I used to think the DAC was 90% of the sound. This sub-$300 DIY project proved me wrong

Tinkering
1

Hey fellow tinkerers,

A quick note: I’m posting this to share a personal journey and my subjective results. I know and respect the “bits are bits” philosophy that’s popular in this community, but this thread is for those of us who enjoy tinkering with software and network topologies to discover possible effects on sound quality (real or imagined).

Here is the direct link to the full instructions:

For years, I was a firm believer that in any digital endpoint, the DAC dictates at least 90% of the sound quality. The transport, I thought, was a minor player.

I was wrong.

After spending the last four months on a new DIY project, my perspective has completely flipped. I am now convinced that typically more than half of endpoint performance can be attributed to the transport.

The project is a dual-Raspberry Pi Roon endpoint using the Diretta protocol. The reason I’m posting is that this is no longer an exotic, high-dollar experiment. The cost to “kick the tires” on this at 44.1 kHz is now squarely in tinkerer territory for many Roon subscribers:

  • ~$225 for the parts (think two RoPieee builds plus an adapter)
  • $69 for the AudioLinux subscription

For under $300, I’ve heard a quantum jump in sound quality, and it’s making me re-evaluate my entire system and thinking around digital audio.

I’ve got a lot more to share on how and why this technology works (turns out that our home networks are “too fast” for audio data), but I’ll start with a simple question:

Has anyone else here felt that a streaming transport made a bigger difference than a DAC upgrade?

Edit: Let’s keep responses civil and friendly. I will be flagging inappropriate and off-topic comments for moderators to remove. Thanks.

12 Likes
5

Can you maybe describe this DIY setup?

I use a raspberry pi streamer (with RoPieee – and a Pi2AES HAT) but know nothing about this Dual-Pi-Audiolinux thing. Sounds intriguing and worth testing.

(Or maybe you have another thread on the forum describing what it is and what you did?)

2 Likes
6

I’m following a similar path, albeit it parallel in the sense it’s using SOtM not RPi

All very positive so far…

And why not… having left vinyl for CD (many years ago) my Mark Levinson 31.5 CD was replaced by a PC, that by a streamer, and all with significant improvements - bring on the next developments ASAP!

2 Likes
7

Sure. I was hoping someone would ask. :smiley:

Here’s what it looks like:

From left to right:

  • Wired Network
    Plugable USB to Ethernet Adapter (connects the setup to your home network, supports checksum processing offload)

  • Diretta Host
    Raspberry Pi 4 computer in an Argon ONE V2 case. Runs Roon Bridge and the Diretta Host software

  • Point-to-Point Connection
    CAT6 Ethernet cable connects the onboard network interfaces on the two computers. These interfaces support auto MDI/MDI-X, so no crossover cable required.

  • Remote Control
    Argon IR Remote (play/pause/next/previous/stop and mute/volume-up/volume-down commands forwarded to Roon Server)

  • Diretta Target
    Raspberry Pi 5 computer in an Argon ONE V3 case. Runs the Diretta Target software and is connected via USB 2.0 to an external DAC or DDC.

The inputs and outputs of this assembly are, essentially, the same as a single Raspberry Pi running RoPieee or similar endpoint solution: Ethernet in / USB out

Not shown are the power supplies. I’m using 5V/5A power supplies on both, even though the RPi4 technically only requires 3A.

5 Likes
8

Yes. I definitely think so. :smiley:

I’ve tried to make the process as easy as possible while also being as transparent about the “hows”, "whats’ and “whys” in the guide I’ve published on GitHub:

The 66 page doc can be intimidating, depending on one’s confidence with computers, but there are a few pages of intro, and you’re up and running with Roon by page 29. I call this a three coffee project:

  1. Hardware Assembly
    Unpack all of the bits and get the Raspberry Pi computers into their cases. Hardest part is peeling tape from the thermal pads and not losing any of the tiny screws!

