Acoustic Delay Measurements in LAN

The only real time measurement is one-way from air2air, microphone in to speaker/headphones out.
I use an expensive analyzer from NTI, the XL2 with CD input. It has the reference signal electrically, the same signal goes into mic in of one sound card and runs through a soundcard only or through sending Raspi, one eth switch and the receiving Raspi. The single switch is a Netgear GS10BE.

  1. Settings: default, bufsize=96 jitterTX=2 jitterRX=2
    • latency shown in web: 13.2ms + 2.2ms jitter buffer
    • from Focusrite to Presonus = 8.2ms
    • from Presonus to Focusrite = 8.8ms
  2. Settings: bufsize=32 jitterTX=1 jitterRX=2
    • latency shown in web: 13.2ms + 2.2ms jitter buffer
    • from Focusrite to Presonus = 4.9ms
    • from Presonus to Focusrite = 7.6ms

This setup is running several days and nights without interruptions.

To find out the delay introduced by the sound card, jack has a nice tool installed by default: jack_iodelay
Start it and jack_connect the in to hw-out and vice versa, and read every second the frames and ms on the command line.
Needs a cable from the output to the input of the sound card!

Hi. How about raspi’s stacked soundcards? I read an article, tht they could have less latency due to i2c connection… Did anybody some research an this?

Hi, any news to this one? Has anybody tested a hifiberry board or maybe audio-injektor board for raspberry? Is i2c „faster“ than usb-aufio interfaces or is it just a ferrytale…
Barny

1 Like

I2C is control only, I2S is the name for audio streaming ports at CPUs and DSPs.
RaspberryPi has only Stereo I2S, no multichannel can ever be possible with shields. XLR for out is existing but out of stock, and XLR input is… everybody waiting for at least a year.
USB has always more latency and particularly high jitter on that, with I2S it’s one word clock cycle - 1/1000 ms min and max, this does never change.

Two months ago I measured PiSound HAT from Blokas, that uses the same I2S interface as Hifiberry does:

PiSound = 4,924 (236.371frames)

Not really impresive.

It depends on the ADC setup, how many samples are buffered before the linux device driver, how low those linux device driver settings can be adjusted without risk.

I have seen sound cards with 48 sample blocksize showing less latency than others with 32Sample block size. Always consider the full chain that is relevant.

1 Like

Measure using linux pc running jack_iodelay