[LAU] Audio over WIFI

[Len Ovens]

On Tue, 20 Jan 2015, Leonardo Gabrielli wrote:

> Thanks Len for the useful feedback.
>  
>       I was going to say 10ms is about where I start to feel disconnected
>       from
>       what I am playing, but I realize I am thinking one way and round trip
>       would be about 20ms. So 16ms might be well workable.
> 
> 
>  Psychoacoustic tests (many of them from the CCRMA SoundWire team and others)
> shows that 21-25ms is acceptable to keep a steady tempo. Larger values make the
> performers slow down, very low values make them accelerate! AFAIK in musical

Hmm, I think though there is a difference between "keep a steady tempo" 
and comfortable. Someone (those with money for whatever they want) who has 
to play every night will choose less clean if it is more comfortable to 
play with. It is hard to market something the "pros" don't use. What that 
means to me is that it has to be not perceptible. I have to not notice it. 
It is one thing to be able to play with it, but I think my playing (or 
artistry) hurts when I have a feeling of something "not right".

I also think there is a difference depending on which instrument is played 
and what effects are used. A guitar through an amp suffers only the delay 
the sound takes to get from the amp to the ears pretty much. A pipe organ 
will have a much longer delay from key down till sound hits the palyers 
ears, both from mechanically opening the pipe but also the pipe will be 
farther away (16 foot pipes just seem to take a while to get going too). 
The synth player will also be used to some delay and may even add it for 
some swelling kinds of sounds. But the synth player is probably closer to 
his monitor too. So different instrument players will expect a different 
amount of delay.

> instrument design the 10ms figure is taken as a threshold for your instrument
> response (e.g. keypress --> sound), but in performance the values can be higher
> as said.

That would be another problem. designing for higher latency means that the 
player who adds digital effects (or the synth player who has play delay at 
10ms already) will be looking at that delay added to the delay of the 
monitor delay. As a player, my feeling is that "if this new digital system 
makes it harder for me to play, I will stick with my analog one thank 
you." So the bar to meet is not what is good enough, but rather what will 
not be noticed. The various wired digital transports are all under 5ms 
latency, 3ms is common for one direction, but "as low as 1ms" is what most 
claim. With a digital mixer, the delays add up. mic in to pcm is .5ms, to 
FOH is 3ms. Mixer effects will add time because they use 3 or 4 DSP boards 
for eq/comp/whatever in each mix strip as well as whatever the master 
throws at it. Every time the audio goes to another DSP, there is a buffer 
worth of time lag. I would assume each strip stays within the same DSP, 
but will move to another one for master or monitor. SO there may be delay 
there. then it has 3ms back to the speakers/inear along with DAC delay.

A keyboard player may already be looking at noticable delay at that point, 
or at least on the edge of it with the delay in their synth thrown in.

In such a case, it would not be possible to add a WIFI that went to AP 
then snake then FOH mix then monitor as there would be close to 30ms. The 
WIFI would have to be a dirrect input to the FOH mix.

In the case with the three players all on different boats, it may have 
actually been easier because of the separation of things. There would be 
little audio from an acoustic path... plus there was less choice to make 
it work :)

SOme digital FOH mixers do not actually have the PCM travel all the way to 
the mixer and back. The "mixer" is really a controller and the mixer is in 
the box that looks like the stage end of the snake. So the only delay is 
the network to monitor box delay. This kind of system may be more 
workable... it is just out of my price range so I don't think of it too 
much :)

So my feeling is that WIFI is not worth using in the studio anyway where 
cables are easy and sure. To work on stage though (where they would be of 
good use) they need to have lower latency. I think shooting for minimal 
AES67 compliance latency would be something that is not noticable.

Minimum AES67 is: 
48k/24bit 48words per packet with 3 packets worth of buffer. (I can't find 
my paper just now so I can't check, but axia's page says that their 
compatibility mode uses 48 sample packets which they call 1ms, but the 
AES67 paper suggests buffering 3 packets at a time for stable operation) 
This is 3ms transport delay. If WIFI can hit that, things will work. I 
know all the blurbs say that AES67 is less than 10ms, but the reality is 
that it is 3ms at the first compatibilty mode. This is what gives the 
widest range interconnectivity as there are already some products out 
there where this is the only mode available.

It would be interesting to see what happened to two AES67 devices 
connected by a fast WIFI link with various amount of radio noise and audio 
channels. I would imagine the maximum number of channels in any WIFI link 
would be 4 (2 in each direction) though 1 or 2 might be more common.

It would, of course, be possible to use larger buffers in the 
WIFI link itself and smaller buffers in the AP to wired end... in other 
words use something other than AES67 and then convert. However, I think it 
would add more latency where less is needed.

A system approach is needed where the whole system where WIFI will be 
used, is analysed as a whole and the max latency is based on the system 
rather than just the transport itself.


--
Len Ovens
www.ovenwerks.net