Listen To/Download The Audio Version Of This Article:
Ever since the first appearance of digital audio technology in the field, latency has been a topic of discussion. How much is noticeable? How much is acceptable for a given application? Is less always better? As with any quantifiable property of audio equipment, latency is described using a plethora of nonspecific marketing claims, such as low, ultra-low, imperceptible, near-zero, and so on. In order to demystify some of these claims, let’s get back to the numbers and understand them in context.
Latency is the time it takes a device to take in, process, and output an audio signal, or in other words, the time it takes for signal to pass through a device or a system. If that device has analog inputs and outputs, this will include analog-to-digital conversion at the input, time spent performing operations on that digital signal, then digital to analog conversion at the output.
Digital I/O complicates things slightly; while it skips the conversion step, there will be latency associated with whatever digital transmission protocol is in use. The latency of a device is typically measured in milliseconds, a unit of time equal to 1/1000 seconds and abbreviated as “ms” or “msec.” (In some contexts such as audio processing plugins, it is also common for latency to be stated in samples.) For the math-averse, rest assured that we can use simple addition to find the total latency of a series of devices in a signal path.
How much latency can we accept in a particular scenario? To understand this, we should discuss latency from acoustic propagation delay – the time it takes a sound wave to travel through the air. Sound waves travel at about 1125 feet per second, or 1.1 ft per ms, though this number varies slightly with temperature and humidity.
That means in a typical midsize venue, it takes 15-20 ms or more for direct sound from drums and amps to cross the downstage edge and enter the audience. As a result, we can accept quite a bit of latency from the signal chain for the front of house mix, and from the PA itself.
For musicians on stage, who are seldom far from their instruments, acceptable latency is much lower. And for vocalists, whose voice arrives almost immediately to their ears via bone conduction, latency is even more critical – monitoring creates a comb filter when summed with bone conduction.
Wedges are more forgiving than in ear monitors because the path length to the listener is not fixed; as a performer moves around, so do the peaks and nulls of this comb filter. With in-ear monitors, the comb is fixed, and the lowest possible latency is the only tool we have to reduce its effects. It’s generally accepted that monitoring latency, measured from the input transducer, through the console and processing, then back out through a wedge or in ear monitor, should not exceed 5 ms.
Digital wireless systems, even well-regarded ones, can introduce a surprising amount of latency. A Shure SLX-D wireless microphone adds 3.2 ms of latency, meaning that the rest of the monitoring chain has to be very quick to stay under 5 ms. Spending more money won’t necessarily buy you more time, either; Shure’s top-of-the-line Axient Digital systems still add 2 ms of latency. As a result, analog wireless is often preferred by guitarists as well as for IEM systems.
Let’s start with the simplest possible scenario: an analog signal traveling through a copper wire. Electromagnetic waves propagate through typical audio cables around two-thirds the speed of light; for our purposes, an analog signal traveling through a cable offers “zero latency.”
As we move on to digital signals, things become more complicated. Their latency depends on the protocol in use, because each has a different way of “packaging” the data to be sent and includes a different number of samples in each transmission.
For example, AES50 has a latency of 0.063 ms, which is on the quicker side. Digital audio over IP is inherently slower because it must play by the rules of computer networking; still, default latency for Dante is 1 ms, and it can be set lower in many situations.
Receive, Process & Transmit
We’ve established that analog and digital signal transmission are both blindingly quick, but they’re also quite boring without things to connect them to. Let’s get into devices that receive, process, and transmit audio.
Analog consoles, outboard gear, amplifiers, and even wireless transmission systems do their jobs in an instant, but the same is not true about their digital posterity. For example, the ubiquitous Behringer X32 digital console has a throughput latency of 0.83 ms, measured from analog input to analog output using a 48 kHz sample rate. For Allen and Heath SQ series and Avantis consoles, this figure is under 0.7 ms.
Further, with the added step of transmission to and from a digital stage rack, most modern mixing platforms stay under 2 ms. This means that if you’re mixing without outboard processing or external plugins, your latency should be acceptably low for any application, even on budget-friendly consoles.
What if you’re using more than just a console and I/O rack? Analog outboard gear is still common in digital workflows, requiring another trip through D/A and A/D conversion if run as an insert. Waves can result in varied latency depending on implementation and plugin selection. Consoles with SoundGrid I/O can send and receive signal from a Waves Server in about 1 ms, but many of the available plugins can double that, or worse.
On platforms without a SoundGrid I/O, several conversions may be necessary to get in and out of Waves, so the process will take longer. Many FOH engineers carry a system processor; an LM44 adds about 1 ms. In addition, modern loudspeaker systems have manufacturer-provided processing and amplification presets, which can add several ms to the signal chain.
To draw a useful conclusion: modern audio devices from reputable manufacturers have acceptably low latency, and in simple use cases will not cause any trouble. At front of house, especially in larger venues, complex signal chains with lots of latency can be implemented with little consequence. In monitoring applications, however, this practice quickly becomes prohibitive.
If latency is a concern, reducing the number of devices in your signal chain and selecting the right tool for the job is a more effective strategy than upgrading to the latest and greatest gear.