Audio class - FIFO feedback not clear

[gmikitiuk]

Hello,

I'm facing an issue where my audio-clock is slightly higher than the USB one so I would expect the feedback endpoint to care of it so I don't run into underrun situation after some time.

Currently I want to use FIFO feedback and if I'm understanding the principle behind it correctly I should remove data from FIFO(using tud_audio_read) with a rate of my main audio clock - so if I remove from FIFO faster than expected then HOST should send more data, if I remove slightly slower than expected, HOST should send less. Is my understanding correct?

What I've noticed is that I cannot force HOST to send more data and I always end up in underrun - for test purpose I read all the audio data from the fifo after tud_task() - so fifo should be constantly empty, which in theory should lead to more data sent by the host, but that's never happening.

What am I missing here?

[HiFiPhile]

Currently I want to use FIFO feedback and if I'm understanding the principle behind it correctly I should remove data from FIFO(using tud_audio_read) with a rate of my main audio clock - so if I remove from FIFO faster than expected then HOST should send more data, if I remove slightly slower than expected, HOST should send less. Is my understanding correct?

Yes.

Are you testing with uac2_speaker_fb example ? The fifo size needs to be at least 4 times the EP size.

tinyusb/examples/device/uac2_speaker_fb/src/tusb_config.h

Line 152 in 2571889

#define CFG_TUD_AUDIO_FUNC_1_EP_OUT_SZ_MAX TUD_AUDIO_EP_SIZE(CFG_TUD_AUDIO_FUNC_1_MAX_SAMPLE_RATE, CFG_TUD_AUDIO_FUNC_1_N_BYTES_PER_SAMPLE_RX, CFG_TUD_AUDIO_FUNC_1_N_CHANNELS_RX)

tinyusb/src/class/audio/audio_device.h

Line 469 in 2571889

// Note about feedback calculation

The adaptive capacity is limited to +- 1 sample so if your clock drift is too high it will still under/overrun. You can read more details in:

• Universal Serial Bus Device Class Definition for Audio Data Formats or Frmts20 final.pdf chapter 2.3.1.1
• USB 2.0 specification chapter 5.12

[gmikitiuk]

Hi!

Thanks for the feedback
Yes, I'm running almost unmodified uac2_speaker_fb example. My FIFO is more than 4 times the EP size.

My current setup is USB HS, 48kHz, 2ch, 16bit so according to my calculations I should get 24B per microframe and this seems to be the case, sometimes I receive 20B, but never more than 24B which I would expect to happen.

What's surely a good hint is that adaptive capacity is limited to +- 1sample, my difference is quite high but anyway I would expect HOST to send more data(even if at the end I run into underrun)

[gmikitiuk]

Nevermind, I was running at 44.1kHz so my calculations were wrong, now I can see more data sent by the host