HOLO Audio MAY DAC

[Quadman]

2 minutes ago, barrows said:

Not if you feed it 44.1 data and run it as NOS though...  there will be obvious issues in its output in that case

With the measurements JA performs with the big AP analyzers, he only measures up to 100Khz, but his final words are as below. He measures as NOS with 44.1 and 48 K data.  That said I strongly stand by my love for 1024.

 

"In almost every way, the HoloAudio May (Level 3) is the best-measuring D/A processor I have encountered, rivaled only by the Weiss DAC502 and MBL N31.—John Atkinson"

https://www.stereophile.com/content/holoaudio-may-level-3-da-processor-measurement

 

[barrows]

22 minutes ago, Quadman said:

With the measurements JA performs with the big AP analyzers, he only measures up to 100Khz, but his final words are as below. He measures as NOS with 44.1 and 48 K data.  That said I strongly stand by my love for 1024.

 

"In almost every way, the HoloAudio May (Level 3) is the best-measuring D/A processor I have encountered, rivaled only by the Weiss DAC502 and MBL N31.—John Atkinson"

https://www.stereophile.com/content/holoaudio-may-level-3-da-processor-measurement

 

I have read the review.  See figure 3 (or 16) in NOS mode, not good, of course.  These are reasons oversampling is important.  

[taipan254]

10 hours ago, barrows said:

I have read the review.  See figure 3 (or 16) in NOS mode, not good, of course.  These are reasons oversampling is important.  

 

Don't most people run this DAC with HQP or PGGB?

[John Hughes]

1 hour ago, taipan254 said:

Don't most people run this DAC with HQP or PGGB?

I doubt that.  Most owners enjoy the May without all this extra fiddling (like myself). The fans that hang out and post to theses forums are a special minority :)  

 

I do plenty of extra audio fiddling myself, just not on a PC. 

[Miska]

22 hours ago, barrows said:

I have read the review.  See figure 3 (or 16) in NOS mode, not good, of course.  These are reasons oversampling is important.  

 

0 - 22.05 kHz sweep at 44.1k NOS:

 

0 - 22.05 kHz sweep, with upsampling to 705.6k:

 

0 - 22.05 kHz sweep, with upsampling to DSD256:

 

[sledwards]

@miska

I clearly see the improvement in the progression you have shown, but still am having a hard time understanding the significance in varying high-frequency response over say, 50 kHz as related to audible results. Could you plot the three charts form 0 to 50 or 100 kHz? Or is the comparison not relevant over that range?

 

Steve

[sworksone]

49 minutes ago, sledwards said:

@miska

I clearly see the improvement in the progression you have shown, but still am having a hard time understanding the significance in varying high-frequency response over say, 50 kHz as related to audible results. Could you plot the three charts form 0 to 50 or 100 kHz? Or is the comparison not relevant over that range?

 

Steve

I’d also be very interested if possible to see 1.4 MHz compared to DSD256 given the change from 44.1 to 700 kHz.  

[Miska]

1 hour ago, sledwards said:

@miska

I clearly see the improvement in the progression you have shown, but still am having a hard time understanding the significance in varying high-frequency response over say, 50 kHz as related to audible results. Could you plot the three charts form 0 to 50 or 100 kHz? Or is the comparison not relevant over that range?

 

This shows the level of reconstruction accuracy. It takes couple of hours to produce the plots, so I don't have time to redo these measurements at different scale right now.

 

For 44.1k NOS output rate, the distortion starts right from 22.05 kHz and goes all the way up to MHz range as you can see from the plot.

 

At 705.6k rate (16x digital filter), good digital filters create empty gap and the first image is around 705.6k sampling frequency from 683.55 kHz to 727.65 kHz you can see in above plot and is already much lower level. At 1.4/1.5M sampling rate it is 2x higher frequency and disappears in the ultrasonic noise floor.

 

At DSD256 rate (256x digital filter), there are no images at all. But a tiny bit of random ultrasonic hiss as left over from 300 - 600 kHz. At DSD512 and DSD1024 these are also 2x and 4x higher frequency and disappear in the ultrasonic noise floor.

