CHANGSTAR: Audiophile Headphone Reviews and Early 90s Style BBS
Lobby => Amp and DAC Measurements => Topic started by: Hands on December 06, 2013, 12:28:01 PM
-
Test Hardware and Setup
1. Acer Aspire 5560G Laptop - Locked at 1.4GHz and 0.9125V via K10Stat, dedicated GPU disabled, internal fan disconnected (low temps and stability verified), most Fidelizer tweaks applied, and other hardware, software, and services disabled.
2. JKSPDIF MK3 - Input with generic USB cable, output with 75ohm RF attenuator and 75ohm BNC to RCA Coaxial cable to Metrum DAC (looks like Blue Jeans? Cable came with used products, not sure)
3. Metrum Acoustics NOS Mini Dac Quad - Blue Jeans RCA cable, w/ stock external 7V PSU using generic 3-prong power cord from a computer PSU
4. Creative X-FI HD USB (SB1240) - Takes line-out signal from Metrum DAC to line-in for recording purposes, powered by USB bus, connected to Acer laptop. Unit is limited to 48/96KHz input/output. Measurements of this unit are in post 6.
Software
Test software used was RightMark 6.3.0. All 16-bit tests were done in DirectSound and Kernel Streaming modes at 48KHz and 96KHz. 24-bit tests would not work in Kernel Streaming mode for some reason, so they were limited to DirectSound only. You will see in the results that these do produce slightly different results. When listening to music, I have no issues running at 24-bit, 96KHz or greater, and with Kernel Streaming. When testing, I matched the input/output sound devices to the same bit-depth and sampling rate in the sound options to what I had set in RightMark. Also, I'm using the free version of RightMark, so I'm somewhat limited when it comes to testing options.
I also ran one test where I used XXHighEnd with Arc Prediction enabled as the playback software and RightMark for recording/measuring.
Update 12/7: Used ARTA to run some other measurements. See 2nd page for updates. These included impulse response, FR, phase, sine wave at various sampling rates, square wave, spectrum, etc.
Misc.
I will also note a few things. First, I'm fairly certain the Creative SB1240 does not measure flat down to 20Hz. There is a slight roll-off. I have measured the Metrum DAC via the line-in on my desktop to confirm it performs flat down to 20Hz. That's the only purpose that particular line-in served, as it was otherwise on a noisy motherboard and gave poor results.
Second, I only ran these tests once each after figuring out optimal settings. Considering that and the cheap, simple nature of my setup, you can expect to see some issues and artifacts that might not be repeatable, especially on a nicer setup.
Third, do notice that the scale and range of the graphs can change depending on the test configuration. Keep that in mind. Some of the images are not attached to posts in the same order as well.
I am attaching all of the Metrum DAC measurement result files and HTML versions to this first post and will touch on a few select examples in following posts.
I'll continue to update this thread assuming I try out other testing methods.
Links
Metrum Acoustics Quad NOS DAC - http://www.metrum-acoustics.nl/NOS_mini_DAC_English.html (http://www.metrum-acoustics.nl/NOS_mini_DAC_English.html)
JKSPDIF MK3 - http://www.johnkenny.biz/home-1/mk3-hiface (http://www.johnkenny.biz/home-1/mk3-hiface)
Creative SB1240 - http://us.creative.com/p/sound-blaster/sound-blaster-digital-music-premium-hd (http://us.creative.com/p/sound-blaster/sound-blaster-digital-music-premium-hd)
RightMark Audio Analyzer - http://audio.rightmark.org/index_new.shtml (http://audio.rightmark.org/index_new.shtml)
ARTA: http://www.artalabs.hr/ (http://www.artalabs.hr/)
-
16-bit/48KHz Kernel Streaming Mode Measurements:
Frequency response (from 40 Hz to 15 kHz) | dB +0.17, -1.49 | Average
Noise level | dB (A) -94.9 | Very good
Dynamic range | dB (A) 91.6 | Very good
THD | % 0.037 | Good
THD + Noise | dB (A) -66.7 | Average
IMD + Noise | % 0.044 | Good
Stereo crosstalk | dB -92.6 | Excellent
IMD at 10 kHz | % 0.216 | Average
General performance | Good
Graphs attached to post
-
24-bit/96KHz DirectSound Mode Measurements:
Frequency response (from 40 Hz to 15 kHz) | dB -0.01, -0.59 | Good
Noise level | dB (A) -98.6 | Excellent
Dynamic range | dB (A) 95.1 | Excellent
THD | % 0.054 | Average
THD + Noise | dB (A) -63.7 | Poor
IMD + Noise | % 0.056 | Good
Stereo crosstalk | dB -91.5 | Excellent
IMD at 10 kHz | % 0.082 | Good
General performance | Very good
-
This is another 16-bit/48KHz test, but this one is using DirectSound mode. See how it compares to Kernel Streaming mode in my second post:
Frequency response (from 40 Hz to 15 kHz) | dB +0.10, -1.56 | Average
Noise level | dB (A) -90.0 | Very good
Dynamic range | dB (A) 86.4 | Good
THD | % 0.054 | Average
THD + Noise | dB (A) -63.7 | Poor
IMD + Noise | % 0.059 | Good
Stereo crosstalk | dB -89.5 | Excellent
IMD at 10 kHz | % 0.108 | Average
General performance | Good
-
Here are measurements using XXHighEnd for playback instead of doing playback and recording both through RightMark. 24/96 file was generated in RightMark. Had a difficult time getting levels matched properly on this, so I can't vouch for the accuracy of this one! (Though, I do think XXHighEnd sounds good...preferring it to JRMC so far...)
This was with Engine 3, which is based on WASAPI. I normally listen with Engine 4, which is based on Kernel Streaming, but the option was unavailable due to the measurement setup and configuration.
Frequency response (from 40 Hz to 15 kHz) | dB +0.06, -0.52 | Good
Noise level | dB (A) -102.5 | Excellent
Dynamic range | dB (A) 102.9 | Excellent
THD | % 0.037 | Good
THD + Noise | dB (A) -66.6 | Average
IMD + Noise | % 0.045 | Good
Stereo crosstalk | dB -98.4 | Excellent
IMD at 10 kHz | % 0.253 | Average
General performance | Very good
-
FOR REFERENCE: Creative X-Fi HD USB (SB1240) external loopback measurements at 24/96 using Blue Jeans RCA cable
Frequency response (from 40 Hz to 15 kHz) | dB +0.01, -0.08 | Excellent
Noise level | dB (A) -98.4 | Excellent
Dynamic range | dB (A) 98.6 | Excellent
THD | % 0.0016 | Excellent
THD + Noise | dB (A) -89.5 | Good
IMD + Noise | % 0.0045 | Excellent
Stereo crosstalk | dB -90.9 | Excellent
IMD at 10 kHz | % 0.0049 | Excellent
General performance | Excellent
It's a solid little unit! They go on sale sometimes for around $60. Still, it's worth factoring these results into the Metrum Quad measurements. Consider that a loopback test is not optimal, nor do I have TOTL equipment and setups behind it. According to this link, its real-world/maximum line-in performance can go beyond this (about -104dB with noise and DR):
http://marlene-d.blogspot.com/2012/01/creative-labs-soundblaster-x-fi-hd-usb.html (scroll down a bit below half way for a table)
-
As to be expected, the NOS treble roll-off is evident in these measurements. At 88.2KHz or above, this becomes much less of an issue (no more than -1dB at 20KHz, with 96KHz sampling). I find it helps me listen more comfortably and develop less fatigue over time. I do prefer a slightly dark sound as-is. The roll-off really isn't that noticeable at 88.2KHz or above, but you can definitely hear it at 44.1 or 48KHz. Setting the hardware to a higher sampling rate and using software upsampling can bring some very noticeable benefits. At that point, you can get into the fun of picking different upsampling methods. I'm currently using XXHighEnd's Arc Prediction method, as that is supposed to have no pre or post ringing.
You'll also notice the weird IMD measurements, though these do almost match the 0.04% distortion as posted on Metrum's website. Based on what I've learned about NOS DACs recently, I had been told to expect issues with harmonic and IM distortion. THD does show spikes at each 1KHz interval but is clean otherwise. THD + Noise was generally rated as average or poor.
