PIC32 Standard Libraries Help

PIC32 DSP Library

Overview
The PIC32 DSP library consists of a set of functions that are applicable to many multimedia
application areas. Most of the functions, like vector operations, filters, and transforms, are
commonly used in many DSP and multimedia applications.
Some functions are designed to be used in specific applications such as video decoding or voice
compression. It is beyond the scope of this manual to describe the operation of such applications.
Functions whose performance is considered critical are implemented in assembly and tuned
where appropriate for a particular processor pipeline implementation and instruction set
features. When a function is typically not considered to be performance critical, or the benefit
from an assembly implementation is not significant, it is implemented in C. Often such functions
perform initialization of data structures and are used only once during the lifetime of an
application.
Table: General Purpose DSP Library Functions by Category lists all the functions currently
available in the DSP Library, arranged by category, with the available implementation versions.
All general purpose functions work with data in 16-bit fractional format, also known as Q15.
Some of the functions also have a version that operates on 32-bit data in Q31 fractional format.

TABLE: GENERAL PURPOSE DSP LIBRARY FUNCTIONS BY

CATEGORY
Category

Function Name

Description

mips_vec_abs16/32

Compute the absolute value of

each Q15/Q31 vector element.

mips_vec_add16/32

Add the corresponding

elements of two Q15/Q31
vectors.

mips_vec_addc16/32

Add a constant to all elements

of a vector.

Vector Math
Functions mips_vec_dotp16/32

Compute dot product of two

Q15/Q31 vectors.

mips_vec_mul16/32

Multiply the corresponding

elements of two Q15/Q31
vectors. Can be used for
applying windows.

mips_vec_mulc16/32

Multiply all elements of a

vector by a constant.

mips_vec_sub16/32

Subtract the corresponding

Fi
elements of two Q15/Q31
vectors.

mips_vec_sum_squares16/32 Calculate the sum of squares

of elements of a vector in
Q15/Q31 format.

Filters

Transforms

mips_fir16

Applies a block FIR filter to a

Q15 vector.

mips_fir16_setup

Prepare the filter coefficients

for the mips_fir16 function.

mips_iir16

Single-sample IIR filter.

mips_iir16_setup

Prepare the filter coefficients

for the mips_iir16 function.

mips_lms16

Single-sample LMS filter

mips_fft16

Compute the complex FFT of

a vector containing Q15
complex samples, i.e., 16-bit
fractional real and imaginary
parts.

mips_fft16_setup
(deprecated)

Create a vector of twiddle

factors used by the mips_fft16
function.

mips_fft32

Compute the complex FFT of

a vector containing Q31
complex samples, i.e., 32-bit
fractional real and imaginary
parts.

mips_fft32_setup
(deprecated)

Create a vector of twiddle

factors used by the mips_fft32
function.

mips_h264_iqt

Inverse quantization and

transform for H.264 decoding.

mips_h264_iqt_setup

Create inverse quantization

matrix used by the
mips_h264_iqt function.

mips_h264_mc_luma

1/4-pixel motion compensation

for luma pixels in H.264 video
decoding.

Video

Fxed-Point Types
Input and output data for most functions is represented in 16-bit fractional numbers, in Q15
format. This is the most commonly used data format for signal processing. Some function may
use other data formats internally for increased precision of the intermediate results. The Q15 data
type used by the DSP functions is specified as int16 in the C header files supplied with the
library. This data type is defined in the common dsplib_def.h header file.
Note that within C code care must be taken not to confuse fixed-point values with integers. To
the C compiler, objects declared with int16 type are integers, not fixed-point, and any arithmetic
performed on those objects in C will be done as integers. Fixed-point values have been declared
as int16 only because the standard C language does not include intrinsic support for fixed-point
data types.
Saturation, Scaling, and Overflow
In the majority of DSP applications, overflow or underflow during computation is not desirable.
It is best to design for appropriate scaling of the data path and avoid the possibility of overflow
and underflow. However, such scaling can significantly limit the usable data range. Hence, many
algorithm implementations relax the scaling and introduce saturation operations that clip the
values that would otherwise overflow to the maximum or minimum limit of the data range.
Some of the general purpose DSP library module functions accumulate a series of values before
producing the final result. Examples of these accumulations could include the vector dot product
calculation, the FIR filter, the sum of squared values and even the FFT transform. All of these
functions, with the exception of the FFT, include a parameter that controls the output scaling,
i.e., additional amount of right shift applied when the result is converted to a Q15 value. The
FFT results are automatically scaled down by 2 log2(N).
Array Alignment and Length Restrictions
For the sake of efficiency, most functions require that array pointer arguments are aligned on 4byte boundaries. Arrays of the int16 data type declared in C will be correctly aligned.
Furthermore, there are often restrictions on the number of elements that a function can operate
on. Typically the number of elements must be a multiple of a small integer (e.g., four or eight),
and must be larger than, or equal to, a specified minimum. Note that to improve performance, the
functions do not verify the validity of their input parameters. Supplying incorrect parameters
may lead to unpredictable results.

