I2S does not work as expected on nRF52832

[NewtoM]

Hello,

some time ago I was struggling with interfacing an I2S microphone with the nRF52832. That remained unsuccessful. My conclusion back then was that the selected I2S microphone and the nRF52832 were incompatible (the mic wanted 32-bit word size, which the nRF did not support).

In a current project, I -again- need to interface an I2S device to A/D convert the analog data from a MEMS microphone. This time a WM8731 codec, which is known to be very versatile and universally interfaceable. Since the WM8731 specification states that it supports different word lengths (16, 20, 24, or 32 bit), I was very positive that it can be used with the Nordic chip.

It seems I was wrong.

I took a small test board consisting of the WM8731 and the most necessary external components as recommended by the manufacturer and connected it with a Knowles analog MEMS microphone. This setup works flawlessly with a SAM4L development board.

With the Nordic nRF52 development board (PCA10040, v1.1.0) and the latest SDK (v13.0) the TWI communication (for setting up the codec) works and the codec also starts providing the samples over I2S, but I only see scrambled data in the Nordic's I2S buffers. I double and triple checked all my methods as well as the clock and codec configuration - they should work, but they don't.

I also played with different bitstream formats, which both sides (the codec and the nRF52) "speak". I made some recordings with a 1 kHz sine as input. Here are the results:

REC_I2S.wav

REC_LEFT-ALIGNED.wav

REC_RIGHT-ALIGNED.wav

As for the recording, I write the data from an intermediate buffer to an SD-Card (based on the SD-Card example). This uses SPI (and EasyDMA). Some parts of the recordings sound like as if the SPI and the I2S transfers would disturb each other. I hope that is not true.

I also tried different MCL speeds (4, 8, 10.6, 16 MHz), different left-right-clock ratios (CONFIG.RATIO), bit-depths (CONFIG.SWIDTH), but nothing works right. If the recording sounds fine for 50 ms, then it gets totally distorted right after that (as if the sync between nRF52 and the codec would be gone).

Here is a simple question to the crowd out there: did anyone manage to interface any audio codec that works perfectly with the nRF52832?

If yes, please tell me which codec it was and what settings worked.

If not, I am tempted to believe that the I2S interface of the nRF52832 is buggy and hence useless.

[NewtoM]

Hello again,

after much invested time I came to the conclusion that the WM8731 I was using was damaged! I don't know when and how it happened, but after swapping the codec, things worked better.

So at this point I'd like to formally apologise for my previous assumption that the I2S interface of the nRF52 would be useless. (Whether it is buggy or not, I am not sure yet.)

I am now able to record over line-in through the WM8731, when the nRF52 is master. I used the following setup:

#define TWI_frequency     NRF_TWI_FREQ_400K

#define I2S_BUFF_LENGTH 512
uint32_t m_i2sBuffer[I2S_BUFF_LENGTH]= {0};

nrf_drv_i2s_config_t const I2S_config=
{ 
  .sck_pin      = (uint8_t) PIN_I2S_SCK,
  .lrck_pin     = (uint8_t) PIN_I2S_LRCK,
  .mck_pin      = (uint8_t) PIN_I2S_MCK,
  .sdout_pin    = (uint8_t) NRF_DRV_I2S_PIN_NOT_USED,
  .sdin_pin     = (uint8_t) PIN_I2S_SDIN,
  .irq_priority = (uint8_t) APP_IRQ_PRIORITY_LOW,
  .mode         = (nrf_i2s_mode_t) I2S_CONFIG_MODE_MODE_Master,
  .format       = (nrf_i2s_format_t) I2S_CONFIG_FORMAT_FORMAT_I2S,
  .alignment    = (nrf_i2s_align_t) I2S_CONFIG_ALIGN_ALIGN_Left,
  .sample_width = (nrf_i2s_swidth_t) I2S_CONFIG_SWIDTH_SWIDTH_16Bit,
  .channels     = (nrf_i2s_channels_t) I2S_CONFIG_CHANNELS_CHANNELS_Left,
  .mck_setup    = (nrf_i2s_mck_t) I2S_CONFIG_MCKFREQ_MCKFREQ_32MDIV8,
  .ratio        = (nrf_i2s_ratio_t) I2S_CONFIG_RATIO_RATIO_128X
};