IMD really looks interesting, though. I verified it looked this weird on my Auzentech sound card's line-in as well. I don't understand all the science behind this, so I'm not sure if this is typical of NOS DACs, unique to the Metrum implementation, or even how this will necessarily affect the music we listen to. I'm also not sure how to read IMD swept tones measurements nor do I know if they really tell us much. I'll need to do more reading on IMD...
Other than what I've noted, measurements show a lot of good qualities as well. You can see situations where the DAC pushes the limits of the A/DC. It might not be measuring at reference-level quality, but it does well enough for the most part to not leave me worried as long as it sounds good to my ears.
Unfortunately, I'm not at a position where I can give good, meaningful, and trustworthy impressions about DACs or the differences between them. I'd have to hear a lot more gear and do many more listening tests before I would feel comfortable making anything more than a guess.
That said, I can say with the Vali or Leckerton UHA-6S Mk.II and modded Tascam TH-02 (XXHighEnd software), I'm not hearing anything that's off about the sound right away. In fact, during my Vali listening tests, I found the Vali with Tascam did a fantastic job resolving fine, quick details with drums and cymbals (think fast progressive metal, anything that is easily smeared over). It did well outside of drums and cymbals, of course, but I was most struck by that right away. I found it very easy to track and place almost all aspects of what I was hearing. And, yet, I did not notice anything sounding harsh or unnatural (outside of the usual harshness of my go-to bands and genres). Very nice. Also noticed good dynamics, plenty of layering and depth to the sound with nice space between instruments and vocalists. Again, very happy overall so far...had a lot of fun listening and digging into the fine details of the music.
Now, I figure that if the Vali with Tascam was sounding that good, the DAC behind it had to be at least just as capable in order to present that. It might just be a good combo that works well together. I can't say for sure, and my listening tests have been limited to just the Tascam so far with those 2 amps. Take all of this with a grain of salt, but so far I am pleased. I'm also not necessarily saying it's the bestest setup I've heard either, just to make that clear. I'll have to listen more.
If anyone has any ideas on other tests I can and/or should run with my current setup (don't really want to pay money for more hardware or software...), let me know! I believe it was ultrabike that mentioned ARTA to me, so I might try that and REW to see what results they provide. Anything else? Questions? Thoughts? Tips?
-
Interesting results. Thanks!
I've been looking into NOS DACs myself as well. Would be good if you could measure impulse response, seeing as how that's one of the supposed advantages of NOS DACs. Should be a way to do that with ARTA IIRC.
-
I would like to figure out ways to test impulse response, sine and square wave response, jitter, etc. Not sure what, if any, additional hardware or software (not free) I'd need for that. I'll look more into what ARTA can do.
Edit: Looks like ARTA might offer this functionality. I'll experiment with it tonight and post my results.
-
+3 karma for you.
-
Thanks Hans!!!
For spectrum stuff ARTA does a pretty good job IMO. For time domain stuff the free version TrueRTA seems pretty decent.
Interesting results with the Metrum DAC. :money:
-
Thanks.
I don't understand that after centuries with better tech people are still buying this.
It just breaks the very basis, the sampling theorem, and therefore fails to reproduce the signal unless it has a very low frequency (relative to the sampling rate).
If someone prefers the looks of a narrow spike as impulse in the time domain then that someone should seriously consider reading a book about digital audio.
-
Mind giving a brief explanation xnor?
-
Digital audio is bandlimited. In case of CD audio to below 22.05 kHz.
An infinitely thin spike (the perfect impulse) contains all frequencies from 0 to ∞ Hz. When there is an impulse on a CD it is bandlimited to below 22.05 kHz, but non-oversampling DACs don't do bandlimited interpolation so they don't reproduce what's on the CD. Instead they usually follow the samples in a stair-step like fashion. So the narrow spike is actually a lot of extraneous spectral content aka images.
At high frequencies relative to the sampling rate the attempt of reproduction (can't call it reconstruction) breaks down.
Briefly: they break the sampling theorem, they produce images which can lead to nasty IMD (with intermodulation products in the audible range)
-
I don't understand how that is an issue exclusive to NOS DACs. Doesn't the LPF take care of the images regardless?
-
Well, this certainly does not exhibit what I would call pretty IMD according to measurements. That said, I'm not sure what to listen for to detect IMD. So far with listening I haven't noticed anything particularly out of the ordinary, but I also can't say I've listened long enough or compared with other gear directly to say that definitively. I would consider loaning it out to some of the more experienced folks around here for more critical listening tests if interested.
If I'm remembering correctly, many of the inherent issues of a NOS DAC can be mitigated by setting the sampling rate high and either using "HD" music or going the software upsampling/oversampling route (I still forget the technical differences between the two terms, TBH). These measurements show little difference in terms of IMD, but I do believe I remember seeing it greatly help clean up the form of sine waves up to 20KHz. And, as mentioned, going the software upsampling/oversampling route gives you greater control in picking out the filtering methods.
I also read that many NOS DACs are sensitive to feedback and reflections. One fellow posted measurements showing how impedance mismatches across the chain can cause reflections:
http://www.head-case.org/forums/topic/9905-metrum-acoustics-octave-a-nos-digital-filter-less-dac/?p=496655
Not sure how that might affect the measurements I did or how it would affect music playback/what we hear. I believe he replaced the DAC's coax jack with a BNC one. I've heard it's easy and recommended to do.
There is without a doubt inherent compromises with NOS DACs and a good bit of disagreement as to whether those compromises can be worth it. I think these measurements show that perhaps NOS DACs can be designed in such a way to minimize the amount of compromises taken, though I'm wondering if the best strategy will also pair this with a strong software back-end with good upsampling methods. That's the idea behind XXHighEnd and the Phasure NOS DAC, at least.
It's unfortunate that the Metrum's DAC chips themselves have the tops scratched off. More can be said about other NOS DACs because the chips they use are known. From all my research, no one has yet definitively figured out which chip is in the Metrum DACs, though there are some good guesses and possibilities.
-
By the way, I've attached all four measurements I did of the SB1240 to this post, since I did match all bit depths and sampling rates during tests. Just in case anyone wants to get more in-depth when looking at these.
-
I don't understand how that is an issue exclusive to NOS DACs. Doesn't the LPF take care of the images regardless?
Nope, only when you were to use a brickwall that is just above the 1/2 fs.
When a DAC can do multiple sample rates you either have to shift that filter or pass it though another filter.
No one does that and IF they did the squarewave and needle reproduction would not look as nice any more and that is what attracts buyers thinking that they have superior 'attack'. The squarewaves/needles would all be showing (severe) post ringing.
In Hans's case there is a brickwall filter in place though just above 12kHz and it is called TH-02 ::) that will at least get rid of the generated 'bit steps' as the driver simply cannot 'follow' those fast transients (ultrasonic sample frequency steps) but the (low level) generated harmonics of all signals below 5kHz will still be present (not really audible though as the distance is too big).
The higher the sample rate the higher the frequency of the steps.
At say 88kHz no headphones will be able to 'follow' those steps but some amplifiers may have issues with it.
When you use the DAC with higher (upsampled) sample rates you 'shift' part of the garbage to higher frequencies (its where DeltaSigma is based on) so the DAC will perform better when you upsample redbook to 176 or 352.
In an upsampling DAC that basically does the same but the algorithm is locked in its firmware.
Hans has a point where he mentions you can choose the upsample algorithm to be used with a NOS DAC and is clearly shown that Peter's algorithm does this better than the ones in the PC by looking at the numbers.
As a side note... Both the Metrum and XXHighEnd are designed by fellow countrymen (Dutch) :-Z
On sound reproduction I am on Xnors camp...
All sounds on a CD are brickwalled and transients faster than a 22kHz sinewave can reach are NOT recorded but when looking at the bits (impulse and squarewave) do not have this 'limit' and reproduce a relatively slowly rising 20kHz sinewave as a 1MHz squarewave with all its harmonics.
BUT that sharp edge has NEVER been present in the original signal and if it has it was lost in the encoding so now ALL bit level changes (those in sinewaves as well) always produce very high harmonics that are NOT present in the original signal.
Yes these DAC's produce very nice squarewaves and needle pulses BUT these signals are NEVER present in ANY audiosignal and don't even sound nice. But it LOOKS really good and 'fast' on scope images.
