HAL1x Latency Details

Ah, the classic “Latency” question. How much time does it take audio signals to travel from a given input to a given output? Keep in mind that, generally, latency only really matters when a reinforced acoustic source also exits a nearby loudspeaker. This got quite long. But it’s complete…

 

HAL4 latency is 2.2 milliseconds. See the HAL System Data All Specifications, page 50.
HAL3s is 2.39 milliseconds – analog input to analog output, see page 49.
HAL2 is 2.39 milliseconds – analog input to analog output, see page 46.

 

The rest of this discussion is about Latency in the HAL1x and its various Expansion (EXP) and RAD devices:

 

In the HAL System Data, page 6 there is a Latency drawing (below) that shows the different latencies depending on the physical path the audio takes, input to output (almost always through the DSP Core in the HAL1x).
HAL1x, EXP & RAD Transport and Conversion Latency

HAL1x, EXP & RAD Transport, and Conversion Latency

Example #1:  HAL1x analog input (through the DSP Core) out to the HAL1x analog output is one path:
   833.5 µs @ HAL1x input A/D
+ 354.8 µs @ DSP Core
+ 624.6 µs @ HAL1x output D/A
= 1.8129 milliseconds (ms)

 

Example #2: RAD input directly to HAL1x RAD port, out HAL1x analog output is a different latency.
   445.2 µs @ RAD A/D
+ 354.8 µs @ DSP core
+ 499.2 µs @ RAD D/A
= 1.2992 ms

 

Example #3: (Still using the same Latency drawing) – EXP5x analog input, through an EXP3x and through an EXP1x to the HAL1x’s DSP. Then back to the EXP3x analog output through the EXP1x is a third latency.
   758 µs @ EXP5x A/D
+ 0.73 µs @ EXP5x Exp. Bus
+ 0.73 µs @ EXP3x Exp Bus
+ 0.73 µs @ EXP1x Exp Bus
+ 62.4 µs @ HAL1x Exp Bus
+ 354.8 µs @ HAL1x DSP Core
+ 62.4 µs @ HAL1x Exp Bus
+ 0.73 µs @ EXP1 Exp Bus
+ 0.73 µs @ EXP3x Exp Bus
+ 793 µs @ EXP3x A/D
= 2.034 ms

 

LATENCY CONTINUED
However, if you draw two audio paths in Halogen – each from HAL1x analog Input to the HAL1x analog output – and each path with the same DSP blocks, these are the same latency (of course). Similarly (but not as obvious), if you draw two paths thru the same two HAL1x analog I/O paths but place 25 DSP blocks in the first path, and zero DSP blocks – just a wire – in second path, these paths will still be the same latency. They are both the same latency as the previous 2-path example, AND the latency is adjusted by Halogen to be the same for each path.

 

MORE ON LATENCY
Even if you were to move the path that has 25 DSP Blocks to a RAD that is several EXP devices away down the HAL1x Expansion Bus, once you subtract out the Latency from the drawing showing conversion & transport latency, the DSP path latencies will still be the same. Meaning, if two signals go through the DSP core, the DSP core does indeed perform latency compensation, so paths match. Halogen does not, however, adjust for the various physical transport and conversion paths – only within the DSP core do we adjust latencies.

 

EVEN MORE ON LATENCY
When the audio path exits the EXP3x output’s DSP (or enters the EXP5x input’s DSP), we do not compensate any EXP’s DSP with either the DSP core inside the HAL1x, nor across EXP DSPs. The Latency drawing includes these latencies.

 

THE BOTTOM LINE REALITY OF LATENCY IN THE REAL WORLD
In general, we’ve found that customers almost always use an audio analyzer that quickly & automatically calculates the “time of flight” from any given system input to any system output (e.g., loudspeaker). These analyzers are very accurate and help system designers know and measure the complete latency through everything: wireless mics, digital consoles, networks, FX devices, computers, DSPs like the HAL System, amps, loudspeakers, time of flight through the air. Those concerned with latency must take this “measure it all” approach to achieve galactic latency happiness and harmony.

 

After measuring latencies (rather than the inconvenient calculation above), use the Delay blocks inside Halogen to adjust the measured system latency to compensate for the complete (not just HAL) paths of concern. At this point, sometimes a minor issue becomes the system latency measurement is often much more resolute or accurate than the minimum 21-microsecond resolution steps available in the Halogen delays blocks. This is where “close enough” alignment times are often quite acceptably accurate.

Leave a Reply