 

[Superdad]

2 hours ago, sledwards said:

I clearly see the improvement in the progression you have shown, but still am having a hard time understanding the significance in varying high-frequency response over say, 50 kHz as related to audible results. Could you plot the three charts form 0 to 50 or 100 kHz? Or is the comparison not relevant over that range?

 

Yes, considering that no power amp or speakers are putting out much of ANYTHING above 50KHz (leaving aside some Spectral zero-to-light amps that need filtering interconnects just to stay stable ), I have never understood @Miska's publishing of graphs that go to 5 Megahertz.

 

I'm supportive of upsampling and DSD (NOS fold-down aliasing can be nasty), but it would be more helpful to the cause to see an expanded relevant bandwidth...

 

Here is the human-relevant portion of each of his graphs. 

 

[barrows]

10 minutes ago, Superdad said:

 

Yes, considering that no power amp or speakers are putting out much of ANYTHING above 50KHz (leaving aside some Spectral zero-to-light amps that need filtering interconnects just to stay stable ), I have never understood @Miska's publishing of graphs that go to 5 Megahertz.

 

I'm supportive of upsampling and DSD (NOS fold-down aliasing can be nasty), but it would be more helpful to the cause to see an expanded relevant bandwidth...

 

Here is the human-relevant portion of each of his graphs. 

 

Alex, many amplifiers and preamplifiers do not respond well to having these high frequency components at their inputs.  This can produce IM products at their outputs within the audible bandwidth.

[Miska]

21 minutes ago, Superdad said:

Yes, considering that no power amp or speakers are putting out much of ANYTHING above 50KHz (leaving aside some Spectral zero-to-light amps that need filtering interconnects just to stay stable ), I have never understood @Miska's publishing of graphs that go to 5 Megahertz.

 

Pretty much all power amps should be flat to 100 kHz, or they have a really horrible phase response. In addition, the power amps I've seen have typically 1st order low-pass on the input, meaning 6 dB/octave which in turn means that it has practically no effect with such inputs.

 

Few other points to consider is that putting that much of distortion energy to the amplifier will have impact also on the amplifier's feedback loop, and intermodulation between those series of fully correlated images appear back in the audio band as signal dependent distortion. In addition, if you are using class-D amplifier, those are susceptible to aliasing just like ADC when the input frequency gets closer to half of their switching frequency.

 

Note that in case you get intermodulation with any if DSD's random left-over noise, the intermodulation products are also random that result in just hiss, like background tape noise or FM radio between the stations.

 

21 minutes ago, Superdad said:

I'm supportive of upsampling and DSD (NOS fold-down aliasing can be nasty), but it would be more helpful to the cause to see an expanded relevant bandwidth...

 

If you have accurate reconstruction, there are no ultrasonic images left. With 44.1k NOS output, you can see that there's no reconstruction at all. Or you can look it from another perspective, example from another NOS DAC...

 

19 kHz sine output of NOS DAC running at 44.1k rate:

 

19 kHz sine output of the same DAC, same source data, but with 384k upsampling rate:

 

Which one looks more like 19 kHz sine?

 

For transients, depending on your sample timing vs the actual transient shape and position, at 44.1k NOS you may have pretty much the entire transient lost, or at wrong place and amplitude. Because the reconstruction through convolution with sinc function (aka over/up-sampling filter) is missing (which is required by the sampling theory).

 

[Superdad]

4 minutes ago, Miska said:

Pretty much all power amps should be flat to 100 kHz, or they have a really horrible phase response. In addition, the power amps I've seen have typically 1st order low-pass on the input, meaning 6 dB/octave which in turn means that it has practically no effect with such inputs.

 

Few other points to consider is that putting that much of distortion energy to the amplifier will have impact also on the amplifier's feedback loop, and intermodulation between those series of fully correlated images appear back in the audio band as signal dependent distortion. In addition, if you are using class-D amplifier, those are susceptible to aliasing just like ADC when the input frequency gets closer to half of their switching frequency.

 

Note that in case you get intermodulation with any if DSD's random left-over noise, the intermodulation products are also random that result in just hiss, like background tape noise or FM radio between the stations.

