REVIEW / MEASUREMENTS: HiFiBerry DAC2 HD HAT for Raspberry Pi

It has been awhile since I've measured any of the Raspberry Pi DAC HAT boards. In fact, it was back in 2016 that I got the HiFiBerry DAC+ Pro which I still use for basic streaming in the living room these days. Thanks to blog reader Doug Gardner, he sent me both his HiFiBerry DAC2 HD (~US$100) and DSP add-on (~US$60) boards you see above for testing. Doug was also very helpful in providing some background information and discussions on subjective impressions.

The DAC2 HD was released last year (around June 2020) and is based on the TI/Burr-Brown PCM1796 "Advanced Segment" DAC chip, an upgrade from the TI/BB PCM5122 found in the DAC+ Pro board. With the majority of my DAC measurements recently being ESS and AKM products, it's good to have a listen and look again at a TI/Burr-Brown-based product. Note that the PCM1796 is not a new component by a long shot! It has been available since 2004 I think (earliest specs sheet Dec 2003). My TEAC UD-501 and ASUS Xonar Essence One from 2013 were both based on dual PCM1795, a close sibling.

While I will be looking at the DSP add-on daughterboard (version 1.1, dated April 2020) more next time, for now, suffice it to say that this is based on the Analog Devices ADAU1452 chip, a 32-bit, 295MHz DSP.

When assembled as a unit, the DAC2 HD + DSP looks like this:

And it ends up being a triple-decker with the Raspberry Pi 3 when all stacked together...

No problem stuffing all that into a convenient HiFiBerry Pi 3-compatible case (here's a link to a Pi 4 case) in a neat package size of 2 5/8" x 1.5" x 3 3/4". Obviously be mindful of heat with all that technology packed together especially if running Pi 3 B+ and Pi 4 boards that run warmer and need a bit more power.

Remember that the DAC2 HD board itself without DSP is compatible with other Raspberry software like Volumio and Ropieee just like the DAC+ Pro predecessor.

If you have the DSP add-on, you'd want to install HiFiBerryOS as per instructions here which will enable various features and extensions related to signal processing. 

Whether you have the DSP or not, as an audio streamer, HiFiBerryOS is quite full-featured and can be instructed to use a number of music sources:

I used both Roon and direct Music playback off a USB stick. I was not able to get Squeezelite working unfortunately. It just went into a screen say "Starting Squeezelite" or something like that for a long time so I didn't bother persisting.

Since I suspect many users will be sticking with HiFiBerryOS, that's what I used primarily for the tests here.

By the way, here's the difference between having the DSP board attached and without when looking at the features in the "Sound" menu:

Without the DSP, there is a basic 10-band graphic EQ available as well as the ability to measure frequency response with a USB microphone for room compensation but in order to build and run the correction filters, you will need the DSP and its advanced parametric EQ capabilities. (The 10-band EQ has been disabled by default in 2021 HiFiBerryOS builds - see here to reconfigure.)

As you can imagine, there's quite a bit of stuff we can talk about with the DAC+DSP combination. For today's post, let's start by taking off the DSP board and test just the DAC2 HD HAT board. Next time, we'll explore the combination of DAC2 HD + DSP.

So... What kind of performance should we expect from the HiFiBerry DAC2 HD HAT-based DAC?

I. MEASUREMENTS

For these measurements, the primary set-up looks like this:

Raspberry Pi 3 B+ "Touch" with HiFiBerry DAC2 HD (generic 5V, 3A switching power supply used) - test signals on USB stick, WiFi connected for web controls --> 6' shielded RCA --> RME ADI-2 Pro FS R ADC --> USB cable --> Analysis/Measurement computer (Intel NUC 6i5SYH)

Here's how the HiFiBerry OS screen looks on my Raspberry Pi 3 B+ "Touch" setup on the test bench - the software automatically detects the touch screen and works reasonably well although there are still some rough areas:

By the way, due to the orientation of the RCA output downwards, it could be inconvenient for cables, so I used a 90° male-female adaptor to reorient the output (a small adaptor like this does not cause any objective deterioration to the sound quality). As you can see, I only have the DAC2 HD HAT board on for these tests without the DSP add-on:

While for many of the tests, I'm using the Raspberry Pi 3 B+, I have attached this HAT to both an old Pi 3 (non-B+) and also the Pi 4 "Touch" - I'll show you some results below for completeness from those also. Unless mentioned otherwise, Roon streaming of audio data will be over WiFi for convenience; the tests are happening in my basement where signal strength is very good. Remember, years ago I have tested WiFi vs. Ethernet and found no difference other than lower 60Hz hum with WiFi here - I trust there's no need to rehash all that stuff we've already learned in the previous decade! ;-)

As you'll see below in the results, I also took out my old HiFiBerry DAC+ Pro for comparison:

A. Oscilloscope:  0dBFS 1kHz Sine + -3dBFS bandlimited/unlimited square waves

Okay, right off the bat, we see that the DAC outputs a 2.04Vrms signal from the RCA jacks. The two channels overlap very well suggesting excellent stereo balance. This should give us a nice and clean center image when listening.

