NMSIS-DSP  Version 1.2.0
NMSIS DSP Software Library
Standard deviation

Calculates the standard deviation of the elements in the input vector. More...

Functions

void riscv_std_f16 (const float16_t *pSrc, uint32_t blockSize, float16_t *pResult)
 Standard deviation of the elements of a floating-point vector. More...
 
void riscv_std_f32 (const float32_t *pSrc, uint32_t blockSize, float32_t *pResult)
 Standard deviation of the elements of a floating-point vector. More...
 
void riscv_std_f64 (const float64_t *pSrc, uint32_t blockSize, float64_t *pResult)
 Standard deviation of the elements of a floating-point vector. More...
 
void riscv_std_q15 (const q15_t *pSrc, uint32_t blockSize, q15_t *pResult)
 Standard deviation of the elements of a Q15 vector. More...
 
void riscv_std_q31 (const q31_t *pSrc, uint32_t blockSize, q31_t *pResult)
 Standard deviation of the elements of a Q31 vector. More...
 

Detailed Description

Calculates the standard deviation of the elements in the input vector.

The float implementation is relying on riscv_var_f32 which is using a two-pass algorithm to avoid problem of numerical instabilities and cancellation errors.

Fixed point versions are using the standard textbook algorithm since the fixed point numerical behavior is different from the float one.

Algorithm for fixed point versions is summarized below:

    Result = sqrt((sumOfSquares - sum2 / blockSize) / (blockSize - 1))
    sumOfSquares = pSrc[0] * pSrc[0] + pSrc[1] * pSrc[1] + ... + pSrc[blockSize-1] * pSrc[blockSize-1]
    sum = pSrc[0] + pSrc[1] + pSrc[2] + ... + pSrc[blockSize-1]

There are separate functions for floating point, Q31, and Q15 data types.

Function Documentation

◆ riscv_std_f16()

void riscv_std_f16 ( const float16_t *  pSrc,
uint32_t  blockSize,
float16_t *  pResult 
)

Standard deviation of the elements of a floating-point vector.

Parameters
[in]pSrcpoints to the input vector
[in]blockSizenumber of samples in input vector
[out]pResultstandard deviation value returned here
Returns
none

◆ riscv_std_f32()

void riscv_std_f32 ( const float32_t *  pSrc,
uint32_t  blockSize,
float32_t *  pResult 
)

Standard deviation of the elements of a floating-point vector.

Parameters
[in]pSrcpoints to the input vector
[in]blockSizenumber of samples in input vector
[out]pResultstandard deviation value returned here
Returns
none

◆ riscv_std_f64()

void riscv_std_f64 ( const float64_t *  pSrc,
uint32_t  blockSize,
float64_t *  pResult 
)

Standard deviation of the elements of a floating-point vector.

Parameters
[in]pSrcpoints to the input vector
[in]blockSizenumber of samples in input vector
[out]pResultstandard deviation value returned here
Returns
none

◆ riscv_std_q15()

void riscv_std_q15 ( const q15_t *  pSrc,
uint32_t  blockSize,
q15_t *  pResult 
)

Standard deviation of the elements of a Q15 vector.

Parameters
[in]pSrcpoints to the input vector
[in]blockSizenumber of samples in input vector
[out]pResultstandard deviation value returned here
Returns
none
Scaling and Overflow Behavior
The function is implemented using a 64-bit internal accumulator. The input is represented in 1.15 format. Intermediate multiplication yields a 2.30 format, and this result is added without saturation to a 64-bit accumulator in 34.30 format. With 33 guard bits in the accumulator, there is no risk of overflow, and the full precision of the intermediate multiplication is preserved. Finally, the 34.30 result is truncated to 34.15 format by discarding the lower 15 bits, and then saturated to yield a result in 1.15 format.

◆ riscv_std_q31()

void riscv_std_q31 ( const q31_t *  pSrc,
uint32_t  blockSize,
q31_t *  pResult 
)

Standard deviation of the elements of a Q31 vector.

Parameters
[in]pSrcpoints to the input vector.
[in]blockSizenumber of samples in input vector.
[out]pResultstandard deviation value returned here.
Returns
none
Scaling and Overflow Behavior
The function is implemented using an internal 64-bit accumulator. The input is represented in 1.31 format, which is then downshifted by 8 bits which yields 1.23, and intermediate multiplication yields a 2.46 format. The accumulator maintains full precision of the intermediate multiplication results, but provides only a 16 guard bits. There is no saturation on intermediate additions. If the accumulator overflows it wraps around and distorts the result. In order to avoid overflows completely the input signal must be scaled down by log2(blockSize)-8 bits, as a total of blockSize additions are performed internally. After division, internal variables should be Q18.46 Finally, the 18.46 accumulator is right shifted by 15 bits to yield a 1.31 format value.