There are many audio codecs available on the market and each requires a driver to work with the audio class. The driver is referred as the Audio Peripheral Driver in the the general audio class architecture. The amount of code necessary to port a specific audio codec to the audio class greatly depends on the codec’s complexity.
...
All audio peripheral drivers must declare different instances of the appropriate driver API structure as global variables within the source code. Each API structure is an ordered list of function pointers utilized by the audio class when device hardware services are required. Each structure will encompass some functions belonging to a category: common, Output Terminal, Feature Unit, Mixer Unit, Selector Unit and AudioStreaming (AS) interface. The API structure will then be passed to the appropriate USBD_Audio_XX_Add() functions. Theses different API structures offers two possibilities to handle the terminal and unit IDs management within a given codec driver function:
...
In order to better understand the use of this stream API, we will consider the typical audio stereo codec configuration shown by 106070237. Moreover, a DMA controller used by the I2S controller will be assumed. 106070237 summarizes the use of stream functions for a playback stream. Please refer to Figure - Playback Stream Dataflow as a complement to know what is happening in the Audio Processing module.
...
| Panel |
|---|
|
(1) The host has opened the stream by selecting the operational AS interface. It then sets the sampling frequency (for instance, 48 kHz). The function USBD_Audio_DrvCtrlAS_SamplingFreqManage() will be called for that operation. The sampling frequency is configured by accessing some codec registers. The register access will be accomplished by sending several I2C commands. (2) Once the playback stream priming is completed within the Audio Processing module, that is a certain number of audio buffers has been accumulated, the function USBD_Audio_DrvStreamStart is called. Usually, you may have to enable playback operations within the codec through some registers configuration. Here again, I2C controller will be used. The function USBD_Audio_PlaybackTxCmpl() is called by the driver to signal the audio transfer completion. The driver can call USBD_Audio_PlaybackTxCmpl() up to the number of buffers it can queue. | Tip |
|---|
The audio peripheral driver should support at least the double-buffering to optimize the playback stream processing. |
(3) The playback task will receive an AudioStreaming interface handle and will submit to the audio peripheral driver a ready buffer by calling USBD_Audio_DrvStreamPlaybackTx . The initial DMA transfer will be prepared with the first ready buffer. Note that the driver should start the initial DMA transfer after accumulating at least two ready buffers. This allows to start a sort of pipeline in which the audio peripheral driver won't wait after the playback task for providing a ready buffer to prepare the next DMA transfer. Once the pipeline is launched, any subsequent call to USBD_Audio_DrvStreamPlaybackTx() should store the ready buffer. Any buffer memory management method may be used to store the ready buffer (for instance, double-buffering, circular buffer, etc.). Depending on the DMA controller, you may have to configure some registers and/or a DMA descriptor in order to describe the transfer. The DMA controller moves the audio data from the memory to the I 2 S controller. This one will forward the data to the codec that will play audio data to the speaker. (4) A DMA interrupt will be fired upon transfer completion. An ISR associated to this interrupt is called. This ISR processes the DMA transfer completion by freeing the consumed buffer. For that, the function USBD_Audio_PlaybackBufFree() is called. This function updates one of the indexes of the ring buffer queue. The ISR continues by signaling to the playback task that the audio transfer has finished with USBD_Audio_PlaybackTxCmpl . Once again, the playback will provide a ready buffer via USBD_Audio_DrvStreamPlaybackTx() as described in step (2). The ISR will get a new ready buffer from its internal buffer storage. A new DMA transfer is prepared and started. If no playback buffer is available from the internal storage, you may have to play a silence buffer to keep the driver in sync with audio class, that is you want to continue receiving DMA transfer completion interrupt to re-signal the audio transfer completion to the playback task. The silence buffer is filled with zeros interpreted by the codec as silence. The silence buffer can be allocated and initialized in the function USBD_Audio_DrvInit() . | Tip |
|---|
The DMA implementation in this example processes the playback buffers one after the other using a single interrupt. Depending on your DMA controller, it may be possible to optimize the performance by using several DMA channels. In that case, you could pipeline the DMA transfers. The DMA controller may also offer to link DMA descriptors. In that case, you could get several ready buffers and link several DMA descriptors. |
(5) The host decides to stop the stream. The function USBD_Audio_DrvStreamStop is called. You should abort any ongoing DMA transfers. You don't have to call USBD_Audio_PlaybackBufFree() to free any aborted buffers nor to free ready buffers stored internally in the driver and not yet processed. The buffers are implicitly freed by the audio class during the ring buffer queue reset. You can also disable the playback operation on the codec if it is needed. |
Figure - Record Stream Functions summarizes the use of stream functions for a record stream. Please refer to Figure - Record Stream Dataflow as a complement to know that is happening in the Audio Processing module.