VECTOR MATH FUNCTIONS

mips_vec_abs16
Description:

Computes the absolute value of each element of indata and

stores it to outdata. The number of samples to be processed
is given by the parameter N.
Mathematically,
outdata[n] = abs(indata[N])

Include:

dsplib_dsp.h

Prototype:

void
mips_vec_abs16
(
int16 *outdata,
int16 *indata,
int N
);

Argument:

outdata: Output array of 16-bit fixed-point elements in Q15

format.
indata: Input array with 16-bit fixed-point elements in Q15
format.
N: Number of samples.

Return
Value:

None.

mips_vec_abs32
Description: Computes the absolute value of each element of indata and
stores it to outdata. The number of samples to be processed
is given by the parameter N.
Mathematically,
outdata[n] = abs(indata[N])
Include:

dsplib_dsp.h

Prototype:

void
mips_vec_abs32
(
int32 *outdata,
int32 *indata,
int N
);

Argument:

outdata: Output array of 32-bit fixed-point

elements in Q31 format.

indata: Input array with 32-bit fixed-point

elements in Q31 format.
N: Number of samples.
Return
Value:

None.

Remarks:

The pointers outdata and indata must be aligned on

4-byte boundaries.
N must be larger than or equal to 4 and a multiple of
4.

mips_vec_add16
Description:

Adds each element of indata1 to the corresponding element

of indata2. The number of samples to be processed is given
by the parameter N.
Mathematically,
outdata[n] = indata1[n]+indata2[n]

Include:

dsplib_dsp.h

Prototype:

void
mips_vec_add16
(
int16 *outdata,
int16 *indata1,
int16 *indata2,
int N
);

Argument:

outdata : Output array of 16-bit fixed-point

elements in Q15 format.
indata1 : First input array with 16-bit fixedpoint elements in Q15 format.
indata2 : Second input array with 16-bit fixedpoint elements in Q15 format.
N : Number of samples.

Return
Value:

None.

Remarks:

The pointers outdata, indata1, and indata2 must be

aligned on 4-byte boundaries.
N must be larger than or equal to 4 and a multiple of
4.

mips_vec_add32
Description:

Adds each element of indata1 to the corresponding element

of indata2. The number of samples to be processed is given
by the parameter N.
Mathematically,
outdata[n] = indata1[n]+indata2[n]

Include:

dsplib_dsp.h

Prototype:

void
mips_vec_add32
(
int32 *outdata,
int32 *indata1,
int32 *indata2,
int N
);

Argument:

outdata : Output array of 32-bit fixed-point

elements in Q31 format.
indata1 : First input array with 32-bit fixedpoint elements in Q31 format.
indata2 : Second input array with 32-bit fixedpoint elements in Q31 format.
N : Number of samples.

Return
Value:

None.

Remarks:

The pointers outdata, indata1, and indata2 must be

aligned on 4-byte boundaries.
N must be larger than, or equal to, 4, and a multiple
of 4.

mips_vec_addc16
Description:

Adds the Q15 constant c to all elements of indata. The

number of samples to be processed is given by the parameter
N.
Mathematically,
outdata[n] = indata[n]+c

Include:

dsplib_dsp.h

Prototype:

void
mips_vec_addc16
(
int16 *outdata,

int16 *indata,
int16 c,
int N
);

Argument:

outdata : Output array of 16-bit fixed-point

elements in Q15 format.
indata : Input array with 16-bit fixed-point
elements in Q15 format.
c : Constant added to all elements of the vector.
N : Number of samples.

Return
Value:

None.

