MEASUREMENTS: Raspberry Pi 3 & HiFiBerry DAC+ Pro (Audio Streaming)

Last week in the preview, I introduced you to the Raspberry Pi 3 Model B with the HiFiBerry DAC+ Pro HAT board. You see above a picture of the boards assembled but in the "naked" state before installation in the case. As I mentioned it doesn't take much time to put this all together on the hardware side, and just a bit of tinkering with Volumio 2 was all that was needed afterwards to get the DLNA/UPnP streaming working over the wired ethernet (a little fiddling with the Volumio hotspot if you want to use WiFi - not recommended for hi-res audio). Since the software is still in development, I noticed that messing with network settings could lead to connection issues. But once it works, it has been stable.

As promised then, let us now put this device to the test and see how it performs objectively. What level of "high fidelity" does this streamer/DAC combination perform at?

Before getting to the results, let's start with identifying the test set-up. It's really quite simple, the Pi 3/DAC+ Pro combination is running off the latest Volumio 2 version for Raspberry Pi - 0.979 release date 20-08-2016, essentially running off default settings. I'll be streaming off the wired ethernet, generic Cat-5e cabling used to connect to an inexpensive D-Link DGS-1008G gigabit switch (remember the Pi 3 is 100Mbps only) which is then connected by Cat-6 to my main router in the basement, a NETGEAR Nighthawk AC1900 R7000.

On the server side, I have my usual Windows 2012 R2 server running JRiver 22 on my "old faithful" AMD A10-5800K-based computer, 16GB DDR3 RAM, put together back in 2012 when that CPU first came out.

Literally what I'm measuring! Pi 3/DAC+ Pro, generic Fujifilm-branded switching power supply (5V, 1A) from an old camera charger, generic USB-A to micro-USB cable. Regular (Nexxtech "Radio Shack" to be exact) 3' RCA cable.

Measurement chain:

Pi 3/DAC+ Pro --> 3' RCA cable --> RCA-to-TRS adaptor --> Focusrite Forte ADC line level in --> shielded 6' USB cable --> Win10 laptop

Remember, this is being measured with my new Focusrite Forte device so don't compare these results directly with the old Creative E-MU 0404 measurements over the years - results will be close but not exactly an "apples to apples" comparison. All comparisons I make on this page will of course be consistent in terms of measurement technique.

JRiver 22 streaming to Pi 3/HiFiBerry DAC+ Pro/Volumio.

Part I: Digital Oscilloscope, Digital Filter, Impulse Response

With the output volume set at 100%, we start with a look at the 1kHz 0dBFS square wave on the oscilloscope:

As you can see, the peak voltage is around 2.95V or ~2.1Vrms. Left and right channels directly overlay on each other, demonstrating precise channel balance. Given the symmetrical shape of the square wave transition, we're likely looking at a standard linear phase digital filter.

Here's the impulse response:

Indeed. Linear phase impulse response with symmetrical pre- and post-ringing. Given the amount of ringing, we can expect this to be a reasonably sharp filter. Notice that polarity is maintained (input signal was a positive impulse).

A look at the "digital filter composite" then:

Yes, we are in fact looking at a "sharp" anti-aliasing digital filter. It's not the cleanest FFT I've seen but I'm not one to claim too much about the audibility of digital filters based on what I have heard and measured. Note that there is evidence of intersample overload with the high amplitude white noise signal using the HiFiBerry. Though not ideal performance, remember that I am using white noise with peaks up to 0dBFS which is more demanding than the -4dBFS noise signal used in the Stereophile article where Juergen Reis introduced this form of testing. Realize that this is not necessarily a problem with good mastering allowing overhead for full dynamic range, but I suppose for poorly mastered audio with amplitudes near clipping for large portions of the music, this digital filter could exacerbate distortion.

Part II: RightMark Tests

Okay, let's have a look now at the noise/dynamic range and distortion levels at various samplerates and bit-depths.

16/44.1:
Good old CD resolution. Any decent DAC must be able to do this well!

