Sonnet Pasithea DAC Review and Measurements

Discussion in 'Digital: DACs, USB converters, decrapifiers' started by purr1n, Jun 17, 2022.

  1. purr1n

    purr1n Burned out

    Staff Member Friend BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    80,585
    Trophy Points:
    113
    Location:
    Padre Island CC TX
    Sonnet Pasithea
    Balanced Output, Coaxial SPDIF In
    THD (no N) vs output level (dbFS) 37Hz, 1kHz, and 9.3kHz sine
    20-20kHz filter
    upload_2022-6-19_22-28-29.png

    Sonnet Pasithea
    Balanced Output, Coaxial SPDIF In
    Linearity 1kHz (run x8 times)
    upload_2022-6-19_22-34-58.png

    This is actually pretty good. Linear down to -110db. Way better than necessary for 16-bit content (-96db).
     
    Last edited: Jun 19, 2022
  2. purr1n

    purr1n Burned out

    Staff Member Friend BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    80,585
    Trophy Points:
    113
    Location:
    Padre Island CC TX
    Sonnet Pasithea
    Balanced Output, Coaxial SPDIF In
    THD (no N) vs Frequency (at four different output levels)
    90kHz bandwidth
    upload_2022-6-19_22-52-15.png
     
  3. purr1n

    purr1n Burned out

    Staff Member Friend BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    80,585
    Trophy Points:
    113
    Location:
    Padre Island CC TX
    Let's try 1kHz sine waves at different output levels.

    Sonnet Pasithea
    Balanced Output, Coaxial SPDIF In
    -30dbFS 1kHz Sine, 48kHz sampling rate, 200kohm load
    upload_2022-6-19_22-55-17.png

    Sonnet Pasithea
    Balanced Output, Coaxial SPDIF In
    -60dbFS 1kHz Sine, 48kHz sampling rate, 200kohm load
    upload_2022-6-19_22-56-1.png

    Sonnet Pasithea
    Balanced Output, Coaxial SPDIF In
    -90dbFS 1kHz Sine, 48kHz sampling rate, 200kohm load
    upload_2022-6-19_22-56-33.png
    upload_2022-6-19_22-57-32.png

    Nice! And consistent with the linearity test.
     
  4. Gazny

    Gazny Friend

    Friend
    Joined:
    May 11, 2020
    Likes Received:
    1,259
    Trophy Points:
    93
    Location:
    SoCal
    Hey Marv, I wanted to ask about the cross over point* and is it audible?
    I've read many times some get a real kick with fixing this quirk of many R2R dacs, and some just let it ride.

    What does it show us in the grand scheme of things when it comes to r2r?

    *I believe that is what they call it LSB/MSB cross over point, the name escapes me but I once read someone on DIYA yelling about this and I just didn't get it.
     
  5. skem

    skem Friend

    Friend
    Joined:
    Nov 1, 2017
    Likes Received:
    1,865
    Trophy Points:
    93
    Location:
    Charles River
    I’ve been living with Pasithea for six months now, so perhaps its time I write something about it. During the last six months I have put this DAC in head-to-head competitions with
    • Bricasti M1 (non-MDx)
    • COS D1
    • Metrum Adagio with DAC2 chips (e.g., Pavane Level 3)
    • Metrum Adagio with new DAC3 chips
    • Exogal Comet
    • Abbas DAC2.3
    • Rockna WaveDream SE (via Texas Crew, see their review here)
    I’m no longer in a position to write six detailed DAC-to-DAC comparisons—too much time has passed. But suffice to say, I still own Pasithea and it’s now my primary DAC after all these showdowns. Exactly why is complicated. Better I think for me to try to communicate the virtues and limits of this particular device. It is a singular product that is more than one might initially think.

    If you have a good digital source, extended first impression will be that Pasithea’s strength is that it retains the natural timbre of NOS without being overly soft like most NOS DACs. Cees’ previous DACs (e.g., Adagio) were among the best NOS DACs made, but Pasithea easily betters them with more treble and speed. I’ve spent some time talking to Cees about this, and he asserts that the secret to Pasithea’s “high bandwidth” presentation is basically ohm’s law. You have to remember that the R2R-ladder architecture is used for resistor DACs because the ladder exhibits the same “Thévenin” equivalent resistance regardless of which subset of resistors the signal is passing through. Ideally, only the voltage changes, making it appear as a voltage source. The secret of Pasithea is that the new modules run on higher voltage (24V rails) and have a low effective R2R ladder impedance of only 16 ohms, which helps make the ladder output more closely simulate a perfect voltage source. As a result, LRC filter effects associated with connecting gain stages to the ladder are minimized. Phase is therefore more perfectly preserved, and high-frequency content is better retained reproduced. This is very obvious when you put Pasithea up against any other NOS ladder DAC, and even subtly obvious against OS ladder DACs like Rockna if you know what to listen for.

