Phasure NOS1 24/384 DAC Specs and graphs

[PeterSt]

This topic will tell you about the specifications of the Phasure NOS1 24/384 D/A Converter. It will be extended over time.
It will also explain (NOS1) DAC behaviour in general.

Do notice : This is about the original Phasure NOS1 (not sold anymore), *not* the Phasure NOS1-USB. Look here for the 24/768 USB version. Or here for Mac specifics.

Specifications in short :
(price EUR 2900 ex VAT (if applicable) and shipping. Due price, however, will be EUR 3040 because of some necessary upgrade (yet to lay out)).

All measurements at the end of a 2m cheap interlink, done with a dScopeIII digital analyser :

• Frequency response 1 Hz to 192000 KHz (384KHz sample rate). Flat within 0.19dBr in the audio band (DC not sustainable).
  
  
• Input of 16 or 32 bits for all Sample Rates (24 bit input is converted to 32 bits).
  
  
• Input Sample Rates (Hz) : 22050*, 44100, 48000, 88200, 96000, 176400, 192000, 352800*, 384000*, all auto-select without glitches.
  * Under control of XXHighEnd only.
  
  
• Output Sample Rates always equal the Input Sample Rates, but notice that in software (PC, XXHighEnd) each native “file” Input Rate can be upsampled to each accepted higher Output Rate (like 96000 to 384000 or 44100 to 176400 etc.) and next is output to the DAC. Output bit depth is 24 - or 16 bits when Input bit depth is 16 bits.
  
  
• 100% Filterless Non Oversampling, using 8 x PCM1704U-K Burr Brown (Texas Instruments) multibit 24 bit D/A converter chips. Filtering to be applied in software (officially provided by XXHighEnd through Arc Prediction which supports all modes), or not using filtering at all, at own choice.
  
  
• Output level for Single Ended Mode (RCA) is 1.5VRMS (-3dBFS relative to normal Full Scale). Output level for Differential Mode (XLR) is 2.7VRMS (+3dBFS relative to normal Full Scale).
  
  
• Output Impedance is 33 Ohm. It is the explicit advise to use the NOS1 without pre-amp and without analogue volume control for optimum results (long interlinks can be driven).
  
  
• The Clock section consists of replaceable oscillators, separately for the 44100 Hz and 48000 Hz base, both separately fed by shunt regulators. The provided oscillators carry an overall jitter of better than 0.5ps and phase noise of better than 120dBc/Hz at 100Hz and better than 108dBc/Hz at 50Hz.
  
  
• Although it has not been measured (lacking an digital input suitable for the analyser) the net jitter should equal the fed jitter (of 0.5ps) because there is nothing in the path from the oscillators to the DAC chips. Additionally the analyser does not show any spurs of sidelobes which thus will be under the noise level.
  
  
• All terminals are official Neutrik.
  
  
• The Phasure NOS1 feautures a PC interface which does not imply a synchronous connection, nor does it imply an asynchronous connection. Instead, it is completely direct by means of extending a PCIe slot from the PC to inside the DAC housing, the one and only clock section being 12cm away from the DAC chips, allowing for a direct i2s connection from there with no single one hurdle (like a receiver chip) to degrade jitter figures.
  
  
• Requirements PC : A small PCIe (x1) slot or an ExpressCard slot for laptops.
  Notice that for good sound quality (and conveniency) a laptop is never advised;
  Regarding the best sound quality XXHighEnd *is* advised, especially because it allows for the input sample rate of 384KHz.
  Please notice that a MAC has not been tested, but should work up to 24/192 input (as it does not work with the XXHighEnd software. Same for Linux based PCs.
  
  
• Explicitly for Driver stability reasons, a modified and OEM licensed ESI based soundcard is used for the actual communication with the PC; It allows for ultra low latency when needed, which is explicitly supported by the Phasure NOS1 in combination with XXHighEnd – and which goes as low as around 16 samples at 32/384 Input/Output. Notice that this also depends on the processor power of the PC, and even lower is very well possible.
  
  
• Output terminals are provided for Single Ended (SE, RCA) as well as Differential (Balanced, XLR) by standard. No switches are in the signal path and no transformers are involved.
  
