COS Engineering? Anyone heard of them?

Discussion in 'Digital: DACs, USB converters, decrapifiers' started by Rockin_Zombie, Mar 27, 2017.

  1. Rockin_Zombie

    Rockin_Zombie Facebook Friend

    Joined:
    Jan 2, 2017
    Likes Received:
    117
    Trophy Points:
    43
    Location:
    Toronto
    Was lurking Audiogon and bumped into this all-in-one unit. Retail seems to be around 2500 USD.
    They are from Taiwan.
    http://www.cosengineering.com/products/h1/

    [​IMG]

    They have a DAC only version too, which is much more expensive ($7000 I think).

    [​IMG]

    A review here:

    http://inearspace.com/2017/02/09/cos-engineering-d1-h1-beauty/

    Had to jump through all the flowery mumbo-jumbo to figure out they are using PCM1792, so nothing original I guess. The two unique features that they boast on the website:

    "COS’s proprietary algorithm with a linear-phase delay FIR can up-sample original data to 176.4K or 192K, and this massive computation is handled by a powerful 3648 MMAC/second DSP."


    And the second one that I found slightly more intriguing:

    "H1 stores digital data in a one-second buffer before converting to analog signals, under the coordination of a crystal oscillator with a jitter precision less than 1ps and a dedicated re-clocking circuitry."

    They actually have an option to turn off the 1 sec delay for videos. So what are they doing with such a huge buffer? Is this all snake oil?
     
  2. FallingObjects

    FallingObjects Pay It Forward

    Pyrate
    Joined:
    Oct 15, 2016
    Likes Received:
    2,235
    Trophy Points:
    93
    To me, it sounds more snake-oily than actual improvement. If you can't dazzle them with brilliance, blind-side them with bullshit.

    Edit: Logic is as follows; you can't really improve music beyond what the source file is, so anything that can happen in that one second buffer isn't going to make your recording part the heavens and sing like a chorus of angels.
     
  3. Grahad2

    Grahad2 Red eyes from too much anime

    Pyrate
    Joined:
    Jan 26, 2017
    Likes Received:
    1,162
    Trophy Points:
    113
    Location:
    Singapore
    So that input jitter is hopefully isolated from the system. Like what async USB is supposed to do.

    6ads have reviewed it as well: http://www.6moons.com/audioreviews2/cos3/1.html
     
  4. Rockin_Zombie

    Rockin_Zombie Facebook Friend

    Joined:
    Jan 2, 2017
    Likes Received:
    117
    Trophy Points:
    43
    Location:
    Toronto
    1 sec is a lot of data. I have had ASIO drivers cut the first second of the first song off, but this one seems to always have a one second delay (that's why the option to disable it for video). So this does not seem like the typical ASIO delay. I haven't heard any other DAC flaunting this "feature" (or flaw more like), that's what piqued my interest.
     
  5. Grahad2

    Grahad2 Red eyes from too much anime

    Pyrate
    Joined:
    Jan 26, 2017
    Likes Received:
    1,162
    Trophy Points:
    113
    Location:
    Singapore
    This is not delay but a 1 second data buffer on the DAC itself.
     
  6. Rockin_Zombie

    Rockin_Zombie Facebook Friend

    Joined:
    Jan 2, 2017
    Likes Received:
    117
    Trophy Points:
    43
    Location:
    Toronto
    What I am saying is the buffer is introducing some form of delay, they have a dedicated switch to turn the buffer off for video. This is from the manual (I am bored so I actually downloaded their ZIPPED manual and unzipped it, who uploads manual, a single .pdf file, in .zip format?):

    **********
    There is a buffer switch on the back panel, and it should be turned on for optimum performance. Sometimes digital music data do not move along and get converted in perfect tandem, which causes jitters, and even a few micros-seconds' timing error is enough to perturb the ears and frustrate the mind. Therefore, H1 uses a buffer of one-second depth, along with an independent and accurate clock, to receive data, align them and send them out in precise time frames for conversion.
    For videos, the buffer should be switched off. This selection makes H1 reduce the depth of the buffer a little to ensure video-audio synchronization. Switching the buffer on or off takes effect immediately.

