How do I know I2S transfer finished using SDK's I2S driver

[Howard]

When using SDK v13's I2S driver, the data to send is filled via data_handler callback. How do I fill last block data? Assume i2s buffer size is 2000, then every 1000 words sent I have to prepare the next new block data. But what if last block was not just 1000 words, which may actually be 300 bytes. For this case what should I do? should I still fill the rest 700 words?

And more, after I filled last block data in data_handler callback, when do I stop I2S? I filled last block data but I don't know when it finished and when to stop I2S.

Please help.

[endnode]

I2S driver cannot process more than 0xFF = 255 samples, does it?

[Jørgen Holmefjord]

You don't need to fill the buffer, but then the receiver should know the total size and ignore the 700 last words. The whole TX buffer passed to nrf_drv_i2s_start()will be transmitted. You should know in your application if you do not have more data to fill. Alternatively, you can setup a new, smaller buffer, for the remaining data. The best solution will depend on your application.

You can stop I2S the next time data_handler is called, as this is called when new data is required for transmission, or data is received. You should preferrably set a flag in the callback and stop I2S in main context.

[Howard]

Thank you for your reply! but My receiver is an Audio DAC, which does not know the total size and not able to ignore the last 700 words. About your alternate solution, how can I setup a new buffer for the remaining data while the previous data is still in transferring?

[Jørgen Holmefjord]

The driver does not support this. You will have to stop the transfer and start it again with the new buffer. I guess this is not a good solution for your DAC either. Another alternativ can be to call the I2S HAL function used by the driver (in nrf_drv_i2s_start()) to set the MAXCNT register (transfer size) and buffers, nrf_i2s_transfer_set(), from inside data_handler with pointers to a smaller buffer. I have not tried this, but I can't see anything in the driver preventing your from doing this.

[endnode]

What I did in my I2S application was simply demanding input stream to be padded by "silence" (whatever it is in your encoding) to the closest transfer "block" size or if not possible doing it inside the FW. It's not super-elegant but few milliseconds of driven silence instead of real silence didn't harm anything;)