...
| Panel |
|---|
|
(1) The host has opened the stream by selecting the operational AS interface. It then sets the sampling frequency (for instance, 48 kHz). The function USBD_Audio_DrvCtrlAS_SamplingFreqManage() will be called for that operation. The sampling frequency is configured by accessing some codec registers. The register access will be accomplished by sending several I2C commands. (2) The Audio Processing will call the function USBD_Audio_DrvStreamStart() to start record operations on the codec side. Operations consists in enabling record operations within the codec through some registers configuration. The I2C controller will be used for that. Then, the function USBD_Audio_RecordBufGet() is called by the driver to obtain an empty buffer. This function also specifies the buffer length. The driver does not have to figure out how many bytes is needed depending on the sampling frequency the number of channels and the bit resolution. This is taken into account by the Audio Processing layer. For sampling frequencies such 22.050 kHz, 44.1 kHz not producing an integer number of audio samples per milliseconds, a data rate adjustment is used (refer to Record Stream for more details about this data rate adjustment). With all the buffer's information, you should prepare the initial DMA read transfer. Depending on the DMA controller, you may have to configure some registers and/or a DMA descriptor in order to describe the transfer. The DMA controller moves the audio data from the I2S controller to the memory. | Tip |
|---|
If the DMA offers multiple channels or is able to link several DMA descriptors, you can call USBD_Audio_RecordBufGet() to obtain several buffers. |
(3) A DMA interrupt will be fired upon transfer completion. An ISR associated to this interrupt is called. This ISR processes the DMA transfer completion by signaling to the record task that a buffer is ready with the function USBD_Audio_RecordRxCmpl . The ready buffer should be stored in an internal buffer storage. Any buffer memory management method may be used to store the ready buffer (for instance, double-buffering, circular buffer, etc.). The ISR continues by getting a new empty buffer with USBD_Audio_RecordBufGet() . A new DMA transfer is prepared and started. If no empty record buffer is available after calling USBD_Audio_RecordBufGet(), that is a null pointer is returned, you may have to get some record data using a dummy buffer to keep the driver in sync with audio class, that is you want to continue receiving DMA transfer completion interrupt to re-attempt to get an empty buffer. The record data stored in the dummy buffer is basically lost. The dummy buffer can be allocated in the function USBD_Audio_DrvInit() . | Tip |
|---|
The audio peripheral driver should support at least the double-buffering to optimize the record stream processing. |
| Tip |
|---|
The DMA implementation in this example processes the record buffers one after the other using a single interrupt. Depending on your DMA controller, it may be possible to optimize the performance by using several DMA channels. In that case, you could pipeline the DMA transfers. The DMA controller may also offer to link DMA descriptors. In that case, you could obtain several empty record buffers with USBD_Audio_RecordBufGet() and link several DMA descriptors. |
(4) Upon reception of the signal, the record task will call the function USBD_Audio_DrvStreamRecordRx . It will get a ready buffer from the driver's internal buffer storage and submit it to the USB side. (5) The host decides to stop the stream. The function USBD_Audio_DrvStreamStop() is called. You should abort any ongoing DMA transfers. You can also disable the record operation on the codec if it is needed. |
...
As described in the section Audio Statistics, the audio class offered some stream statistics that may be useful during your development. An audio statistics structure ( USBD_AUDIO_STAT ) specific to each AS interface can be retrieved by the application and consulted during debugging. 106070237 describes all the fields of USBD_AUDIO_STAT. Among them, there are four interesting for the driver:
...