    ***********

    Either their clock circuitry/code isn't optimal to ensure synchronization or they are doing something new.
     
  7. Clemmaster

    Clemmaster Friend

    Pyrate Contributor
    Joined:
    Sep 28, 2015
    Likes Received:
    3,268
    Trophy Points:
    113
    It's nothing new. Although, the fixed 1s duration is both high and probably unnecessary.
    • The Genesis Digital Lens already featured something similar, with one big exception in that the actual delay depends on the input jitter: the lower the jitter, the less delay.
    • Madrigal features something similar in the Proceed AVP (which, IMO, is the greatest value in multi-function DACs today. Can be had for ~$450. It used to cost $6k).
    • Pretty sure the digital flywheel® of Enlighted Audio Design works similarly.
    • MSB uses something similar, as well, but it is disabled for 48kHz contents (typically rate of video/TVs output), to alleviate the sync issue.
    • The ALLO Kali I2S reclocker (for Raspberry Pi) works the same, with (fixed) 0.7s of buffering.
    • ...
     
  8. Clemmaster

    Clemmaster Friend

    Pyrate Contributor
    Joined:
    Sep 28, 2015
    Likes Received:
    3,268
    Trophy Points:
    113
    Schiit's Manhattan Project might have you reconsider.
     
  9. FallingObjects

    FallingObjects Pay It Forward

    Pyrate
    Joined:
    Oct 15, 2016
    Likes Received:
    2,235
    Trophy Points:
    93
    I can understand having a clean digital source for output, I just fail to understand how an arbitrary duration for a buffer is better than just having a large buffer store all the information at once, then output it as needed.
     
  10. Clemmaster

    Clemmaster Friend

    Pyrate Contributor
    Joined:
    Sep 28, 2015
    Likes Received:
    3,268
    Trophy Points:
    113
    You don't want to have to wait 30s between the moment you start playback and the moment it actually plays on the speakers.

    I remember the times of Airfoil that I used to stream music from my windows PC to the Airport Express. It had approximately 3s delay, which made navigating in the songs and the overall playback of audio a horrendous experience (you start the song and it doesn't play right away. It's very disconcerting).
     
  11. FallingObjects

    FallingObjects Pay It Forward

    Pyrate
    Joined:
    Oct 15, 2016
    Likes Received:
    2,235
    Trophy Points:
    93
    Yeah, that's what I mean though. If you can just increase the total buffer size (So store 100mb of info at once, for example), instead of just storing a flat 3 seconds of audio (and have that scroll through), couldn't you just have continuous and instantaneous playback?

    If the concern is 'get away from jitter' then that should be able to be done more or less instantaneously by the DAC, no?
     
  12. Clemmaster

    Clemmaster Friend

    Pyrate Contributor
    Joined:
    Sep 28, 2015
    Likes Received:
    3,268
    Trophy Points:
    113
    Only if your device with 100mb is your playback device. You're just shifting the problem here.

    The buffering that matters, ultimately, is the one performed just before conversion to analog, inside the DAC. This is the one that can* remove jitter induced by all the components in the chain downstream.

    (*) "can". It doesn't necessarily remove everything. Also, the process of reducing jitter might generate other issues inside the DAC, like ground noise, which are also detrimental.

    Now, I'm not saying that buffering before this critical element (the DAC) is irrelevant. Devices that aim at reducing jitter between the transport and the DAC can have positive effects, reducing the total amount of transport jitter (and also provide electrical isolation/filtration, which is always a good thing).

    From experience, the DACs that feature intelligent buffering are the least sensitive to transport jitter. They tend to sound good even with the crappier sources.
    They are usually more expensive, because it adds complexity to the design and requires more knowledge than simply following the cook book and reference design of a DAC chip (which any EE can do). It also adds parts to the design, so the BOM is higher, too.
     
