Topic: Bus speed and AD conversion latency

Hello,
I would like to address a technical question related to my other post on a more general level:

I am confused about the role of bus speed in AD conversion. There seems to be everything from slow USB 2, Firewire, Thunderbolt, PCI, up to superfast PCIe.

If latency is a key issue, why do I read many reports that USB 2 audio interfaces are doing a good job?
Is the amount of data transferred between CPU and DAC relatively small, so that bus speed is not the limiting factor?
In other words, is a fast DAC on a slow bus just fine, don't worry about the bus?

Or are the USB 2 interfaces not really great for Pianoteq in real time play, and faster buses give lower latency?

Focusrite spec sheets say latency for USB 2 interfaces around 2.5 ms, Thunderbolt 1 ms. That is a big difference in relative terms, but not in absolute latency time.
I guess 1.5 ms are not a game changing advantage worth hundreds of euros.

Thanks for your help,.
Andreas

Re: Bus speed and AD conversion latency

PianoTeq is relatively efficient so does not require expensive computing power. In my experience, you just need good ASIO drivers and a decent computer to run PianoTeq at very low latencies.

How shall we define latency? For digital pianos, a practical approach is a simplified "round trip", or the time it takes between pressing a key and sound arriving at the eardrum.

How much latency matters for digital piano? There are a lot of general latency studies and results might differ by person, activity, skill level, etc. Maybe a beginning pianist is less sensitive whilst a professional drummer might be more sensitive to latency.
Maybe less than 10ms "round trip" is difficult for the average person to perceive.

Calculating that "round trip" latency is a bit difficult as there are a lot of pieces to the puzzle. The "latency" figures reported by a VI or DAW or ASIO driver alone are rather useless. Some of the components you just can not change. A simplified picture for a very fast system might look like:

-Trigger key sensors and send midi data to audio interface (>1ms)
-USB and ASIO driver buffers - in & out + VI processing (>4ms) ****
-D/A conversion (~0.5ms)
-Sound from speaker to eardrum (~3ms per metre)

So "round trip" latency between key trigger and headphone output might be ~6ms (or ~9ms with loudspeakers).

This varies for every system. For example, a very fast system (good computer & good drivers) might allow you to get PianoTeq to run at say a sample rate of 48KHz with a buffer size of 48 (1ms "reported" by PianoTeq). A slower system might requre a larger buffer size to prevent dropouts/crackling sounds (say a buffer size of 512 (10.7ms "reported" by PianoTeq).

Also, if your digital piano has USB MIDI, you might not bother with an audio interface (although maybe something like ASIO4ALL could boost performance).

Apple has some good plug-and-play audio performance (there are some hidden buffers but still good performance).

Linked below is a comprehensive study of latencies with different audio interfaces (optimized drivers = low latency). PCIe is "best" but really overkill for most of us. There is not much ThunderBolt gear available and some "underperforms" (Thunderbolt still requires a clunky controller board inside your computer, drivers are inmature...).  USB interfaces with good drivers perform fine for most of us (USB3 is just as fast as USB2 for our purposes). Firewire and PCI interfaces are not generally sold anymore:

https://www.gearslutz.com/board/music-c...-base.html

Presonus wrote an article providing a simple summary of typical processes with computer music latencies. It is marketing and it is old so take it with a grain of salt: https://forums.presonus.com/viewtopic.php?p=95740

DAC speed should not concern you as a general matter. Assume 0.5ms. However, some FPGA are slow (e.g. Chord does not recommend some of its DACs for VIs). On the other hand, RME uses FPGAs to produce very low latency interfaces.

Of course, there is some natural latency with a grand piano. For example, it takes some time for the key stroke to activate mechanics inside the piano to produce sound (there are some interesting studies on that). And it takes time for sound from say the strings and soundboard to travel to the ears (say 3ms per metre).

So, jitter may be more important than latency. That is tougher for users to measure and evaluate. RME interfaces, for example, market jitter reduction as a key feature.

Last edited by music_guy (06-04-2018 17:22)

Re: Bus speed and AD conversion latency

music_guy wrote:

So "round trip" latency between key trigger and headphone output might be ~6ms (or ~9ms with loudspeakers).

Have you ever measured the over-all-latency ("round trip") of your Pianoteq-chain in real life? I mean from touching a key to eardrum?

I would highly recommend.

That is all, that matters for the pianist latency-wise. Not estimations.