Sound consists of sinewaves and they are NOT accurately reproduced by any NOS DAC instead they look like this:
(http://www.computeraudiophile.com/attachments/f6-dac-digital-analog-conversion/4045d1359969766-should-i-nos-dac-steps01.png)
the picture is from Computer Audiophile and uploaded by PeterSt (the XXHE developper)
Pick your poison... a perfect looking squarewave or needle pulse as a sales argument (as they are never present in music, and also not in 'attack' of any instrument nor are these attacks encoded as Xnor explained briefly) but terrible at sinewave reproduction and high levels of distortion / HF energy.... versus ... excellent sinewave reproduction (music consists of sinewaves) and less good looking squarewave/needle pulses that do not exist in music and have harmonics HIGH above the audible band and WON't be reproduced by the transducers anyway. Well maybe some good speakers but do you really want high amounts of supersonic, UNRELATED to the audiosignal, frequencies 'energy' in your tweeter nor in your amp.
What is clearly seen in the provided plots (I have an SB1240 myself as well b.t.w.) is the added harmonics all across teh audible band reaching -70dB and NOT decaying for higher harmonics.
This is just 1 simple tone imagine what amount of harmonics you would be getting when you play music with a plethora of frequencies between 60Hz and 10kHz mainly... they will ALL be producing harmonics at THESE level DISTANCES so a signal at -20dB will also have harmonics but 20dB below the other ones that are louder.
This basically means that when you do a S/N ratio test you will get impressive numbers but as soon as you play music the noise level rises to about -70dB !
Fortunately it only rises when music is present so in quiet passages there is still a >70dB S/N ratio but the S/N ratio embedded in the recording (mic noise) will always be present and almost at the same level so might 'mask' that added noise.
Distortion levels < 70dB are inaudible anyway (acc. to blind tests not acc. to sighted tests).
So even though the numbers appear quite bad I highly doubt it will be audible.
What is audible is the HF roll-off (certainly when playing redbook) but a lot of people do not find that objectionable (I do).
These DACs can still sound good simply because our hearing isn't as good as we think it is. It certainly isn't because of rise times.
Take a good look at the 'attack' in any music piece at 'step level' in something like Audacity and see what 'rise times' are actually present in that signal.
Good luck in hunting for steps that are faster than 1/2 fs. You will find even fast transients in music take a few samples at least.
Just my thoughts...
I would be using that expensive DAC with upsampling only and aim for the highest possible bit rate / bit depth it allows.
XXHE seems to deliver a better upsampled 'image' than the default windhose drivers do...
-
Yup.
I just started reading the Metrum linky marketing info:
http://www.metrum-acoustics.nl/NOS_mini_DAC_English.html (http://www.metrum-acoustics.nl/NOS_mini_DAC_English.html)
It seems they do not use any sort of analog filtering to smooth out the Sample & Hold (S/H) out the DAC. The S/H deal applies a sinc filter to the output, but it's a bit crude. It will have some amount of ultrasonic stuff which may or may not cause IMD issues in the audible range (like Xnor pointed out).
For background they link to this page:
http://www.sakurasystems.com/articles/Kusunoki.html (http://www.sakurasystems.com/articles/Kusunoki.html)
IMO the paper has quite a bit of issues:
1)Through out his discussion, it is not 22.7 s or 22.7 ms, it's 22.7 us: 1/44.1 kHz.
2) I fail to see the jitter issues he is alluding too. And his "maximum limit of the acceptable error (maximum limit of the jitter)" makes little sense to me. I think he somewhat feels 0.5 LSB error is equivalent to 173 ps for 16-bit @ 44.1 KHz and 1.35ps for 20-bit @ 8x that. IMO it is not. It depends on the signal. To me jitter performance is data, noise, clock and PLL BW dependent. Adding more bits or interpolating should not introduce significant changes in the audio signal bandwidth, clock or PLL BW. It may add quantization noise (may not depending), but don't think that would be an issue if things are properly implemented.
3) A FIR filter will require more processing bits depending on stages. But 32-bits seems plenty if the FIR is properly implemented. His opinion that a 16-bit NOS is more accurate than FIR processed 20-bit 8x OS is not necessarily accurate IMO. The ENOB of the whole conversion process might be more telling. This can be calculated from the signal SINAD and PAR. If THD+N and IMD don't look too hot, neither will ENOB... so there goes that.
4) K's (Kusunoki) discussion of FIR filtering operation is also somewhat flawed. The delay taps are 22.7 us apart for 44.1 kHz... and 2.8 us effectively apart for a classic 8x interpolation filter... And even if the filter length was several thousands of samples, the length of the filter alone does not say much about what the filter is doing to the signal...
5) There is not trade off in time domain performance vs. frequency domain performance. If one is effed-up, so will the other. Going to the pulse and square wave plots at the original Metrum link, I wish they had included the sine wave plots at 1 kHz from K... They would look like a Mayan pyramid in the time domain (see Solderdude's post above). The "obvious notches" as he calls the S/H operation might not be detected by human ear, if not for issues with IMD and stuff (as pointed by Xnor), which goes back to ENOB.
6) IMO his PLL discussion based on the xyz ps criterion is flawed... because it's based on the xyz ps criterion which I felt was flawed.
Anyhow, while I don't agree with half of what K said, that does not mean that the Metrum DAC performs horribly. Hanns did a great job describing what he heard, and makes sense to me:
As to be expected, the NOS treble roll-off is evident in these measurements. At 88.2KHz or above, this becomes much less of an issue (no more than -1dB at 20KHz, with 96KHz sampling). I find it helps me listen more comfortably and develop less fatigue over time. I do prefer a slightly dark sound as-is. The roll-off really isn't that noticeable at 88.2KHz or above, but you can definitely hear it at 44.1 or 48KHz. Setting the hardware to a higher sampling rate and using software upsampling can bring some very noticeable benefits. At that point, you can get into the fun of picking different upsampling methods. I'm currently using XXHighEnd's Arc Prediction method, as that is supposed to have no pre or post ringing.
That said, I can say with the Vali or Leckerton UHA-6S Mk.II and modded Tascam TH-02 (XXHighEnd software), I'm not hearing anything that's off about the sound right away. In fact, during my Vali listening tests, I found the Vali with Tascam did a fantastic job resolving fine, quick details with drums and cymbals (think fast progressive metal, anything that is easily smeared over). It did well outside of drums and cymbals, of course, but I was most struck by that right away. I found it very easy to track and place almost all aspects of what I was hearing. And, yet, I did not notice anything sounding harsh or unnatural (outside of the usual harshness of my go-to bands and genres). Very nice. Also noticed good dynamics, plenty of layering and depth to the sound with nice space between instruments and vocalists. Again, very happy overall so far...had a lot of fun listening and digging into the fine details of the music.
Now, I figure that if the Vali with Tascam was sounding that good, the DAC behind it had to be at least just as capable in order to present that. It might just be a good combo that works well together. I can't say for sure, and my listening tests have been limited to just the Tascam so far with those 2 amps. Take all of this with a grain of salt, but so far I am pleased. I'm also not necessarily saying it's the bestest setup I've heard either, just to make that clear. I'll have to listen more.
-
I'm really enjoying the discussion and information, even if some of it goes over my head! Lots to learn. I went into the NOS DAC purchase fully aware of the inherent issues (at least, as far as I understood it all). I just wanted to try something different that wasn't Delta Sigma (even if worse in most regards), and the buzzwords thrown around about it sounding "not digital" helped rope me in, as I seem to get listening fatigue rather easily and wanted to try a different avenue entirely. I was also disappointed at the lack of NOS DAC measurements and wanted to do something about that. In the end, I figured it would be a fun project of sorts! I was legitimately just too curious to not try one. The Metrum was a good option, because it sat right in between the price of the crappy eBay NOS DACs and the ones costing well over $1K, plus it came used with the JKSPDIF. It was more just timing and luck than me specifically picking out this unit, though I did do at least some research before buying!
Here is a post I found on sine wave response with a varying sampling rate:
http://www.computeraudiophile.com/f6-dac-digital-analog-conversion/metrum-acoustics-nos-mini-dac-octave-10867/index3.html#post151984
Still not perfect, but noticeably better. When listening, I go beyond 96KHz. I was just limited to that in measurements because of the SB1240.
solderdude did bring up the point that the Tascam TH-02 is certainly a limiting factor in my tests. While they do sound surprisingly good at times, I wanted to be transparent about everything I was doing simply because of how limited and imperfect my tests and environment are. It's a bit hard to find time to listen to the HE-500 when everyone else is asleep (I work 3rd shift and keep the same sleep schedule even on weekends). I will continue to say to take my subjective impressions with a grain of salt.