Remarks:

The pointers outdata and indata must be aligned on

4-byte boundaries.
N must be larger than or equal to 4 and a multiple of
4.

mips_vec_addc32
Description:

Adds the Q31 constant c to all elements of indata. The

number of samples to be processed is given by the parameter
N.
Mathematically,
outdata[n] = indata[n]+c

Include:

dsplib_dsp.h

Prototype:

void
mips_vec_addc32
(
int32 *outdata,
int32 *indata,
int32 c,
int N
);

Argument:

outdata : Output array of 32-bit fixed-point

elements in Q31 format.
indata : Input array with 32-bit fixed-point
elements in Q31 format.
c : Constant added to all elements of the vector.
N : Number of samples.

Return

None.

Value:
Remarks:

The pointers outdata and indata must be aligned on

4-byte boundaries.
N must be larger than or equal to 4 and a multiple of
4.

mips_vec_dotp16
Description: Computes the dot product of the Q15 vectors indata1 and
indata2. The number of samples to be processed is given by the
parameter N. The scale parameter specifies the amount of right
shift applied to the final result.
Mathematically,

Include:

dsplib_dsp.h

Prototype: int16

mips_vec_dotp16
(
int16 *indata1,
int16 *indata2,
int N,
int scale
);

Argument:

indata1 : First input array with 16-bit fixed point

elements in Q15 format.
indata2 : Second input array.
N : Number of samples.
scale : Scaling factor: divide the result by 2scale.

Return
Value:

Scaled result of the calculation in fractional Q15 format.

Remarks:

The pointers outdata and indata must be aligned on 4byte boundaries.

N must be larger than or equal to 4 and a multiple of 4.

mips_vec_dotp32
Description: Computes the dot product of the Q31 vectors indata1 and
indata2. The number of samples to be processed is given by
the parameter N. The scale parameter specifies the amount of
right shift applied to the final result.
Mathematically,

result

N 1

indata_1 indata_2

n
n
scale
2
1

n 0

Include:

dsplib_dsp.h

Prototype: int32

mips_vec_dotp32
(
int32 *indata1,
int32 *indata2,
int N,
int scale
);

Argument:

indata1 : First input array with 32-bit fixed point

elements in Q31 format.
indata2 : Second input array.
N : Number of samples.
scale : Scaling factor: divide the result by 2scale.

Return
Value:

Scaled result of the calculation in fractional Q31 format.

Remarks:

The pointers outdata and indata must be aligned on 4byte boundaries.

N must be larger than or equal to 4 and a multiple of 4.

mips_vec_dotp32
Description: Computes the dot product of the Q31 vectors indata1 and
indata2. The number of samples to be processed is given by
the parameter N. The scale parameter specifies the amount of
right shift applied to the final result.
Mathematically,

result

N 1

indata_1 indata_2

n
n
scale
2
1

n 0

Include:

dsplib_dsp.h

Prototype: int32

mips_vec_dotp32
(
int32 *indata1,
int32 *indata2,
int N,
int scale
);

Argument: indata1 : First input array with 32-bit fixed point elements in
Q31 format.
indata2 : Second input array.
N : Number of samples.
scale : Scaling factor: divide the result by 2scale.
Return
Value:

Scaled result of the calculation in fractional Q31 format.

Remarks:

The pointers outdata and indata must be aligned on 4byte boundaries.

N must be larger than or equal to 4 and a multiple of 4.

mips_vec_mul16
Description:

Multiplies each Q15 element of indata1 by the

corresponding element of indata2 and stores the results to
outdata. The number of samples to be processed is given by
the parameter N.
Mathematically,
outdata[n] = indata[n] x indata2[n]

Include:

dsplib_dsp.h

Prototype:

void
mips_vec_mul16
(
int16 *outdata,
int16 *indata1,
int16 *indata2,
int N
);

Argument:

outdata : Output array of 16-bit fixed-point

elements in Q15 format.
indata1 : First input array with 16-bit fixedpoint elements in Q15 format.
indata2 : Second input array.
N : Number of samples.

Return
Value:

None.

