accept that the effort it takes to explain SDM shows how much the music is manipulated before it's converted to that voltage and how this is changing the sound all over. This, opposed to genuine PCM which doesn't change a single bit/thing. And if the upsampling/filtering is done per XXHighEnd means, then even the original data does not change and only samples are injected (genuine interpolation). Notice though that this would be pure Arc Prediction, as the Filters we talk about in this topic change the original data slightly (and any PCM based converter (e.g. MSB) with normal filtering changes it heavily).

But to answer your question :
The iFi is SDM based. So now you know.


Peter

[xp9433]

Peter

Sorry Peter, I don't really follow.

So I researched the make up of my "in-bed/in-office" headphone system further, which is:
Laptop > Fidelizer optimizer > JRiver MC21 (upsamples 16/44.1 wav to 32/705.6 - includes EQ for my headphones and uses the iFi Asio driver) > iFi Gemini 90ohm cable > iFi iUSB Power unit > iFi Gemini 90ohm cable > Uptone Regen > iFi iDSD Micro Dac/Headphone amp > Headphones.

iFi state that their iDSD Micro handles PCM 768/705.6/384/352.8/192/176.4/96/88.2/48/44.1kHz - ALL with native decoding, no internal hardware conversion.

In the light of that expanded description, does this change you answer to my question -
Peter "No, not especially for NOS1(a) owners. However, it is obvious that the NOS1(a) is the only D/A converter which is NOS/Filterless (not Oversampling and no Analogue Filter) with an input possibility of 32/768Khz (in effect 24/768)."

Frank "Can you please explain a little further, for a technical digital novice like me, how the NOS1 differs from something like the budget iFi iDSD Micro that will accept 705.6 and 768 PCM input. The iFi doesn't upsample or down sample that input as far as I understand."

Thanks Frank

[PeterSt]

Hi Frank,

I am sorry but ask the iFi guys. Post the answer here if you like.
Start with asking them what D/A chip they use. Next I can enlighten you with the datasheet of that, but possibly you can do that yourself. I am not asking, but it is that simple.
And if I am wrong I will easily admit.

Regards,
Peter

[PeterSt]

Frank,

Of course this bugs me a bit. So just before heading to bed over here :

http://www.technoheadphone.com/amplifier-dac/ifi-audio-micro-idsd-features/24/

The DAC Chip we use in the iDSD nano offers a rather unusual way to handle things. It uses a 6-Bit Multi-bit DAC for the upper 6-Bits of PCM Audio and delivers the warmth and slam Burr Brown Multi-bit DAC’s are so famous for. Any bits below this are converted with a low-order 256 Speed Delta Sigma modulator (in effect DSD256) giving PCM playback the smoothness Delta Sigma DAC’s and DSD are famed for.

Does this resemble my story or what ?
What you may fall for is this :

The platform we chose for the micro iDSD (and the nano and rest of iDSD range) was the Burr-Brown. Why? Because it is one of the few chipsets that handles DSD and PCM without internal conversion. It keeps things unchanged and whether your music file is DSD or PCM, the whole chain remains as close to original.

(this is about the DSD1700 (DSD1793 actually))

All what that says is that the DSD1700 can be used without its internal filter (it can be shut off).
It is still an SDM chip (see my little story about that).
There no such thing as DSD *and* PCM which both do not convert. If that would be so then the DAC as such would need to carry two different chips - one SDM for DSD and one full multibit (like 24 or 20 if you like) for PCM.

You don't need to agree of course as I just tried to answer your question.

Regards,
Peter

[PeterSt]

Some truth - what the guys says himself :

http://www.head-fi.org/t/711217/idsd-micro-crowd-designed-and-the-new-firmware-flavours-are-here-page-138/60

Further, with this chip we retain a true Multibit PCM Core, true, only the upper 6 Bits are converted using multibit conversion, while anything below the top 6 Bits goes through a 256FS low order delta sigma modulator, in many ways this particular configuration is the best arrangement currently available,

Sure, if you anticipate that nobody knows that at some stage a PCM1704 existed or if you disregard that over here we have sufficient to feed the world.
Some times it is just how one likes the story to be told.

Now I'm off .
Peter

[xp9433]

So iFi when they talk about true native PCM at 705.6 they are using a touch of marketing hyperbole? That is, the Micro Dac accepts a 705.6kHz input without manipulation before coverting to analogue, but when converting that 705.6 signal directly to analogue the DAC doesn't process the signal in the purest way (the way your Dac does)?

iFi also state that their bit-perfect PCM filter (taken straight out of their AMR DP-777) is the same as non-oversampling! See why I get confused.