    Relative to the Adagio, there is a price to pay for this transparency. Pasithea gives up some of the gorgeous vocal romance. It’s still romantic compared to other DACs, but I actually think it might be more correct than the more sterile presentation we are accustomed to hearing from most DACs. Maybe I’m wrong here, we would need studio masters to really decide the matter (Incidentally, that is reportedly how Cee's tuned Paisthea, by A/B'ing live music in a studio) In any case, I haven’t sold my two old Adagios for this very reason—the delicious vocals of Adagio are definitely warmer and more colored, and also more beautiful—but Pasithea is a much more well-rounded product, and thus my daily driver for a wider array of genres.

    That said, although Pasithea has lots of good treble compared to most NOS DACs, it is still softer than oversampling DACs. Why? In short I don’t know. Generally, on most NOS DACs there is a filter to cut-out the ultrasonic aliases, and if you are running RedBook content, the anti-aliasing filter will necessarily bleed into the audible part of the audio band, causing some phase shift and amplitude roll-off in the very last octave (where the ‘air’ Lives). Consequently, high-frequency components will not line up quite right (the phase part of this story), and so it will not be possible to reconstruct the fast signal rise times needed to reproduce good ‘attack.’ As such, most NOS DACs running from a RedBook source are going to have softer attack and slightly diffuse imaging—mathematically, with an analog filter in place, there is just no way around it. Pasithea eliminates this effect by skipping the filter, but it still doesn’t slam like Dave reportedly does, so I assume there is another distortion phenomenon that psychoacoustically accentuates attacks in OS DACs, but I couldn’t prove it. I grant that gentle attack sucks if you like to listen to metal and other such highly percussive or aggressive genres. On the other hand, with NOS the rest of the audio band remains more intact. There isn't nearly as much phase drift [signal fuckery like ringing] in a NOS DACs as there is in an oversampling DAC—although we can debate the importance of this. In any case, oversampling DACs always have timbral defects to my ear. Yes, even with linear-phase reconstruction filters I still defects in timbre on the attack. Linear-phase reconstruction filters in OS DACs ensure that all frequencies are properly ‘in phase’, which helps a lot the harmonics of a sustained note… but there is still a pre-ringing issue that spoils timbre in chromatic percussive instruments like piano.

    Of the oversampling DACs, the Bricasti and COS D1 were among the best I’ve heard. I’d take COS for beauty and simplicity, and Bricasti for sound and versatility. But, while both of these have good energy and drive, and the images are sharper—the timbral defects bother me, and they are pretty miserable with their 2-dimensional staging (compared to Pasithea).

    OK…so does this mean that Pasithea is a good choice if you value timbre and stage, but a poor choice if you like energy, imaging, and attack? Mmmm... maybe. But I really think there’s another chapter in the Pasithea story.

    Audiophiles like to mix and match components: a favorite source, a favorite DAC, a favorite pre-amp—or for vinyl heads, a favorite turntable, maybe a few arms and a few cartridges—all driving one or more amps, and one or more transducers. Synergy is the name of the game, and we are ever chasing after some personal flavor. But in all this we tend to think of DACs as unitary objects. Now, for the first time in my audio journey, I’ve come to view Pasithea not as my favorite DAC, but as my favorite R2R ladder! The rest of what we normally regard as being integral to a DAC—the oversampling reconstruction filters and digital interfaces—are something that can be glommed on using separate boxes, and doing so produce some pretty interesting results.

    Wait… why? It helps to understand why all DACs kind of suck compared to analog. It’s because they are all trying to grapple with a set of inescapable trade-offs that are mathematically imposed by the near impossibility of achieving perfect analog reconstruction from a digital sample. Yes… in theory RedBook recordings have all the information that you need to do a perfect reconstruction, but in fact this requires SINC(x) functions that extend infinitely in both directions of the time domain. That means perfect reconstruction is impossible in a realtime device like a DAC. However, new outboard resampling tools like HQPlayer and PGGB are bringing us closer and closer to perfect reconstruction by doing the heavy time-domain computations on more powerful computers (in realtime in the case of HQPlayer, or in non-realtime with PGGB), and then feeding the DAC higher-sample rates that are more dense with information per unit time, making the DAC’s job much easier.

