At the end of last year I completed a project to build a Rasbperry-Pi-based DAC streamer. Effectively it was building a power supply for a Raspberry Pi zero and the connected Hifiberry DAC+ pro HAT. The idea of that power supply was to provide clean power to the DAC to ensure the best possible sound reproduction. Cutting a long story short: I misused that power supply a little for something else and because of that broke one of the TPS74A7 regulator modules which I chose because of their low noise capabilities. With the idea of rebuilding that channel with something new and with a little more power I started searching for a regulator with a good reputation for audio applications.
It didn't take long and I was reading a lot about regulators and power supplies in various DAC forums. Now here comes the disclaimer: I don't listen to high definition audio files - all my material is 16bit/44.1kHz FLAC data. Also, I don't claim to have audiophile hearing capabilities. So if my FLAC data would be up-sampled to 24bit/96kHz then I would definitely not be able to hear the difference - unlike some people who claim that there is a clearly audible difference. Finally, I like to listen primarily to popular music - which has been mastered completely different through the last decades: Perfect for vinyl (with all it's limitaions set by RIAA) in the 70s/80s, perfect for CD in the 90's and early 00's, and much more compressed since then to allow the song to be recognised even when my daughter plays it on the speaker of her mobile phone.
And then I read this post from a guy who claimed that with a certain power supply to power a specific DAC streamer he can hear the oboes clearly more center stage than with another power supply. And several commentators agreed or offered alternative audiophile power supplies to achieve an even better or different sound. The streamer in question costs around 500 EUR and the power supplies compared were at around 300 EUR. Now, my naive assumption would be that a 500 EUR DAC streamer would have it's own power supply section which ensures that incoming power is regulated and filtered to an extend that there is no impact on the generated audio signal - filtered with regards to EMC but also to ripple on the voltage signal. But obviously not...
Well, which elements of a power supply do influence the output signal of a DAC streamer? The experts in the various forums say ripple / noise on the voltage, the impedance of the power supply and the ability of the power supply to handle fast load changes. Especially the impedance part makes complete sense to me. Regarding the load changes, however, I struggle to believe that this has too much impact - simply because bottom-line a DAC streamer is a generator and not an amplifier. Therefore I decided to have a closer look at the load changes that a power supply of a DAC streamer has to endure.
For this, my Hifiberry DAC offers a great testing ground as it doesn't have it's own power section on the board (unlike it's more premium successor Hifiberry DAC2 HD). This means that it even uses the 3.3V from the converters of the Pi. Consequently, an unsuitable power supply should have a direct negative impact on the generated audio signal. The music data I streamed for these tests was sent to the Pi/DAC through WiFi, so no LAN cable with a potential source for noise was used. For the tests I powered my combination of a Raspberry PI zero and a Hifiberry DAC+ pro with four different power supplies:
- the cheapest 1A USB wall adapter I could find in our house
- an iFi iPower Audio 5V ultra low noise power supply with 2.5 A
- my homemade 1A TPS74A7 based power supply
- a Rohde & Schwarz HMP2030 lab power supply with 5A per channel
Of course I tried to find differences by listening to the playback with these power supplies. For this I used a "Head Box S2" headphone amplifier from Project Audio as well as "Amiron Home" headphones from Beyerdynamic. The song I used was "Rather Be" from "Clean Bandit". Why? Because it starts with a very distinctive violin and I find a single violin perfect to spot "audio imperfections".
For the technical comparison I borrowed a little bit of gear from HIOKI Europe's demo pool (which is easily possible for me as I work there): A MR6000 memory recorder with a 16 bit / 1 MSamples high resolution voltage unit, and a CT6711 current sensor, which for this purpose has a bit of an "overkill bandwidth capability" of 120MHz but at the same time offers a perfectly fitting 500mA current range.
For each power supply I recorded the (5V) supply voltage to the Raspberry Pi as well as the current while streaming music. In the set range of the current sensor 500mA of measured current would be shown as an output voltage signal of 5V. The MR6000's sampling rate was set to the maximum capability of the input unit, so both voltage as well as current values were recorded with 16 bits and sampled 1 million times per second. This should be more than sufficient to catch even the fastest load changes of a USB power supply.
Above you can see the voltage in blue as well as the measured current in red. The action starts with the Raspberry starting to play the music. But even in idle mode you can see the current constantly changing between 100mA and 150mA, shown on the red scale as voltage values between 1V and 1.5V. In line with the current changes the voltage reacts to this by moving between 4.95V and 5V. Once the music streaming starts you can immediately see the voltage dropping more when the current changes. I'm sure you are not surprised when I reveal that the image above shows the measurement of the cheapest USB wall charger I could find in the house. Lets zoom in a bit and look at one of those load changes:
Can you see how "unclean" the 5V voltage is? As you can see, the noise is unrelated to the load changes - but again, this is the cheapest 5V USB wall adapter I could find in the house. Also, you can really see how the voltage drops when the load increases. But I wasn't at all surprised to see these results. In fact, I saw what I wanted to see.