In the image above, I've shown overlapping bandwidth limited and unlimited square waves at -3dBFS to demonstrate the "ringing" when not bandlimited. Notice that the ringing is symmetrical indicating that HiFiBerry is using a linear digital reconstruction filter in the DAC; good to see.

B. Digital Filter Impulse Response & "Digital Filter Composite"

So if we examine the digital filter closer, we can see that it's a moderately steep filter with an impulse like so:

Looks good. A straight forward linear phase impulse response. Absolute polarity is maintained.

In the frequency domain, here's the "Digital Filter Composite" with a number of overlaid graph to show the limits of this filter (this is an extension of the technique Stereophile uses as per the discussion with Juergen Reis back in 2013):

Yup, steep filter as expected. The DAC will "overload" with high amplitude content in the form of "intersample overs" as shown in the yellow 0dBFS white noise plot compared to the red -4dBFS one. We see some 2nd and 3rd harmonics with the -6dBFS 19 & 20kHz sine tones.

Notice a bit of "needling" at the base of the 19 & 20kHz tones below about -110dBFS. This suggests that the dual -6dBFS signals are slightly "hot" and encroaching beyond the maximum comfortable output from the DAC. These anomalies will go away if we just pull the volume down by -1dB.

C. RightMark PRO Tests (16/44.1 / 24/96 / 24/192)

Next, let's move on to examine how well this DAC compares in the RightMark battery. Although the software hasn't been updated since 2018, IMO it's still a useful benchmark so long as one has experience with it taking care to avoid some pitfalls and bugs. In the last while, I've been using it in concert with my old Intel NUC 6i5SYH computer which has proven to be a very stable platform with adequate processing speed. Data can be acquired quickly and it's nice to be able to compare multiple devices on the graphs.

--- 16/44.1:

Standard CD-resolution sound quality...

Notice the comparison here with the HiFiBerry DAC+ Pro, the recently reviewed SMSL M100 Mk II, and the 2 Topping DACs - D10 and DX3 Pro.

As with most modern DACs, there's often little to find at 16-bits and 44.1kHz these days. Having said this, notice that the DAC2 HD has a tendency to very slightly accentuate the frequency response up to 20kHz (no big deal just +0.25dB). Notice that the DAC+ Pro has highest distortion of all the DACs and this is well illustrated in the IMD+N sweep graph.

So we can already say that the HiFiBerry DAC2 HD is an objective improvement over the older DAC+ Pro.

--- 24/96:

Let's get more interesting. Here's the high-resolution comparison summary at 24/96...

It's in these high-resolution measurements where we see the different DACs dissociate from each other in terms of performance. Generally the HiFiBerry DAC2 HD performed similar to most of the other DACs but doesn't really distinguish itself. We see that it still has a little high-frequency "hump" around 10-30kHz on the frequency response before rolling off. Noise level is very low. A nice, low, and consistent crosstalk level across the frequencies. Very reasonable IMD+N sweep as well - not the best but certainly respectable.

As you can see, I've included the results from the RME ADI-2 Pro FS Black Edition XLR output to demonstrate the difference with a much higher priced DAC capable of balanced output.

To the far right of the summary table above, notice I've included the HiFiBerry DAC2 HD using the HiFiBerry OS software streaming the test signal off Roon. Notice that while the other results compare similarly to the same DAC playing off a USB stick, the measured noise level and dynamic range was reduced by up to 8 dB streaming from Roon! What I found was this interchannel difference in noise level:

Notice that the left channel was a bit noisier than the right for some reason! I have never found such an anomaly in previous testing streaming the test signal over Roon (1.7). I've checked to make sure I turned off all DSP in Roon. I wonder if there might be a bug with the device drivers; maybe Roon Bridge itself. Given that the USB stick playback was normal, this must be some kind of software issue...

--- 24/192:

The DAC2 HD is capable of a maximum 192kHz samplerate. Even though many DACs these days can do stuff like 384kHz or even 768kHz, realize that those are mainly used for upsampling like if you want to play around with HQPlayer. You might be able to download some native DXD (352.8kHz) material, but this is uncommon.

For completeness, here's a summary table:

Since the DAC2 HD doesn't do DSD, it makes my life easier as an objective reviewer and I don't need to measure DSD output. :-)

D. THD(+N) FFTs

Let's have a closer look at some high-resolution THD(+N) FFTs for this DAC. Here's 0dBFS and -6dBFS 24/96 1kHz sine waves:

Clean looking tracings with generally a predictable "cascade" of harmonics with low-order components (2nd and 3rd) highest. As suspected from the RightMark results, this DAC isn't going to win awards for the lowest THD+N however.

