Problem with feedback on UAC2

[duddie]

Hello,

I have problem with feedback on UAC2 and I cannot tell if the problem is with my descriptor, with the STM32H750 driver or maybe with all endpoint configuration. So fwar I have managed to get both input and output of audio fully working however I have small clicks with audio coming from host because my DAC frequency slightly differs (driven from STM32H7 MCLK) from 48000 Hz audio.

Now the problem, whenever I try to enable Feedback EP host stops streaming audio after several frames. I can see queued EP 83 (Feedback EP) messages but not anymore audio packets. Also, Feedback EP callback is called only once when device is initialised and is not being called repeatedly. While most likely problem is on my side I would really appreciate feedback on feedback (pun intended).

As I cannot attach .h files I have attached one .c file with audio descriptor and other important settings (4 channels audio, but same problems for 2 channels). Also attached is log at level 3 that clearly shows problem.

problematic_descriptor.c
usb_log.txt

Any help will be greatly appreciated! Thanks for amazing work on TinyUSB!

I am using the latest master branch.

Kind regards, Duddie

[HiFiPhile]

Hi, the behavior is indeed abnormal, even if the feedback value is incorrect the host will still send more data.

• What's your host OS ?
• Have you connected the board directly to the PC without a hub, since ISO transfer in full-speed occupies a large portion of bandwidth.
• Have you tried stock uac2_speaker_fb example ?
• feedback_param->sample_freq should be set to target sample rate 48000

[duddie]

Hi,

• Host OS is macOS (latest).
• Its direct connection. The audio in/out works fine. It only stops sending packets when FEEDBACK is enabled. Normally I have good flow of packets and for audio going to host I have managed to get all good compensation with buffering even despite slight difference in audio clocks between host and device. For the audio coming from host without feedback I managed to reduce clicks to minimum so but they are stil there so I need feedback. So there is no problem with cable, communication or basic descriptor.
• I could not get uac2_speaker_fb to work on my board.
• feedback_param->sample_freq I tried setting it to 48kHz, a few Hz less, a few Hz more. I tried to set threshold to different values for FIFO. I tried deifferent methods (DISABLE (send fb from audio task), FIXED FREQ. FLOAT FREQ, FIFO - all same... feedback callback gets called once and then audio packets just stop arriving.

I will work some more on your example but I had to switch UAC2 from HIGH SPEED to FULL SPEED and maybe I am missing things. I will put more attention to that.

Not sure if the problem is around DCD2 as other STM32 chips seem to use fsdev.

[HiFiPhile]

If you are on mac you have to use UAC1 for FS due to some feedback format (3 bytes vs 4 bytes) shenanigans acrossing different OS.
A quirk was used to guess host OS for the feedback to work correctly but removed once UAC1 is added: https://github.com/HiFiPhile/tinyusb/blob/6a67bac47c0f83eebd63ca99654ed26e51b21145/examples/device/uac2_speaker_fb/src/quirk_os_guessing.c
#2328

feedback_param->sample_freq needs to be set to target sample rate to allow the internal calculation to be done correctly.

[duddie]

Thank you for your feedback, this is giving me a great understanding of things I wasn't aware. Although forgive my my ignorance as I am new to TinyUSB but I would like to still use UAC2 (same descriptor) across FS/HS however here are my questions, if those are stupid questions please help me understand why they are stupid:

1. I need 4 channel audio and I think UAC1 does not allow me to do so.
2. If UAC2 implementations differ for different platforms for feedback mechanism on FS platform would be good to have userland raw_feedback_ep_cb() to handle those differences if handling this at the TinuUSB level is not appropriate
3. Your "os quirks" - could you provide little hint on how do I get them to work with FS UAC2? I see what they do but I have no idea how to integrate or use them - going back to question 2 - probably raw feedback handler in userland would help.

I do appreciate your work on TinyUSB audio part and thanks for making this happen!

[HiFiPhile]

1. I need 4 channel audio and I think UAC1 does not allow me to do so.

Nope, not sure where did you heard that ;)

In FS bandwidth limit is 1023 bytes/ms which translate to more than 4CH 16bit/96kHz ((96+1)*2*4) = 778 < 1023, no matter you are UAC1 or UAC2.