As you can see, I'm comparing the HiFiBerry with a number of other devices like the TEAC UD-501 desktop DAC, the Light Harmonic Geek Out V2 USB DAC, the PonoPlayer mobile device, and the Focusrite Forte's DAC itself as an example of "pro" gear.

What can I say? They're all performing very well numerically. Essentially ideal low noise level and excellent dynamic range across the board. This is to be expected as they're all capable of hi-res performance. This is of course not to say they all sound the same :-):

As you can see there are some idiosyncrasies among the devices. The PonoPlayer has high frequency roll-off. The Focusrite Forte's stereo crosstalk isn't as nice as the others, and the IMD+N distortion is higher for the HiFiBerry DAC+ Pro. In regards to that, realize that the IMD+N distortion is still measured as averaging just 0.032% so this is more interesting from an academic perspective rather than a worrisome finding in actual listening.

24/96:
Time for high-resolution. Here's the summary:

As above, here are some graphs taken from the data:

Interesting. In terms of noise characteristics and dynamic range, the HiFiBerry DAC+ Pro is really quite excellent. It's actually similar to the more expensive TEAC UD-501 and Focusrite Forte than the PonoPlayer and Geek Out V2. Certainly very close to the noise floor limit of the Focusrite Forte around -110dB averaged across the audio band. What is also interesting again are the higher distortion results for the HiFiBerry and most evident in that IMD+N sweep graph (you see this in the THD distortion number also). Again, we're not talking about a massive load of distortion by any means, but it is significantly higher than I've seen with most DACs these days.

24/192:
This is the highest resolution setting for the HiFiBerry DAC+ Pro.

Just a couple of the graphs (RightMark doesn't handle the 192kHz graphs well):

Again, there's a bit more measurable distortion - in both THD and IMD in the HiFiBerry DAC+ Pro than compared to the likes of the TEAC UD-501 or the Geek Out V2. Otherwise we're still looking at a device with excellent dynamic range with concomitantly low noise level.

Part III: Jitter

As I mentioned last week, the I2S interface is still vulnerable to jitter depending on the quality of the master clock being supplied to the DAC. The HiFiBerry DAC+ Pro incorporates its own high quality oscillators for the 44.1kHz and 48kHz families of samplerates. In principle, this should give good results on tests of jitter...

That looks excellent. The 16-bit test shows the expected jitter-modulation tone in the LSB. The 24-bit J-Test (24/48 signal) looks good, completely free from spurious tones from 5-18kHz with the primary signal at 12kHz. Okay, how about we add further "stress" to the system by accelerating the jitter test tone to 24/96 - I started looking at this with the Google Chromecast Audio a number of months back.

This looks excellent. There's a noise peak down below 14kHz. The fact that the primary signal at 24kHz is arising without much "tenting" and the lack of symmetrical sidebands is testament to the excellent timing precision. (Just look at the Chromecast Audio using TosLink output to see what jitter will look like with this rather extreme version of the J-Test.)

Part IV: Battery Power!

Well, 5V lithium ion batteries are easily accessible these days to charge USB devices. As I showed a few months back with the ODROID-C2, a little battery like the Duracell above has enough juice to power a computer like the Pi 3 for probably a couple of hours (depending on how much processing being done of course). And since these days my Focusrite Forte can be run directly off the computer's USB port, the measurements can be done completely without plugging into a wall socket.

So, does running off battery power result in any improvement to the analogue output?

Let's look at the 24/96 and 24/192 results for hints:

As you can see, the "Batt" tests are with battery power versus a 5V switching power supply. Results are almost exactly a match... You can maybe make a case for insignificant reduction in the distortion results when run off the battery. People talk about switching power supplies causing noise, so let's scour the 24/192 noise floor:

Note: That noise at 37kHz originates from the ADC as previously documented and is not an issue with the Pi/DAC+ Pro.

Nope. Nothing to see... Notice the 60Hz hum (peaking way down at -130dB) is exactly the same whether using the switching power supply or lithium ion battery! [Remember this... I know where this comes from and you'll see this discussed in another blog post later :-)]. Plus there's no evidence of extra noise in either case all the way to at least 50kHz before we see the ADC noise floor rise!

