Hello,
I'm working on IMX8M Mini EVK with Android Pie (p9.0.0_2.3.0) OS.
I have integrated CX9000 synaptics audio amplifier and CX20921 Dual MIC voice processor to the EVK.
There are two DMICs connected on the CX20921, one for capturing left channel data and other for capturing the right channel data. A 10W speaker is connected on the CX9000 speaker amplifier.
The connection between NXP processor, CX9000 and CX20921 are as shown in below diagram.
The CX9000 and CX20291 are detected successfully on the EVK. I am also able to play and record audio on the device using tinycap and tinyplay commands.
Then I created a custom WebRTC based Android application to implement VoIP functionality. Using this app I can connect to other Android based device running the same application. When I tested the VoIP functionality, I hear a repetitive noise on the other end i.e. the other Android device.
I have tested the application with several Android devices like Samsung tablets, Google tablets, mobile phones, etc and the same issue is observed.
When I use the application between two mobile devices or tablets or a combination of those, the noise is not produced. Only when IMX8 EVK is connected to other Android device, the noise is produced.
I also tried using other applications from PlayStore like Linphone and Jitsi-Meet to test VoIP functionality, but the same noise is heard on them too.
For reference I have attached a recording file containing the noise that is heard on the other end.
Audio capture and playback using the default recorder and music applications in Android works fine. No noise is observed while using them. Similarly, no noise is observed during playback/capture through tinycap and tinyplay.
I tested the same use-case on a Qualcomm EVM with Snapdragon 625 processor, but in that device the noise is not observed during VoIP. In this case, the HW connections, Codec drivers and the Android applications used were same as that were used in IMX8MM.
From the above observations, it seems the issue is introduced in the Android framework.
How can I go about to remove the noise during VoIP call ?
Hello namanthaker,
From the table#3, the playback clock is provided by Host, the capture clock is provided by CX20921.
From the below image, the playback clock is provided by CX9000, the capture clock is hard to say provided by whom.
I don't know which one is the correct one.
From my experience, to make AEC work, capture and playback clocks need to be synced. But in either case, it can not be ensured that the clocks are synced. The best way to be sure is to measure the clock signal while running VoIP...
Best Regards,
Suresh
Hello,
Thanks for your inputs !
In our system the AEC is controlled by CX20921. We in contact with Synaptics for disabling the AEC and checking the use-case to see the impact it has and see if tuning can be performed on it.
1. NXP Host side issue probability reason :
Host Processor | Audio Play Mechanism | Audio Record Mechanism | Test Case: Both Audio Out (play) & In (Record) use case |
NXP i.MX8M Mini EVM | Synaptic CX9000 | Synaptic CX20921 | Periodic Noise Issue is present |
NXP i.MX8M Mini EVM | Synaptic CX9000 | NXP PDM Direct Input | Periodic Noise Issue is NOT present |
NXP i.MX8M Mini EVM | NXP Default Headphone WM5424 | NXP PDM Direct Input | Periodic Noise Issue is NOT present |
NXP i.MX8M Mini EVM | Bluetooth Speaker | Synaptic CX20921 | Periodic Noise Issue is NOT present |
QCOM 620 EVM | Synaptic CX9000 | Synaptic CX20921 | Noise is not present |
Table #1
2. Tinycap is working, Where-as VoIP has issue ruled out :
[Test recording is attached here]
Test setup : NXP i.MX8M Mini EVM + Synaptic CX9000 (Speaker) + Synaptic CX20921(Mic)
Test timeline | Actions other than tinycap recording | Add on action | Test Results |
00-05 Seconds | No tinyplay | Quiet environment | Periodic noise not observed |
06-10 Seconds | tinyplay for 5 sec | Speak 1, 2, 3 (for initial 3 seconds), last 2 secs quite | Periodic noise observed |
11-15 Seconds | No tinyplay | Speak 1, 2, 3 (for initial 3 seconds), last 2 secs quite | Periodic noise not observed |
16-20 Seconds | tinyplay for 5 sec | Speak 1, 2, 3 (for initial 3 seconds), last 2 secs quite | Periodic noise observed |
Table #2
3. B CLK & WS_CLK summary for setup:
Codec | Use-Case | Signal | PAD | Qualcomm Setup | NXP Setup | Clock Source |
CX9000 | Playback | BCLK | I2S_RX_CLK | 3.075 MHz | 3.075 MHz | Host |
WS_CLK | I2S_RX_LRCK | 48 KHz | 48 KHz | Host | ||
VoIP | BCLK | I2S_RX_CLK | 3.075 MHz | 3.075 MHz | Host | |
WS_CLK | I2S_RX_LRCK | 48 KHz | 48 KHz | Host | ||
CX20921 | Recording | BCLK | I2S_TX_CLK | 3.075 MHz | 3.075 MHz | Self |
WS_CLK | I2S_TX_LRCK | 48 KHz | 48 KHz | Self | ||
VoIP | BCLK | I2S_TX_CLK | 3.075 MHz | 3.075 MHz | Self | |
WS_CLK | I2S_TX_LRCK | 48 KHz | 48 KHz | Self |
Table#3
Summary :
1. As per above Table#1, ROW #1, 2, 3, 4 && Tabel#2, Row#1, 2, 3, 4
It has been evident that noise is present when Synaptic chip AEC feature has been used.
2. However when we compare Tabel#1: ROW#1 & 5, NXP host side issue probability is evident
Please note that Qualcomm based board setup and NXP host based setup has used same Synaptic unit (HW, Firmware and all setting including test environment used, refer below Table#3 for details)
Infect recording using tinycap as well as VoIP call has very good Quality with long range.
3. Due to lockdown we can’t check these clock, whether they are very perfect with more calibrated CROs without going to office labs
To confirm whether the QCOM Vs NXP host provided clocks are perfectly same (i.e. glitches or noise free or not).
Hence does your team can share experience, whether NXP i.MX8M Mini chipset checked/ used with any other ODM with any external Audio CODEC based on I2S data and I2C programming Like Synaptic is being used in our case (Connection being show below, where Voice processor is CX20921).
Indeed, Once we hear inputs from Synaptic, we will let you know the result with AEC disabled.
Hello namanthaker,
From the table#3, the playback clock is provided by Host, the capture clock is provided by CX20921.
From the below image, the playback clock is provided by CX9000, the capture clock is hard to say provided by whom.
I don't know which one is the correct one.
From my experience, to make AEC work, capture and playback clocks need to be synced. But in either case, it can not be ensured that the clocks are synced. The best way to be sure is to measure the clock signal while running VoIP...
Best Regards,
Suresh