Further background:
iFi use the dual-core Burr Brown DSD1793  (x2) multi-bit thermometer code DAC chips for the most important upper six bits, and a low-order 256 Speed Delta Sigma modulator for the 26 lesser bits.  No chip specification sheet available because iFi don’t use the DSD1793 in its default configuration, and the standard specification doesn’t disclose what this older chip can really do.

ifi say, deep inside the silicone, the chip itself is actually capable of “true native” hi-res decoding for PCM and DSD separately. The way iFi say they are able to explore the DSD1793's full potential by employing a their own ground-up customization of the latest U-series XMOS Octa(8-Core) chip.

XMOS implementation that allows the DAC to do those native decoding function that are otherwise hidden by BB/TI, as well as optimising core loading, and using their own star clocking scheme.

For sound commercial reasons, iFi don’t give away how they have achieved their results. They do say they used the DSD1793 because they liked sounded better than the ESS for example.

However, you would have to agree they are pretty clever to deliver a surprising good sounding PCM768/DSD512 DAC + high power Headamp using a BB DSD1793 Dac chip for only $500!

Thanks
Frank

[PeterSt]

Frank,

I feel you are pushing me in corners where I don't want to be - probably because I don't know what you want to achieve.

So let's start over with the knowledge we have now.
Here's your original question again :

Can you please explain a little further, for a technical digital novice like me, how the NOS1 differs from something like the budget iFi iDSD Micro that will accept 705.6 and 768 PCM input. The iFi doesn't upsample or down sample that input as far as I understand.

Well Frank, what to say. A first thing could be that one needs to wonder what IfI is going to do with that sampling rate possibility; no material exists for that. So for them it could be about the high numbers and win some souls. And yes, I recall an email from them, asking whether they could use XXHighEnd for testing their gear. Maybe now we understand why.
So notice that the strategy is in XXHighEnd, thinking "filtering".

An answer from another angle ... see first picture below.
The iFi uses the DSD1793 which is rated to work at 192KHz. But look at the THD figures; possibly you are not into such numbers, but I am and do all with them - for a base. So for 192KHz the THD already is 0.003% undoubtedly (read : 100% sure) including a heavily ringing filter (otherwise the 44.1 plot would show 0.04% while it now shows 0.001%).
If we'd be allowed to extrapolate this to 768KHz we'd arrive at something like 0.04% again (just expand the lines upwards in linear fashion) which implies headaches. Plus :
we are not allowed to extrapolate because things will collapse (closely look at the distance between 96 and 192 which already is larger than the distance between 44.1 and 96).

So 786KHz ? nice story. Also nice that someone tells that it sounds nice, but I'll only believe that when I heard it myself (which I did not).
Those numbers are too important to me.

I can cough up the numbers of the NOS1(a) but I assume they are known, because in this forum at several places.  But for 24/768 it is 0.00074% THD+N which sadly is also my analyser limit. Watch out, because this is measured at the end of the interlink with a fully operative DAC. Compare with the plot below which is from a CHIP.

For me, a noise line is even more important; Look at the 2nd picture below. That is quite hurting (looks like partial DSD sh*t WHILE this is PCM mode (see datasheet in the Note under that plot - http://www.ti.com/lit/ds/symlink/dsd1793.pdf)).
I don't think I have put up plots at -60dB but you'll hopefully take my word for it if I tell you that the noise then is close to -160dB and super straight up to 96KHz.

With all this I'd have to tell you that "my" figures are with a zero ringing filter.

If you add all up then there will be no single way that I would use such a chip for any design which is to sound good. It is just too poor from all angles. Remember once again : what you see in the datasheet is the specs from the chip, which tends to eliminate all noise sources etc. normally present in a ready D/A converter.
NOS1(a) is more expensive though. But hey ...

Kind regards,
Peter

[manisandher]

I have an iFi iDSD Nano here and although not as good as the Micro on paper, I suspect it uses the same DAC chip. The Nano really doesn't play in the same league as the NOS1a. Even if the Micro is substantially better than the Nano, I doubt it would be in the same league as the NOS1a either. No other DAC that I have ever heard is.

That's not to take anything away from iFi - great products for the price.

Mani.