 

All terrific points!  And illuminating graphs.

 

Still, an expanded 250KHz or 500KHz bandwidth would be far more relatable that your 5 MegaHertz wide charts. Just sayin'.  

[Miska]

1 minute ago, Superdad said:

Still, an expanded 250KHz or 500KHz bandwidth would be far more relatable that your 5 MegaHertz wide charts. Just sayin'.  

 

I sometimes store also 500 kHz plot, if there's something interesting there. But usually just the 5 MHz bandwidth one to see extent of the images. Usually 5 MHz is enough to see the point where the output drowns in the noise floor due to the DAC's analog output filter. 500 kHz is not enough to see first image of 705.6k or 768k converter rate. 5 MHz is enough to see if there's any image left when the converter is running at 1.4/1.5 MHz rate. Sometimes I also have 100 MHz or 200 MHz plots, but usually such are not needed for audio outputs (hopefully).

 

Primary reason for storing some of the results is not for posting online, but for my own development purposes.

 

[Kalpesh]

i'm doing my best trying to adopt upsampling and preferably with the cleaner DSD 256, be a good boy, listening kosher, savvy,  Nevertheless, I'm very much in line with John Atkinson when he said he could swap NOS and OS depending on source : NOS vs DSD 256 upsampling is very much in the the same ball park as comparing ps mq3 MP vs xla in exemple : flavors, not extreme distorsion, rains of frogs or crickets invasions. Often, I find NOS more natural while upsampled version sound like brightened by a unifying varnish

if instead of picking a 19 K exemple you picked a say 12 (I'm most certainly quite deaf above) , would the sine wave be nice, NOS or OS ? does the couple speakers/amps matter ? that is if it cuts sharp above 20K are we not exposed to the problems you mention ? How about NOS for SR above 88.2 or even 192 ?

[Miska]

At 44.1/48k rate you get treble roll-off starting from about 1 kHz and reaching maximum at 22.05/24 kHz (this is mathematical function of S/H). That's is one reason why upsampled sounds brighter. You could add similar roll-off with EQ to the upsampled data if you like. Plus there is some lack of detail due to missing reconstruction.

 

At 705.6/768k rate the output begins to get decent. Still just over 12-bit worth accurate.

[acg]

2 hours ago, Miska said:

Still just over 12-bit worth accurate.

 

 Which is all you need if your gain structure is good.  Not many playback rooms offer 72dB dynamic range...

[Kalpesh]

4 hours ago, Miska said:

At 44.1/48k rate you get treble roll-off starting from about 1 kHz and reaching maximum at 22.05/24 kHz (this is mathematical function of S/H). That's is one reason why upsampled sounds brighter. You could add similar roll-off with EQ to the upsampled data if you like. Plus there is some lack of detail due to missing reconstruction.

 

At 705.6/768k rate the output begins to get decent. Still just over 12-bit worth accurate.

Most of my comparisons are made with 96 sources. This being, the impression of natural vs varnished/smoothed over persist with 192 sources . It's not the bright size that troubles me it's a sort of standardisation of sound, of loss of texture with upsampling that puzzles me. I almost wrote a lack of accident. I don't buy the euphony theory that would apply to vinyl the way I do it or to NOS ; but maybe... accidents, imperfections bring me closer to live impressions. Once again, for those who don't dare trying, NOS or upsampled differ just like flavors it's often subtle and I certainly don't hear differnces such as illustarted by the 19 K sine

[Kalpesh]

How awful ! I just found out I had the same crisis a few months ago and found out while searching Closed form.... thing is that I'm back to CF16M as preferred alternative to NOS. Difference with few months ago is that now I dig it with DSD256 7ECV2 rather than PCM LNS15.

It's not a Jussi's favorite nor recommended but seems to the be the sweet spot where by inability to prefer a better technical solution without going totally in the bushes lies

 

As of when the apodizing counter goes berserk, I think I will shamelessly use the mq3/mp filter even at N SR, rolling-off something bad can't be too bad, can it?

[Miska]

I tried once again with couple of my test tracks, listening through Sennheiser HD800 headphones. Just want to emphasize that this is my personal listening result and thus it is just my personal opinion.

