 Hello and welcome to this presentation of the STM32 Serial Audio Interface, or SAI. It covers all the features of this interface, which is widely used to connect external audio devices. The SAI integrated inside STM32 products provides an interface, allowing the microcontroller to communicate with external audio devices, such as amplifiers, ADCs, DACs, or audio processors. This interface is fully configurable and supports most audio standards, allowing easy connection to existing audio devices. Thanks to internal synchronization features, the amount of IO pins is reduced to its minimum. The SAI can be programmed in four different modes. Free protocol mode allows the SAI to support standards such as I2S, PCM, TDM, etc. Thanks to its flexibility, it is possible to customize the serial interface if needed. SPDIF protocol mode allows the SAI to transmit audio samples using the IEC 6958 standard. PDM interface mode allows the SAI to connect up to eight digital microphones for beamforming or simple speech capture applications, or AC97 protocol. The SAI supports all the usual audio sampling rates, according to the crystal frequency used for the application. In addition, the SAI supports the master and slave modes in half-duplex or full-duplex communication. It is also possible to synchronize several SAI interfaces together. The SAI also provides a FIFA buffer of eight samples and up to two interrupts and DMA interfaces. The SAI is composed of two independent sub-blocks, sub-block A and B. Each sub-block has its own APB interface, clock generator, FIFA buffer, DMA interface, and interrupt interface. Each sub-block can be configured in receiver or transmitter mode and in master or slave mode with its own protocol. Internal and external synchronization allows two sub-blocks to be synchronized or two SAI interfaces to be synchronized. Each sub-block can handle up to four IOs. For each sub-block, FS is the frame synchronization, SCK is the bit clock, SD is the serial data, and MCLK is the master clock. In addition, a PDM interface allows the connection of up to eight digital microphones. The STM32MP1 embeds four SAI instances. It offers a large choice of kernel clocks. The SAI can receive a kernel clock from the DIVQ output of PLL3 or PLL4, from the DIVR output of PLL3, from the HSI, CSI, or HSE oscillators, or from an input pad, I2S, CKIN. In addition, the SAI2 can receive the symbol clock of the SPD IFRX block, allowing the synchronization of the STREAM received by the SPD IFRX with the SAI2. Note that the symbol clock frequency is 64 times the sample rate. Note as well that the kernel clock is used by the SAI to generate the timing of the serial audio interface when configured in master mode. All PLLs can be programmed in fractional mode, allowing very accurate audio sampling rates. Each SAI IP can be assigned either to the Cortex A7 non-secure or to the Cortex M4. Please refer to STM32MP1 wiki pages for details. The free protocol mode makes it possible to emulate most of the common audio standard interfaces thanks to the flexibility of changing the behavior of several parameters, such as data justification, data size and position, frame size, frame period, frame polarity, sampling edge for the clock and number of slots. The following example shows some of the possibilities of the interface for the I2S-like protocols. In an I2S-like protocol, each edge of the frame synchronization, or FS, is used to align the slot positions. The frame length, the duty cycle and polarity can be adjusted. The clock data strobe edge can be selected as well. The position of the slots with respect to the frame edges can be selected. The size of the slots can also be adjusted. There must be an even number of slots per frame in I2S-like protocols. The following example shows some of the possibilities of the interface for the TDM-like protocols. In a TDM-like protocol, only one edge of the frame synchronization, rising or falling, is used to align the slots position. The frame length, the duty cycle and polarity can be adjusted. The clock data strobe edge can be selected. The position of the slots with respect to the frame active edge can be selected. The size of the slots can also be adjusted and the amount of slots per frame up to 16. The SAI is able to handle up to 16 slots and each slot can be individually activated or not. The inactive slots can be set in high Z. The slot size is always bigger than or equal to the data size. The SAI allows control of the position of the data inside each slot and to set the unused parts of the slots to high Z if needed. This function can be helpful when the data line is shared between several devices. In master mode, the SAI can generate the master clock or MCLK depending on the audio system configuration. This master clock provides a reference clock to the external audio codecs. In master mode, the SAI generates the frame synchronization signal or FS and the bit clock or SCK. The data line SD can be either input or output. In slave mode, the MCLK signal is not used. In slave mode, the SAI receives the frame synchronization signal or FS and the bit clock or SCK from another device, external or internal. The data line SD can be either input or output. In master mode, it is up to the SAI to generate the appropriate timings to provide the correct sampling rate. In slave mode, the sampling rate is provided by the external audio device. The clock generator is needed for master mode communications. It is used to adjust the sampling rate of the serial audio interface. The clock generator provides the root frequency for the MCLK, SCK and the FS. When the master clock or MCLK is generated, the frame length must be a power of 2. The ratio between the FS frequency and the MCLK frequency is set to 256 or 512 according to the OSR bit. The clock SAICK is provided by the STM32 MP1 RCC block. When the MCLK is not generated, the frame length can take any value from 8 to 256. In this case, the frequency of the SCK bit clock is directly given by the clock received on the SAICK input divided by the MCK div value. The internal synchronization can be used for communications needing