Hi @Walid ZRELLI
Thanks for taking interest and reaching out to help me.
After spending some time and conducing various test, I have further details on it.
I am using X-CUBE-USB library for Audio Class 2.0; Having USB 2.0 and External phy for High speed configuration.
Where audio configuration is set to use speaker_with_feedback and microphone.
When I just use the Microphone EP from PC for recording (nothing is streamed on speaker), it generates very few Iso-in incomplete events (which is handled correctly) and there is no abrupt loss of communications.
However, If I play speaker along with recording - I am having abrupt loss of communications either in feedback or in mic or in both ( iso-in comm is lost).
Hence, The observation is that, iso-in implementation is appropriate and "Just recording with mic only EP" is working appropriately but whenever two iso-in endpoints are active simultaneously the communication is lost abruptly on any of them.
Whenever, the speaker with feedback EP is started - the iso-in incomplete events occurs a lot for feedback EP.
The feedback EP polling rate is set for 8 micro-frames (1 MS) - so out of 8000 Micro-frames in a second - 1000 micro-frames are served appropriately - but stm32 USB core stack generates 7000 iso-in incomplete events which are falsely handled by USB audio 2.0 class driver - which might be causing communication to fail abruptly.
The above speculation is verified by making feedback polling rate to 1 micro-frame - which led to very few iso-in incomplete events. With feedback polling rate to 1micro-frame, there is very few iso-in incomplete events and very rare loss of communication.
However, for polling rate of 1 micro-frame, facing issue with figuring out which EP generated the iso-in incomplete event - the Reference Manual does not provide much help.
I found a forum thread Here which stats that the existing implementation is flawed and need something better in the programming.
With required polling rate as 8 micro-frames, Need to figure out the handling of iso-in incomplete events -
Also need to know why feebdack EP is having 7000 iso-in incomplete events when polling rate is set to 8 micro frames -
Following is code snippet for handling the iso-in incomplete event.
/**
* @brief USBD_AUDIO_IsoINIncomplete
* handle data ISO IN Incomplete event
* @param pdev: device instance
* @param epnum: endpoint index
* @retval status
*/
static uint8_t USBD_AUDIO_IsoINIncomplete (USBD_HandleTypeDef *pdev, uint8_t epnum)
{
USBD_AUDIO_EPTypeDef *ep;
USBD_AUDIO_HandleTypeDef *haudio;
uint16_t current_sof;
haudio = (USBD_AUDIO_HandleTypeDef*) pdev->pClassData;
/* @TODO check if the feedback is responsible of event */
for(int i = 1; i<USBD_AUDIO_MAX_IN_EP; i++)
{
ep = &haudio->ep_in[i];
current_sof = USB_SOF_NUMBER();
if((ep->open)&&
((ep->ep_usage == USBD_AUDIO_FEEDBACK_EP)||(ep->ep_usage == USBD_AUDIO_DATA_EP))
&& IS_ISO_IN_INCOMPLETE_EP(i,current_sof, ep->tx_rx_soffn))
{
epnum = i|0x80;
USB_CLEAR_INCOMPLETE_IN_EP(epnum);
USBD_LL_FlushEP(pdev, epnum);
ep->tx_rx_soffn = USB_SOF_NUMBER();
#if USBD_SUPPORT_AUDIO_OUT_FEEDBACK
if(ep->ep_usage == USBD_AUDIO_FEEDBACK_EP)
{
USBD_LL_Transmit(pdev,
epnum,
ep->ep_description.sync_ep->feedback_data,
ep->max_packet_length);
continue;
}
else
#endif /*USBD_SUPPORT_AUDIO_OUT_FEEDBACK */
if(ep->ep_usage == USBD_AUDIO_DATA_EP)
{
USBD_LL_Transmit(pdev,
epnum,
ep->ep_description.data_ep->buf,
ep->ep_description.data_ep->length);
}
}
}
return 0;
}