 

Artist: Porcupine Tree, Album: Fear of a Blank Planet, Song: Anesthetize

Artist: Opeth, Album: Pale Communion, Song: Eternal rains will come

 

44.1k NOS:

• I can hear constant irritating high frequency sheen
• Really rough, coarse top end, like listening to square waves
• Dense passages get congested
• Sound is flat without air, like listening to cheap plastic portable radio
• Bass is lacking depth, body and kick
• Drums don't have body, hard to hear instrument materials
• Reminds me of 80's CD players

 

I would not be able to listen entire album this way. Makes my ears ring.

 

When I switch back to my usual settings, I get all the detail and finesse back, nice and clean while keeping the roughness/crispiness when necessary, not getting congested and I can recognise each cymbal and drum type, material and hit separately. The "cling" sound early in the Anesthetize sounds cleaner and has consistent character and level.

 

[Miska]

6 hours ago, acg said:

 Which is all you need if your gain structure is good.  Not many playback rooms offer 72dB dynamic range...

 

I listen mostly through headphones.

 

But if you have 72 dB dynamic range it means you could hear at least -100 dB discrete tones.

 

[semente]

13 hours ago, Kalpesh said:

i'm doing my best trying to adopt upsampling and preferably with the cleaner DSD 256, be a good boy, listening kosher, savvy,  Nevertheless, I'm very much in line with John Atkinson when he said he could swap NOS and OS depending on source : NOS vs DSD 256 upsampling is very much in the the same ball park as comparing ps mq3 MP vs xla in exemple : flavors, not extreme distorsion, rains of frogs or crickets invasions. Often, I find NOS more natural while upsampled version sound like brightened by a unifying varnish

if instead of picking a 19 K exemple you picked a say 12 (I'm most certainly quite deaf above) , would the sine wave be nice, NOS or OS ? does the couple speakers/amps matter ? that is if it cuts sharp above 20K are we not exposed to the problems you mention ? How about NOS for SR above 88.2 or even 192 ?

What speakers are you using?

Maybe that’s the real cause of your brightness.

[Quadman]

Here's a couple of interesting graphs which may help show the Holo Mays high frequency issues at NOS a bit better then Jussi's 5Mhz graphs above.  Below are two measurements from John Atkinson of Stereophile.  The first is the DCS Vivaldi Apex system which is in the current March 23 issue.  The second is the Holo May KTE.  Explanation of each measurement (they are both measuring the same thing is below each picture.

 

 

Look how clean and tidy the DCS is, the signal drops sharply at the Nyquist frequency of 22.05Khz and stays very low (around -130 dB up through 100Khz).  The holo OTOH has two null points at 44.1 and 88.2 Khz the rest of the measurement band it's -40 to -75 dB.  Much more noisy than the DCS which obviously is using very good filters to clean up this high frequency mess.  Impressive.

 

Sure NOS (with the May) may sound good on certain systems, but as clearly shown above it is not the optimal way to listen to the Holo May.  Me, I am firmly in the 1024DSD w AMSDMEC7  +512fs camp, last night mostly with Sinc S.  The Holo May is absolutely incredible at these rates.  I warm up my system at DSD256 with same filter and when I switch to 1024 I cannot believe how different they are, 1024 is another level or three above 256.

[sledwards]

7 hours ago, Quadman said:

 Me, I am firmly in the 1024DSD w AMSDMEC7  +512fs camp, last night mostly with Sinc S.  The Holo May is absolutely incredible at these rates. 

 

Listening to your combination now and it is definitely a step up from 512DSD with EC7v2. I like the sinc-S and the sinc-M sounds great too.

[Miska]

Nice thing about Holo DACs is that being bit-perfect and supporting both PCM and DSD, they give you total freedom to use them in various different ways.

 

[Kalpesh]

3 hours ago, sledwards said:

 

Listening to your combination now and it is definitely a step up from 512DSD with EC7v2. I like the sinc-S and the sinc-M sounds great too.

I like Sinc Mx too but with Mac MM1 @256 I have to resort to ASDM5EC at most, or PCM 1.5