  
• S/N ratio in-(audio)band is at –132dBFS at full scale (-0dBFS) differential (XLR) output.
  Out of band (from off 31000Hz and up to 96000Hz which is the measureable limit) this is better than –144dB, and at the limit of the analyser operating at 24 bit and 192KHz.
  For SE (RCA) output at –6dbFS (the maximum for the 1.5VRMS output) the noise is at –140dBFS or 134dB down. Here, the noise from off 31000Hz and up is at –144dBFS or 138dB down.
  
  
• Mains related products for SE measure 17uVRMS at –121dBFS (measured at 50Hz mains). Notice this will be because of the pickup of noise of poor interlinks. Better interlinks will show 8 uVRMS at –127dBFS.
  Mains related products for Differential measure 2.2uVRMS at –139dBFS (measured with 50Hz mains).
  
  
• SE (RCA) output measures in(-audio)band Unweighed 0.0018% THD+N at 16 or 24 bit input and an 1KHz test signal with a sample rate of 44100Hz and Arc Prediction filtered into 176400Hz.
  The THD+N from Differential (XLR) is beter than 0.0015% which is under the typicle specs of the DAC chips.
  
  Notice : Actual figures of both SE and Differential will be better when (XXHighEnd) Arc Prediction filtered into 352800Hz, but the used analyser in our development rooms won’t allow measuring above 192000Hz.
  The same 1KHz test signal measured unweighed over the total analyser’s useable range (0 – 96000Hz) shows a THD+N of better than 0.0035% (differential).
  A-Weighed this is better than 0.0013% and C-Weighed better than 0.0009%.
  In-band A-Weighed is better than 0.0012%, C-Weighed better than 0.0009% again.
  All measurements have been done at the end of 2m interlinks, the input impedance of the analyser set to 100K. This gives the real honest figures going into your amplifier.
  
  
• In addition to the above, all in-band harmonics stay well under –100dB, and out of band only residuals at an equal distance to the sample rate of the test tone can be seen at –95dB (like for a 44100 base sample rate and 1000Hz test tone, at 43100 and 45100.
  
  
• A unique means of channel separation has been applied which goes as far as in the digital domain (by means of XXHighEnd only); Although not measureable, this is the most audible; As far as this is possible to measure without the digital means of an analyser, the FFT graph shows a channel separation of better 125dB (130dB with differential (XLR) output) when one channel is muted.
  
  
• Suitable for 110-130V or 220-240V.
  Watch out : Units with a Serial Number of 10001 u/i 10200 will be internally wired for 110-130V or 220-240V on your indication at ordering. It will need the change of this wireing to make the unit suitable for that other mains. Please take good notice of this when taking (or selling) the unit to country with this other mains;
  The re-wireing will take 5 minutes to change, but you may not be able to do it yourself. Units with Serial Numbers of 10201 and up will contain a user-switchable power inlet. And careful : The currently used power inlet looks to be switchable as well, but it really is not !! It will only indicate the target mains on the outside, and while this by itself is user-switchable this should not be done in any occasion, without changing the inernal wireing accordingly !
  
  
• Double secured with 500mA (1000 for 110V) fuses (non “audiophile” for hot and cold); The second fuse is not a spare, but actually used.
  No mains filter of any kind is applied.
  
  
• Some parts in the inside are vibration protected with special (blue) silicone material. Do not remove this, because they influence sonic performance, and in the mean time it protects against unwanted ground connections or otherwise, up to shortcuts.
  
  
• Similarly you may find wireing running strange or looking like “I would have done that differently”. Also you may find protective material in the iside which is not really fixed to one place (with screws or anything) and which you could change easily for its position. Please leave all as you find it, because all contributes to the specs the DAC currently have and which specs equal the DAC chips themselves, even at the end of your interlinks. It is the result of endless measuring and listening.
  
  
• Full color touch display which can be switched off. (Touch) Functionalities vary per XXHighEnd version.
  
  

For those who are interested, or think they can derive sonic quality from it, here are some graphs.
Please notice that not all normal measurements and graphs can be done, because of no SPDIF input being there (which the analyser used needs for those measurements).