So even though the numbers appear quite bad I highly doubt it will be audible. What is audible is the HF roll-off (certainly when playing redbook) but a lot of people do not find that objectionable (I do). These DACs can still sound good simply because our hearing isn't as good as we think it is. It certainly isn't because of rise times.
Yes, I generally prefer a slight roll-off, but I understand why most others might not. But this really is the main thing I'm wondering. How audible are these downsides? How far can software take a properly implemented NOS DAC? You can find a lot of information on how NOS DACs are good or bad, but not much objective analysis on how much of that is audible or matters. It's always very black and white...
-
LOLs!!! Actually I've seen that Kusunoki paper being thrown around at some other forums I think, and it usually goes over people's heads, understandably so. I had to stare at it for some time.
Anyhow, I also think cans tend to be the limiting factor on most rigs anyway. Ha! IMO changing the sampling rate, which will change the LPF effect of the S/H has some cool factor to it. The slight roll-off would be welcome on some cans...
karma points! :)p1
-
Yes, I saw that paper before as well. I read more summaries about it than the paper itself, I think, ha.
I find myself wanting to EQ treble all the time, so the roll-off is really a big plus for my ears. Sometimes I think I should just buy an LCD-2 or HD650 and be done with it, LOL.
-
I'll be attaching some ARTA measurements in my next few posts. Let me know if you have any questions about my methodologies, notice any flaws that need corrected, want to offer other test ideas for me to run, etc.
I also can't necessarily say everything was measured correctly (did my best) or that I necessarily understand what all of these measurements mean or if they hold any value.
1. Periodic Noise Impulse Response 24/96
2. MLS Impulse Response 24/96
3. Sine Sweep Impulse Response 24/96
4. Frequency Response and Distortion generated by above sine sweep impulse 24/96
5. FR, Phase, and Cohesion (no idea if this is right, also no idea what cohesion is...this is a linear average using PN Pink Noise)
Yep, it's a NOS DAC.
-
These are not all to same scale necessarily. Manually zoomed in to get a better picture.
1. 1KHz Sine Wave - 16/48
2. 1KHz Sine Wave - 24/96
3. 10KHz Sine Wave - 16/48
4. 10KHz Sine Wave - 24/96
5. 20KHz Sine Wave - 16/48
6. 20KHz Sine Wave - 24/96
7. 20KHz Sine Wave - DAC set to 16/176.4, A/DC set to 24/96, and ARTA set to 16-bit, sine wave sampled at 192KHz
That last result is particularly interesting. These show how the NOS DAC's sine wave response improves at higher frequencies with a higher sampling rate.
-
1. 25Hz Square Wave 24/96
2. Zoomed-in view of the above. No idea if this has any value or means anything or not, but did notice the rough texture. Higher frequency sine waves would not allow me to zoom in as closely.
3. 25Hz Square Wave 24/96
4. 1KHz Square Wave 24/96
There is a good chance I did not set the square wave tests up optimally.
-
Do note FFT size and also compare with RightMark results. Software was set to 0dB but sometimes reports lower. I can't verify there aren't artifacts in here or not! When listening to music, I set the JKSPDIF to 0 no matter what. I don't think this makes a difference if using WASAPI or Kernel Streaming. But when doing these measurement tests, it was very hard finding the right balance of volume on the DAC/JKSPDIF and A/DC. When adjusting the levels on RightMark, I couldn't set the JKSPDIF and DAC to 0 like normal and had to balance that volume setting with the SB1240. While you could get the levels to match on a few different mixture of settings, some would introduce more distortion than others. I tried to narrow it down to produce the best, most consistent results possible. As such, assume these might contain artifacts due to a less-than-optimal measurement setup, but they should still hold good information and value regardless.
1. 1KHz Sine Spectrum - 0dB - Linear Avg - 24/96
2. 100Hz Sine Spectrum - 0dB - Linear Avg - 24/96
3. 60Hz and 7KHz Dual Sine Spectrum (4:1 ratio) - 0dB - Linear Avg - 24/96
4. 25Hz and 66Hz Dual Sine Spectrum (1:1 ratio) - 0dB - Linear Avg - 24/96
5. Noise - Linear Avg - 24/96
6. Noise - Peak Avg - 24/96
-
The irony is, if you associate old D/A converter performance with "digital sound" or "digitis", then non-oversampling DACs will give you exactly that.
Look at this:
(http://xserv.shell.la/xnor/audio/images/sampling.png)
The green dots are the samples. The blue sine is the input and output when correctly reconstructed.
Purple shows how non-oversampling DACs do it. Red is linear interpolation. Some audio editors display waveforms incorrectly that way.
@ #22: Notice how all the impulse responses look very different. It shouldn't make much difference if you use noise, MLS or a sweep to generate the impulse response but since a non-oversampling DAC just holds the last sample value.... And this is with the impulse response of course excluding any nonlinearities.
-
Play with the 'Wnd' settings box (Arta) that now says 'Hanning' and you will get sharper defined spectrum plots.
Also note that the steps that are (most likely) there in the output of the NOS DAC will be 'ironed' out a little by 'brickwall' filter in the ADC of the SB1240 card as well so may well be more severe in real life than when measured.
A real oscilloscope is the only instrument that can really tell.
The pulse edges are probably a lot 'steeper' than the SB1240 can show.
What does a pulse look like when played through the XXHE upsampling algorithm ?
-
Adding to my previous comment about the looks of an impulse: a CD is not going to contain a perfect impulse. Even if some instrument came close (which is pretty much impossible, but let's say it does) it would still be bandlimited and probably not hit the sampling interval at exactly the center:
(http://xserv.shell.la/xnor/audio/images/sampling-impulse.png)
(This is with 160 kHz low pass filter, hence the blips)
The funny thing is that if the impulse had happened a tiny bit earlier or later, the non-oversampled result would look completely different (again, since it doesn't respect the sampling theorem).
The "perfect impulse" these people are talking about is simply a marketing ploy.
-
Marketing ploy? In audio? I've never heard of such shenanigans!
Thanks, guys. You bring up good points and, as always, I appreciate the information and discussion!
Solderdude, I'll play around with ARTA more...probably sometime in the next few days. I'll have to figure out how to record an impulse response from XXHE (I'm assuming I can do this in ARTA?). Any other suggestions for things to test, especially to test XXHE's filtering/upsampling?
If I were to start measuring more gear seriously, I'd invest in an oscilloscope (and a better A/DC with wider options). Doing what I can for now based on my relatively limited hardware and knowledge, but it is fun!
It's clear that NOS DACs have a host of inherent issues that can easily be demonstrated by simple measurements and analysis, and I do appreciate the hard, technical details everyone has shared. Solderdude touched on this as I've mentioned, but is there anything that can be said about how it all translates into what we hear during music playback? Are the technical downsides really all that unpleasant, or is it hard for humans to really notice (assuming downsides are of similar magnitude to these measurements)? To what extent can software playback and upsampling methods improve on this?
There is gear, after all, that measures poorly in the grand scheme of things but receives widespread praise even from the types of folks that participate in this forum. The Vali is a good example, though I understand it's not a DAC and can't be compared in the same ways we're discussing, but it does tie into my point and my questions. A counterpoint could be made for the ODAC/O2...measures great, but I get the sense that most people here don't think it's the end-all-be-all.
Also, what listening experience do you guys have with NOS DACs? What experience with NOS DACs and various software playback and upsampling methods? I am not very experienced with DACs and, as such, really value the subjective impressions from the experienced members on the site, so this is an honest question. Hoping others can chime in...
-
Re oversampling (resampling) in software: then what's the point of a non-oversampling DAC to begin with.
Sigma-delta DACs have clocks operating at several MHz and oversample by a large ratio. In software you can do what, 4x oversampling with 44.1 kHz, probably even 8x to 352.8 kHz?
I don't know where the THD is coming from in your measurements, but if it's the DAC then you'll always get IMD at according levels plus whatever the zero-order hold reconstruction / DAC adds additionally. IMD is something to minimize in my book, since it's a lot more offensive than harmonic distortion.
-
For one, software upsampling is more interesting to me because it gives me greater control and flexibility in what methods I use. It's interesting to experiment with. In theory, would it not be possible to devise software-based methods that do a better job than hardware-based methods?
