[LAU] Audio interface latency measurements

[Len Ovens]

On Sat, 21 Jun 2014, Robert Jonsson wrote:

> 2014-06-21 19:18 GMT+02:00 Len Ovens <len at ovenwerks.net>:
>> As was already stated, clock speed of the interface is not really relevant.
>> It seems in fact that no one is really interested in USB3 because it does
>> not have any improvement for audio, USB2 is enough. The limitation with USB1
>> is bit depth, bit rate and channel count.
>
> You are probably right about USB3 but with USB1 and USB2 speeds, how
> can the clock speed not be relevant? As an example, wouldn't
> transferring 1024 frames of float on USB1 (12Mbit) take 2,7ms and only
> 68us on USB2 (480Mbit)?
> In raw throughput that is, there are probably other factors limiting the speed?

Measure round trip. That is the only way to find out. As an aside, if you 
have to measure it, does it matter?

In general, the manufacturers seem to have chosen to use the extra through 
put on USB2 to add extra bit depth (24bit instead of 16), extra bitrate 
(96k or more instead of 48k) and more channels. The latency would end up 
being simmilar.

Being realistic... I can set my card to ~.6ms (one way) latency at 48k... 
maybe less at 96k but I haven't tried it. However, The dsp inside the 
audio interface itself already adds ~1ms of latency. That is 1.6ms one 
way. I can double the latency in jack to 1.2ms and the total is now 2.2ms 
not that much higher(this is not measured values BTW so the jackd added 
latency may be even less of the equation). There is a point where the 
latency of the interface itself is the main part of things... and how many 
inches away or towards the speaker do I have to move my head to have the 
same effect? When is a latency decrease really a gain? It seems to me from 
my own experience that a USB1 IF at 64/2 is good enough for live work as 
an effects rack or softsynth. For tighter needs the cost for the interface 
and the computer to run it goes up. Even audio imaging with multi speakers 
using delays for spacial placement is not overly latency dependant so long 
as all the channels are in good sync.


--
Len Ovens
www.ovenwerks.net