Using the STM32F10x DSP library FIR

[denis]

Posted on October 29, 2013 at 16:57

I'm unsure on how to use the FIR interface. From the provided example:

int a[N];/*filter output vector*/

short x[M+N-1] = {x0,x1...,xM+N-1};/*filter input vector*/

short h[M]={h0,h1...,hM-1};/*filter coefficients vector*/

fir_coefs.nh = M; /*Number of Coefficients for FIR*/

fir_coefs.h = h; /*Pointer on FIR coefficient vector*/

fir_16by16_stm32(a,x,&fir_coefs,N);/*performs the FIR filtering*/

Suppose I only have 4 samples (the minimum allowed),  I put them into the first 4 elements of x, and call the fir func.

But if i have continuous stream, can i just load again the next 4 same way?

Trying now but I see wrong stuff seems to come out.

(I find the ARM math library fir interface is more clearer in this respect, it keeps the state and input separate)

[denis]

Posted on November 01, 2013 at 16:26

answering myself,

if using in continuous mode you have to shift delay line yourself, the lib doesnt do it :\

and i only saw now that  the latest/newest sample has to go first in input array array,  and same for output (latest is at 0).

[denis]

Posted on November 08, 2013 at 10:33

If you commend out the f10xxx includes, then it also builds and runs on other models.

I'm currently using it on stm32l15 device .

[frankmeyer9]

Posted on November 08, 2013 at 11:03

The DSP_lib core is basically vendor and device agnostic. It is provided by ARM, and included by ST into its library framework, as done by others (NXP, Infineon, EnergyMicro, ...).

This core DSP functions only differ between M0, M3 and M4, by way of conditional build.

[denis]

Posted on November 14, 2013 at 14:58

Decided to share on using the STM's FIR in my project, vs the simplest (dumb) FIR implementation.      (Still hope to improve on the use if the lib, and not sure using it best way...)

///Performance

(=cycles)

of

FIR

the

dumb

way,

vs

STM's

DSPLib

optimized

FIR

///

|---------------------------------------------------|

///

|

STM32L152xB,

8

MHz,

40

taps,

4

samples

|

///

|

Simple(dumb)

FIR:

|

///

|

Dbg(low

opt)

:

cycles

=

4321

<->

uS

=

540

|

///

|

Rel(full

opt)

:

cycles

=

2819

<->

uS

=

352

|

///

|

STM

FIR:

|

///

|

Dbg(low

opt)

:

cycles

=

1276

<->

uS

=

159

|

///

|

Rel(full

opt)

:

cycles

=

1102

<->

uS

=

137

|

///

|---------------------------------------------------|

///Compiler:

IAR

6.60.2.5449

 // The simplest FIR

//

int

ii;

//

int

accsum;

//

for

(int

i=0;

i

<

NSMPLES;

++i)

//

{

//

shift_arr_[0]

=

smpls_P[i];

//

accsum

=

0;

//

//FIR

//

for

(ii=0;

ii

<

N_TAPS_;

++ii)

//

accsum

+=

static_cast<int32_t>(FIR_FILTER_TAPS_[ii])

*

shift_arr_[ii];

//

output_arr_[i]

=

static_cast<int16_t>(

(accsum

+

HALF)

>>FBITS

);

//

//Shift-delay

//

for

(ii

=

N_TAPS_-1

-

1;

ii

>=0;

--ii)

//

shift_arr_[ii

+

1]

=

shift_arr_[ii];

//

}

//

result_

=

output_arr_[NSMPLES-1];

// FIR with STM lib

//Copy

samples

into

the

shift/intput

arr.

(TODO:

rem

this

copying

by

shift_arr_

->

smpls_P)

std

::

memcpy

(

shift_arr_

,

smpls_P

,

NSMPLES

*

sizeof

*

smpls_P

);

void

*

vp

=

&

COEFFS_

;

fir_16by16_stm32

(

output_arr_

,

shift_arr_

,

static_cast

<

COEFS

*>(

vp

),

NSMPLES

);

//LAME:

have

to

shift

delay

line

ourselvs!

(TODO:

can't

just

modify

that

ST's

asm

?)

for

(

int

ii

=

(

N_TAPS_

+

NSMPLES

-

1

-

1

)

-

NSMPLES

;

ii

>=

0

;

--

ii

)

shift_arr_

[

ii

+

NSMPLES

]

=

shift_arr_

[

ii

];

//Scale

back

all

results.

for

(

int

i

=

0

;

i

<

NSMPLES

;

++

i

)

output_arr_

[

i

]

=

static_cast

<

int16_t

>(

(

output_arr_

[

i

]

+

HALF

)

>>

FBITS

);

//Note:

most

recent

is

at

0.

See

also

filter

init/ctor:

reversed

sample

order.

result_

=

output_arr_

[

0

];

I hope to take their asm fir and modify it so that I don't have to extra shift

delay, outside of their func