In FS UAC2 brings little to no advantage compared to UAC1.

2. If UAC2 implementations differ for different platforms for feedback mechanism on FS platform would be good to have userland raw_feedback_ep_cb() to handle those differences if handling this at the TinuUSB level is not appropriate
3. Your "os quirks" - could you provide little hint on how do I get them to work with FS UAC2?

It's a deliberate choice to drop all quirks/workaround and stick on specification for maintenance sake, it was a nightmare to test all audio examples everytime on my Windows desktop, Linux laptop then mac VM.... Besides there is zero guarantee that the quirk will still work with a OS update.

The quirk was needed because UAC1 supported was missing and there was no other way to make UAC2 works on every OS.

You can easily turn uac2_speaker_fb example into 4 CH:

#define CFG_TUD_AUDIO_FUNC_1_N_CHANNELS_RX 4

diff --git a/examples/device/uac2_speaker_fb/src/usb_descriptors.h b/examples/device/uac2_speaker_fb/src/usb_descriptors.h
index 79f25bbfa..b1fb9c395 100644
--- a/examples/device/uac2_speaker_fb/src/usb_descriptors.h
+++ b/examples/device/uac2_speaker_fb/src/usb_descriptors.h
@@ -97,7 +102,7 @@
   + TUD_AUDIO10_DESC_CS_AC_LEN(1)\
   + TUD_AUDIO10_DESC_INPUT_TERM_LEN\
   + TUD_AUDIO10_DESC_OUTPUT_TERM_LEN\
-  + TUD_AUDIO10_DESC_FEATURE_UNIT_LEN(2)\
+  + TUD_AUDIO10_DESC_FEATURE_UNIT_LEN(4)\
   + TUD_AUDIO10_DESC_STD_AS_LEN\
   + TUD_AUDIO10_DESC_STD_AS_LEN\
   + TUD_AUDIO10_DESC_CS_AS_INT_LEN\
@@ -112,11 +117,16 @@
   /* Class-Specific AC Interface Header Descriptor(4.3.2) */\
   TUD_AUDIO10_DESC_CS_AC(/*_bcdADC*/ 0x0100, /*_totallen*/ (TUD_AUDIO10_DESC_INPUT_TERM_LEN+TUD_AUDIO10_DESC_OUTPUT_TERM_LEN+TUD_AUDIO10_DESC_FEATURE_UNIT_LEN(2)), /*_itf*/ ((_itfnum)+1)),\
   /* Input Terminal Descriptor(4.3.2.1) */\
-  TUD_AUDIO10_DESC_INPUT_TERM(/*_termid*/ 0x01, /*_termtype*/ AUDIO_TERM_TYPE_USB_STREAMING, /*_assocTerm*/ 0x00, /*_nchannels*/ 0x02, /*_channelcfg*/ AUDIO10_CHANNEL_CONFIG_LEFT_FRONT | AUDIO10_CHANNEL_CONFIG_RIGHT_FRONT, /*_idxchannelnames*/ 0x00, /*_stridx*/ 0x00),\
+  TUD_AUDIO10_DESC_INPUT_TERM(/*_termid*/ 0x01, /*_termtype*/ AUDIO_TERM_TYPE_USB_STREAMING, /*_assocTerm*/ 0x00, /*_nchannels*/ 0x04, /*_channelcfg*/ AUDIO10_CHANNEL_CONFIG_LEFT_FRONT | AUDIO10_CHANNEL_CONFIG_RIGHT_FRONT, /*_idxchannelnames*/ 0x00, /*_stridx*/ 0x00),\
   /* Output Terminal Descriptor(4.3.2.2) */\
   TUD_AUDIO10_DESC_OUTPUT_TERM(/*_termid*/ 0x03, /*_termtype*/ AUDIO_TERM_TYPE_OUT_DESKTOP_SPEAKER, /*_assocTerm*/ 0x00, /*_srcid*/ 0x02, /*_stridx*/ 0x00),\
