Repository navigation
[BUG] [MTL] AEC functionality depends on Echo reference endpoint to be started #8075
Description
With the AEC topology on main, the following endpoints get listed.
i see the following o/p:
rex-rev1 ~ # arecord -l
**** List of CAPTURE Hardware Devices ****
card 0: sofrt5682 [sof-rt5682], device 0: Headset () []
Subdevices: 1/1
Subdevice #0: subdevice #0
card 0: sofrt5682 [sof-rt5682], device 2: Bluetooth () []
Subdevices: 1/1
Subdevice #0: subdevice #0
card 0: sofrt5682 [sof-rt5682], device 27: DMIC0 RTC AEC () []
Subdevices: 1/1
Subdevice #0: subdevice #0
card 0: sofrt5682 [sof-rt5682], device 28: DMIC0 RTC AEC NS () []
Subdevices: 1/1
Subdevice #0: subdevice #0
card 0: sofrt5682 [sof-rt5682], device 29: EchoRef () []
Subdevices: 1/1
Subdevice #0: subdevice #0
card 0: sofrt5682 [sof-rt5682], device 99: DMIC Raw () []
Subdevices: 1/1
Subdevice #0: subdevice #0
rex-rev1 ~ #
This sequence doesnt work:
1)start speaker playback
2)start dmic capture: arecord -Dhw:0,27 /tmp/tmp.wav -c4 -r48000 -fs32_le -vv
there is a strong dependency on echo reference endpoint to be started.
While this works:
1)start speaker playback
2)start echo ref capture
3)start dmic capture: arecord -Dhw:0,27 /tmp/tmp.wav -c4 -r48000 -fs32_le -vv.
There is no need for the echo reference endpoint to be opened from the user space, while speaker playback the data can be fed to AEC without the FE.Kind of hostless stream.
Activity
lgirdwood commentedon Aug 23, 2023
@RDharageswari the intention is that connected PCM endpoints are meant to be started in any order with any timing, since its all driven by userspace. There is a topology flag that's meant to tell any shared modules to ignore internal XRUN errors in such cases and I'm wondering if we dont have this set (as this was a common issue when enabled with IPC3 + M4).
kv2019i commentedon Aug 23, 2023
@RDharageswari I need to investigate the topology definitions, but it does sound this is working as expceted. The long history and rationale is documented here:
#2564
But, but, if the AEC is on the DSP, then we actually should not have the user-space echo ref PCM in topology (or it can be there, but then it has to be used).
kv2019i commentedon Aug 23, 2023
@yongzhi1 It seems the speaker ref part of the topology is from you? Is the speaker ref to user-space here intentionally?
ranj063 commentedon Aug 24, 2023
@kv2019i I know the root cause for this issue. I am in the provess of removing the host PCM 27 for echo ref adn this should take of this issue as well.
yongzhi1 commentedon Aug 25, 2023
Yes, the original code is to enable feedback path from speaker, same idea for SDW_AMP_FEEDBACK.
ranj063 commentedon Aug 25, 2023
@yongzhi1 @RDharageswari PR for the modified topology is here: #8101
This change also requires a kernel change in thesofproject/linux#4550
RanderWang commentedon Sep 22, 2023
Please include these patches for this feature
Rimage side: thesofproject/rimage#180
Topology: #8101, #8246, #8204
Kernel: thesofproject/linux#4593
mengdonglin commentedon Sep 25, 2023
@RDharageswari @yongzhi1 @ranj063 FYI
kv2019i commentedon Oct 10, 2023
@RDharageswari @RanderWang @ranj063 can we close this as all the patches referred here are now merged?
RanderWang commentedon Oct 12, 2023
Let me close it