The Truth about Latencies

General

Latency is unavoidable when working with digital signal processing:

  Delay equivalent distance *)
A/D converter 0.5 ms 0.17 m
D/A converter 0.5 ms 0.17 m
Buffering for digital signal processing **) typically
3.5 - 30 ms
typically
1.16 m - 9.90 m

*) For example as the velocity of sound is approx. 330 m/s, the sound needs e.g. 1 ms to go 0.33 m.

**) for example at a sampling rate of 44.1 kHz and a buffer size of 1024 samples you need at least 1024/44.1 * ms =  23.2 ms to fill the buffer.

 

So what?

... you might think. But: 10 ms is the maximal delay you (I) can handle to exactly hit the metronome beat.

And that also meets my experience on large stages: You have to make sure that no musician is farther away from the drums (and bass) than 3 meters.

Otherwise you´ll not get tight and this can muck up your whole concert. The other instruments can easily be monitored with the nearby monitor speakers.

So that´s what I think about acceptable delays *):

Delay equivalent distance Feeling to hit the metronom
4.5 ms 1.5 m Excellent
6 ms 2.0 m Good
10 ms 3.3 m More or less Ok
15 ms 5 m Difficult
> 15 ms > 5m Awful


*) I´m, talking about playing guitar so guitar players are used to delays up to 10ms because of the distance from the amp or guitar to the ear.

For singers things are quite worse:

Because singers hear their own voice directly in their head, the smallest delay result in a comb filter effect which means that some frequencies will be extinguished and their voice sounds like being on the loo. For real singers a no-go since they learned how to fill the room with their voice and in fact they feel like being in the laundry. But of course also singers are used to this effect from the rehearsal room unless they use analog mixers and in ear monitor systems or headphones.

For those who still think you cannot hear effects resulting from 2ms delays (that´s what today´s digital mixers are able to provide) some extra lessons in physics:

The formular for the resulting amplitude of 2 overlapped equal signals where one is delayed (dt=2ms):

A(f) = 2 * abs[cos(2*pi*f*dt/2)]

And this means extinctions (A=0) at 250Hz, 750Hz, 1250Hz, 1750Hz, 2250Hz,... (250 Hz * 1, 3, 5, ...)

So you better go to the acoustician if you don´t hear that. For the older devices (dt=10ms) it´s even worse:
Extinctions (A=0) at 50Hz, 150Hz, 250Hz, 350Hz, 450Hz,... (50 Hz * 1, 3, 5, ...)

So when shall we trash our good old analog devices?

I would say when delay times are lesser than 0.033ms (33µs). This leads to:
Extinctions (A=0) at 15kHz, 45kHz,... that no one will hear.

 

What to do?

There are several ways to avoid too great delays:

  • Hardware monitoring (analog signal will be put through): Most of the sound cards (even the cheaper ones) have some sort of direct analog recording out you can put through to your head phones. Of course your instrument is then "extra dry" (without any effect)
  • Using an analog mixer between your instrument and the sound card device and split the signal, one to the sound card and the other to your head phones.

 

Measure the delay - don´t trust the manual!

This is an easy way to measure the delay:

Metronome


1st Microphone

Head phones


2nd Microphone

Use a separate metronome and record the metronome beat with the 1st microphone

Send the sound to the head phones and record the output with a 2nd microphone (as near as possible).

Compare the two records: Here we got 40ms delay