  2. Image Preparation
    After purchasing AudioLinux ($69 covers both computers), you’ll use Raspberry Pi Imager to flash the images to two blank microSDXC cards. Pop the cards into the RPi computers, boot them up and install updates.

  3. Point-to-point Networking
    After creating a few network config files, you’ll power down both computers and reconnect them as I showed in my photo, with that CAT6 cable between them. Power them up, install audio software (Roon Bridge and Diretta), and configure the new Zone in Roon.

I’ve added QA scripts in my doc to help you check your work along the way. There are six other appendices for configuring optional goodies, including:

  • Argon40 Fan Control
    Disables the case fans unless things get really hot (above 65C). Has not happened for me so far.

  • IR Remote Control
    Adds a service on the Diretta Host that relays IR remote control signals to Roon Server. I really like this because I tend to listen with my eyes closed. If a track isn’t doing it for me, instead of waking my phone or tablet to skip, I can reach for the IR remote without getting out of “the zone.”

  • Purist Mode
    Reduces background processing and network traffic on the point-to-point link. Some say this mode sounds ever-so-slightly better while others insist on it.

  • Web UI
    I’ve written a little web UI and Flask API to control Purist Mode, manage the Roon Zone name for the IR Remote control, and for getting to the underlying AudioLinux UI conveniently from one page. This runs on the Host computer.

  • Advanced Performance Tuning
    Incorporates CPU isolation for audio-related processes and interrupts. With these enhancements in place, we can use more aggressive Diretta protocol timing (a smaller time interval between Host to Target data transmissions). Everyone who has tried these attests to a clear improvement to sound quality.

  • Replace Timers with Process Hooks
    This is an enhancement to the previous tuning step. By default, on AudioLinux, timers wake up once/minute to enforce CPU core affinity, moving processes to specified CPU cores as needed. However, on the Raspberry Pi platform, I observed that these timers can consume up to 20% of the CPU to do their work. Not cool. So, I added hooks for audio apps to handle CPU core isolation just once at the time they start (or restart). This eliminates the need for these demanding timers. It’s the little things, right?

A lot of information, but after you’ve had time to read through it all, including a detailed parts list (sorry, links are for US shops), let me know if you have questions. Thanks for having a look. I hope you will consider building a Host + Target pair to try or for a friend.

5 Likes
9

In no subtle ways, YES!

I’m kicking in open doors here, but i have seen/heard similar leaps in audio quality lately.
Actually in two ways, one is powering the digital transport (or at least parts of it) with absolutely overkill supercapacitors from IAN Canada:

IMG_1677
IMG_16771920×1359 167 KB

This is powering (3.3V) the digital transport board which is controlled by a Pi (separately powered).

The second way of elevating sound quality is utilizing the already licensed and packaged Diretta available in my SOtM sMS-200 devices. In conjunction with Roon on the back end this is revelatory as well. In my setup the separate network card in the Win based server, used as a dedicated Diretta Link to the sMS acting Diretta Target is a very clear improvement.

5 Likes
10

Thanks for documenting this so well, will have a good read and think about building that. What DAC is this driving (apologies if you have mentioned this and I’ve missed it), and do you get similar results using different DACs?
The reason for my interest and question is that I have just purchased what I hope is my endgame DAC (yes really, stop laughing at the back…) a T+A DAC 200. This is supposed to have high precision clocks and buffer everything coming in and therefore be immune from upstream influences. It is currently fed from a RasPi3+ running RoPieee using USB out to the DAC 200. I know the Pi3+ has issues with regard to the shared USB/Ethernet interface so an upgrade has never been out of the question and it is getting on for 8 years old now.

3 Likes
12

Hi @Anthony_B. This is an excellent question, and it gets to the heart of the matter.

I’ve been testing this Diretta project with five different DACs (plus a DDC), all ranging in price from $200 to $750:

  • Topping DX3 Pro+
  • SMSL D6-s
  • Topping DX5 II
  • Gustard X16
  • Gustard X18

Diretta substantially elevated the subjective performance of all of them in obvious ways. They all sound like they were replaced by DACs costing $1k to $5k more.