    NOS DACs are interesting here. For one, you don’t want to oversample your oversample. At the same time, feeding high-sample-rate data to a NOS DAC pushes the aliasing up in the frequency so that it no longer interferes with the ‘air’ octave, and consequently the quality of percussive attacks are restored without ultrasonic energy in the region near to the audio band. We are basically adding back the oversampling that NOS is missing, but using a much higher quality outboard implementation.

    In such a case, a very high quality R2R ladder becomes especially interesting, because we can do this with fewer hardware-imposed defects to the spectral response. Suddenly, you can see why Pasithea looks interesting. It’s not that NOS is better; I want oversampling. It’s that I get a high quality DAC that lets me add high quality oversampling. On the downside, Pasithea doesn’t do much to clean up noisy digital signals. I’ve heard substantial effects by swapping digital cables and digital sources with Pasithea. I am not a cable person so I was pretty shocked by this. Pasithea needs a clean (and evidently reflection-free) digital setup.

    How well outboard oversampling works depends on how fast you can feed data to the DAC. The hot-dog eating king here is Holo, which can go up to 1.5MHz, but is otherwise not an especially good-sounding DAC ladder implementation (or so I’ve inferred, since I haven’t demo’ed the Holo DACs myself). Unfortunately, the SPDIF and I2S interface on Pasithea crap out at 192 kHz, which is barely good enough to start tasting the real benefits of outboard resampling tools. The USB interface on Pasithea goes to 384kHz, but historically Cee’s products have used shitty USB interfaces. I honestly don’t know how good the one on Pasithea is, since I don’t have a good USB source handy to try it with. Nevertheless, even at 192 kHz, I have experienced some amazing sounds that make Pasithea blow any other DAC I’ve heard out of the water.

    Perhaps the most striking example of this was when I upsampled the RedBook copy of Eagles’ Hotel California to 192kHz/24bit using a minimum-phase reconstruction filter (yes, I said minimum phase, gasp!). The staging will blow your mind. On my system, the sound stage extends well past the speakers in the room, by which I mean it goes to the right of the right speaker and to the left of the left speaker. The audience, which applauds throughout much of the start of the song actually sounds far away… like its coming from many hundreds of feet behind the rear wall of my room. When one experiences something like that, you really do sit up and take notice. This is not something to be dismissed.

    The problem is, I haven’t figured out how to make Pasithea sound wonderful all the time. My upsampling algorithm for Hotel California fails miserably with heavily mixed studio recordings. It emphasizes all kinds of artifacts that are normally not audible when limited to NOS mode. Compare this filter schizophrenia to Bricasti and their relatively straightforward menu of built-in options. Bricasti never sounds anywhere as close to being live like Pasithea does with a well-chosen OS filter, but Bricasti nevertheless sounds pretty darn good almost all of the time. I can see why a reasonable person would choose an oversampling DAC like Bricasti (or Rockna or COS) and then turn to vinyl when they want the best possible sound. But I can’t stand all the fussing associated vinyl and I am relatively happy running Pasithea in NOS mode as well. So… while elite audiophiles may roll arms, carts, and tubes to attain temporary nirvana… I roll filters. And I bet you that on the rare occasions when I get it right, most analog setups won’t stand a chance.
     
    • Like Like x 11
    • Epic Epic x 4
    • List
    Last edited: Jun 22, 2022
  6. jexby

    jexby Posole Prince

    Staff Member Friend Contributor
    Joined:
    Sep 28, 2015
    Likes Received:
    7,154
    Trophy Points:
    113
    Location:
    gone
    Not sure any Holo Spring3 KTE or Holo May owners would agree with that statement in any way, especially with HQP.
    still- a very stellar write up of well explained and detailed impressions!
    kudos for the insight.
     
    • Agreed, ditto, +1 Agreed, ditto, +1 x 4
    • List
  7. purr1n

    purr1n Burned out

    Staff Member Friend BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    80,585
    Trophy Points:
    113
    Location:
    Padre Island CC TX
    @skem: There is some technical stuff that you said which is just plain wrong, based on misconception, or torn from the typical NOS book of technobabble.

    I could care less about less than ideal measurements because it's about the sound. But wrong understanding of math, science, and engineering is a bad way to justify what one prefers.

    To be continued.
     
    Last edited: Jun 19, 2022
  8. skem

    skem Friend

    Friend
    Joined:
    Nov 1, 2017
    Likes Received:
    1,865
    Trophy Points:
    93
    Location:
    Charles River
    Interesting. My preferences are based purely on acoustic evaluation, but I’ll be curious to learn from you what science I got wrong.
     