And as for the distortion graph (the IMD+N Sweep):

No significant difference between battery or switching power supply either...

Is this surprising? I don't think it should be. This is the same result I got with the ODROID-C2 previously but given that the ODROID was just streaming to an outboard powered DAC (the TEAC UD-501), perhaps that wasn't unexpected. This time, we see no difference even with an internal DAC running off the same switching power supply.

Part V: Conclusions

As I noted in the preview last time, the combination of a Raspberry Pi 3 with the HiFiBerry DAC+ Pro is an easy-to-assemble box which as I have configured with Volumio acts as a low-power and reliable DLNA/UPnP audio streamer capable of 24/192 output to one's hi-fi system. Of course, one could use it like the ODROID-C2 and send the audio out using the USB interface instead without purchasing the DAC daughterboard.

I've had this unit running for about 3 weeks now and it does sound great as I noted last week. It has been reliable running 24/7 in the soundroom and is reported by my Kill-A-Watt P3 meter as sipping at most 4W when booting up, and typically uses only ~50mA.

Let's then run through a list of some "take home" points from this series of measurement results:

1. The HiFiBerry DAC+ Pro, despite being a daughterboard connected over a Raspberry Pi 3 computer puts out very low noise, high dynamic range analogue unbalanced RCA output. It's not the absolute best I have measured but certainly very close to much more expensive DACs. Clearly, the DAC+ Pro is capable of benefiting from true, high quality, high-resolution 24-bit audio files.

2. Excellent channel balance noted on the oscilloscope reading. Though not shown, this precise L-R amplitude balance was maintained throughout the volume range. Maintaining channel balance is of course very important for soundstage "image" stability.

3. This DAC's digital filter does things "by the book" with a "ruler flat" audible frequency response. The antialiasing filter is an "orthodox" linear phase steep filter, likely one of the presets in the TI/BB PCM5122 chip and I suspect the "digital filter composite" graph is a reflection of the Burr-Brown design. As I mentioned above, it's not the nicest measuring digital filter with evidence of intersample clipping in the oversampling process; a technical finding which doesn't necessarily correlate with audibility. (For an example of what an essentially picture perfect "digital filter composite" graph looks like, check out the Light Harmonic Geek Out V2 based on the recent 2015-released ESS SABRE9018AQ2M DAC chip.)

4. When making comparisons, it is interesting that the HiFiBerry DAC+ Pro's harmonic and intermodulation distortion results are higher than others I've measured over the years. Realize of course that the magnitude is still very small compared to the amount introduced by one's speakers or room acoustics (is anyone going to hear 0.03% THD+N distortion?) - but worth noting as a fidelity "perfectionist". I wonder if this is a limitation of the PCM5122 chip itself. On the other hand, I am impressed by the very low stereo crosstalk with this DAC; one of the best I've seen. Not that this is audible of course, rather it speaks to the quality of the device and the well controlled noise levels.

5. Jitter is not a problem. Higher jitter can be found on S/PDIF interfaces and old synchronous USB devices. These days with typically asynchronous USB, ethernet, and this I2S board with its own oscillators, there's generally nothing to worry about.

6. Raspberry Pi 2 vs. Pi 3 for audio - distortion from wireless transmitter? Some have said that the inclusion of WiFi and Bluetooth in the Pi 3 makes it less suitable for audio applications due to distortion... Realize that I have made all of the above measurements with the WiFi module "on"! Yes, I have measured the noise floor with WiFi off also:

Raspberry Pi 3 & HiFiBerry DAC+ Pro: WiFi ON vs. OFF... Perhaps I'm mistaken, but I don't hear a difference in the noise floor :-).

As you can see, it makes no difference to the DAC+ Pro's analogue output (nor a difference to distortion results). IMO, this is another one of those "possible" issues raised by anxious audiophiles that really has no basis in reality... If it did, I think it would be nice for someone to document this so we can all avoid such gear.