E. Jitter

Moving along, let's talk about the Audio Boogeyman known as "the dreaded jitter". Here you go with 1M-point FFTs, sampled at 192kHz, focused on the 11kHz or 12kHz primary -6dB signals:

Looks great. Nothing scary at all in regards to jitter. It looks like the Boogeyman has been vanquished without even any special power supply, and even with the DAC HAT board strapped behind the Raspberry Pi with a touchscreen running to boot!

F. Final miscellaneous tests: "DAC" vs. "Softvol" volume control? Does Pi model matter?

To end, there were a couple questions worth trying to answer.

First, there was a question that Doug raised of whether using the DAC's "device" (possibly hardware) volume control resulted in better sound quality than simply software volume reduction. I've seen others ask this question as well over the years with folks who seem to think that software volume control of 24-bit or 32-bit devices sounding "inferior".

To test this, I ssh'ed into HiFiBerryOS so I have access to Alsamixer and the "DAC" volume control which is the main device volume setting adjusted in dBs. While in there, I can also adjust the "Softvol" setting which is as you imagine, the software level control adjusted in percentages. Here's a screenshot of the 2 forms of volume control, both set to -20dB:

I used the RME ADI-2 Pro FS's digital output meter to get the levels approximately the same (<0.5dB difference) between the DAC's dB attenuation and the percent level on the software volume control. For the record, -10dB was ~89%, -15dB around 84%, and -20dB around 78% on the "Softvol" control.

So at each level, let's run RightMark at 24/96 to see what happened to the signal resolution: 

Notice some significant changes in the Stereo Crosstalk measurement and the massive change in the IMD+N Sweep shortly above 20kHz. These are probably just a reflection of the levels of the test signals dropping too low for the software; likely down into the noise level and could no longer be calculated accurately.

The point however is that it really did not matter whether we used the "DAC" or "Softvol" volume controls, there was no change in the tonality (ie. frequency response), the noise level did not change substantially, and overall, no evidence for a change in distortion. As discussed before, I feel that the value of hi-res DACs is in maintaining quality given the extra headroom when volume attenuating, normalizing, or running other DSP processing like room correction.

Remember that this DAC does not implement any special hardware output level circuitry like with the RME ADI-2 range of devices (ADI-2 Pro or ADI-2 DAC) where dynamic range is maintained over wide reference levels (see my ADI-2 Pro FS R BE preview with the SINAD vs. Volume Control graph from Matthias Carstens for what this looks like).

Another question I have been asked is whether the sound changes depending on which Raspberry Pi one uses. Well, let's have a look with the Pi 3 B, Pi 3 B+, and Pi 4 - notice the Pi 3 B did not have a touchscreen attached:

The answer is simply NO - it doesn't matter which Raspberry Pi I used. Even with a generic switching power supply, notice how clean the noise level is; there's barely even a 60Hz mains hum to be detected. Seriously folks, with bit-perfect performance, low noise level, and jitter performance shown above that's essentially "perfect", "Bits Are Bits" when it comes to transmitting the data to the DAC HAT regardless of the "source" Raspberry Pi's performance.

II. SUBJECTIVE

As per usual procedure, I had a listen to the HiFiBerry DAC2 HD for a couple of evenings before putting it through the paces on the test bench so as to develop some subjective opinions before seeing objective results.

To be honest, over the years, it has become more and more difficult to speak about the subjective qualities of DACs. Compared to something like speakers where you have multiple dB's of variation through the audible spectrum or potentially >1% THD at normal output levels typically at specific frequencies, the impression one might get from a DAC is IMO at best subtle these days unless it's designed to not be "accurate". When I read subjective reviews or write these sections, I've seen them as more of an opportunity to share some music I've listened to recently and stuff to listen for rather than being convinced that subjective differences would hold if we actually were to run a controlled listening test.

Having said this, subjectively, I've enjoyed the TI/Burr-Brown "sound" over the years. While I have heard some folks complain about a "glare" to ESS DACs (which I disagree with), I've always psychologically associated the TI/BB sound as a little more "meaty", a little more "fleshed out". On the first evening of listening to my Raspberry Pi 4 "Touch" with the HiFiBerry DAC2 HD attached, I enjoyed visiting an old favourite album - Eric Clapton's Unplugged (1992). This album won the Grammy for "Album of the Year" and IMO is well-deserved for the artistry of the live performance and its expression of emotion. The DAC2 HD had no problem translating the performance and emotions as it fed the sound system. Little details like Clapton's strumming, foot taps, the 3D depth of the background singers and audience participation were rendered beautifully. I've always loved the energy and humor in the track "San Francisco Bay Blues".