Forget audiointerface "jitter" when you speak of latency. Jitter is lower than microseconds in DACs, far away from milliseconds.

Re: Bus speed and AD conversion latency

I measured about 6 ms between key contact and headphone out on myPianoTeq system. Even setting latency a bit higher, PianoTeq plays fine.

One of PianoTeq's secret weapons is the very "consistent" response timing across the keyboard and throughout the dynamic range. Today's sampled VIs don't come close. I think the on-board sounds of some digital pianos can be relatively consistent, albeit lousy sounding.

Computer system jitter is a bit different. It should be vanishingly small. But in complex, real-time computer music systems how would you know? I "speculate" the mind could adapt to longer-latency whilst it might go bonkers with randomly changing response times from every keystroke.

Since this is all very tough to measure and not a centre of scientific study, I just tried to optimize my computer. And picked up a interface which seemed to offer good jitter performance and good drivers.

Re: Bus speed and AD conversion latency

music_guy wrote:

I measured about 6 ms between key contact and headphone

Please define "key contact". -- Is anything about your measurement documented?

Computer system jitter is a bit different.

... you wrote: "RME interfaces, for example, market jitter reduction as a key feature."

As I said, jitter of the cheapest Audiointerfaces/DACs is much lower than microseconds (often measured in nanoseconds/picoseconds ranges). It cannot be heard as latency.

Re: Bus speed and AD conversion latency

Nothing is documented for audit purposes. Easiest and relatively consistent measurement for my purposes was record sound of finger hitting key with mic on one channel and note from of headphone jack on interface. Then compare in audacity.

Re: Bus speed and AD conversion latency

music_guy wrote:

[...]sound of finger hitting key with mic on one channel and note from of headphone jack on interface.

That is prone to errors as you compare two different input channels (line vs. mic) with unknown latencies itself (as long as not specified and verified).

Some older posts around these topics:
http://www.forum-pianoteq.com/viewtopic...63#p937263
http://www.forum-pianoteq.com/viewtopic...44#p937144

Re: Bus speed and AD conversion latency

groovy wrote:
music_guy wrote:

[...]sound of finger hitting key with mic on one channel and note from of headphone jack on interface.

That is prone to errors as you compare two different input channels (line vs. mic) with unknown latencies itself (as long as not specified and verified).

Some older posts around these topics:
http://www.forum-pianoteq.com/viewtopic...63#p937263
http://www.forum-pianoteq.com/viewtopic...44#p937144

Also tried same mics next to key and next to headphone driver with essentially same results. Nothing here is scientific but it was good enough for my purposes. Hopefully this responds to the original poster's query. Thanks

Re: Bus speed and AD conversion latency

Dear all,
thanks for your feedback.

While shopping around on various websites in search of the perfect gear set, I came across the simple answer for the USB-2 vs "faster" buses question:

In my question, I have inadvertently been mixing up the speed of signal transmission with bandwidth. The bandwidth of the USB-2 bus is well large enough for several channels of high definition audio signals. So, the audio signal is not getting to the target any faster on a bus system with higher bandwidth. USB-3 "feels" faster when transferring very large video files, because more bits are travelling in parallel. But the individual bit, audio or video, is not any faster than on USB-2.

So, as bandwidth of USB-2 is sufficient,for latency it does not matter at all which bus is used to connect an audio interface. And that's why so many high quality audio interfaces are "still" running on USB-2.

So, one issue sorted but tons of new questions, which will be adressed in other posts.
Cheers, Andreas

Re: Bus speed and AD conversion latency

Andreas wrote:

[...] for latency it does not matter at all which bus is used to connect an audio interface.

Proof? For example I measured 5 ms longer latency from USB-Audio compared to PCI-Onboard-Audio here:

https://www.forum-pianoteq.com/viewtopi...14#p939514

Cheers

Last edited by groovy (15-04-2018 18:38)

Re: Bus speed and AD conversion latency

Thank you very much for pointing this out. I think it does not contradict the USB-2 argument, but answers another latency question I have in mind, see below.

What I was referring to above was the conceptual design question wether USB-3, Firewire or Thunderbolt  are "faster" for audio than USB-2.

