You remind me of another change since joining the Diretta cult last summer: while I still prefer to turn down lighting, I find that 3D imagining within the expanded soundstage is even more tangible with my eyes open. Maybe because “seeing” that individual instruments palpably occupy their own spaces without regard to where the speakers physically exist?
Clean up on aisle 5 complete…I think 
2 Likes
I’ve avoided posting on the other thread, but this thread seems to confirm my suspicions. Namely: when you use an inexpensive general purpose system on a board with a general purpose operating system, the amount of noise being generated due to processing of network packets is not really statistically relevant- especially since the TCP/IP stack of Linux is already heavily optimized.
Anyways, I was honestly hoping to be wrong… I mean, if a pair of RasPi’s really could make a real difference sign me up! Perhaps custom hardware and software (definitely skip Linux or any OS for that matter), you could see a difference in the signal, but obviously gonna be $$$.
A couple of points:
- The OS was AudioLinux, which is supposedly optimized for audio, whatever that means, and for which I had to pay. You may consider that a general purpose system, but most dedicated streamers use some modified version of Linux.
- I used Diretta versio 147_13, which is supposed to use Layer 2 directly to send packets from host to target, so no TCP or UDP, only Ethernet.
So AudioLinux is built upon RTOS Linux- the idea is that it is “real-time” which important for latency and to avoid jitter. I can see that being valuable/useful since normal Linux can put a task on pause in order to service other tasks. But that has nothing to do with “noise” created by “processing packets” or anything like that.
As a networking geek, I can say with a great deal of confidence that parsing the IP and UDP headers is quite simple (I’m going to ignore a bunch of “options” that can make things complicated, but you’re not going to use in an audio streaming environment) and already extremely optimized/efficient on Linux. It’s really trivial actually. The idea that removing those bytes and processing is going to provide some kind of dramatic improvement is a bit hilarious to me.
But yeah, the big thing was you’re running the same OS as the vast majority of streamers on COTS hardware that was not designed around any kind of noise suppression whatsoever. Maybe processing network packets does introduce some level of noise (some of your graphs show some difference afterall), but my gut was always that this is trying to optimize the last 1% of the problem away while ignoring the first 99%.
1 Like
The idea that fiddling with the OS or the network protocol in any way is going to provide any kind of improvement is 100% hilarious to me. That is, of course, when a decent DAC - i.e. a DAC that does what it’s supposed to do - is used. I did however put that aside and took measurements because nobody provided any kind of objective evidence that Diretta in particular makes any difference in sound quality. Not only did it make no difference in the DAC’s output, it didn’t make much difference in the power rail noise either.
6 Likes
Real time and low latency kernels are not the same. Typically, low latency is used in audio (recording and mixing), and real time in industrial or financial settings.
A real time kernel ensures a specific process occur son time at the expense of all other processes (incorrectly set up, real time can actually slow things down on a general purpose computer.)
Both require the application to communicate with the kernel schedule, and we have no idea if Diretta actually communicated with the scheduler in Audio Linux (or other products and firmware.)
4 Likes
As you say, the important feature of real time kernels is not low latency. Indeed it is quite possible for a real time kernels to have worse average latency although, with carefully designed real time systems, worst case latency will likely be better.
The defining characteristics of real time kernels is deterministic and thus the predictable behaviour combined with the ability to control relative priorities of all tasks (even system tasks) so that a system can be designed that will satisfy the real time constraints.
With lightly loaded systems, as a Roon Bridge system should be, there will be little benefit to an RTOS because the system will always be able to respond to the arrival of data in a timely manner. In fact, because of buffering in various parts of the system, the deadlines in an audio system can be very lax indeed. Witness the reports of Roon endpoints continuing to play for a number of seconds after the network connection has been removed. Thus ‘timely’ can often be surprisingly relaxed.
An RTOS becomes necessary when you have multiple tasks being performed some of which have tight deadlines and others more relaxed deadlines. The deterministic behaviour of the RTOS allows you to prioritize and organise tasks so that deadlines can always be met (unless, of course, you start hitting the limits of your processors capability).
In addition, RTOS’s often provide schemes to mitigate conditions that are hard to predict but can cause a very detrimental affect on the ability to meet deadlines. These schemes can actually degrade latency. The classic example of this is ‘priority inversion’ where a high priority task cannot proceed because it requires a resource that is currently held by a low priority task - but the low priority task can’t proceed to the point where the resource is released because a mid priority task is running. Thus the mid-priority task appears to be running at a higher priority than the high-priority task.
Finally, taking a piece of software designed for a general purpose OS (or even multiple such OS’s) and running it on an RTOS does not typically convey any advantage. To take advantage of the benefits of an RTOS, applications have to be designed to employ the mechanisms offered by that RTOS.
Edit: As an aside, I spent the last 15 years or so of my professional career working on a real time system with time constraints markedly tighter than any audio system but which ran on the standard Linux kernel and not the RT kernel. Whilst I, personally, was not involved in the investigation that lead to this decision (I was too late to the party), there were very sound reasons for preferring the non-RT kernel.
6 Likes
And, we don’t know, Diretta does do this (I suspect not.) A standard package runs on Audio Linux, GentooPlayer, and some firmware offers Diretta support without any configuration (unless I missed this in the guide.)
What evidence is there that supports the need for a RTOS? Quite important since the claim in the original thread is party based on using a RTOS (my question there was left unanswered.)
2 Likes
A post was split to a new topic: Using AI content
@marian did you ever post the results of your listening test? I’ve scanned the two long threads and I haven’t been able to find them.
Hi Bruce, not yet. Apologies for the delay. I completed stage 2, one stage remaining. My kids seem to be in short supply. When done, I’ll update the first post with the results, so you won’t need to search for it.
2 Likes
Cool
Fast switching is essential
Long term listening is essential
Only those who’ve passed the ‘Harmon Cardon Listening Tool’ test
Some cat owners who expressed a preference
I’m on the edge…
Each stage has short and long listening btw. I need 3 stages to make it as scientific as possible.
2 Likes
I’ll just note that if my Orch-OR quantum consciousness theory is correct and that Diretta is coupling directly to our brains, then we should expect both no conventional measurement success and ABX hearing support for Diretta…obviously!
In the umbra of this eventuality, we need to look towards neo-conventional verification methods based on the theoretical work of brilliant polymaths like Arthur Koestler in his incredible The Roots of Coincidence that posits that quantum consciousness interactions rule both conventional modes of consciousness and paranormal phenomena. It’s virtual particle exchanges everywhere!
Therefore, @Marian, if you have a positive signal from the subject tests, please follow-up with a battery of paranormal tests of your children, including all the tests that you are familiar with for farsight, intuition, randomness, and don’t forget pyrokinesis!
1 Like
Let’s wait and see what the results of the listening tests performed by @Marian are before we start proposing explanations (especially such left-field ones) why such tests might show a difference when the measurement tests don’t appear to.
1 Like
Well, I’m still waiting for a Christmas Miracle.
2 Likes
It’s coming today, but I can’t promise miracles.
3 Likes
I’m on tenterhooks! My bets are on strong evidence of omphalomancy!
2 Likes
I finally added the listening test results in the first post of the thread. Just go to section 2. As usual, let me know if you have any comments.
Happy holidays!
8 Likes