1

Unscientific DAC Measurements

David C. Snyder

October 05, 2020 00:33

None

Follow

I've been comparing the new HiFiBerry DAC2 HD to the Katana Tone Board this week…only because I recently purchased both and they happened to arrive this week. It's been an interesting journey as I've been running both continuously to get them burned in (for those who believe in such things).

I felt that there was a fairly obvious difference in tone between the two…not glaring, but easily detectable, even with my 50+ year-old ears. So, I decided to connect them to my Focusrite Scarlett 2i2 audio interface and do some measurements. Now, this is not the most precise measurement device…its natural response slopes slightly down on both ends of the spectrum, but measurements are consistent from run to run, so it's adequate for comparing DACs.

I wanted to understand trends, so I measured each of the following DACs:

• HiFiBerry DAC2 HD
• Khadas Tone Board
• Orchard Audio PecanPi
• iFi Audio micro iDSD
• iFi Audio micro iDAC2
• iFi Audio nano iDSD
• Allo Katana

Here are my plots:

Note that most of the DACs follow that downward bend in response above 3 kHz, with the HiFiBerry DAC2 HD being the notable exception; highs seem to be tipped up roughly 0.2 dB relative to the median of the other DACs. I would not have expected that this would be enough to for me to hear, but it's not difficult to pick it out of the bunch.

I thought I'd also have a look at step response to get a sense of what's happening in the time domain. I found another surprise with the DAC2 HD there also. Relative to all six of the other DACs, it's phase is inverted:

Here's what I'm seeing with the other six DACs:

Fortunately, it's easy enough to use Roon's Procedural EQ (under DSP settings) to invert the phase for a zone. Not sure that I can actually hear absolute phase, but for measurements, it's nice for everything to be run the same way. Note, I used the same cable for all seven DACs, so it's not a wiring issue.

I also created 2-band PEQ filter in Roon to get the frequency response of the DAC2 HD to closely match my reference DAC. Here are the settings:

Band 1:

• Type: High Shelf
• Frequency: 17000 Hz
• Gain: -0.17 dB
• Q: 0.2

Band 2:

• Type: Peak/Dip
• Frequency: 17500 Hz
• Gain: -0.13 dB
• Q: 0.6

Here's the resulting plot comparing my reference DAC to the HiFiBerry DAC2 HD:

I've saved this as a DSP preset in Roon so that I can enable it whenever I need to. Please share similar plots if you have done them. Let me know if you have questions about how I created these. Otherwise, Cheers and happy listening. 

• Facebook
• Twitter
• LinkedIn

19 comments

• 0
  
  David C. Snyder October 05, 2020 00:52

  Oh, here's a REW file containing all 30 of the DAC plots that I made today. Interesting stuff.
  https://dsnyder.ws-e.com/ts/dac_comparisons.mdat

• 0
  
  David C. Snyder October 05, 2020 22:36

  Thanks, clivem. You're already seeing more options in alsamixer than I (including Phase). Would be nice to be able to cycle through filter selection like I can with my DAC+ Pro. There, I seem to have:

  • DSP Program [Ringing-less low latency FIR]
  • DSP Program [Fixed process flow]
  • DSP Program [High attenuation with de-emphasis]
  • DSP Program [Low latency IIR with de-emphasis]
  • DSP Program [FIR interpolation with de-emphasis]

  Nothing like this currently exposed for the DAC2 HD in alsamixer as far as I can tell.

  There's usually not a huge difference in the sound of different filters, but they can be fun to experiment with. Here are the responses that I measured from my Allo Katana while cycling through the seven different filter options available with the ESS Ess9038q2m:

  

• 0
  
  David C. Snyder October 05, 2020 22:58

  Got it. Thanks for sharing. It's easy enough to have Roon invert the phase and apply PEQ to adjust the tone.

  What do you think of the sound of the DAC2 HD so far? I'm quite happy with it, but there is definitely a bit more sizzle on top than my other DACs, which can affect choice of headphones and other downstream components.