This is the key point. My assumption was exactly the same as yours: the better the DAC, the smaller the difference Diretta would make. I figured that high-end DACs, like your T+A, would be immune.

A friend brought over a $7,500 DAC for us to try. To our complete surprise, the differences between a single Raspberry Pi (Roon Bridge) and the dual-Pi Diretta setup were even more obvious!

I have since helped folks with $30,000 MSB DACs get Diretta working. I told them upfront that I had no idea if it would make a difference, as MSB goes to heroic lengths to purify the incoming signal. Again, to my surprise, they are delighted with the improvements to musicality and listening engagement that this <$300 project brought to their elite systems.

Based on these experiences, it’s very likely that you’ll experience similar improvements with your excellent T+A DAC 200.

A Note on What to Listen For

There is an interesting phenomenon when A/B testing Diretta. It’s common for some folks to initially prefer the “before” sound. They often describe the Diretta sound as “calmer,” “more relaxed,” or even that it has the effect of “slowing down the tempo.” They may miss the more energetic and “detailed” original sound.

However, upon further listening, what they realize is that they were mistaking low-level digital glare for “detail”.

Our brains (subconsciously) work overtime to make sense of digital music when that noise is present. When the noise is finally gone, our brains need a few minutes (a track or two) to adjust to this new, low-effort presentation.

Once that adjustment is complete, most listeners find that the sound with Diretta has brought:

  • Greater harmonic density
  • The ability to follow melody, harmony, and rhythm with far less effort
  • A more open and defined soundstage

In general, people report that their listening sessions have become longer and more enjoyable, even if they can’t immediately put into words how the sound is different.

6 Likes
13

Hi Tony.

Like you, I started with RPi3B+ (Allo DigiOne) years ago, with different DACs and operating systems along the way since then. I installed a pair of Meridian DSP9 active speakers last year, which accept up to 384kHz @ 24-bit digital input, with a RPi4 running as Diretta Target -without the shared USB/Ethernet issue- as source. Great sound!

As wonderful as the single RPi4 w/Diretta sounded (using superior cabling and power of course), I found that installing a second Pi as a separate Diretta Host in August, rather than running Host along with Roon Server on my NUC as I had been doing, gave me dramatically even-greater performance: the biggest jump since installing these $85k MSRP speakers.

I doubt that any claims of DAC immunity from upstream influences will save you from this relatively inexpensive revelation. Go for it!

2 Likes
14

Hi,

sounds good (literally it seems) so I have to try this then. I would really love (more need I think) to know what is going on here to make this difference. The difference in sound as described seems similar to the claims made for DSD via HQPlayer? I am going to try that next as I have a Mac Mini to use for HQPlayer and to drive the DAC200 with but don’t have it all connected up yet, only just got the DAC 200 last week so early days. I need to baseline the sound I have before making changes.
@Brad_Burnside thanks for the additional report.

1 Like
15

Has anyone ever tried to measure the alleged reduction in noise in the DAC’s analog output and published the results?

5 Likes
16

Yep, not knowing what Harmonic density is :grinning_face:
I’ll definitely agree with these two

And I’m hoping the direct crossover Ethernet connection will improve things even more (haven’t had the chance to do it yet - hopefully tomorrow)

1 Like
17

Hi @Marin_Weigel,

That is a perfectly fair and logical question.

I’m not aware of any simple, third-party “before and after” measurement of a DAC’s analog output that definitively quantifies this. (If anyone has one, I would genuinely love to see it!)

The theory behind Diretta, as I understand it, is that the “noise” being reduced isn’t in the digital signal itself, but is the electrical noise (RFI/EMI) generated by the transport’s own CPU in response to “bursty” network data. The goal is to create an “electrically quiet” environment for the DAC by smoothing out the transport’s processing load, which in turn keeps the DAC’s internal power rails cleaner.

