LinuxMusicians

I've tried comparing results from jack_iodelay and LSP Latency Meter. They're very close but not identical.

Using JACK in sync mode, 48kHz, 2 periods per buffer, 64 frames per period the theoretical value for this setup is :

nominal latency = ((1 + number of periods) * number of frames per period) / sample rate

nominal latency = ((1 + 2) * 64 ) / 48000 = 4ms.

For jack_iodelay : For Latency Meter :



OK, so there's ~1.3ms extra latency. It's my hardware latency which I set earlier at 31 frames on the input and 31 frames on the output. Adding that in to the calculation :

latency = (192 + 62) / 48000 = 5.333 ... ms.

That's pretty accurate although I must say I was expecting the measured value to be higher than the calculated value. The difference between jack_iodelay and Latency Meter is 0.007ms. Now trying different buffer sizes :

With a 16 frame buffer : latency = (48 + 62) / 48000 = 2.292 ms

jack_iodelay : Latency Meter :



Again the difference is 0.007 ms.

Now with a 1024 frame buffer : latency = (3072 +62) /48000 = 65.292 ms.

jack_iodelay Latency Meter :



The difference is 0.007 ms.

The measured values are consistently lower than the calculated values. I think that the reason for this is that I originally set JACK's latency compensation using a sample rate of 96kHz. Some of the hardware latency is due to a buffer, and that will result in a latency time that goes down as the sample rate goes up. Some of the hardware latency is a fixed time that doesn't vary with sample rate. So the amount of extra latency changes with sample rate, but it's only a couple of frames.

Please remember that built-in DAC and ADC of the sound card can provide some additional latency to the signal because they require some set of clock cycles to convert analog audio to digital and vice verse.

Also please notice that 0.007 ms * 96 samples/ms = 0.672 samples < 1 sample
So results are almost the same, the difference between them is the constant error value (probably, rounding error).