    Last edited: Mar 29, 2017
  13. FallingObjects

    FallingObjects Pay It Forward

    Pyrate
    Joined:
    Oct 15, 2016
    Likes Received:
    2,235
    Trophy Points:
    93
    Mmm, I think I'm starting to piece it together more now. Thanks!
     
  14. bengo

    bengo Friend

    Pyrate
    Joined:
    Apr 17, 2017
    Likes Received:
    1,371
    Trophy Points:
    93
    Location:
    Divisive Kingdom
    Home Page:
    Some of the Chord DACs also do this RAM buffering, from memory I think the DAVE allows 1s or 4s buffer, or you can switch it off.
    Even my little iFi SPDIF iPurifier has an (unspecified size) buffer, and that is a lot cheaper than a DAVE.
    As written above, the purpose is to reduce jitter. With a longer buffer, you can average out jitter with a longer period.

    I've actually heard the COS D1. If you like Burr Brown it sounds pretty nice.
    It's a big solid thing with some fancy metal casework.
    Is it worth £7k or more? Only you can answer that.
     
  15. Torq

    Torq MOT: Headphone.com

    Pyrate
    Joined:
    Jan 15, 2016
    Likes Received:
    8,193
    Trophy Points:
    113
    Location:
    Seattle, WA
    It doesn't really work like that - you don't "average out" jitter.

    Buffered re-clocking schemes only need a long enough buffer to ensure that the buffer doesn't empty during the re-clocking process. And all that process is, or has to be, is to take data out of the buffer under the control of the most accurate clock possible. Re-clocking accuracy does not, and in fact cannot, improve simply because the buffer is longer. The size of the buffer is entirely immaterial with regard to the rate and accuracy with which sample data is clocked out of it and fed to the DAC chip.

    Anything that happens IN the buffer is always going to be limited by the output from it ... rendering anything you do upstream of that irrelevant to jitter performance.

    If your source can't keep up with feeding data at a suitable rate then a larger buffer will reduce the incidence of drop-outs, at the cost of increased latency, but improving jitter performance ... that a bigger buffer won't do. And, outside egregiously inaccurate/flaky sources, with absolutely horrendous clocking (say, an early CD player attempting to use the embedded disc servo data as the "clock") a few dozen samples is more than a sufficiently large buffer.

    If your source remains accurate to within just one sample per rate period, which in practice would mean jitter on the order of 5µs (or 100 times worse than what a typical audio-grade clock or simple CoTS PLL will manage today and five to six orders of magnitude worse than some of the better clocks), you only need a two-sample buffer to ensure the re-clocking process won't cause drop-outs AND is correcting jitter to the maximum possible degree. And that's a "damn-near-broken" level of clock performance for any reasonably modern, entry-level, source.

    Bear in mind this sort of thing is done by EVERY USB 2.0 Async compatible DAC as a matter of necessity.
     
    Last edited: May 3, 2017
  16. GoodEnoughGear

    GoodEnoughGear Evil Dr. Shultz‎

    Pyrate
    Joined:
    Oct 25, 2015
    Likes Received:
    3,070
    Trophy Points:
    113
    Location:
    Cape Town, South Africa
    @Torq, you have the patience of Job. You're a class act.
     
  17. bengo

    bengo Friend

    Pyrate
    Joined:
    Apr 17, 2017
    Likes Received:
    1,371
    Trophy Points:
    93
    Location:
    Divisive Kingdom
    Home Page:
    Thanks @Torq. I guess I had (wrongly) assumed that the buffer output clock was generated by tracking the input clock, a "running average" if you like. So in this scenario, the bigger buffer would allow you to average out any input clock frequency variation over a longer period. For example if you have periodic jitter at 60Hz coming from a bad power supply. That's what I meant anyhow, even if it is a load of ****
     

Share This Page