-
I have no (auditive) experience with NOS DAC's.
software upsampling (both in frequency and bit depth) will yield smaller 'steps' and the frequency of those steps will be higher as well moving the HF energy far above the audible range and the headphone amp / headphone will act as a reconstruction filter.
The spikes you see in the THD plots (which in music would translate in a raised noise floor WHEN PLAYING will be lower in amplitude.
However, it will ALL depend on who has made the algorithm.
Most upsampling filters will have post and ore-ringing and all advantages of the NOS concept will have gone.
The roll-off in the highs will be gone, the exemplary squarewave and needle pulse behaviour will be gone, pre-post ringing will be back etc.
Of course as Xnor explained these pulses do NOT exist in audio and 'attack' of instruments is NEVER a jump from LSB to MSB.
An NOS DAC without a proper reconstruction/brickwall filter 'invents' things that aren't in a music signal and have never been in the original music signal as well.
What you will have gained is flexibility in that you can try different algorithms/ideas from different people which may lead to measurable and perhaps even audible differences (I will leave that to golden eared people, I honestly can't tell which seems a blessing and saves me loads of money)
There will be no difference between hardware and software 'upsampling' when the same algorithm is used in the same way.
Of course it is easier to change/revise software than it is to change hardware.
Yes, you could flash hardware with other firmware but in software one could do tricks that may not be possible with existing hardware chips.
Haven't experimenting with recording files for Arta but don't think it would be impossible and really interesting to see the differences with other upsampling algorithms.
Afterall, Peter uses it on his own NOS DAC as well and that may yield measurable differences that may not be measurable with OS DAC's at all.
As I have questioned before the audibility of (pre) and post ringing at inaudible frequencies (well above 20kHz even for redbook) that incidentally are NOT related to the audio signal, baffles me.
I could go along with HF energy that has a relation to fundamentals/harmonics in certain music providing transducers can actually reproduce that but I have never seen research that 'links' improved or degraded sound with unrelated HF energy above the audible limit.
I still hope someone can show me pre and post ringing artefacts in say a 5kHz sinewave or even lower frequencies.
I once asked Archimago and he did do such a test and could find no ringing at these frequencies.
But as you say... very fun to experiment with and suggest to now and then test 'blind' when in doubt about findings.
To me that is just as enlightening as suitable analysis equipment.
Just remember most test are 'limited' in what they can show and in most cases only show one or two 'measurable' aspects.
You need a plethora of tests to draw conclusions.
-
There will be no difference between hardware and software 'upsampling' when the same algorithm is used in the same way.
Of course it is easier to change/revise software than it is to change hardware.
Yes, you could flash hardware with other firmware but in software one could do tricks that may not be possible with existing hardware chips.
Haven't experimenting with recording files for Arta but don't think it would be impossible and really interesting to see the differences with other upsampling algorithms.
I still hope someone can show me pre and post ringing artefacts in say a 5kHz sinewave or even lower frequencies.
I once asked Archimago and he did do such a test and could find no ringing at these frequencies.
Trimmed some of that. That was more what I was getting at. You should have more flexibility and ease of development with software-based methods, or, at least, this is what I assumed from my background with programming (B.S. in Computer Science - but I think programming is too frustrating! haha).
I'll look into recording external samples with ARTA. I'm fairly sure it's possible and might have even noticed options for it when originally playing around.
I'm willing to test nearly anything someone asks so long as I have the means and time to do it. I might not always know how to do it immediately or what it necessarily means, but I'm willing! The more information we have, the better we can objectively assess something, even if just single elements.
Would it be at all possible to devise a measurement test that takes an actual music file for recording and measurement purposes? I mean, I know I only listen to square waves and impulses (hence the NOS DAC purchase...kidding!), but I think that would tell more about a DAC's musical performance than anything else. I could see that being a very complex process, though.
-
Would it be at all possible to devise a measurement test that takes an actual music file for recording and measurement purposes? I mean, I know I only listen to square waves and impulses (hence the NOS DAC purchase...kidding!), but I think that would tell more about a DAC's musical performance than anything else. I could see that being a very complex process, though.
This would mean a nulling test with either the original digital file (but that only represents sample points in timespacing) or the original analog waveform. The original analog waveform has to pass through a brickwall at the ADC stage and thus alters slightly anyway during the encoding.
Archimago has made a 'correlation' test with a few 'representative' audio samples.
The problem with such null testing is the clocks of the recording and playback step that will be compared at a digital level (and thus only partly relevant by itself) differ as the sample points differ in time of that analog waveform (see Xnors plot about sampling frequency and input frequency not being synchronised).
This causes some errors which will always show up and ensure perfect nulling is NOT possible in practice nor theory.
Nulling 2 pure analog waveforms can reach better results IF you can actually null the 1kHz extremely exactly.
Once you know the limits of this test you can make an educated guess of how different the sound would/could be and you can actually have a LISTEN to the differences and see which volume level of the original music is actually needed before you can actually hear the differences.
Be ready for some surprises.
Everyone should decide for themselves which 'method' they prefer based on their own experience.
-
I see. Even more difficulties than I initially assumed!
-
This is not directly related to the Metrum, BUT I did just find out the HiFiMan HM-601 and 602 both use the TDA1543 NOS DAC. Found RightMark measurements for the 602:
http://sonove.angry.jp/Comparison_HM602_HM801_iPhone4.htm
The 602 does not measure as well as the Metrum Quad (not sure if this is due to the 16ohm load or not?), but the thing that stuck out most to me was how similar the IMD measurements looked. Interesting, and that might validate what my measurements show as well.
In measurement terms, it seems like the Metrum sits somewhere in between NOS DACs that measure fairly poorly, like the TDA1543, and the more expensive options that show better numbers, generally those running PCM1704 chips. Again, it's unfortunate the chips used in the Metrum DACs have their labels scratched off.
-
Another curiosity is the stronger odd-order harmonics, which are said to be more offensive.
Rightmark seems to use a sweep with two tones 1 kHz apart for the swept IMD test. I wonder how high the IMD is with 5 tones, 10 tones, 100 tones. It could be much worse with a more music-like test signal with more tones.
-
Given the interesting results I got with my Auzentech Bravura sound card's coax out with the NOS1704 DAC (see that thread, and it'll have more details on my desktop rig and measurement setup), I wanted to try it with the Metrum Quad. I had only tried the JKSPDIF prior. For these measurements:
Desktop rig -> Auzentech Bravura sound card's coax out -> Metrum Quad DAC -> SB1240 ADC -> tweaked laptop on battery
24/96
I'm also posting revised ARTA spectrum measurements, so it's easier to see details. This is also on top of better narrowing down on DAC and ADC volume settings in Windows to get the best, most reliable measurements. They'll look nearly identical but with slightly better numbers, generally.
The first two files are JTest jitter measurements played in JRiver Media Center using a 24/48 file. The Metrum has a noticeably lower floor than the NOS1704. The first picture is just standard playback through JRMC. The second is what happened when I enabled upsampling to 24/96.
Spectrum files are all 24/96 and in order are 1KHz sine, 100Hz sine, 60Hz + 7KHz at 4:1 ratio, and 25Hz + 66Hz at 1:1 ratio.
-
RMAA tests at 24/96 using the same Auzen coax out setup (keep in mind the results look better than earlier ones because I better adjusted volume settings):
Frequency response (from 40 Hz to 15 kHz), dB | +0.03, -0.84 | Good
Noise level, dB (A) | -102.6 | Excellent
Dynamic range, dB (A) | 102.9 | Excellent
THD, % | 0.037 | Good
THD + Noise, dB (A) | -66.7 | Average
IMD + Noise, % | 0.044 | Good
Stereo crosstalk, dB | -97.9 | Excellent
IMD at 10 kHz, % | 0.170 | Average
General performance | Good
-
Same tests as above but with the JKSPDIF, 24/96. Also re-did the RMAA test in DirectSound mode. I'll just post the numerical results, because the graphs look about the same as you've seen already. First JTest is straight playback, second is upsampled.
