[Digidesign Mbox 3] Does ucm SplitPCM add latency?

[ManuLinares]

#237

I've correctly configured ucm profile, but I notice a small latency added when using UCM vs not using it (having the default alsa analog-surround4.0 profile)

Is this the expected behavior of SplitPCM in UCM?

[perexg]

Do you use PA ? Do you have installed alsa-lib version 1.2.8 ?

[ManuLinares]

pipewire 1:0.3.59-5 (pipewire-jack, pipewire-pulse)
alsa-lib 1.2.8-1

I noticed when using carla to connect my mic to output, using the same latency config, when using UCM a little more lag.

I'm going to measure the latency now.

[ManuLinares]

Tested latency.
"speakers -> microphone", measured with "LSP Latency Meter" in carla (Buffer size 64, Sample rate 48000), Latency in ms.

Using UCM:

38.604
38.416
38.437
38.395

Without UCM:

14.312
14.395
13.916
14.375
13.979

[perexg]

The latency is configurable. For USB profiles, the 10ms period size is the default. You may lower this value in /usr/share/alsa/ucm2/USB-Audio/USB-Audio.conf.

[ManuLinares]

The latency is configurable. For USB profiles, the 10ms period size is the default. You may lower this value in /usr/share/alsa/ucm2/USB-Audio/USB-Audio.conf.

Indeed, lowering that variable to 1 lowered the latency to 12ms

Put this below line 233 of USB-Audio.conf
True.Define.SplitPCMPeriodTime 1 # 0.001ms

Gonna test it, and submit a patch if I have no issues.
Thanks.

[perexg]

The minimal sample chunk size for the USB audio devices is 1ms (1000) and I think that PW expects the total buffer to be more large, so values 4000-6000 may give almost similar results with the lower CPU overhead. Could you test?

[ManuLinares]

SplitPCMPeriodTime	Latency
4000	28ms
1000	18ms
1	12ms

And actually setting SplitPCMPeriodTime to 1 gives me the lowest cpu usage. 🤷 ¯_(ツ)_/¯

[perexg]

Could you show output from cat /proc/asound/card0/sub0/pcm0*/hw_params files when the PW is running with value 1? Replace card0 with your USB card...

[ManuLinares]

I think the "pcm0*" comes first, sub0 comes after

With 10000

$ cat /proc/asound/card3/pcm0*/sub0/hw_params

cat /proc/asound/card3/pcm0*/sub0/hw_params
access: MMAP_INTERLEAVED
format: S24_3LE
subformat: STD
channels: 4
rate: 48000 (48000/1)
period_size: 480
buffer_size: 24000
access: MMAP_INTERLEAVED
format: S24_3LE
subformat: STD
channels: 4
rate: 48000 (48000/1)
period_size: 480
buffer_size: 24000

with 1

$ cat /proc/asound/card3/pcm0*/sub0/hw_params
access: MMAP_INTERLEAVED
format: S24_3LE
subformat: STD
channels: 4
rate: 48000 (48000/1)
period_size: 24
buffer_size: 24000
access: MMAP_INTERLEAVED
format: S24_3LE
subformat: STD
channels: 4
rate: 48000 (48000/1)
period_size: 24
buffer_size: 24000

[ManuLinares]

updated my comment

[perexg]

So the lowest possible value is 500 (0.5ms) -- 1000000/(48000/24). But this is too low value for the normal operation.

If you set the direct profile (or don't use UCM) which values do you see here?

[ManuLinares]

Without UCM, the values are as follow:

$ cat /proc/asound/card3/pcm0*/sub0/hw_params
access: MMAP_INTERLEAVED
format: S24_3LE
subformat: STD
channels: 4
rate: 48000 (48000/1)
period_size: 32
buffer_size: 32768
access: MMAP_INTERLEAVED
format: S24_3LE
subformat: STD
channels: 4
rate: 48000 (48000/1)
period_size: 32
buffer_size: 32768

[perexg]

So it's setup for 1000000/(48000/32) -- config value approx. 667 . It's really close to 500.

[perexg]

BTW: Do you use the default pipewire configuration or do you modified the latency (quantum) in PW ? It seems that the configuration is not default where the period size is usually around 1024 samples (21.3ms at 48000Hz rate).

[ManuLinares]

I was testing always live with PIPEWIRE_LATENCY="64/48000" carla

So i have (Mic1_in --> "LSP Latency Meter" --> speaker_out) in carla, my physical microphone at 5cm of my speaker.

Carla shows (Buffer size 64, Sample rate 48000)

PD: I really thank you for your help, not very much docs for alsaucm.