   

• 0
  
  David C. Snyder October 09, 2020 22:44

  I just got another DAC in today (S.M.S.L. Sanskrit 10th v2). By all accounts, it measures quite well, so I thought I'd update my measurement protocol and run another comparison between HiFiBerry DAC2 HD and the new SMSL. Here's what I'm seeing (best plots for each, plus my Khadas Tone Board and iFi micro iDAC2, as a tie-breakers):

  

  As you can see, the SMSL and the other three DACs have nearly identical plots…same horseshoe shape, which is likely showing the native response of my test gear. However, the HiFiBerry DAC2 is a significant outlier, with rising response above 1 kHz, peaking almost a quarter of a dB above the others at ~16.8 kHz. That may not sound like much, but it's easy to hear after adjusting for loudness.

  As I mentioned before, phase (and step response) is inverted relative to the other DACs as well, but that's less of a concern. I'm curious to know if this tipped-up high-frequency response was an intentional part of the design or if my board is a bit off and should be exchanged. Please advise.

• 0
  
  David C. Snyder October 09, 2020 23:00

  HiFiBerry team - ^^

  Thanks.  :)

   

• 1
  
  HiFiBerry team October 11, 2020 12:15

  Zooming in into frequency response plots like this will show you details that aren't really useful. 0.25db isn't really something you will hear (I know you say you do ;-). You might hear the effect of a different phase response - and this will also happen if you add filters, e.g. in Roon. Our filter has been optimized for an optimal phase response. However, let me check this here in the next days to make sure that there is no incorrect setting somewhere in your setup.
  
  The phase can be inverted in software, I'll see that I can give you a command to do this. Please note that it might take a few days as we're quite busy.

• 1
  
  HiFiBerry team October 13, 2020 08:30

  I just checked the design documents. The +0.25 is exactly how it's supposed to be. This way we allow the phase response to be most linear.
  
  

• 1
  
  HiFiBerry team October 13, 2020 14:48

  Ok, now the phase. You can invert the output using:
  
  amixer sset 'DAC Invert Output' on

  You can also switch between the sharp and the slow rolloff-filter using amixer:
  
  amixer sset 'DAC Rolloff Filter' on
  amixer sset 'DAC Rolloff Filter' off
• 0
  
  David C. Snyder October 13, 2020 14:56

  I can respect your decision to prioritize phase response. Thanks for looking into it and your explanation.

  I'll experiment with the amixer settings and share my results later today. Cheers.

• 0
  
  David C. Snyder October 13, 2020 16:07

  Okay, I've run the invert output command. Here's what I see:

  root@dac2hd:~# amixer sset 'DAC Invert Output' on
  Simple mixer control 'DAC Invert Output',0
  Capabilities: pswitch pswitch-joined
  Playback channels: Mono
  Mono: Playback [on]

  And, the step response is still inverted.

  

  No change if I re-measure after running this command:

  root@dac2hd:~# amixer sset 'DAC Invert Output' off
  Simple mixer control 'DAC Invert Output',0
  Capabilities: pswitch pswitch-joined
  Playback channels: Mono
  Mono: Playback [off]

  Although that control does not seem affect the results, the filter change does, especially for lower sampling rates. For example:

  

  The effect is smaller than I was expecting, but it's definitely measureable. :)

  Confirming my current settings (after attempting to invert phase and enable the filter) with this command:

  root@dac2hd:~# amixer contents
  numid=2,iface=MIXER,name='DAC Invert Output Switch'
  ; type=BOOLEAN,access=rw------,values=1
  : values=on
  numid=1,iface=MIXER,name='DAC Playback Volume'
  ; type=INTEGER,access=rw---R--,values=2,min=0,max=240,step=0
  : values=240,240
  | dBscale-min=-120.00dB,step=0.50dB,mute=1
  numid=3,iface=MIXER,name='DAC Rolloff Filter Switch'
  ; type=BOOLEAN,access=rw------,values=1
  : values=on

  Note: Using Roon's procedural DSP to invert the phase works as expected (correct step response below), so I'm not sure why the command did not have an effect (tried both "on" and "off" for 'DAC Invert Output').