All test signals are at 1000Hz and with a signal of -3dBFS, unless mentioned otherwise.
The graphs shown are not all there can be, but are oriented towards the actual virtues of the Phasure NOS1 DAC, that being a non-oversampling DAC with complete filterless design. It is therefore good to notice that the test signals themselves are produced by means of playback in XXHighEnd, that sofware taking care of the "filter" stage to eliminate the "stepping distortion" because of the inherent too low resolution of redbook CDs; since the test signals are created with a sample rate of 44.1Khz/48KHz -16 bits (unless noted otherwise), they represent the playback of redbook CD - and which is the objective of the NOS1 : to do that as well as possible.
Also notice that the playback software can use any means of filtering serving the purpose of eliminate the stepping distortion, but that the type used will influence sonic quality, as it will influence measurements;
Here the Arc Prediction filtering of XXHighEnd has been used, which was explicitly designed for the NOS1. Again unless noted otherwise it was used at "8x", or in other words creating an output of 32/352.8 (384) towards the DAC, which thus takes that as the input. 24 bits are actually used for the input, and the other 8 are truncated (but should optimally be 0's in the first place, the feeding software taking care of that).

Important : Since the Arc Prediction Filtering used is not subject to any pre or post ringing (while no phase changes occur throughout), we indeed *are* looking to the results of a CD playing as if this were not "periodic" waves (but they unavoidedly are). Therefore the graphs (or figures) can not be compared to any OS type D/A converter, because there the graphs and figures *depend* on the periodicy of the wave; It needs subsequent samples (before or after the actual sample playing) in order to reconstruct the original wave, and it is exactly this where the ringing emerges (it is a necessity to let operate the filter well);
Since normal music is hardly about periodic waves, measurements of OS type DACs (or ringing filters), say exactly nothing. They will tell relative differences amongst DACs, but they will NOT tell one thing about the quality of the filter. As a matter of fact, the longer a filter "is", the better its measurements results will be, but the longer it rings. And, theoretically an inifitely long filter will produce a 100% original result (read : an original wave with infinite resolution (sample rate) will be reproduced at the analogue side of the DAC with infinite high resolution again, BUT, it will also infinitely long "echo" throughout the music in normal life).
The whole point is, this ringing can't be shown in FFT graphs (that we know of), unless it is shown as an "impulse response", and which latter by itself tells you nothing but "will I hear that ?". Well, if an impulse (one sample) is really there even only 5 more times (but at a lower level), why not ? If we don't, something else must be wrong.

Outside of FFT graphs we can show the ringing by means of an impulse response graph. Here is one which comes from the output of the NOS1 while a common means of filtering is used (AI Filtering from XXHighEnd in this case) :

These are actually one sample pulses, positive signal only. It doesn't look like single samples at all, and it is the filtering which clearly "unsharpens the sound".

But now look how it can be done too :

This is Arc Prediction filtering, and the only ringing you see is the electrical "swing" needed to get the pulse going.
Let's keep in mind that the necessary filtering as such is still applied. But no ringing at all.

The real message is : when we look at the graphs below, we do can interpret them as "this is how the music will play too", and this is because no ringing is in there. And : don't compare with similar graphs which will inherently ring under the hood, *unless* you look at graphs from another filterless (can only be NOS) design.

Where applicable, measuring figures are always presented unweighed, with the notice that e.g. "C-Weighing" or "A-Weighing" will produce better numbers, and that this Weighing often is used by manufacturers (and we're not saying that this is done to improve the numbers -> could be for any reason).

What we see here is the test signal of 1000Hz, and how its harmonics emerge throughout the audio band (up to 10000Hz here, but it really won't get worse beyond that).
Althoug the DAC plays at -0dBFS output, for good interpretation keep in mind that the signal itself is at -3dBFS. This means that where you see the harmonics (always "false") emerge at where you see them, but that 3dB should be added really. So, the second harmonic (which is the one at 2000Hz) shows at ~ -112dB, but it really is at -109dB.
We could say here that the worst case is the third harmonic (commonly appreciated as nice-sounding for distortion if it were to be there anyway), but it really is so much "down" that it's totally inaudible. For theories at least.

For those who can interpret it : there are no spurs of sidelobes to the test signal, indicating very low jitter. or at least nothing that can show here.