    • Agreed, ditto, +1 Agreed, ditto, +1 x 2
    • Epic Epic x 1
    • List
    Last edited: Jun 19, 2022
  9. skem

    skem Friend

    Friend
    Joined:
    Nov 1, 2017
    Likes Received:
    1,865
    Trophy Points:
    93
    Location:
    Charles River
    Try adjusting your bandwidth limit between 20Hz and 29Hz and see if those lower sidebands bounce around. I wouldn’t be surprised if they are artifacts of the analyzer’s digital sampling frequency interacting with the signal and not actually present in an analog sense. I’ve seen such behavior before in other analyzers near a (imposed) nyquist limit.

    edit: yeah, those are beat frequencies because the ADC is sampling below the nyquist limit of the analog signal, which in this case is essentially unbounded because of aliases.
     
    Last edited: Jun 20, 2022
  10. purr1n

    purr1n Burned out

    Staff Member Friend BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    80,585
    Trophy Points:
    113
    Location:
    Padre Island CC TX
    Here, I did one better. I had planned to do this anyway because I felt that more realistic measurements, this running it through a real component chain, where the various bandwidth limited components would be responsible for the analog reconstruction filter. The last thing I wanted to hear was objectivist weirdos bitching about this or that (although it likely doesn't matter). I blasted a the same 15kHz signal (generated from the AP) into a Grado RS1X via a Magni 3+ on high-gain and measured with the a microphone.

    The only limitations are the 48kHz sampling rate and microphone bandwidth. However, in this case, I feel they subtracting anything from the results or taking away from the general gist. However, it's entirely possible we could get different results with a high-bandwidth system with high-bandwidth transducers.

    15kHz signal ~102db
    Sonnet Pasithea -> Magni 3+ -> Grado RS1X
    upload_2022-6-20_8-35-56.png

    No aliasing artifacts. LPF / analog reconstruction is performed by downstream components. Small spikes seen are environmental noise and do not coincide with any of the spurs (aliased down) seen on the AP.

    Now the smartass objective may say that perhaps there still some aliasing going on, but the measurement system aren't good enough to capture this. My response is this: seriously, you can hear -75db down from the signal? How loud do you listen to?
     
    • Like Like x 2
    • Agreed, ditto, +1 Agreed, ditto, +1 x 1
    • List
    Last edited: Jun 20, 2022
  11. purr1n

    purr1n Burned out

    Staff Member Friend BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    80,585
    Trophy Points:
    113
    Location:
    Padre Island CC TX
    Ok, makes sense. Keep the output impedance as low as we can to lessen the effects of Miller capacitance, thereby less roll-off on the high-end.

    I don't have any other NOS DACs on hand to test, so I cannot confirm this claim against other NOS DACs.

    However this claim is FALSE when compared to OS ladder DACs, the Rockna WDS, Yggdrasil LIM, MMB, etc.

    Sonnet Pasithea
    upload_2022-6-20_9-33-53.png

    Yggdrasil LIM
    [​IMG]

    Rocka WDS (from Goldensound)
    [​IMG]

    MMB
    upload_2022-6-20_9-35-45.png
     
  12. skem

    skem Friend

    Friend
    Joined:
    Nov 1, 2017
    Likes Received:
    1,865
    Trophy Points:
    93
    Location:
    Charles River
    Actually, all of those plots show frequency responses that imply terrible phase response near the cutoff. You just made my point. Plot the phase error and let’s have a look.
     
  13. purr1n

    purr1n Burned out

    Staff Member Friend BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    80,585
    Trophy Points:
    113
    Location:
    Padre Island CC TX
    That's another point you made of which I will return. I'm arguing against your specific claim of better high-frequency retention, which is not true. You are shifting your argument. It's a great rhetorical and debate technique, but it will not work. Nice try.
     
  14. skem

    skem Friend

    Friend
    Joined:
    Nov 1, 2017
    Likes Received:
    1,865
    Trophy Points:
    93
    Location:
    Charles River
    No, I started my sentence with emphasis on phase (not amplitude) and I never used the word extension. You misread my intent. Nice try, though.
     
    Last edited: Jun 20, 2022
  15. purr1n

    purr1n Burned out

    Staff Member Friend BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    80,585
    Trophy Points:
    113
    Location:
    Padre Island CC TX
    Well, it's confusing, so perhaps you should delete or clarify this statement:

    upload_2022-6-20_9-50-36.png
     
  16. skem

    skem Friend

    Friend
    Joined:
    Nov 1, 2017
    Likes Received:
    1,865
    Trophy Points:
    93
    Location:
    Charles River
    OK, fixed so the intent is clearer. Thanks.
     