Remarks:

The pointers outdata, indata1, and indata2 must be

aligned on 4-byte boundaries.
N must be larger than or equal to 4 and a multiple of
4.

mips_vec_mul32
Description:

Multiplies each Q31 element of indata1 by the

corresponding element of indata2 and stores the results to
outdata. The number of samples to be processed is given by
the parameter N.
Mathematically,
outdata[n] = indata1[n] x indata2[n]

Include:

dsplib_dsp.h

Prototype:

void
mips_vec_mul32
(

int32 *outdata,
int32 *indata1,
int32 *indata2,
int N
);

Argument:

outdata : Output array of 32-bit fixed-point

elements in Q31 format.
indata1 : First input array with 32-bit fixedpoint elements in Q31 format.
indata2 : Second input array.
N : Number of samples.

Return
Value:

None.

Remarks:

The pointers outdata, indata1, and indata2 must be

aligned on 4-byte boundaries.
N must be larger than or equal to 4 and a multiple of
4.

mips_vec_mulc16
Description:

Multiplies each Q15 element of indata by the Q15

constant c and stores the results to outdata. The number
of samples to be processed is given by the parameter N.
Mathematically,
outdata[n] = indata1[n] x c

Include:

dsplib_dsp.h

Prototype:

void
mips_vec_mulc16
(
int16 *outdata,
int16 *indata,
int16 c,
int N
);

Argument:

outdata : Output array of 16-bit fixed-point

elements in Q15 format.
indata : Input array with 16-bit fixed-point
elements in Q15 format.
c : 16-bit fixed-point constant.
N : Number of samples.

Return
Value:

None.

Remarks:

The pointers outdata and indata must be

aligned on 4-byte boundaries.
N must be larger than or equal to 4 and a
multiple of 4.

mips_vec_mulc32
Description:

Multiplies each Q31 element of indata by the Q31 constant

c and stores the results to outdata. The number of samples to
be processed is given by the parameter N.
Mathematically,
outdata[n] = indata1[n] x c

Include:

dsplib_dsp.h

Prototype:

void
mips_vec_mulc32
(
int32 *outdata,
int32 *indata,
int32 c,
int N
);

Argument:

outdata : Output array of 32-bit fixed-point

elements in Q31 format.
indata : Input array with 32-bit fixed-point
elements in Q31 format.
c : 32-bit fixed-point constant.
N : Number of samples.

Return
Value:

None.

Remarks:

The pointers outdata and indata must be aligned on

4-byte boundaries.
N must be larger than or equal to 4 and a multiple of
4.

mips_vec_sub16
Description:

Subtracts each element of indata2 from the corresponding

element of indata1. The number of samples to be processed
is given by the parameter N.
Mathematically,
outdata[n] = indata1[n] indata2[n]

Include:

dsplib_dsp.h

Prototype:

void
mips_vec_sub16
(
int16 *outdata,
int16 *indata1,
int16 *indata2,
int N
);

Argument:

outdata : Output array of 16-bit fixed-point

elements in Q15 format.
indata1 : First input array with 16-bit fixedpoint elements in Q15 format.
indata2 : Second input array with 16-bit fixedpoint elements in Q15 format.
N : Number of samples.

Return
Value:

None.

Remarks:

The pointers outdata, indata1, and indata2 must be

aligned on 4-byte boundaries.
N must be larger than or equal to 4 and a multiple of
4.

mips_vec_sub32
Description:

Subtracts each element of indata2 from the corresponding

element of indata1. The number of samples to be processed
is given by the parameter N.
Mathematically,
outdata[n] = indata1[n] indata2[n]

Include:

dsplib_dsp.h

Prototype:

void
mips_vec_sub32
(
int32 *outdata,

int32 *indata1,
int32 *indata2,
int N
);

Argument:

outdata : Output array of 32-bit fixed-point

elements in Q31 format.
indata1 : First input array with 32-bit fixedpoint elements in Q31 format.
indata2 : Second input array with 32-bit fixedpoint elements in Q31 format.
N : Number of samples.

Return
Value:

None.

Remarks:

The pointers outdata, indata1, and indata2 must be

aligned on 4-byte boundaries.
N must be larger than or equal to 4 and a multiple of
4.

mips_vec_sum_squares16
Description: Computes the sum of squared values of all elements of
indata. The number of samples to be processed is given by
the parameter N. The scale parameter specifies the amount of
right shift applied to the final result.
Mathematically,

result

N 1

scale

2
indata_1

n 0

Include:

dsplib_dsp.h

Prototype:

int16
mips_vec_sum_squares16
(
int16 *indata,
int N,
int scale
);

Argument:

indata Input array with 16-bit fixed-point

elements in Q15 format

N Number of samples
scale Scaling factor: divide the result by 2scale.
Return
Value:

Scaled result of the calculation in fractional Q15 format.