Frequency response (from 40 Hz to 15 kHz), dB | +0.00, -0.59 | Good
Noise level, dB (A) | -102.5 | Excellent
Dynamic range, dB (A) | 102.9 | Excellent
THD, % | 0.036 | Good
THD + Noise, dB (A) | -66.9 | Average
IMD + Noise, % | 0.043 | Good
Stereo crosstalk, dB | -97.8 | Excellent
IMD at 10 kHz, % | 0.166 | Average
General performance | Good
-
XXHighEnd jitter tests for the lulz. Compare to JRMC. I have not yet tried XXHE through the Auzen out on my desktop. File is 24/48, software and hardware settings at 24/96.
1. XXHE - From what I can tell, upsampling and all those fancy related options disabled. It's hard to tell what's going on with this UI.
2. XXHE Linear Interpolation upsampling
3. XXHE Arc Prediction upsampling - According to the developer, this should be perfect for NOS DACs...
4. XXHE Anti-Image upsampling
-
These are experimental and likely useless REW measurements.
As ultrabike mentioned in the SB1240 measurement thread, it's questionable whether these measurements have any sort of usefulness when it comes to DAC measurements. I can't say I know how to read some of these (DAC or not) or if any information in them is valuable, but I can say that the measurements are producing different results among my DACs. I'll post them just in case someone thinks there's good info to be found in these, but they're likely not too helpful. I apologize that these are not always scaled and matched across my other DACs, or if I'm missing a measurement.
1. FR and Phase
2. Harmonic Distortion - not aligned on the graph properly, but you can see basic % values at the bottom
3. Impulse Response
4. Filtered IR
5. Group Delay
6. Waterfall
7. Spectrogram
-
The irony is, if you associate old D/A converter performance with "digital sound" or "digitis", then non-oversampling DACs will give you exactly that.
Look at this:
(http://xserv.shell.la/xnor/audio/images/sampling.png)
The green dots are the samples. The blue sine is the input and output when correctly reconstructed.
Purple shows how non-oversampling DACs do it. Red is linear interpolation. Some audio editors display waveforms incorrectly that way.
That is a massive misunderstanding of how proper anti-aliasing and reconstruction filters work. Reconstructing the signal (in this case, analog frequency being ~ 0.43 sampling rate) is NOT a matter of connecting the dots (linear interpolation) - even with no oversampling.
Please stop making stuff up.
-
These are experimental and likely useless REW measurements.
Now you understand why I don't bother with DAC measurements. Some people freak out and see a DAC with -3db at 20kHz and scream "what a POS" when in fact that DAC sounds better than this:
(http://i.imgur.com/axUUg7dl.jpg)
Which probably measurements perfectly in FR. There way too much which can't be measured with two or three steady statement measurements. DACs are not like headphones where the errors are gross.
-
Haha, I don't mind. I have a lot of fun doing measurements like this simply for the sake of science, knowledge, information, etc. If it gets someone riled up, that's their problem. I have my tastes (these posts are not indicative of my tastes necessarily), and I understand my tastes don't always fall within the bounds of excellence from a scientific, objective perspective.
The NOS DAC concept intrigued me enough to try one, hence the Metrum purchase. I wanted to start somewhere decent. I also noticed a lack of easily accessible or robust measurements for NOS DACs and figured I could help fill some of those gaps. And now I'm just interested in measuring anything I can!
-
I don't think the REW measurements are useless. Specially if they yield consistent different results among DACS. But frequency response you already get with ARTA and RMAA. THD and IR you can also get with ARTA. However, the calibration process for ARTA and REW is different. In the end, if properly calibrated and w same drivers, both programs should yield similar results.
I think is a good idea to show results among the different SW packages, particularly if the same measurement yields different results, and/or if different drivers and modes were used perhaps due to card compatibility or whatever other reason. It's also a sanity check.
Group Delay sort of measures latency.
Waterfall and Spectrogram are proly similar to CSD and should be pretty "fast" for DACS relative to headphones. Still useful tho.
As far as the S/H operation, I also feel that the reconstruction filter should produce a decent tone. I think the difference between the impulsive dots and stair wave approximation is out of band images.
And really nice job Hanns. I can say for myself that I think I have learned a lot when doing my own measurements and is always fun to correlate what I get w my impressions and some ol stuff I learned.
-
The Metrum does have that particular...hump or ridge demonstrated in the waterfall and spectrogram plots. It starts at the high frequencies and moves down to the lower frequencies over time. The NOS1704 and SB1240 did not exhibit this. I only ran these tests once, so I'd have to see if I can repeat this. If I can, I'd be interested to know what's causing that.
-
Thought I'd post my subjective Metrum Quad impressions w/ Octave mk1 PS.
Sounds smooth and plasticky. Rounded timbre. All songs have the same timbre coloration so it's harder to tell how well they were mastered and each instrument sounds more similar next to the other. Kind of a Super Shure 500/530 plus signature. Very relaxing but fake. No sibilants or digital hash. Sound seems compressed in SS w/ instruments superimposing over others, very 2D and small L/R blob signature. Sounds a bit slow, smeared and muddled in the upper mids. Imaging and resolution are not the best. Vocals do have that liquid seduction which is enticing. Makes strings sound almost like an unmodded Abyss, incoherent. Guitar and drums sound plastic for sure. Background is not black and as clear as it could be.
Switching back, the Uber BiFrost via uber USB is more clear, more black, more resolving, more spacious and more accurate sounding.
-
...Wait, is this the part where I get upset or something because I like the Quad? ;) Thanks for the comments!
I hope to get my hands on the Bifrost or Gungir soon so I can compare, but I quickly found myself fond of the Metrum's relaxed, smooth sound. The Hex steps it up a notch and is noticeably less laid-back, but it still shares those traits I like. I'm doubting it'd make enough of a difference for your tastes. The fact that you and Marv both said vocals are good or seducing on it is particularly interesting, because vocal abilities is one of my musical strengths. That wasn't something I had really considered before you guys mentioned it, but I'm wondering if that might have something to do with why I like its sound? It might just be due to it slightly taking the edge off music, like metal, that can otherwise fatigue me quickly and easily.
Did you listen to this on Marv's setup? Just curious to know the chain behind it all if that wasn't what you used, and, in particular, I'm curious if there was an appreciable difference between software and upsampling methods if you happened to experiment with that. Genuinely curious because I like to experiment with it and like to see what others find. I'm not trying to imply anything one way or the other about your setup or your impressions.
-
Did you listen to this on Marv's setup? Just curious to know the chain behind it all if that wasn't what you used, and, in particular, I'm curious if there was an appreciable difference between software and upsampling methods if you happened to experiment with that. Genuinely curious because I like to experiment with it and like to see what others find. I'm not trying to imply anything one way or the other about your setup or your impressions.
No, I didn't use Marv's rig which is too forgiving for me. I used my own test/listening rig which I am intimately familiar with. 98% of all my impressions are done the same way except when they are of a specific vendor room at an audio show (obviously I can't control that).
Denon CDP (SPDIF)>Metrum>Super7 (modded)>HD800 (modded)
PC (JRiver ASIO)>UberFrost (uber USB)>Super7 (modded)>HD800 (modded)
I didn't use SPDIF on the UberFrost as I find it noticeably inferior to it's latest USB implementation.
-
If you ever get a chance to try a NOS DAC with/without software upsampling (or if you have already), I'd be curious to see if your impressions match what some others have noticed or not.
-
If you ever get a chance to try a NOS DAC with/without software upsampling (or if you have already), I'd be curious to see if your impressions match what some others have noticed or not.
I very rarely use upsampling, especially for reviewing gear. Hardware or otherwise. I can't imagine upsampling doing anything w/ the Metrum except making it even more of what it already is which I think would be going in the wrong direction. Could be wrong.
-
I see. Well, many claim software upsampling "fixes" aspects like you noted. I've noticed some level of difference but wanted to get opinions from people here to verify.
-
I see. Well, many claim software upsampling "fixes" aspects like you noted. I've noticed some level of difference but wanted to get opinions from people here to verify.
Reviewing source gear using software upsampling is like reviewing headphones using DSP and EQ. Pointless. Are you reviewing gear capability critically or listening for pleasure?
Having said that, IME upsampling fixes things like digititus, glare, grain and other types of effects. I've never heard software upsampling increase fine resolution, improve air and separation. Quite often the opposite which is something the Metrum doesn't need IMHO. Now if the effect is the opposite w/ a NOS dac, that would be interesting and news to me. Ultrabike has the Metrum, you could try asking him to give it a go.