  

   

• 0
  
  David C. Snyder October 13, 2020 16:13

  You may download my REW measurements from here:

  dsnyder.ws-e.com/ts/dac2hd-filters.zip

• 1
  
  Douglas Gardner November 06, 2020 01:25

  David C. Snyder Thanks for doing those tests. I have DAC 2 HD too and I've also noticed that it sounds just so slightly different than my other two DACs. The others are Chord Qutest and the built-in DAC in my McIntosh C48. Maybe it's that slight .25db bump. I can't quite figure it out. I agree that I shouldn't hear that little bit of a difference but something is different. I prefer the DAC 2 HD on a lot of tracks. 

  I've also added the DSP board. That gives me some additional options in the HiFiBerry OS. There are selections to invert polarity in the UI so I can take care of that part there. Not sure I can hear any difference though. Also various EQ settings to change the character of the DAC. This will come in handy in the second system where I intend to use the DAC 2 in lower volume listening scenarios.

  Thanks again for the work you've done. 

• 0
  
  HiFiBerry team November 06, 2020 06:10

  We think that the difference in phase response is what you are hearing, not the 0.25dB difference (10dB mean about twice the perceived volume). If you add a filter with the DSP, it will also affect frequency and phase response. 

  While in general it it possible to use FIR filters on the DSP that will only affect frequency or phase response, this is very complex and there is no easy one-click UI for this. Most people use Mathlab or similar to design this kind of filter. It really requires a lot of know-how in DSP filter design to do this.
  
  However, even without this, just trying small IIR filter changes will let you experiment with this. Especially with small changes in the range of <1db, changes will be very subtle and not everybody might hear these changes at all. 

• 0
  
  HiFiBerry team November 06, 2020 11:45

  @David C. Snyder: 
  
  I just wanted to check why the phase invert control doesn't work as it should. The test is easy: 
  
  1. Create a file with an impulse:

  sox -b 16 -Dr 44100 -n out.wav synth 1s square pad 88199s repeat 30
  
  2. Play back impulse:
  
  play out.wav

  3. Connect the DAC output to my measurement PC and record it using Audacity

  4. Inverse phase using amixer:
  
  # amixer sset "DAC Invert Output,0" on
  Simple mixer control 'DAC Invert Output',0
  Capabilities: pswitch pswitch-joined
  Playback channels: Mono
  Mono: Playback [on]

  5. Play/record the response to this again

  6. Time align the measurements
  
  This is how this looks like:
  
  Row 1: Impulse
  Row 2: Impulse response with DAC Invert Output set to off
  Row 3: Impulse response with DAC Invert Output set to on
  
  
  (ignore the fact that these responses don't look perfect, the had been recorded with a low level with 16 bit and scaled up)
  
  Not only is the original impulse response not phase-inverted, but the "Invert output" setting is also working as expected. 
  My guess is that you already had this set to "on" before doing the first measurement.

  Edited by HiFiBerry team November 06, 2020 11:56
• 0
  
  Douglas Gardner November 06, 2020 15:22

  I agree that phase inversion seems to be biggest factor for the DAC 2 with DSP card sounding different. I also noticed that channel Left and Right are swapped. After correcting absolute phase, swapping left and right in Beocreate within HiFiBerry-OS, and doing the "dsptoolkit write-reg 0xf146 1" command line thingy I'm getting more expected results. Will need to do some level matching listening but I think it will be much harder to hear differences now. DAC 2 + DSP board is sounding really good. 

• 0
  
  HiFiBerry team November 06, 2020 15:24

  @Douglas Gardner: See my picture above: There is no phase inversion

  Edited by HiFiBerry team November 06, 2020 16:49
• 0
  
  Douglas Gardner November 06, 2020 16:08

  Got it; thx. Apologies for perpetuating that notion. Thx for clearing it up.

• 0
  
  David C. Snyder November 06, 2020 17:56

  If there's no phase inversion, what would cause the step response to be inverted?

• 0
  
  HiFiBerry team November 06, 2020 18:00

  I simply can't say. As the "invert phase" did not have any effect in your setup, I guess, this has to do with something completely different. You might try the steps I did above, just record the response and have a look at the recording with both phase inverted and not inverted. You should get the same results as above.