I just wanted to share what is not my insight but taken from the website of a highly regarded audio interface company (I've read so many in the last days that I forgot which one it was) that "speed of transfer" is not to be confused with bandwidth, that bandwidth feels like speed when you are transferring a high bandwidth video, but that even high definition video signals use rather narrow bandwidth. They use the analogy that passengers in a private business jet with 20 seats arrive at the same time as passengers in an Airbus 380.

In your example, I am not surprised that upstream, between the processor and audio chip, the PCI bus is wired in faster than the USB bus (and has higher bandwidth, too ;-). Doesn't the PCI bus have the most immediate access to the core machinery of all buses?

This would be a strong argument to use a PCI or PCI-E soundcard, as you are saving time at a level where other buses are not even in the game yet.

Re: Bus speed and AD conversion latency

USB2/USB3 can provide excellent results from high-quality interfaces (see TAFKAT's low latency interface database thread for a clear idea). You don't need to spend a lot of money and as you note, USB3 will not provide better latency performance for the home-gamer.

Thunderbolt is still a bit immature but can get closer to PCIe speeds in theory (good engineering still required - see TAFKAT's database for his preliminary thoughts). Currently, there are zero TB3 consumer audio interfaces. And I believe that there are zero 2018 computers sold with TB2 ports. So you might need to buy a TB3-TB2 adapter cable. I see some TB interfaces work well on Mac but some are still beta on Windows so do your research to avoid any "problem children". At the computer level, all TB computers require an external controller; rumours indicate TB will be incorporated into the CPU c.2019; ideally, that could improve performance and reduce issues.

PCIe is for professionals.

Firewire support is disappearing.

Re: Bus speed and AD conversion latency

Roland - Mobile UA

USB Audio Interface

EXPERIENCE THE BEST SOUND POSSIBLE, NO MATTER WHERE LIFE TAKES YOU

Music production used to mean being tied to a studio. Today, you can create high-quality music almost anywhere. You might come up with an idea in one place, edit tracks on the go, and mix the whole thing in yet another place. The studio is wherever you are and wherever inspiration strikes. But sound quality is always an issue. You need to know that your audio interface is performing to professional standards, while not weighing you down when you’re on the move. With the Mobile UA, you don’t have to compromise audio fidelity to be creative outside the studio.

- High-quality PCM and DSD playback with support for a variety of formats

- Roland's proprietary S1LKi DSP technology makes everything sound its best

- Powerful headphone amp provides plenty of volume in noisy environments

- Supports up to four outputs; configurable as two stereo channels for main and monitor

- VS Streaming technology for extremely stable low-latency performance

- About the size of a deck of cards, it goes anywhere

Re: Bus speed and AD conversion latency

Leandro Duarte, I frankly wonder if you are being paid to advertise Roland products, because that's just a post of marketing material from Roland's website.

You're not relating any personal experience or knowledge, just (in effect) advertising.

StephenG

Re: Bus speed and AD conversion latency

music_guy wrote:

...Currently, there are zero TB3 consumer audio interfaces.

That depends on your definition of consumer, I suppose. The Universal Audio Arrow is a $499 Thunderbolt 3 bus powered interface which might be considered for casual use by some professionals. I’m thinking Zoom or Focusrite May come out with a sub-$300 TB3 interface this summer.

Re: Bus speed and AD conversion latency

Hi all,
Leandro's post was Roland marketing text, but I am grateful that he pointed this Interface out to me as I hadn't known about its existence.
I might get one 2nd hand from eBay and give it a test drive.
Best wishes
Andreas

Re: Bus speed and AD conversion latency

Andreas

Currently I use a Behringer UMC202, and I do not recommend it to my friend. I recommended Mobile UA because I think you deserve good equipment.

Re: Bus speed and AD conversion latency

Hi all and hi Leandro,
I ordered the Roland Mobile UA and I am VERY happy. Thank you very much for the recommendation.

It offers 4 independent channels recognized by Pianoteq and working (tested with headphones). As I only have two speakers connected right now, I tested playability in a two channel setup. Very impressive, to my ears no discernible latency at 256 samples. I test latency by having the Casio's internal system playing in parallel to the MIDI-Pianoteq sound. This is not a numerical measurement, but if I can't distinguish two different sounds on a keystroke I figure it will not affect my playing at all. I can hear a slight latency at 512 samples.

So: sleek package not larger than an mobile phone, easy setup, great sound, minimal latency, 4 channels, for €100 -> this is a very clear recommendation.
The only downside is that it is just out of production, I've picked it up on eBay.

Best wishes
Andreas