AURIX™ MCU: Data formats supported by Signal Processing Unit (SPU) – KBA237912
Community Translated: AURIX™ MCU: 信号処理ユニット（SPU）がサポートするデータフォーマット - KBA237912
The Signal Processing Unit (SPU) native format is a 32-bit integer. However, the SPU can read and write FFT input and output data in the following formats:
- 16-bit precision integer
- 32-bit precision integer
- 16-bit half-precision floating point
- The conversion to and from half precision floating point is automatic using the algorithm defined in the user manual, see “126.96.36.199.4 Half Precision Floating Point Format” section when setting the bit BEx_ODP_CONF (x=0-1).HPFP to 1B.
- The conversion from 32-bit to 16-bit integer requires that the ODP_CONF.SCALE bitfield be set to a value that prevents overflow while preventing unnecessary loss of precision. The statistical and histogram functions of the SPU provide information that can be useful in deciding what value to use in this field.
- The conversion back from 16-bit to 32-bit precision on read is controlled using the LDR_CONF.EXPNT bitfield as described in 188.8.131.52.1 Data Reformatting” section.
Take note of the following:
- The floating-point support was added primarily as a method of data compression (and the internal 32-bit precision integer does not enable full use of the range offered by single precision floating point).
- The different sub-modules within the SPU will expand the data read from any buffer into a 32-bit precision integer before using it in any of their different functions.
- The setting of the following bitfields should be used to avoid conflict when processing the data:
- Leave the EXPNT bitfield as the default when setting HPFP.
- Do not set the SCALE bitfield when HPFP is set.
Note: This KBA applies to the following series:
- AURIX™ TC3xx series