In my opinion, the benefits of the faster Pi 3 (~50-60% increase on a processor intensive benchmark like Whetstone), inclusion of the 64-bit instruction set, with built-in wireless ability I think are worthwhile upgrades over the Pi 2. By now, it's also even cheaper than the Pi 2.

7. Value. Remember what we're talking about here. A self contained streaming device that can operate in WiFi or through wired ethernet, demonstrably capable of high-resolution playback up to 24/192, uses minimal electrical power, known to be reliable, easily purchased through on-line storefronts, and is priced currently just over US$100! Realize that I'm putting this little computer & DAC board up against devices costing much more... Even the Light Harmonic Geek Out V2 DAC costs ~$300 and you'll still need to have a computer to go along with it. A self-contained PonoPlayer would cost ~$350 but it's for a totally different purpose. I think it's also worthwhile remembering that there's an opportunity to repurpose the Pi 3 in the future should you want to try making it into a video streamer for example (like with Kodi). Furthermore, with the power of open-source development, you could convert this to be a Squeezebox-like streamer or Roon endpoint.

Looking around the internet, I see this video review from Hans Beekhuyzen for the DAC+ Pro. He claims the sound was poor without the use of a linear power supply (starting at 2:50 - "Not the sound I expected... Not by a long shot... I don't want to review stuff at this low a level...")!? Hmmm... Unless there has been a change in the board revision (mine says "HW 2.2" on the board), or he used some remarkably poor switching power supply, I see no evidence of a problem personally. Maybe he should take a moment and use the testing equipment in the background of the video and see what he finds... IMO, there's no reason to bring up questionable pet theories like the "water organ" analogy and claims of "voltage of the power supply not constant" as the rationale for vague subjective impressions. What evidence is there to support the claim of "stunning" improvement such that he's suddenly "enthusiastic" to compose a review!?

Linear power supplies are expensive, they're also much less efficient and larger - all characteristics very much antithetical to the benefits of the small, efficient, and inexpensive Pi/HiFiBerry that can be left on 24/7! Myths are plentiful in this world and I see no need to create or perpetuate ones around engineered products without actual evidence. I do recommend reading Benchmark's post on the "Switching Power Supplies are Noisy" myth for a rational second opinion on this. Make sure to watch the video also.

Based on these results and my subjective impressions discussed last week in the preview, I think it goes without saying that I'm impressed at the sound quality at the affordable price point. Sure, one could spend more for better measuring gear as per the comparison charts above but if we're just talking about sonic fidelity, I think we could put the Pi3/DAC+ Pro in a blinded, volume controlled session against other much more expensive streaming devices and DACs and not feel embarrassed in the least!

Of course, as discussed in the past, people will spend money for non-utilitarian reasons beyond fidelity and that's fine too...

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

In the last few days before publishing this, I've been running piCorePlayer (latest 3.02 version, September 2016) and using the Pi3/DAC+ Pro combo as a "Squeezebox replacement" in a system upstairs. Kudos to the team that brought this together! Very stable so far and excellent web-based set-up. Great stuff. (And yes, I ran a quick measurement demonstrating identical results at 24/96 and 24/192 as with Volumio.)

As much as I enjoy the power and DSP abilities of JRiver, good old Squeezebox/Logitech Media Server remains rock solid, running strong and stable for years now. Even better since I got digital room correction running earlier this year with BrutefirDRC.

Over the last couple nights, I have been listening to Macy Gray's recent Stripped (2016) album, which includes jazzy covers including her own song "I Try". No doubt the recording quality is excellent from the Chesky label. However, the "binaural+" dummy head recording is clearly meant for headphone listening. It really sounds strange through a speaker system with instruments appearing well placed but the vocals diffused without a sense of being able to imagine the person singing in the front soundstage! I guess Chesky must be targeting the headphone audience with these recordings as a reflection of the rise of mobile music consumption. Anyone know if there is a non-binaural mix out there? It would be a shame if there wasn't a concomitant live recording of the performance meant for standard stereo speaker playback (a multichannel version would also be very interesting I think!).

Happy October everyone! Here's wishing you many joyful listening sessions through the autumn/spring...