|
These days, anyone getting into the music technology field quickly learns about latency in audio paths passing through a computer. We also learn that this is a problem for host-based systems, with the delay being a necessary 'breathing space' to allow our CPU a chance to run all the maths behind signal mixing and processing. What's more, latency is all about the problem of delays in the monitoring chain. Except it's not: even in 'real-time' DSP systems there's potential for delays to cause problems with feel and foldback monitoring, and in particular phase misalignment in mixing. And we haven't even mentioned DirectConnect, HTDM and MIDI latencies yet! First let me point out that latency has always been an issue in multitrack audio recording. In a studio without any computers or digital gear, you've still got acoustic latency to deal with: the time it takes for sound to propagate through the air (and other materials in fact) to the mic. With close miking this delay is small, but if you have two mics open and they both pick up the same sound, then even tiny differences in distances from the source will cause phase cancellation (a hollow, flangy effect) when the signals are mixed together. Engineers learn to account for this problem when setting up for a multi-mic recording session through careful mic placement, isolating sounds as best as possible, and even introducing artificial delays to keep all signals in phase. If, like me, your recording tends to stick to soft synths and maybe a single mic, you can still hear this problem on TV quite often a good bet is a studio discussion programme, where two boom mics may both be open at some points. Getting back on topic, in an analogue studio latency is no longer an issue once the audio reaches the mic and becomes an electrical signal. Audio travelling through analogue circuitry, such as a mixer channel strip followed by an insert or send/return, might take a few microseconds. This is too short a delay to cause phase problems within the human audible frequency range. In a digital system, however, you have to be on guard for further signal delays. Firstly, in any digital system it takes time to go through analogue-to-digital convertors and vice versa. If you're using a Pro Tools TDM system with an 888/24 it takes about 80 samples to make a round trip from analogue to digital to analogue, so there's a minimum monitoring latency of a little under 2ms. Now this is a pretty short period of time, but it could have a slight noticeable effect on what a performer hears. First, at low headphone levels the interaction between the direct 'in the room' sound and the cue signal might sound phasey or even flammy. Additionally, there's the question of timing, and whether you think the through-delay could be big enough to force a musical difference when overdubbing. If in doubt, you need to think about monitoring overdubs through an analogue mixer rather than Pro Tools. Interestingly, the delays are measured in samples, so running at higher sample rates reduces latency in actual time. But to put this all in perspective, I have to say that self-monitoring delay has never come up as a problem in any TDM sessions I've been involved in, and 2ms of delay is the same as standing two feet from the source... It's Just A Phase The real issue is in being aware of delays when routing and mixing in the TDM mix environment. Delays occur through internal buss sends, external sends and inserts, and plug-ins. Obviously, an insert or send/return via analogue connections to an external outboard processor will introduce latency of the same order as the A/D/A round-trip mentioned above. Internal sends and plug-ins are mostly just a few samples (six for a send, three for a d2 EQ). For the most part the delays are so tiny as to be insignificant, but in certain circumstances they have effects that need addressing: When you have a mixer group with the faders at different levels, take care not to option-click any of the faders (as you might do routinely to 'zero' faders). With a group, option-clicking sets all the faders to unity and loses the relative relationships. Scenario 2: processing individual channels of a multi-mic recording. A drum kit or ensemble is recorded with a number of mics, and there's a large amount of spill between the mics. If any single track has a plug-in inserted, the few samples of delay will put the recorded spill between mics out of phase with each other. Additionally, if it's an insert via an external analogue unit, then the delay might cause subtle timing problems, or flamming with the spill. The best way to handle this situation is to get into the habit of matching delays across tracks that need to remain exactly aligned. The extravagant way to do this is to put the same plug-ins on all of the tracks in question. However, if it's a DSP-hungry plug-in, or you're using an external device, then you can use the Time Adjuster plug-in on the other tracks to match everything up. Pro Tools helps by keeping track of the latency of each track, measured in samples. Command-click twice on a track's volume display to switch to the delay readout. Adjust each track's TA plug-in to match the biggest delay (see screenshot). I should note that the delay indicator only shows plug-in delays, not hardware inserts. To account for latency incurred by routing out of Pro Tools, you can use the values given in the Pro Tools manual for each particular Digidesign interface. Notice that a round trip through the interfaces via digital connections is much quicker than going through D-A and A-D converters, so using effects devices with digital I/O is a way of reducing latency (unless the effects box imposes lots of latency internally!). Scenario 3: high-delay plug-ins, and/or working to picture. Certain TDM plug-ins cause an unusually large delay. These tend to be ones that use 'look ahead', such as Maxim and DINR. This kind of plug-in is often used on the Master Fader, and as it affects the whole mix at once, won't cause timing problems in this context. However, this would not be acceptable if you were working against timecode, because your final output would be out of sync. Similarly, if you use such a plug-in on a single track then you will introduce timing errors. An example of both these problems at once would be mixing the sound for a DVD of a live performance. If you used a very high-delay process, such as Wave Mechanics' Pitch Doctor, on the lead vocal track, you'd push the vocals late by nearly 4000 samples. This is approximately 90ms, and over two frames of video: the vocal would be both audibly and visibly out of time. In this situation, you can't delay all the other tracks, not only because it would be incredibly tedious, but because it would also put everything out of time with the video. So the best option is actually to move all the audio in the lead vocal track earlier by 4000 samples. Simply select all the audio in the track, choose Shift from the Edit menu, click Earlier and enter the number of samples. So why can't Pro Tools just look at the TDM delay in each track (it already knows that, after all) and adjust the playback time in the timeline to compensate? This is exactly how the new MTS (MIDI Time Stamping) system will work on MIDI tracks in PT6. Eventually PT will probably be able to do something similar with audio, but it's actually a bit trickier than it sounds. Firstly, an audio track might have sends going to more than one place, incurring different delays along the way. Also, some sound elements (or even all) may not be audio tracks, but external sources coming in via auxes: You can't shift them earlier without introducing a global mixer buffer (and destroying the whole point of having a TDM system) or travelling backwards in time... RTAS, DirectConnect, HTDM, StreamManager Other times that latency can creep into the world of the TDM user are when using HTDM plug-ins or DirectConnect-based software, all of which use the host CPU instead of Digi's DSPs. From what I can tell, tracks using RTAS plug-ins are read ahead of time during playback, going via the hardware buffer that you set in the Hardware Settings page, and are not delayed with respect to other tracks. I guess this is why you can't mix and match TDM/RTAS plug-ins on one track, or use RTAS plug-ins on auxes in a TDM mix. HTDM plug-ins are processed by the host, then re-routed into your (TDM) mixer by the StreamManager extension, incurring latency. (All this happens without you knowing about it of course.) Stuff coming from a DirectConnect application also comes in via StreamManager. By default StreamManager has a buffer of 512 samples, so anything HTDM or DC will incur an 11ms delay. Many of the programs that use this format are soft synths, so this is getting towards intolerable when it comes to percussive instruments. If you have a really fast computer, you can replace the default StreamManager extension with a 128-sample buffer version. This reduces the delay by 75 percent, but the trade-off is that you'll be able to run fewer HTDM plug-ins at once. The download URL for the alternative extension is download.digidesign.com/ support/digi/mac/pt/Stream Manager_128v531r2.hqx. For more information about the issues raised in this column, check out John Klett's 1999 article, 'Delay In Large-Format Digital Music Consoles', which you can read on-line at www.technicalaudio.com. Windows Published in SOS February 2003 | Sunday 8th November 2009 November 2009
Click image for Contents
Photos too small? Click on photos, screenshots and diagrams in articles to open a Larger View gallery. |