As of now, my opinion about upsampling and other software conversion from original masters is if you like it, it's usually a pleasing coloration, not an accurate rendition. I've heard the same thing from some well known folks about converting DXD PCM to DSD 'adding weight' and sounding more 'natural'. Sorry, the samples I've heard from say someone like 2L are simply colored and more compressed sounding versus their original PCM masters. If air and micro detail disappear after conversion, it simply cannot be more accurate.
-
I'm not so much talking about critical reviewing or comparing gear on strict terms, but I think software upsampling (of which you can, in theory, choose nearly any method you'd like) coupled with a NOS DAC is a viable way to listen critically and for pleasure. It's taking a commonly used DAC function and moving it to a different part of the chain, though you have the option to opt out entirely if you so choose. In my mind, I'm hesitant to liken it to something like EQ because of this, but I can understand if you disagree. I personally like having such an expanse of options to dial in whatever works best for my ears.
And, as mentioned, I have read other subjective reports, of which I can't vouch for their credibility, containing similar impressions to yours that were later mitigated with the use of software upsampling. I distinctly remember someone saying the Octave sounded "mushy" at 16/44.1 and without software upsampling. The potential variability of software and upsampling methods used does raise further doubts and questions (not to mention they could have bad ears or just be idiots), but the relative consistency of the reports is interesting enough that I think it warrants further research. And it's not to hard to believe, really, considering a higher sampling rate does lessen the treble drop-off in the audible range. That alone has a noticeable impact on the final sound.
Most of you here seem to be a bit more critical and honest in your evaluations, hence the reason I was particularly interested in gathering thoughts from anyone that might have also experimented with software sampling and NOS DACs. I can hear a difference with/without software upsampling, but I still have a while to go before I'd feel comfortable making definitive statements one way or the other, so I really value others' impressions as well.
-
I distinctly remember someone saying the Octave sounded "mushy" at 16/44.1 and without software upsampling.
And if some of my test tracks are 24/96, 24/352 and DSD which sound the same as 16/44 on the Metrum to me, where does software upsampling fit into that theory?
-
And if some of my test tracks are 24/96, 24/352 and DSD which sound the same as 16/44 on the Metrum to me, where does software upsampling fit into that theory?
Well, if that is indeed the case, I'd say that is evidence against the claims others have made, and that's the sort of information I was curious about gathering and think is valuable to all listeners/potential buyers. I think software upsampling can bring some benefits to red book music on a NOS DAC, but what I've heard is close enough it could just be placebo (perhaps minus the slight treble lift). Certainly not the dramatic change some make it out to be that will make all the difference in the world to those that don't use it.
-
Put together something quick to test the Metrum:
Sony PS2 (SPDIF) > Metrum > Focusrite 2i2 (Line In) > HD600
I used mostly CDs I had here and there (Eurythmics, Aerosmith, Red Hot Chili Peppers, Jon Secada, Celine Dion, and stuff). These are limited to by 16 bit / 44.1 kHz sampling rate. Might try later using my senile Xbox 360 (old enough to have the proprietary connector w SPDIF instead of HDMI).
IMO the bass and the mids were good. The treble might be rolled off. Things indeed seem relaxing. I think however that the rig revealed how noisy and crappy my Jon Secada CD sounded. Which is good, but also depressing. The Dion and Eurythmics recording did not seem to do that bad. Will put together some better recording CDs for the next round and compare to the internal DAC of the 2i2 (that's the best I have at the moment).
I did put the Line-in through the ARTA spectrum analyzer at 96 kHz sampling rate, and that revealed a lot of > 20 kHz stuff going on probably due to the NOS nature of things.
-
I preferred the Metrum Quad with JRMC up sampling. Seemed less hazy, more clear in the bass, and slightly more detailed. I wonder if this is because the ultrasonic shit is pushed up further where the analog filter would be more effective?
Similar in some respects to DSD to PCM conversions where I prefer the most aggressive ultrasonic filters. The DSD to PCM conversions with the DSD hash intact from 24k upwards sounds similarly hazy to NOS and no software up sampling on the Quad.
-
My PS2 is too old and I think it only supports 44.1 and 48 kHz sampling rate. Even my ol' XBOX 360 may only go up to 48 kHz due to Windows evil voodoo inside. Seems only few entry level audio cards support 96 kHz out through optical (like Creative X-Fi).
I'll stick with 44.1 and 48 kHz for now, but it makes sense that proper up-sampling might make things better by pushing ultrasonic stuff out, and by possibly reducing the treble roll-off.
-
Did some more listening. There is roll off at the upper tremble, but many cans (and speakers) proly do much worse.
Here is a quick spectral snapshot of a song I was hearing (had issues w Audacity/Windoze doing a decent job generating 20 kHz band-limited white noise waveform):
(http://www.changstar.com/index.php?action=dlattach;topic=1260.0;attach=5455;image)
There seems to be some ultrasonic stuff, but things did not sound awful IMO though.
-
Right, and the roll-off becomes less severe the higher the sampling rate you use. I'm assuming you're still limited to the setup(s) you mentioned earlier. Or, if you're like me, the roll-off can be desirable at times.
That's an interesting test you did there! I might see if I can do something similar in the future.
-
Yes, I think the Creative X-FI is a sweet piece of equipment for these kinds of tests given the optical out (mic in, hp out, and nice 5.1 pre outs for $50 current street price!). It would be nice to see the same tests with white noise. I tried by creating a CD with some Audacity generated 20 kHz BW white noise files, playing them back through my Xbox 360 (my old PS2 didn't seem to recognize CD-Rs) optical out, and running ARTA and/or REW spectrum analyzer at 96 kHz sampling rate.
However, while I got somewhat decent frequency response plots, I was getting unexpected results around the 20 kHz area w the Audacity generated noise files (wouldn't be the first time Audacity under Windows gave me some weird results). I'll see if I can get better white noise files.
-
Here is the spectrum of a band-limited white noise signal (brickwalled maybe around 16 kHz) when using the following signal path:
WDTV-Plus (optical pass through) > Metrum (44.1 kHz) > 2i2 (96 kHz)
Finally got sick and tired of making CDs, dealing with my PS2 CD-R impotence, and other issues w my senile Xbox 360...
(http://www.changstar.com/index.php?action=dlattach;topic=1260.0;attach=5480;image)
I think a NOS DAC (zeroth order S/H) should generate a mirror image (brickwall and all) starting at 22.05 kHz and centered at around 44.1 kHz. This might explain the sharp gap between 16 to 28 kHz (the gap gets mirrored from 16 to 22.05 kHz into 22.05 to 28 kHz).
Moreover, the S/H operation should also modulate a sinc function into the FR, with nulls occurring every integer multiple of the 44.1 kHz sampling rate. This explains the null at around 44.1 kHz.
The images and nulls will continue to repeat at every integer multiple of 44.1 kHz, but more attenuated by the sinc function shape (a result of the S/H operation).
Here is another white noise signal spectrum using the same set up, but this time the noise does not seem that brickwalled but proly alias a little bit (notice the discontinuity around 20 kHz). This I think shows the sinc deal better.
(http://www.changstar.com/index.php?action=dlattach;topic=1260.0;attach=5484;image)
While sinc function related losses around 20 kHz and 15 kHz are proly about 3 and 1.6 dB respectively, I liked the laid back nature of the DAC. This is a worst case 44.1 kHz mode situation, at 96 kHz (thu SW upsampling or so) the losses might be closer to 0.6 and 0.3 dB (20 and 15 kHz). These theory numbers seem to correlate well w another similar NOS DAC discussed here (http://www.changstar.com/index.php/topic,1425.msg37416.html#msg37416). The effects of the ultrasonic stuff probably depends on the amp, but I also did not seem to have problems with that.
-
Very interesting. Thanks! I've had a lot of fun experimenting with these NOS DACs and leaning about them. I too haven't noticed any ill effects from the ultrasonic stuff, but I'm also not sure what to listen for, nor do I understand electronic design well enough to know whether or not a specific amp or anything else in the chain would/would not handle it well.
-
Here's what a pure 10k sine wave looks like from the Quad, vs the gamma2 (a delta-sigma WM8741 D/A chip)
(http://i.imgur.com/vCFTLJV.jpg)
(http://i.imgur.com/zo6aJAU.jpg)
Interesting to note the stair-steps. Even with a 1k sine wave they were visible on my 'scope, although I can't seem to find those pics that I took.