My suspicion is that this type of internal, high-frequency processing noise is incredibly difficult to measure at the DAC’s final analog output, as it’s not a traditional signal-level artifact.

This is exactly why I find the topic so fascinating, and why I posted this in ‘Tinkering’ to begin with. The focus of this thread is to explore the subjective listening experience—we’re sharing notes on software and topology changes that affect how listening sessions make us feel. It seems to be a case where these experiential results are, for now, far easier to detect with our ears than they are to capture with a standard audio analyzer.

It’s a great question, and I’d be thrilled if someone did manage to measure it definitively someday!

2 Likes
18

Hi @Andrew_Stoneman,

That’s a totally fair point! It’s one of those audiophile terms that’s tricky to define. Let me try to explain what I mean by it.

The timbre of an individual instrument or voice is definitely part of it—what makes a cello sound like a resonant wooden body instead of a flat “etched” sound.

But, for me, “harmonic density” is even more apparent when multiple voices and instruments are playing at the same time.

It’s the ability to both distinguish each individual voice… and more deeply appreciate their combined sound. You can more easily follow how each instrument is contributing to the chords, or how octaves are adding weight and energy to the sound.

It’s about being able to hear what’s going on musically with less effort.

Thanks for sharing your results! I’m really looking forward to hearing your report after you try the direct crossover connection.

2 Likes
19

Of course we would all love to know what is going on from a technical perspective but that isn’t what this thread is about (as made clear at the start). If you at all follow ASR you’ll know that the majority of audio equipment measures in a ‘blameless’ manner these days, with SNR’s and distortion measurement way below what we should be able to hear, especially in a less than ideal home environment. The designers of Diretta must have a way to measure the results of their protocol, but the last time I looked at their website it was poorly presented and not exactly encouraging. I would also be interested to know how this compares with the other professional studio standards for audio transmission, RAVENNA and Dante etc (AES67). They will all be bit perfect so should sound identical in David’s ‘Pi-Pi’ configuration.

2 Likes
20

When only allowing for a subjective approach in evaluation, everyone should be critically aware that
a) they are prey to all the various cognitive biases and dissonances, which don’t evaporate by simply ignoring them,
b) strict scientific and engineering approaches are the foundation of the underlying technology they are using.

Diretta’s inventors only ever showed the effect on the allegedly evened out network transfer graph and infer audible improvements on these grounds.

These days, any noise artifacts in the analog domain are easily measurable orders of magnitude below human auditory perception.

Food for thought …

8 Likes
21

Hi @Marin_Weigel,

Your sarcasm makes it clear you find the entire premise of this discussion “ridiculous,” and that’s fine.

As I stated in my very first post, this thread is “for those of us who enjoy tinkering with software and network topologies to discover possible effects on sound quality (real or imagined).” It is not a thread to debate the “bits are bits” philosophy or demand empirical proof for subjective reports.

I’ve already answered your question by stating that I’m not aware of the measurements you’re looking for.

This line of argument is off-topic for the discussion I’m trying to host. I have flagged it for moderation and will not be engaging with it further.

7 Likes
22

Ah, ok, that makes sense

I agree, it’s great when you can finally unravel what seemed like a noisy hash (what was the producer thinking?!) and hear the different components separated out - so it suddenly makes sense

And yes, my system was pretty good at that (I have a few reference tracks specifically for that e.g. Radio Days by World Party) and Diretta does unpack them even further

1 Like
23

Thanks Dävid, for sharing this well documented experiment. I saved it for later experiments of my own. If you are looking for more improvement of your DACs, see if you can have a DAC with I2S input and get the I2S signals from your 2nd RPI, feed them through a FiFoPi with good clocks, than straight into your DAC and you will have your next leap.

best,
doede
blog.dddac.com (pops up as not “secure” - but don’t worry :wink: )

1 Like
24

Whilst everyone in the community must remain courteous and respectful, it is absolutely fine for disagreement and discussion. In this forum, everyone has the freedom of expression so long as they adhere to the community guidelines.

6 Likes