   /* Feature Unit Descriptor(4.3.2.5) */\
-  TUD_AUDIO10_DESC_FEATURE_UNIT(/*_unitid*/ 0x02, /*_srcid*/ 0x01, /*_stridx*/ 0x00, /*_ctrlmaster*/ (AUDIO10_FU_CONTROL_BM_MUTE | AUDIO10_FU_CONTROL_BM_VOLUME), /*_ctrlch1*/ (AUDIO10_FU_CONTROL_BM_MUTE | AUDIO10_FU_CONTROL_BM_VOLUME), /*_ctrlch2*/ (AUDIO10_FU_CONTROL_BM_MUTE | AUDIO10_FU_CONTROL_BM_VOLUME)),\
+  TUD_AUDIO10_DESC_FEATURE_UNIT(/*_unitid*/ 0x02, /*_srcid*/ 0x01, /*_stridx*/ 0x00, \
+    /*_ctrlmaster*/ (AUDIO10_FU_CONTROL_BM_MUTE | AUDIO10_FU_CONTROL_BM_VOLUME), \
+    /*_ctrlch1*/ (AUDIO10_FU_CONTROL_BM_MUTE | AUDIO10_FU_CONTROL_BM_VOLUME), \
+    /*_ctrlch2*/ (AUDIO10_FU_CONTROL_BM_MUTE | AUDIO10_FU_CONTROL_BM_VOLUME), \
+    /*_ctrlch3*/ (AUDIO10_FU_CONTROL_BM_MUTE | AUDIO10_FU_CONTROL_BM_VOLUME), \
+    /*_ctrlch4*/ (AUDIO10_FU_CONTROL_BM_MUTE | AUDIO10_FU_CONTROL_BM_VOLUME)),\
   /* Standard AS Interface Descriptor(4.5.1) */\
   /* Interface 1, Alternate 0 - default alternate setting with 0 bandwidth */\
   TUD_AUDIO10_DESC_STD_AS_INT(/*_itfnum*/ (uint8_t)((_itfnum)+1), /*_altset*/ 0x00, /*_nEPs*/ 0x00, /*_stridx*/ 0x00),\
@@ -126,7 +136,7 @@
   /* Class-Specific AS Interface Descriptor(4.5.2) */\
   TUD_AUDIO10_DESC_CS_AS_INT(/*_termid*/ 0x01, /*_delay*/ 0x00, /*_formattype*/ AUDIO10_DATA_FORMAT_TYPE_I_PCM),\
   /* Type I Format Type Descriptor(2.2.5) */\
-  TUD_AUDIO10_DESC_TYPE_I_FORMAT(/*_nrchannels*/ 0x02, /*_subframesize*/ _nBytesPerSample, /*_bitresolution*/ _nBitsUsedPerSample, /*_freqs*/ __VA_ARGS__),\
+  TUD_AUDIO10_DESC_TYPE_I_FORMAT(/*_nrchannels*/ 0x04, /*_subframesize*/ _nBytesPerSample, /*_bitresolution*/ _nBitsUsedPerSample, /*_freqs*/ __VA_ARGS__),\
   /* Standard AS Isochronous Audio Data Endpoint Descriptor(4.6.1.1) */\
   TUD_AUDIO10_DESC_STD_AS_ISO_EP(/*_ep*/ _epout, /*_attr*/ (uint8_t) ((uint8_t)TUSB_XFER_ISOCHRONOUS | (uint8_t)TUSB_ISO_EP_ATT_ASYNCHRONOUS), /*_maxEPsize*/ _epoutsize, /*_interval*/ 0x01, /*_sync_ep*/ _epfb),\
   /* Class-Specific AS Isochronous Audio Data Endpoint Descriptor(4.6.1.2) */\

[duddie]

Thank you for claryfying that. If UAC1 can handle my needs then indeed it's better to stick with the specification and not being dependant on hacks that are fragile and suspectible to changes.

Your help is really really appreciated and now I know how I am going to approach my tiny project.

Thank you so much for all your hard work on TinyUSB and thank you for your time to explain me those nuances.

Cheers!