Remarks:

The pointer indata must be aligned on a 4-byte

boundary.
N must be larger than or equal to 4 and a multiple of
4.

mips_vec_sum_squares32
Description: Computes the sum of squared values of all elements of
indata. The number of samples to be processed is given by
the parameter N. The scale parameter specifies the amount of
right shift applied to the final result.
Mathematically,

result

N 1

scale

indata_1

n 0

Include:

dsplib_dsp.h

Prototype:

int32
mips_vec_sum_squares32
(
int32 *indata,
int N,
int scale
);

Argument:

indata : Input array with 32-bit fixed-point

elements in Q31 format.
N : Number of samples.
scale : Scaling factor: divide the result by 2scale.

Return
Value:

Scaled result of the calculation in fractional Q31 format.

Remarks:

The pointer indata must be aligned on a 4-byte

boundary.
N must be larger than or equal to 4 and a multiple of
4.

FILTERING FUNCTIONS

mips_fir16
Description: Computes a finite impulse response (FIR) filter with
coefficients specified in coeffs2x over the input data samples
in indata. The function updates the delayline, which is used to
initialize the filter the next time mips_fir16() is called. The
number of samples to be processed is given by the parameter
N and the number of filter coefficients is given by K. The
scale parameter specifies the amount of right shift applied to
the final result.
Mathematically, output
n

scale

Include:

K 1

k 0

data_in ( n k ) coeffs k

dsplib_dsp.h

Prototype: void

mips_fir16
(
int16 *outdata,
int16 *indata,
int16 *coeffs2x,
int16 *delayline,
int N,
int K,
int scale
);

Argument:

outdata : Output array with 16-bit fixed-point

elements in Q15 format.
indata : Input array with 16-bit fixed-point
elements in Q15 format.
coeffs2x : Array of 2K 16-bit fixed-point
coefficients prepared by mips_fir16_setup().
delayline : Delay line array holding the last K
input samples.
N : Number of samples.
K : Number of coefficients (filter taps).
scale : Scaling factor: divide the result by 2scale.

Return
Value:

None.

Remarks:

The pointers outdata, indata, coeffs2x, and delayline

must be aligned on a 4-byte boundary.
K must be larger than or equal to 4 and a multiple of 4.

Notes:

The coeffs2x array is twice the size of the original coefficient

array, coeffs. The function mips_fir16_setup() takes the
original coefficient array coeffs and rearranges the coefficients
into the coeffs2x array to enable more efficient processing. All
elements of the delayline array must be initialized to zero
before the first call to mips_fir16(). Both delayline and
coeffs2x have formats that are implementation-dependent and
their contents should not be changed directly.

Example:

int i;
int K = 8;
int N = 32;
int16 coeffs[K];
int16 coeffs2x[2*K];
int16 delayline[K];
int16 indata[N];
int16 outdata[N];
for (i = 0; i < K; i++)
delayline[i] = 0;
// load coefficients into coeffs here
...
mips_fir16_setup(coeffs2x, coeffs, K);
while (true)
{
// load input data into indata
...
mips_fir16(outdata, indata, coeffs2x,
delayline, N, K, 3);
// do something with outdata
...
}

mips_fir16_setup
Description:

Rearranges the coefficients from the input array, coeffs, into

the output array coeffs2x, which is used by the mips_fir16()
function. The number of coefficients to process is given by
the parameter K.

Include:

dsplib_dsp.h

Prototype:

void
mips_fir16_setup
(
int16 *coeffs2x,
int16 *coeffs,
int K
);

Argument:

coeffs2x : Output array holding 2K coefficients

rearranged for mips_fir16().
coeffs : Input array holding K 16-bit fixed-point
coefficients in Q15 format.
K : Number of coefficients.

Return
Value:

None.

Remarks:

None.