The above gives me 31250 Hz, 16 bit, mono recording. This is how I set up the codec:

wm8731_reset(&m_twi);
wm8731_set_master_or_slave_mode(&m_twi, 0);           // set codec to be slave
wm8731_power_mode_all_active(&m_twi);
wm8731_set_sampling_control(&m_twi, 0, 0, 7);         // normal mode, BOSR=0, SR=[0,1,1,1]
wm8731_set_digital_audio_interface_format(&m_twi, 2); // I2S
wm8731_set_digital_audio_data_bit_length(&m_twi, 0);  // IWL_16_BIT
wm8731_adc_input_select(&m_twi, 0);                   // select Line-in
wm8731_left_line_in_volume_set(&m_twi, 0x1F);         // max input volume
wm8731_right_line_in_volume_set(&m_twi, 0x1F);        // max input volume
wm8731_set_left_line_in_mute(&m_twi, 0);              // unmute
wm8731_set_right_line_in_mute(&m_twi, 0);             // unmute
wm8731_enable_adc_high_pass_filter(&m_twi, 1);        // enable HPF
wm8731_set_active(&m_twi, 1);  

And this is how I start I2S:

nrf_drv_i2s_start(m_i2sBuffer, NULL, I2S_BUFF_LENGTH, 0);

Now, I have some comments on the I2S driver. According to the documentation, it splits the pointed buffer into two halves and uses them as ping-pong buffers. Then the documentation says, the last parameter of nrf_drv_i2s_start() is "Size of the buffers (in 32-bit words)." For me it is ambiguous: does it mean the size of each buffer or the total size of the buffers? It turns out, it means the total size.

The next surprise was, that the I2S handler seems to deliver the contents of the sample buffers in a (for me) unexpected order. What I mean is, if the input audio can be split into a sequence of buffers "A,B,C,D,E,F,...", then I get the buffers from the I2S handler in this order: "B,A,D,C,F,E,...". (I really think that I did not mess up things here myself ...)

This is an example recording:

REC_0092_TheKyotoConnection-Hachiko.mp3 (From freemusicarchive.)

At first it sounds okay. No clicks at buffer edges, i.e., buffers of samples are transferred OK.

But if you look at the samples more closely OR if you look at recordings where the input was muted, you will see that there are two problems:

• Every 31st or 32nd sample (alternating, and this is NOT aligned with buffer sizes) is superimposed by a small click, which leads to a 1000 Hz (+harmonics) overlay.

• Every writing to the SD card can be seen (and heard) in form of a low-frequency (as compared to the 1000 Hz one) distortion.

Samples are here (to be viewed as waveform):

• REC_0082_8MHz-SPI_16bit-DMA.wav
• REC_0081_2MHz-SPI_16bit-DMA.wav
• REC_0084_8MHz-SPI_16bit-NoDMA.wav
• REC_0086_8MHz-SPI_24bit-DMA.wav

Here are some facts about these distortions:

• SPI priority (SD card) makes no difference
• SPI speed (SD card) makes no difference (tried 2, 4, and 8 MHz)
• Enabling or disabling EasyDMA for SPI makes no difference
• Bit depth (16 or 24 bit over I2S) makes no difference
• I2S clock (i.e. sampling rate) makes a difference: doubling the sampling rate doubles the distance between clicks
• I2S buffer size makes no difference (tried: 64, 128, 256, and 512 32-bit words)
• I added all relevant external components (mostly capacitors) as recommended, but they also make no difference.

Can anyone suggest a fix to get rid of the distortions?

Thanks, Tamas