Feedback from RT1011 - channels are swapped at every session

[sathyap-chrome]

On CML platform, we have RT1011 smart amp providing feedback. we have 4 CH feedback ( from 4 individual speakers) enabled with attached Topology ( m4 is attached too)

Test procedure:
Play a known audio sample ( preferably sine tone) on Speakers
Record 4 ch data on Feedback PCM ( hw:0,10) for few seconds and make multiple recordings.
Each file shows feedback to be different - ideally they should be same for same input file.

TEST ENV:
CML device with RT1011 Amp.
SOF v0.8 backport release FW and topology changes attached on top.

008_aec.zip

[sathyap-chrome]

@keqiaozhang Please reproduce one time and then we can see how to proceed debug. Codec vendor informs that the codec side, there is no channel swap.

cc @keyonjie @aiChaoSONG

[keyonjie]

@sathyap-chrome can you help clarify:

1. you are playing stereo(2 channels) sine wav output to speakers,

2. you are capturing 4 channels(actually only channel 0/1 has data input) which expected the channel 0/1 getting the exact output of the playback channel 0/1 and 0s at channel 2/3.

3. is it possible to probe the output/input on the SSP pins on your side?

Hi @aiChaoSONG @keqiaozhang let's try if we can reproduce it on our side.

[aiChaoSONG]

@keyonjie could you please have a look at thesofproject/sof#2430, and see if this merged PR has any influence on the abnormal behavior mentioned in this issue?

[sathyap-chrome]

@keyonjie

you are playing stereo(2 channels) sine wav output to speakers

Yes

you are capturing 4 channels(actually only channel 0/1 has data input) which expected the channel 0/1 getting the exact output of the playback channel 0/1 and 0s at channel 2/3

. Its a 4 CH capture since there are 4 individual codecs configured ( 1 in each TDM slot )
Hence feedback will have data on all 4 channels.
The 4 individual codecs play 2 CH playback but provide feedback individually.
Refer TDM slot mapping here:

linux/sound/soc/intel/boards/cml_rt1011_rt5682.c

Line 185 in 30887e1

/*

3. is it possible to probe the output/input on the SSP pins on your side?
   I can try. But i wanted to know if we can get a dump or some s/w means to easily debug without opening up the closed chassis machine.
   Also easy for codec vendors to duplicate any debug points.

@aiChaoSONG I guess not. This issue was seen with SOFv0.7 as well as SOF v0.8.

[keyonjie]

@keyonjie could you please have a look at thesofproject/sof#2430, and see if this merged PR has any influence on the abnormal behavior mentioned in this issue?

the PR is to divide playback/capture channel numbers, I believe it won't fix channel swapping issue.

[keyonjie]

@sathyap-chrome can you help upload the playback and captured .wav files?

[sathyap-chrome]

This has 2 captured files. playback for this case was just normal music playback.
008_aec.zip

If you need sinewave for testing, here it is -

left1KHz_right777Hz.zip

I'll send out the captures from this file in a few minutes

[aiChaoSONG]

On sof-dev + master, seems reocrd nothing. @sathyap-chrome @keyonjie

[sathyap-chrome]

@aiChaoSONG did you make the topology modification ?
I've shared the tplg and also m4 for reference . You need to test only on the Helios device.

Can you try the SOF 0.8 backport on chrome first ?

[aiChaoSONG]

@keyonjie @sathyap-chrome result on 008 backport(Chrome OS) is below.

@sathyap-chrome @keyonjie Now, I understand what the issue is. the wave seems on different channel slot at each recording.

[sathyap-chrome]

@aiChaoSONG please do multiple recordings (during same playback) and check the snaphot.. above is only 1 recording, we need to compare 2 recording.

[keyonjie]

@keyonjie @sathyap-chrome result on 008 backport(Chrome OS) is below. some phase shift?

record4ch.zip

@aiChaoSONG did you play the left1KHz_right777Hz file Sathya shared?

[aiChaoSONG]

@keyonjie The problem is here: we have two channel with sound, and there are 4 recording channels, the two with sound distribute randomly on the 4 recording channels. sometimes on ch1&ch2, sometimes on ch1&ch4. you can see the picture I attached above. file used for playback is the one Sathya shared.

[sathyap-chrome]

@aiChaoSONG this is correct 👍 #1838 (comment)

The feedback should actually get data from SSP Rx and its mapped to codec.
there are 4 individual channels on feedback. its not a playback loopback. if its behaving the way you mentioned, its actually incorrect.

[keyonjie]

@sathyap-chrome @aiChaoSONG let's try to verify if the issue exist with hardware loopback mode(codec playback output jack connected to capture input jack).