Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[Micah McFadden]

There is a benefit to using 96k vs 48k live. The latency of the AD/DA conversions are cut in half.  Latency can become noticeable for a singer.  I have run into this problem more than once, where a singer is complaining of their voice sounding horrible and I listen to the wedges while they are singing and it sounds great.  The moment I talked into the mic, I realize it sounds awful.  I was hearing my voice in my head phasing with my voice in the wedges.  I later found that there was an extra plugin added to the vocal that day which added latency.  Removing this plugin solved the issue.

If we were running the desks at 96k, not only would the AD/DA conversions have been cut in half, the latency of the plugins would have also been cut in half.  I have measured a desk at 48k at 4.1 ms with no plugins.  Another brand of consoles at 96k, I measured at 1.8 ms.  When you begin to add plugins or other digital processing to that number, that is when it really starts to be beneficial to run at 96k.

[Stefan Maerz]

There is a benefit to using 96k vs 48k live. The latency of the AD/DA conversions are cut in half.  Latency can become noticeable for a singer.  I have run into this problem more than once, where a singer is complaining of their voice sounding horrible and I listen to the wedges while they are singing and it sounds great.  The moment I talked into the mic, I realize it sounds awful.  I was hearing my voice in my head phasing with my voice in the wedges.  I later found that there was an extra plugin added to the vocal that day which added latency.  Removing this plugin solved the issue.

If we were running the desks at 96k, not only would the AD/DA conversions have been cut in half, the latency of the plugins would have also been cut in half.  I have measured a desk at 48k at 4.1 ms with no plugins.  Another brand of consoles at 96k, I measured at 1.8 ms.  When you begin to add plugins or other digital processing to that number, that is when it really starts to be beneficial to run at 96k.

I'm curious as to how higher sampling reduces latency. I've heard a number of people say it, but never understood why that is.

I guess my question is why does adding twice the data into a system make it work faster?

[Tim McCulloch]

You're moving 2x data at 2x rate.  Everything internally is processed at the sample level.

[Stefan Maerz]

You're moving 2x data at 2x rate.  Everything internally is processed at the sample level.

2x the data at 2x the rate means the latency should be equal, no?

[John Roberts {JR}]

I'm curious as to how higher sampling reduces latency. I've heard a number of people say it, but never understood why that is.

I guess my question is why does adding twice the data into a system make it work faster?

It depends on what the mechanism is that causes the latency, The higher sample rate will mean a shorter delay interval between each sample. If the latency is caused by some batch operation on a fixed number of samples it seems logical that shorter per sample time would scale the latency down linearly. I suspect there are fixed and variable terms in the total latency time, so some latency components that scale with sample rate and some that don't.   

JR

PS: I am not the digital expert here but i did stay at a holiday inn.

[Samuel Rees]

Good point, latency is clearly proven benefit. Consoles are getting their 48k latencies down pretty low these days though, sub millisecond, will this matter less and less you guys think? Or will IEM engineers keep seeking lower latencies into the microsecond range?

[Jason Lucas]

There is a benefit to using 96k vs 48k live. The latency of the AD/DA conversions are cut in half.  Latency can become noticeable for a singer.  I have run into this problem more than once, where a singer is complaining of their voice sounding horrible and I listen to the wedges while they are singing and it sounds great.  The moment I talked into the mic, I realize it sounds awful.  I was hearing my voice in my head phasing with my voice in the wedges.  I later found that there was an extra plugin added to the vocal that day which added latency.  Removing this plugin solved the issue.

If we were running the desks at 96k, not only would the AD/DA conversions have been cut in half, the latency of the plugins would have also been cut in half.  I have measured a desk at 48k at 4.1 ms with no plugins.  Another brand of consoles at 96k, I measured at 1.8 ms.  When you begin to add plugins or other digital processing to that number, that is when it really starts to be beneficial to run at 96k.

Brand makes a difference, I know of at least one console that operates at 48k and comes in at 2.8ms delay. And Behringer claims a delay of only 0.8ms @ 48k.

While the Midas Pro 2 is capable of latency below 2ms, it can also go as high as 8ms, even at 96k.

[Christian Tepfer]

There is a benefit to using 96k vs 48k live. The latency of the AD/DA conversions are cut in half.
...

Yes, but what really counts is the latency analog in -> analog out, conversion latency is just a small part of the overall latency.

I don't need 96k, I need low latency. I don't care how it's done. Why bother with an insignificant specification?

[Merlijn van Veen]

IEM set aside, anything less than 3 ms is the equivalent of roughly moving your wedges 1 meter away in an analog setup. Surely this isn't insurmountable. I'm aware of one article stating that George Michael's mic signal is split analog and fed through a separate Midas condole for IEM purposes all other mixes for everybody else are done digital. AFAIK the digital Midas consoles allow for plugin delay compensation, which in extreme cases might account for high latencies similar like Waves multirack. For this very reason, I stick to certain plugins with documented low latencies. Waves i.e. publish these values (expressed in samples) in their technical data section.

[John Roberts {JR}]

Good point, latency is clearly proven benefit. Consoles are getting their 48k latencies down pretty low these days though, sub millisecond, will this matter less and less you guys think? Or will IEM engineers keep seeking lower latencies into the microsecond range?

The primary concern about latency in digital platforms IMO is dominantly associated with  IEM applications where the internal path from a singers voice conducts sound energy into the ear that can constructively or destructively interfere with the IEM mix sound.

I suspect there is a threshold latency delay amount where the latency time maps out to the wavelength of the lowest frequency interference. This set of frequencies that can interfere then are factored against the frequency response of the internal conduction path.

From simple (one rat) observation the frequency response of my internal path, or what I hear when I block my ears and make sound with my mouth is mostly the LF energy.

If the latency is short enough that the lowest frequency capable of interference, is already significantly attenuated in the LPF internal conduction path, the IEM feed should sound natural. 

A 0.8 mSec latency maps out to a roughly 625 Hz lowest frequency for interference, how much 600 Hz do we hear when we talk with our ears blocked? Some, but it is well down from the lower frequency energy we hear pretty loudly. To interfere deeply the two signals need to be roughly equal loudness so I suspect the threshold for IEM latency issues could be multiple mSecs.

I asked the early tire kickers about how well the Behringer X-32 performed on IEM mixes, and the testers reported that it passed, so apparently it's latency is low enough. I would advise caution about using plug-ins or efx with an IEM mix that could increase  latency into the audibility range.

I am not aware of research into the frequency response of this inside the head sound conduction path, but expect my personal observation about the frequency response to be reasonably generic.   


JR

 
PS: Coincidentally several decades ago I manufactured a device that intentionally added delay to a headphone feed for use in speech therapy to force stuttering. That unit operated over a 50-300 mSec range to incite stuttering but this is a somewhat different meat computer mechanism from combing that is more simple acoustics. 

[ProSoundWeb Community]