Noise during playback peaks just over -130dB. Later we will see that this is, say, quantization noise of some sort.
Notice that the FFT depth is 256K.

What we see here is again an FFT (256K), but now the DAC is idle. It's normal level is at -150dB.

The right side is the analyser's limit, and the noise is not really going down there. This is not to be confused with the left side (with a theoretical limit of 0Hz = DC), and the creaping up of the noise at that side. It really does. This is of the mains impeeded products, and say, the 50Hz (or 60Hz) coming from there. This shows here at ~ -125dB, and it really is what me must take into account for the real noise level. Look :

Here it shows. Although this expresses the noise in uVRMS by numbers (and "RMS" can be seen as the average from peaks to 0), this would also show a -113dB for the peaks. That the previous picture shows this as -125dB is because of the averaging effect (to noise) of the FFT depth. The picture here just shows "the wave" and its peaks are really 113dB down only.

What you can't see here, is that the "figures" really emerge from a "swing" on the voltage; the red and blue lines ("waves") in reality swing a little around the 0 value, so in the longer term is creates a low frequency wave. This is exactly how the FFT from the previous picture shows the noise to go up at the lower frequencies, and it is exactly why this is of crucial importance for your bass response ...
All is not about the noise as such, but it shows us the general swing the voltage fed to the DAC implies an impact on the lower frequencies; The lower this is, the more tight a bass will sound.

The message : never mind this "swing" is some 113dB (125 in the previous graph) down and is so called inaudible; someone may re-interpret this to the variation in output level in dB and come up with an "impossible", but let's not believe anything but our ears.

Here's the same test signal as the first one, but now the full measuring range is shown (96KHz for this analyser, sampling at max 192KHz).

The peaks around 48KHz are derivals from the native sample rate in the file (not the sample speed of the DAC), and it shows as an offset worth the value of the frequency of the test signal - plus and minus. So, the peaks will be at 47000Hz and 49000Hz.

The obviously noticeable is the decreasing noise which starts somewhat before 20KHz. But now look here :

Here we see exactly the same (intended !) output, but something clearly has changed. Now, suddenly, the roll off of the noise level starts at somewhere before 40KHz. All other peaks / harmonics look fairly the same - if not exactly the same. What happened ?

Well, this is a "hires" test tone ! Instead of the before 48KHz sample rate, this one is comprised from 96KHz. But, it is still filtered (upsampled) towards 384KHz. And, since the filtering algorithm doesn't know much about which frequencies not to touch because they will be quite allright in the first place, now it has attacked (molested) the waves again up to where they were good enough, which in this case is towards the Nyquist Frequency of this sample rate : 48KHz. Dislaimer : this is an own explanation for now, not knowing any better.
So, we could say that the downside of the general filtering which in this case goes up to half of the created sample rate - and which is 192KHz here, is the somewhat degradation into the noise level of something which contained less noise at first.
But how negative is this really (apart from being into a noise level where nobody will perceive it) ? Look :

Well that is strange ...
What we see here is the same 96KHz sampled test signal (1000Hz) from before, but now natively played. Thus, the DAC runs at 96KHz, and the file is played NOS as NOS can be. But the noise is higher towards the right ...
Notice that the DAC does not electrically behave the same as before. So, before it ran at 384KHz, now at 96KHz. Apples and oranges ? maybe, but it looks like there's a trend.

Anyway, where the noise may vary, we can see that the output level (of the (false) harmonics) is the same throughout. Ok, the natively played 96KHz file shows a 10dB worse 3rd harmonic, and we see aliasing at the right side. But further ? To show this better, look at the next two graphs :

Here we can see that the output level of 1000Hz vs. 20000Hz doesn't even differ by 0.1dB. So, we can well say that the roll off of the siginal is just nothing. Still there is no imaging beyond the audio band (there will be, but it's beyond 192KHz), so the filtering does work.

For your information : the output of the native test signal (of -3dBFS) has been decreased here by the Peak Extension function from XXHighEnd. This is in order to not let overshoot the relatively very steep 20KHz test signal (this is an electrical thing, but indirectly covered for by Peak Extenstion).


More later.