two data lanes, such as full duplex I²S. The external synchronization can be used for communications needing more than two data lines, up to four, for example when interfacing HDMI ICs. All the sub-blocks synchronized together must use the same protocol characteristics. In order to reduce the data size, it is possible to insert in the data pass an ALaw or Microlaw compander. Note that ALaw and Microlaw are not lossless compressors. Compending modes are generally used in telephony. The small values are amplified and the big values are attenuated. The SNR tends to be identical for strong and for weak signals. The SAI also provides a mute function. In transmit mode, the user can choose to send zeros on muted slots or the previous transmitted value. The previous transmitted value is limited to configurations having one or two slots per frame. Note that in transmit mode, the TX-FEFO pointer is still incremented, meaning that the data which was present in the FIFO and for which the mute mode is requested is discarded. The receive mute mode can be helpful to detect an amount of consecutive slots having all data reset to zero. The anticipated or late frame error detection function increases the interface's reliability by detecting unexpected frame synchronization misalignment. A status flag is set and an interrupt can be generated as well. The application software will then have to restart the SAI interface. The SAI guarantees the data alignment even if an underrun or an overrun occurs. The SAI supports the audio IEC 60958 standard in transmit mode when configured for the SPDIF protocol. The SAI generates the preambles and the parity bit according to the transmitted data. The software has to handle the CS, U and V bits. In the IEC 60958 specifications, the block structure is used to decode the channel status or CS and user information or U. Each block contains 192 frames. Each frame contains two subframes. The SAI automatically generates the B, M and W preambles. Preamble B detects the start of a new block and the start of a channel A. Preamble M detects the start of a channel A when it is not a block boundary and Preamble W detects the start of a channel B. Each subframe contains 32 bits divided into three fields. A synchronization preamble allows the detection of the block and subframe boundaries, a payload of 24 bits and status bits V, U, CS and P. The FSAICK frequency must be adjusted in order to generate the proper audio sample rate or FS. The data inside the transmit FIFA must be aligned as shown in this slide. The MSB of the data must always be at position 23. The PDM interface remaps the bit stream received from the digital microphones into TDM frames. The PDM interface waits for the reception of 8 bits from each microphone before sending a new TDM frame. In addition, the PDM interface offers an 8-bit delay line for each microphone stream. These delay lines are working with the resolution of the bit stream clock provided to the microphones. It enables beamforming applications and removes constraints on microphone placements. When the PDM interface is enabled, the serial interface of the subblock A cannot be used to connect an external device. This serial interface is connected internally to the PDM interface, and the subblock A must be configured in TDM mode as an RX master. The figure shows an example of connection of four digital microphones. Note that each data line D1, D2, D3, or D4 can be connected to one or two digital microphones. The subblock B is still available for other applications and can be used to connect an external device using TDM, PCM, I2S, or any other supported protocol. With this PDM interface, the bit clock frequency has to be adjusted according to the sampling frequency and the number of microphones. The frame length is also adjusted according to the number of connected microphones. The SAI is able to work as an AC-97 link controller. When this protocol is used, the frame length, the slot number, and slot length are set by the hardware. Several events can be enabled in order to generate interrupts. The WCKCFG event can be used in order to inform the user that the frame length of the SAI has been improperly programmed. This feature only makes sense in master mode. The following table shows an overview of the SAI activity for the various possible power modes. The SAI is active in run and sleep modes, frozen in stop mode, or powered down in standby mode. The SAI needs the bus interface clock, or APB clock, and the kernel clock, or SAICKX, to work properly. For a full duplex master mode, two data lines are needed, so two subblocks need to be used. The master subblock A provides the synchronization to the slave subblock B, using the internal synchronization feature, or IO line management. Note that in this example, the subblock B only uses the SDB. The amount of IOs is reduced to its minimum, thanks to the internal synchronization. This is another kind of full duplex mode, using the TDM protocol. Slot 1 is inactive or not used for subblock A. Slots 2 and 3 are inactive for subblock B. For both subblocks, the frame structure has four slots. Subblock A will generate three samples per frame. Subblock B will receive two samples per frame. This example shows the most important SAI settings in order to capture the samples provided by four digital microphones. In typical applications, the microphones receive a bit stream clock frequency 64 times higher than the wanted audio rate. If the application needs to handle a 16 kilohertz audio stream, then the bit stream clock provided to the digital microphones must be 16 kilohertz multiplied by 64, which corresponds to a clock frequency of 1.024 megahertz. As there are four data streams, the bit clock SCKA must be four times higher than the bit stream clock provided to the microphones, which results in a bit clock frequency of 4.096 megahertz. Using this configuration, the SAIA writes into its RX-FEFO and 8-bit data every time a slot is received. In order to reconstruct the 16 kilohertz audio signal, the software has to perform a low-pass filtering of each microphone stream followed by a decimation by 64.