Next up was the iFi iPower which costs around 70 EUR. iFi says about this power supply: "Add the iPower into your system and you’ll notice music come back to life. It can add in detail originally lost through conventional power supplies." The power supply used above certainly was convential - so let's hook up the iPower and play "Rather Be":
Yes - that's what I wanted to see! Because I spent 70 EUR on this and I would have really been annoyed if this screenshot would have looked like the one with the conventional wall adapter from the first image. Both screenshots show the beginning of the same music track being played - the only thing which is different is the power supply.
iFi specifies the noise of the iPower to be just 1uV - they don't specify at which frequency these 1uV apply, but regardless of that it is a really low value. Lets zoom in to look at a single load change and the cleanliness of the voltage:
Again, the current curve looks much cleaner and the voltage drop is lower with the iFi than it is with the cheap wall adapter. The noise on the voltage is not really any different than with the wall adapter, though. I assume this comes from the voltage regulation on the Pi - but this is just a guess. Anyway, it looks great - but does it also sound great?
Yes, it does (sound great). But so does the music when I use the cheap wall adapter. Honestly, I can't hear any difference. Two possibilities: Either my hearing ability is a lot less audiophile than I would like to admit, or while the curves with the conventional wall adapter look "dramatic" - maybe they don't have any influence on the generated audio signal? After all the voltage drops of the cheapest wall adapter were 70mV compared to around 50mV with the iFi - so overall the voltage drops where in the range of 1% to 1.2% for around 200 to 300 micro seconds. Is that enough to have an audible impact on the output signal of a DAC?
Next up - my home-made power supply based on Texas Instruments' TPS74A7:
"Mr. Scharrmann, don't give up your daytime job" was the first thing I thought when looking at the that image. It's probably not as "dramatic" as the curve of the "cheapest wall adapter in the house" but at the same time it's far away from the iFi. Does it get better when zooming in?
I think it looks much closer to the graphs of the iFi than to those of the cheapest wall adapter in the house. And does it sound different? Nope - all three power supplies result in output signals that sound exactly the same to my ears.
But now... the next and final test candidate is far away from the 70 EUR iFi: The next candidate is a lab power supply of the HMP series from Rohde & Schwarz. Each of the three channels can provide 5A of current as long as the total power stays below 80W - load changes in the range we have seen above should be handled with ease.
I have to say that I was quite suprised about the level of voltage drop. Comparing the graphs even the cheapest wall adapter in my house has less of a voltage drop. But I think there are two things to keep in mind:
- All tested power supplies apart from the HMP2030 are fix voltage power supplies with mostly much smaller current outputs. The HMP on the contrary has a freely selectable voltage range between 0 and 32V and a current range between 0 and 5A per channel. Therefore I might "compare apples with pears" when I put the HMP neck on neck with the purpose built iFi. On the other hand, the iFi retails for 70 EUR including VAT while the HMP2030 retails for 1250 EUR excluding VAT.
- Also, I think it is important to point that I did not use the sense input of the HMP2030: This measures the voltage at the DUT and allows the power supply to counter any voltage drops or rises. Lets zoom into this picture:
Again, I am surprised about the high level of voltage drop. I really didn't expect this. But as mentioned before, I didn't use the sense input. That's why I repeated the same measurement - but this time by using the sense input:
The above graph of the HMP2030 with the sense input used is probably the most "dramatically" looking graph. At the same time it is a great way to explain a sense input function of a power supply because the difference is easily visible when comparing the two curves (one with and one without using the sense input). It even becomes more clear when zooming in:
The actual voltage change is probably the highest of all power supplies in the last graph. But you might already guessed it: Again, there is no audible difference when comparing the sound. In fact I couldn't hear any audible difference with regardless of the power supply used at all. One thing to note is that the load changes are in no way aligned with the sound played - so a more powerful part in the song does not result in higher current being drawn.
What do you think about this? Should I be surprised that the audio playback sounded the same with every power supply, or was that something you think was to be expected? Any flaws in my measurement you spotted? Any details missing? Please let me know about your thoughts...
Thanks everyone for the great comments so far - a very valid comment discussed my setup and the cable resistance that might explain the voltage drop. I have to admit that I completely ignored cables losses in my measurements and with hindsight that was a mistake. It bugged me that much that I went into the office this Sunday morning and borrowed a HIOKI RM3545 resistance meter. I have no possibility to measure the cable resistance with the iFi, because the iFi cable is fixed to the PSU. That said, the iFi has by far the longest cable (the cable length from PSU to USB micro connector is 210cm including a short adapter cable from 2.5mm to USB micro) yet the smallest voltage drop occurs with the iFi.
I could measure the cable resistance for all other power supplies, though:
|Power Supply||Cable Resistance [mOhm]|
|The cheapest wall adapter in the house||225|
|My homemade TPS74A7 based power supply||238|
|The Rohde & Schwarz HMP2030||207|
To me, these values explain a part of the voltage drop, but certainly not all of it. For the HMP2030 (with the lowest measured cable resistance yet the highest voltage drop) the cable loss in my setup is around 25% of the measured voltage drop.
Again, thanks a lot for the great comments - any other thoughts please keep letting me know.