In keeping with a "live music" theme recently, I had a listen to the new remaster of INXS' Live Baby Live (1991, 2019 remaster). I remember listening to the original 1991 CD years ago and thought it sounded a little "dark". While not up to the sound quality of Clapton above, I think the remaster added a bit more life into the sound and I enjoyed the experience more than I remember. This album needed to be listened loud. :-)

The other night, I watched Christopher Nolan's Inception (2010, 4K UHD-Blu-ray of course!) with the family again. It has been years and watching it again reminded me of both plot holes and at the same time the neat concepts and insights into the story that I had missed the first time around. It inspired me to have another listen to the Inception soundtrack. Listening to this reminds me of familiar themes that Hans Zimmer would also infuse into works like The Dark Knight Rises (2012) and particularly Interstellar (2014) without being as bombastic; a bit more subtle, more nuanced here. My favourite tracks on the album are "Dream Is Collapsing" and "Dream Within A Dream" for the sense of urgency these pieces evoked. The album sounded great with the HiFiBerry. There's nothing here I can fault and I loved the majesty and fullness especially in the latter half of "Dream Is Collapsing". [While we probably know him best with movie soundtracks, remember that Hans Zimmer has been around quite a while... Since the late '70s with The Buggles and "Video Killed The Radio Star" days.]

For something totally different, LL Cool J's All World: Greatest Hits (1996) was a fun visit back to old-skool rap. Nice bass and classic hip-hop looped samples featured in favourites like "Mama Said Knock You Out" or "I Can't Live Without My Radio"... Remember the days when "my JVC vibrates the concrete"? :-) Needless to say, this sounds really good through the HiFiBerry with excellent perceived resolution of complex sonic content and tight dynamics.

III. SUMMARY

While a DAC is not just the converter chip, that chip is still the heart of the technology and the resolution it provides remains the core building block for which the ultimate sound quality arises. With the HiFiBerry DAC2 HD HAT board for Raspberry Pi SBCs, we have a device that has maintained excellent low noise level (high SNR), fantastic low-jitter performance, and overall expected low distortion. If you have one of the older HiFiBerry DAC+ Pro boards, this is indeed an upgrade in performance objectively.

Notice however that the distortion level is not the lowest as you might find in some of the latest generation ESS SABRE or AKM AK44XX DACs. Again, remember this is the performance level of a good but 17(!) year-old chip. Seventeen years is an eternity when it comes to high tech products. For example, back in 2004, Intel was still just releasing Pentium 4 and Celeron processors before the introduction of Core CPUs and even before i7/i5/i3 variants familiar to us today! However, I trust that for the reasonable audiophile, we recognize that this isn't really a problem because the performance is still excellent with better than -90dB/<0.003% THD+N. Not many amplifiers, no analogue source devices, and certainly no transducers are capable of that level of distortion-free accuracy. Do not forget the idea of "Good Enough" in audio as matured technologies and the point of diminishing returns can be achieved I believe quite easily these days when it comes to subjective satisfaction even if the DAC isn't exactly achieving award-winning THD+N/SINAD numbers.

I'm not sure what was happening with the Roon streaming and the unusual noise floor anomaly between the right and left channels as shown in the 24/96 RightMark results above. I was left scratching my head for awhile and eventually just went back to testing with a USB stick. I also wanted to check if this was an issue if I streamed with Logitech Media Server but I was unable to get Squeezelite started on the main Pi 3B+ testing platform. Since I have not seen this issue before in other devices, I assume it's something for the HiFiBerry folks to review (for the record, I was using the latest HiFiBerryOS version 20210203 with Roon 1.7 just before the 1.8 update). At this time, HiFiBerryOS still needs a bit of polishing up... For example, when playing from local storage, the touchscreen would not allow selection of individual tracks on an album which was not a problem with the web interface.

Alright then. Next time, let's see what the DSP Add-On can do.

Thanks again to Doug Gardner for his generosity these last few weeks in parting with his DAC+DSP as I poked and prodded the products. :-)

HiFiBerry DSP Add-On reviews/measurements now up.

-------------------

Hmmm, speaking of "diminishing returns", I watched Steve Guttenberg's recent YouTube video on "THE LAW of DIMINISHING RETURNS is BS".

How can that video even convince anyone that he knows what he's talking about? What are the arguments here? What facts are being put on the table for the viewer to consider? Basically it sounds like we're being told to agree with him because he says it's BS while throwing out rather haphazard examples that may or may not have anything to do with price, value, or actual performance - you know, some of the determinants of what constitutes "diminishing returns" for economic input.

Knowing that he's a salesman, is that good enough that we have faith in these kinds of claims!? Is it generally in the job description of salespeople to believe that there exists a plateau in performance-to-$$$ regardless of a manufacturer's claims? I don't think so... There's some BS going on here, that I agree.

Stay safe and enjoy the music, dear audiophiles...