Note:

This function is implemented in C.

mips_iir16
Description: Computes a single-sample infinite impulse response (IIR) filter
with coefficients specified in coeffs. The number of biquad
sections composing the filter is given by the parameter B. The
scale parameter specifies the amount of right shift applied to
the input value of each biquad. Each biquad section is
specified by four coefficients A1, A2, B1, and B2 and has
two state variables stored inside delayline. The output of
each biquad section becomes input to the next one. The output
of the final section is returned as result of the mips_iir16()
function.
The operations performed for each biquad section are
illustrated below:
2-S

Input

Ounput

1/z
B1

-A1

S
1/z
-A2

B2

Include:

dsplib_dsp.h

Prototype: int16

mips_iir16
(
int16 in,
int16 *coeffs,
int16 *delayline,
int B,
int scale
);

Argument: in : Input value in Q15 format.

coeffs : Array of 4B 16-bit fixed-point coefficients prepared by
mips_iir16_setup().
delayline : Delay line array holding 2B state 16-bit state
variables.
B : Number of biquad sections.
scale : Scaling factor: divide the input to each biquad by 2scale.
Return
Value:

IIR filter output value in fractional Q15 format.

Remarks:

T he pointers coeffs and delayline must be aligned on

a 4-byte boundary.
B must be larger than or equal to 2 and a multiple of 2.

Notes:

The coeffs array contains four coefficients for each biquad.

The coefficients are conveniently specified in an array of
biquad16 structures, which is converted to the appropriate
internal representation by the mips_iir16_setup() function. All
elements of the delayline array must be initialized to zero
before the first call to mips_iir16(). Both delayline and coeffs
have formats that are implementation-dependent and their
contents should not be changed directly.

Example:

int i;
int B = 4;
biquad16 bq[B];
int16 coeffs[4*B];
int16 delayline[2*B];
int16 indata, outdata;
for (i = 0; i < 2*B; i++)
delayline[i] = 0;
// load coefficients into bq here
...

mips_iir16_setup(coeffs, bq, B);

while (true)
{
// get input data value into indata
...
outdata = mips_iir16(indata, coeffs,
delayline, B, 2);
// do something with outdata
...
}

mips_iir16_setup
Description:

Rearranges the coefficients from the input array, bq, into the
output array coeffs, which is used by the mips_iir16()
function. The number of biquad sections to process is given
by the parameter B.

Include:

dsplib_dsp.h

Prototype:

void
mips_iir16_setup
(
int16 *coeffs,
biquad16 *bq,
int B
);

Argument:

coeffs : Output array holding 4B coefficients rearranged for

mips_iir16().
bq : Input array holding Q15 coefficients for B biquad
sections.
B : Number of biquad sections.

Return
Value:

None.

Remarks:

None.

Notes:

This function is implemented in C.

mips_lms16
Description: Computes a Least Mean Squares (LMS) adaptive filter and
updates its coefficients. The new coefficients are computed
using the error between the last filter output and the reference
signal ref. The function takes one input sample in and
computes one output sample. The parameter mu controls the
adaptation rate of the filter.
Include:

dsplib_dsp.h

Prototype:

int16
mips_lms16
(
int16 in,
int16 ref,
int16 *coeffs,
int16 *delayline,
int16 *error,
int16 K,
int mu
);

Argument: in : Input value in Q15 format.

ref : Desired (reference) value in Q15 format.
coeffs : Input/output array of 16-bit fixed-point coefficients.
delayline : Delay line array holding the last K input samples.
error : Input/output value indicating the difference between
the filter output and the reference value.
K : Number of coefficients (filter taps).
mu : Adaptation rate in Q15 format.
Return
Value:

LMS filter output value in Q15 format.

Remarks:

The pointers coeffs and delayline must be aligned on

a 4-byte boundary.
K must be larger than or equal to 4 and a multiple of
2.

Notes:

The order of the elements of the coeffs and delayline arrays is

implementation dependent. The delayline array must be
initialized to zero before the first call to mips_lms16().

FREQUENCY DOMAIN TRANSFORM FUNCTIONS

mips_fft16
Description: Computes the complex fast Fourier transform (FFT) of the
input sequence din. The number of samples to be processed is
specified by the parameter log2N: N = 2log2N. The fftc array
holds complex coefficients needed by the FFT algorithm. The
scratch hold intermediate data; its contents are destroyed on
each call to mips_fft16().
Mathematically,
( 2 k ) n

j
N 1

1
N
data_in e

output

n
n

log( 2 N)
2
k 0

Include:

dsplib_dsp.h

Prototype: void

mips_fft16
(
int16c *dout,
int16c *din,
int16c *fftc,
int16c *scratch,
int log2N
);

Argument:

dout : Output array with 16-bit complex fixedpoint elements in Q15 format.
din : Input array with 16-bit complex fixed-point
elements in Q15 format.
fftc : Input array with 16-bit complex fixed-point
twiddle factors in Q15 format.
scratch : Intermediate results array holding 16-bit
complex fixed-point data.
log2N : Logarithm base 2 of the number of
samples: N = 2log2N.

