UAC2 isochronous transfer limits in full speed mode?

[gregory-whaley]

I have successfully developed a USB stereo microphone device on the Raspberry Pi Pico RP2040 MCU using Tinyusb and UAC2. That system used 2 channels at 32 bit samples at 48KHz in full speed mode. The EP buffer size in this case was 384 bytes.

I am trying to expand that system to 4 channels, 32 bit samples at 48KHz. The system compiles and runs and returns no errors or warnings, however the TUSB device stack fails to respond to ISO transfers, but rather responds with an error packet and zero bytes of data. In this case, the EP buffer size is 768 bytes which is within the 1023 byte limit of the USB spec. Here is the wireshark screenshot of an error packet device response to a host ISO request.

To test if this is a buffer size limit problem I took the 4 channel microphone example from the tinyusb examples and altered the data rates. The published version is 4 channel x 16 bit x 48kHz = 384 byte EP size and it works as expected. When I change to 4 channel x 32 bit x 48KHz = 768 byte EP, it fails to transfer properly to host ISO requests.

Here's a table of trials with different transfer rates:

# Chan	#bits	Samp Rate	EP Size	Outcome
4 chan	16 bits	48KHz	384 bytes	Success
4 chan	32 bits	48KHz	768 bytes	Fail
4 chan	24 bits	48KHz	576 bytes	Fail
4 chan	24 bits	40KHz	480 bytes	Success
3 chan	24 bits	48KHz	432 bytes	Success
4 chan	24 bits	46kHz	552 bytes	Success
4 chan	24 bits	47KHz	564 bytes	Fail

So I conclude there is an effective EP buffer limit of between 552 and 564 bytes.

My question is why is this the case? I expected that EP buffers up to 1023 would work, but might have throughput problems and congestion. But this limit is not congestion, rather it seems to be TUSB device code rejecting the EP size at run time. It would be helpful to understand this limit to ensure reliable operation.
Thanks for any comments!

[HiFiPhile]

For FS my suggestion is never exceed ~512 bytes, 1023 is the theorical limits that occupies full bandwidth of a root hub, you need a dedicated USB controller to make it work.

[gregory-whaley]

Thank you for the quick reply. I now see that it is the bandwidth allocation negotiation between the host and device which is dependent on the network topology, i.e. how many hubs are in the path. When I connect the device directly to the root hub I can indeed get a 768 byte EP to work!!