  17. purr1n

    purr1n Burned out

    Staff Member Friend BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    80,585
    Trophy Points:
    113
    Location:
    Padre Island CC TX
    There is something that is way better reproduced on NOS DACs which you did not mention, which I feel could be a strength. More on this later. I'll get to everything.
     
  18. purr1n

    purr1n Burned out

    Staff Member Friend BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    80,585
    Trophy Points:
    113
    Location:
    Padre Island CC TX
    This is also false, at least with DACs that implement linear phase digital filters with "over"sampling / upsampling.

    I'm absolutely fine with the subjective statements. However I strongly feel that subjective statements and personal preferences should not be supported by misunderstandings and misconceptions of science, how this stuff really works. The engineers guys at Sony and Philips and later guys who invented OS, they weren't dummies. These guys were extremely extremely smart and looked at this stuff at all angles, more than any of us can possibly imagine.

    MMB (ladder OS DAC) Phase
    upload_2022-6-20_10-17-2.png

    MMB (ladder OS DAC) Group Delay
    Note increase in group delay in lows is the result of AC coupling in the APx555. Output of MMB has 75-ohm resistors, increasing the effect.
    upload_2022-6-20_10-2-24.png

    Here is the Pasithea for comparison:

    Pasithea (ladder NOS DAC) Phase
    upload_2022-6-20_10-11-3.png

    Pasithea (ladder NOS DAC) Group Delay
    upload_2022-6-20_10-13-51.png
    Note group delay in lows is the result of AC coupling in the APx555.
     
    Last edited: Jun 20, 2022
  19. purr1n

    purr1n Burned out

    Staff Member Friend BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    80,585
    Trophy Points:
    113
    Location:
    Padre Island CC TX
    The reconstruction filter (anti-alias, low-pass, etc.) is something all together different. It's an analog filter.

    I understand why there is this misconception that filters screw phase up. Very early CDs players, and I owned one from Sony, where NOS. They did not oversample. They also sounded like ass. (That would be cool if I can find my old Sony CDP in my parents house and measure it). It was surmised that they sounded like ass because the reconstruction filter was something like an 12-pole 72db/octave (don't quote me, but you guys get what I mean) brick-wall filter. Since the sampling rate was 44.1kHz, there was only octave to give. That's not any room at all! Thus a good part of the transition band (just above the pass band) was going to end up in the top part of the audio band. And the transition band is going to contain most of the massive phase-shift and group delay that comes with a brick-wall filter. Yeah, life sucks for engineers.

    At least this was the foregoing theory of why early NOS CDs sounded like ass. Note that it's only until fairly recently that NOS DACs without analog reconstruction filters came about, letting the bandwidth limitations of the downstream do the work. However, engineers of big companies being engineers for big companies, needed to design for a variety of situations, including the guy with the high-bandwidth solid state amp and tweeter or ribbons that would extend out to 60kHz. They didn't want people blowing up their tweeters, especially with amps that could oscillate when they saw a 44.1kHz signal (remember, digital was new), so they had to implement the brick-wall filters.

    This is where oversampling comes in. We move the sampling rate higher and we end up with much more room to work with. X2, X4, X8. Some of you guys will remember this. The analog filters don't have to be as steep. The transition bands can be moved farther away from the passband. The audio band can sit totally inside the passband. (FWIW, I believe the X8 OS Yggdrasil DACs have a 6db/octave filter around 100kHz). The use of linear filters for the oversampling part will further ensure that the dots don't get jumbled, that there's no phase shift and that group delay from 20-20kHz is consistent. The only downside to a linear filter is lag. They are computationally expensive.

    Minimum phase is great because the lag is minimal. If we want to use digital mixer DSP effects, we have no choice but to use minimum phase processing. However, minimum phase screws stuff up as @ultrabike demonstrated here: this: https://superbestaudiofriends.org/i...n-ringing-and-shit-like-that.2627/#post-70767

    However, and the end of the day, it doesn't matter. It's what sounds good to you. Measurements are interesting. Science is interesting. But we aren't ASR. Just no false science.
     
    Last edited: Jun 20, 2022
  20. skem

    skem Friend

    Friend
    Joined:
    Nov 1, 2017
    Likes Received:
    1,865
    Trophy Points:
    93
    Location:
    Charles River
    Thanks for the corrections Marv! You're right that once I raised linear-phase filters my criticism should've focused on time domain. Technically there are equivalencies, but I did not use the correct language here.
     
    Last edited: Jun 20, 2022

Share This Page