Return

None.

Value:
Remarks:

The pointers dout, din, fftc, and scratch must be

aligned on 4-byte boundaries.
log2N must be larger than or equal to 3.

Notes:

The scratch array must be large enough to hold N 16-bit

complex data samples having 16-bit real part and 16-bit
imaginary part.
Copying fftc to RAM prior to calling this function can be used
to improve performance.

Example:

#include fftc.h // pre-computed coefficients

int log2N = 6; // log2(64) = 6
int N = 1 << log2N; // N = 2^6 = 64
int16c din[N];
int16c dout[N];
int16c scratch[N];
#define fftc fft16c64 // from fftc.h, for N = 64
while (true)
{
// load complex input data into din
...
mips_fft16(dout, din, fftc, scratch, log2N);
// do something with dout
...
}

mips_fft16_setup Function Deprecated

Description:

Calculates the twiddle factors need to compute an FFT of

size N. The twiddle factors are used by the mips_fft16()
function. The number of samples to be processed is
specified by the parameter log2N: N = 2log2N.

Include:

dsplib_dsp.h

Prototype:

void
mips_fft16_setup
(
int16c *twiddles,
int log2N
);

Argument:

twiddles : Output array containing N 16-bit

complex twiddle factors.
log2N : Logarithm base 2 of the number of
samples: N = 2log2N.

Return
Value:

None.

Remarks:

This function requires floating-point support.

Notes:

This function is implemented in C.

mips_fft32
Description: Computes the complex Fast Fourier Transform (FFT) of the
input sequence din. The number of samples to be processed is
specified by the parameter log2N: N = 2log2N. The fftc array
holds complex coefficients needed by the FFT algorithm. The
scratch hold intermediate data; its contents are destroyed on
each call to mips_fft32().
Mathematically,

Include:

dsplib_dsp.h

Prototype: void

mips_fft32
(
int32c *dout,
int32c *din,
int32c *fftc,
int132 *scratch,
int log2N
);

Argument:

dout : Output array with 32-bit complex fixedpoint elements in Q31 format.
din : Input array with 32-bit complex fixed-point
elements in Q31 format.
fftc : Input array with 32-bit complex fixed-point
twiddle factors in Q31 format.
scratch : Intermediate results array holding 32-bit
complex fixed-point data.
log2N : Logarithm base 2 of the number of
samples: N = 2log2N.

Return
Value:

None.

Remarks:

The pointers dout, din, fftc, and scratch must be

aligned on 4-byte boundaries.
log2N must be larger than or equal to 3.

Notes:

The scratch array must be large enough to hold N 32-bit

complex data samples having 32-bit real part and 32-bit

imaginary part.
Copying fftc to RAM prior to calling this function can be used
to improve performance.
Example:

#include fftc.h // pre-computed coefficients

int log2N = 6; // log2(64) = 6
int N = 1 << log2N; // N = 2^6 = 64
int32c din[N];
int32c dout[N];
int32c scratch[N];
#define fftc fft32c64 // from fftc.h, for N = 64
while (true)
{
// load complex input data into din
...
mips_fft32(dout, din, fftc, scratch, log2N);
// do something with dout
...
}

mips_fft32_setup Function Deprecated

Description:

Calculates the twiddle factors need to compute an FFT of

size N. The twiddle factors are used by the mips_fft32()
function. The number of samples to be processed is
specified by the parameter log2N: N = 2log2N.

Include:

dsplib_dsp.h

Prototype:

void
mips_fft32_setup
(
int32c *twiddles,
int log2N
);

Argument:

twiddles : Output array containing N 32-bit complex

twiddle factors.
log2N : Logarithm base 2 of the number of samples: N =
2log2N.

Return
Value:

None.

Remarks:

This function requires floating-point support.

Notes:

This function is implemented in C.