For reference, signal path was:
Asus Laptop -> gamma1 USB-SPDIF -> Quad -> HP Scope
Asus Laptop -> gamma1 USB-I2S -> gamma2 -> HP Scope
It was also interesting to see how the response in the transition band changed with the gamma2's different filters vs. the Quad.
-
unfiltered NOS DAC's are GREAT for reproducing squarewaves and needle pulses but suck at sinewaves.
This is why sellers only show needle impulse and squarewave responses on their website showing how good that DAC works compared to all other (flawed) concepts.
Unfortunately... there are no squarewaves and needle pulses in audio just sinewaves.
The 'attack' in music signals is NEVER as fast in the original signal (looked at many 'attacks' on bit levels) because a: it passed through an anti-aliasing filter before encoding and b: those 'attacks' are high frequencies and they are very low in level so never near 0 or -10dB and thus incredible risetimes are not needed (as the dV in dV/dt becomes smaller)
The fact that most people still find heavily distorted sinewaves 'pleasant sounding' is because our ears cannot resolve those steps (nor can many transducers) and says more about hearing abilities than technical differences IMHO.
It would be interesting to see what a headphone makes of this signal... it sure isn't going to look even close to the electrical signals.
Even more interesting would be to see the differences between the 2 signals when reproduced by the same headphone and measured with a rig that can measure in the 100kHz region.
-
Cool, thanks, OJ!
Solderdude, it would indeed be interesting to see what the signal looks like coming through headphones or speakers. As for the lack of filter ringing on NOS, or when considering filters with just a bit of post-ringing, my upsampling tests with the Hex and comparisons against other DACs make me think this might actually have audible differences. Much more subtle than the many other benefits you gain from upsampling/oversampling, but there were still certain selections where I thought transients were slightly smeared on OS DACs or when using upsampling filters with pre and post-ringing on the Hex. It is very hard to describe...
Of course, YMMV, and it's definitely a possibility I was not actually hearing what I thought I was hearing. If I was, though, it's a trade off. You can definitely hear the rolled-off treble and other "unique" characteristics of NOS DACs, but I don't mind due to easily getting fatigued while listening.
I would be very curious to see what your subjective impressions would be in these situations as well.
-
AFAIK a S/H operation from a DAC results in stair-step behavior. This in turn may result in ultrasonic images, and a sinc window (sin(x)/x) applied to the DAC output signal in the frequency domain. The windowing operation may roll-off the audible high frequencies as a function of the (up or over)sampling rate.
-
Can I ask the question that shouldn't/mustn't/oughtn't be asked in these circumstances.
When you were listening, did you know what type of filter was being used or did someone else set the filters and you listened out for differences ?
I know this is the wrong question to ask here but would like to know the test conditions as they were.
My question is not intended dismiss what differences you (and many others) heard nor to invalidate all the hard labour and many hours of intense listening in any way nor to discredit all the work put into it.
It is more to satisfy my own curiosity as I am one of the persons with a reasonable hearing but cannot hear differences between DAC's and 'properly designed' amps. One might call me cloth eared, biased or deaf but recognise the idea that some may have better hearing than I.
I DO know that there are lots of audible differences between down/up-sampling algorithms but this is because they were not properly designed tested.
This doesn'thave anything to do with the ringing in filters which only happens with steps and not with sinewaves and steps do not occur in music signals.
Bad algorithms show all kinds of aliasing products bouncing back into the audible range.
Some even are barely attenuated in some cases.
There seem to be somewhat more well designed than poorly designed algorithms by the way.
See for yourself: http://src.infinitewave.ca/ (http://src.infinitewave.ca/) (set to sweep)
The part to the left and right of the yellow line should be pitch black (maybe dark blue is accepatble) but bright lines, especially on the left of the bright curve are a big nono.
These are ONLY of downsampled samples going from 96 to 44.1 in which case 96 is not a multiple of 44.1.
I am willing to bet that 96 to 48 would have given better results for most resamplers though.
When someone for instance has used a bad downsampling algorithm and compares a downsampled file with the original one they might easily find the downsampled one sounds worse yet done with a good algorithm may find there are no audible effects (when tested blind)
-
Oh, I did all the testing myself, so I knew what I was listening to. That's why I try to make sure and put disclaimers with anything I say. :) I actually wish I had someone around that was interested enough to put me through blind tests. But, to be honest, I went into these tests expecting I should not hear a difference in these regards, but I'm still fallible and could have fooled myself regardless.
When you say ringing only happens with steps, I'm confused. I have seen graphs of sine waves with pre/post-ringing due to upsampling, but there's a strong chance I'm missing something or misunderstanding entirely. See the link below...I would definitely appreciate an explanation to better understand this!
http://www.cicsmemoryplayer.com/index.php?n=CPlay.Measurements (http://www.cicsmemoryplayer.com/index.php?n=CPlay.Measurements)
That's a cool site, for sure. I came across it earlier when looking into different filter types.
-
What the pictures show is an instantly increasing/decreasing sinewave which is an artificial waveform that doesn't exist in music (unless it is electronically generated and has never been processed along the way) so essentially is forming a step.
Of course an instant 'attack' DOES exist but an instant stop does not... so pre-ringing is a real change to the signal.
What is interesting though is that the sine is stepped as well and not starting from '0' what a real sound would do.
Possibly these plots would not show the pre-ringing so they went with the published plot (really dark thinking of me) that actually shows it is there.
Also it depends on how the filter is made.
What would be more indicative to asses the effects filters have is by downsampling music and upsampling it again to the same file type.
Then digitally null those files and have a look in the audible band for the difference file.
It stands to reason everything above 1/2fs of the lowest sampling rate is also in the difference signal but can most likely tell if it is ringing or US contents by looking at the spectrum.
I am fully aware of the possible dangers in sighted tests and often am convinced I hear 'something' (even when not expecting it) and only after I tested blind it turns out the things I heard (and could even tell with a certain amount of certainty) were not present when tested blind.
The thing is one knows something has changed even when we expect nothing of it.
The most known counterargument is that people independently report the same.
This discrepancy causes the well known debate and will never be resolved I am afraid.
So for this reason my motto is... if it is an improvement, regardless where it comes from, and is worth the effort/money I am all for it.
-
Gotcha, makes sense. Thanks for the explanation, as always! :)p6
And that's why I don't try to pass my ideas off as anything more than they are (subjective, likely fallible thoughts subject to placebo). Regardless of whether or not I'm actually hearing an improvement with attack/transients, my ears like the laid-back NOS sound for long listening sessions along with some other characteristics I find pleasing (subjectively, not necessarily better or worse, but to my tastes). For other times, upsampling on a NOS DAC does a good job when I'm in the mood for it. The attack/transient thing is by no means my highest priority, but rather something I thought I noticed during listening tests while I was listening more for changes in treble and soundstage characteristics.
-
The funny thing about 'attacks' in music is that they aren't really fast in most cases.
That is with respect to the level and amount of harmonics present.
The ultrasound content is down 50 dB or so while the measurements are always done near FS to show 'worst case scenario'.
In fact most 'transients' are often sampled taking several samples and thus way below1/2 fs so do not stress the frequency range.
So a NOS DAC can reproduce transients quite accurately (so can OS DAC's).
At the same time the roll-off (for 44/48kHz) is still present yet the transients are not affected so relatively 'more' present.
Depending on the level of the higher harmonic contents and how long higher frequency tones are present there is a continuously change in roll-off.
The 'severity' of this roll-off is thus NOT always the same and depends on the relation between the sample frequency and the to be reproduced frequencies.
If you were to place a (post) analog filter behind an OS DAC with the same roll-offyou would find using a testsignals on a NOS DAC you would NOT end up with a similar tonal balance.
The NOS DAC would sound brighter and even less rolled-off than would it appear to be the case when looking at plots based on a test-sweep.
Yet... rolled off the NOS DAC will always be which many people prefer and may be largely depending on recording quality/music I would add.
-
I think you might be repeating yourself at this point Solder.
-
Repeating perhaps.
Elaborating and explaining a bit further was the intention.
There is info in there I hadn't touched before...
-
Hans, I think you were talking about the different sort of digital filters that can create pre- and post- ringing. Some might say that that too is inaudible so long as it doesn't effect the passband, but I think differences are there.
I was able to capture those waveforms on my gamma2, but I won't post them until you try listening first. Wouldn't want to bias you of course :P