I2S interrupt priority on nrf5340

[Nishant Kulgod]

Hi,

We are working on nrf5340 with the nrfx I2S implementation. We have configured the I2S IRQ for every 5 miliseconds.
However, we have observed that successive IRQ introduce 100 micro seconds delay which results into a deleay over time.

Is there any way to reduce this delay in the I2S IRQ?

We have observed "#define CONFIG_NUM_METAIRQ_PRIORITIES 0" in the autoconf.h file.
Is this the maximum priority for I2S IRQ? Is there anyway to increase the priority for I2S IRQ?

Thanks,
Nishant

[Simon]

Could you share your project? Then I can try to do some debugging myself.

Best regards,

Simon

[Nishant Kulgod]

Hi Simon,
Thanks for getting back to me.
We have referred following I2S implementation : https://github.com/siguhe/NCS_I2S_nrfx_driver_example/blob/master/src/main.c

This is a github link which we got from this devZone ticket : https://devzone.nordicsemi.com/f/nordic-q-a/63708/i2s-loopback-example-missing-in-installation-folder/260841?focus=true#259789

Thanks,
Nishant

[Nishant Kulgod]

Hi Simon,

#define I2S_DATA_BLOCK_WORDS 512 //How many numbers do we want. time reorded = DATA_BLOCK_WORDS / freq

Above MACRO is configured to store 5ms data. And we are capturing time difference between two successive data handlers which should be exactly 5 ms ideally. We get it as 5.1 ms. Can you please check this interval in your setup?

Thanks,

Nishant

[Simon]

I've been sick lately and haven't been able to answer you until now. I will look into this the upcoming days.

Best regards,

Simon

[Simon]

I'll get ahold of an Adafruit I2S MEMS Microphone tomorrow and will try to reproduce your issue.

Best regards,

Simon

[Nishant Kulgod]

Alright, thanks a lot Simon.

Nishant

[Simon]

By the way, what nRF53 version are you using? And what i2s microphone are you using? An Adafruit i2s mems microphone?

If you are using an nRF53 PDK v0.8.0 or older and the mention sensor (Adafruit i2s mems) it won't work, since 32 bit sample are not supported due to this errata, which the Adafruit sensor requires.

See the comment made by another colleague:

"I2S microphones have a scaling of WS (Word Select) speed relative to that of the BCLK clock speed. This has also been called the data word length. For the microphone used in the I2S MEMS Adafruit breakout, the datasheet specifies: “WS must be BCLK/64”. One word is therefore 64 clock cycles. WS is interchangeable with LRCK. Now, this rate has to be set as a setting for the nRF I2S peripheral, named “CONFIG.RATIO” . Under Master clock (MCK), an important limitation is given: “The MCK/LRCK ratio shall be a multiple of 2 * CONFIG.SWIDTH”. Therefore, since MCK/LRCK = 64, the Sample Width SWIDTH has to be 32. From the top of the I2S section, we see that the old nRF peripheral supports: “8, 16 and 24-bit sample width”. This is the reason that the old I2S peripheral does not support the Adafruit I2S MEMS microphone."

A workaround for this is explained here: https://github.com/gregtomasch/nRF52_24-bit-_I2S_Microphone_Audio_Recording_Utility

I learned this the hard way, thus the delayed response from me.

---

However, if this is not an issue for you, I will get it working on my side and help you with the initial question (100 micro seconds delay).

Best regards,

Simon

[Nishant Kulgod]

Hi Simon,

We are using another codec board that takes care of audio capture.
Our concern lies where if we directly do a busy wait on the register value, we are able to identify faster. Whereas, above implementation gives a delay of ~100 us over 500 ms capture duration. We do not know the cause of this delay.

Thanks,

Nishant

[Simon]

I toggled a pin inside the data handler of the I2S sample, I put it inside the NRFX_I2S_STATUS_NEXT_BUFFERS_NEEDED if statement and watched the time between every call. I got the following results:

1st: 31.738998 ms
2nd: 31.74207 ms
3rd: 31.7382 ms
4th: 31.7446 ms
5th: 31.7386 ms
.
.
nth: 31.74786

Best regards,

Simon