using Hercules DSP library

Miguel Amezcua Peregrina

Other Parts Discussed in Thread: HALCOGEN

I have been struggling using the DSP library examples...

Does anyone has any idea of how to compile them to test them?? I have been trying to open the examples but I cannot open the projects from CCS....

thanks

over 9 years ago

0 Bob Crosby over 9 years ago

TI__Guru 72500 points

I downloaded, imported and did a build in CCS6.1. It almost worked, but had a hardcoded file location for the archiver to put the library. I changed that, but now am hitting what I think is a makefile difference between CCSv6 and CCSv5. I will continue to look at this tomorrow.

0 Miguel Amezcua Peregrina over 9 years ago in reply to Bob Crosby

Genius 4015 points

Hi Bob

amm how you do that? I tried that. Exactly that. And did not work...

regards

0 Miguel Amezcua Peregrina over 9 years ago in reply to Bob Crosby

Genius 4015 points

Hi Bob
ammmm for using the FFT to obtain the median power frequency... how should it work??
I saw that the codes are made for a 1024 input data, but I have 5200... how should I scale it?
Thanks!!

0 Anthony F. Seely over 9 years ago in reply to Miguel Amezcua Peregrina

TI__Guru 69070 points

Miguel,

So you want to run a 5200 point FFT instead of a 1024pt like the examples? Is that right?

-Anthony

0 Miguel Amezcua Peregrina over 9 years ago in reply to Anthony F. Seely

Genius 4015 points

Yes. that right. My data is in a 5200 size buffer .

And obtain the median frequecy

Thanks Anthony

0 Anthony F. Seely over 9 years ago in reply to Miguel Amezcua Peregrina

TI__Guru 69070 points

So you're going to need to do some math work first, because the CMSIS library only supports up to 1024 (complex) point FFTs. Unfortunately we don't have anything that is canned and supports longer FFTs.

You could either code your own transform algorithm for 5200 points, or use the CMSIS library smaller FFTs to build up to a larger DFT result as this article describes: www.dsprelated.com/.../63.php for example.

You might also want to look at http://www.fftw.org/ for ideas; however I don't know of anyone who has run this code on a Hercules. Still I think they look at composing larger FFTs from smaller ones for practical reasons like fitting in caches.

I'd also strongly suggest doing the algorithm work in something like MATLAB or Octave or Python first ... then when you are happy w. the result port to the Hercules.

0 Miguel Amezcua Peregrina over 9 years ago in reply to Anthony F. Seely

Genius 4015 points

I could, without problem, divide the data in different groups and process them and then obtain the median power freq. with them. But does the library gives that? Or i should obtain that with the data the fft gives me??
Thanks!!!

0 Anthony F. Seely over 9 years ago in reply to Miguel Amezcua Peregrina

TI__Guru 69070 points

Miguel,

Look at the file: file:///C:/ti/Hercules/Cortex-R4%20CMSIS%20DSP%20Library/1.0.0/Docs/html/modules.html

(you may need to adjust the path but above is the default install path).

That gets you a list of the functions supported by the library and if you read the description, there are only a small number of supported lengths for the FFTs.

0 Miguel Amezcua Peregrina over 9 years ago in reply to Anthony F. Seely

Genius 4015 points

Yeah... I already looked at it.
I was thinking on using different pointers. I explain myself.
I have data[5120] , which is where I saved the data.
then:
if(i==0)
{
fft function (data[0])
i++;
}
if(i==1)
{
fft function (data[1023])
i++;
}
if(i==2)
{
fft function (data[2047])
i++;
}
if(i==3)
{
fft function (data[3071])
i++;
}
if(i==4)
{
fft function (data[4095])
i=0;
}

So the input data comes from a pointer to a different location from the same arrangement.
what do you think??
Thanks!!

0 Miguel Amezcua Peregrina over 9 years ago in reply to Miguel Amezcua Peregrina

Genius 4015 points

Hi Anthony

maybe I had a confusion but... looking at the example, I saw the data is 2048 samples.

But the FFT is supposed to only handle 1024 data no?? or the sample data has no limit??? Is the output the one that has a limit??

0 Anthony F. Seely over 9 years ago in reply to Miguel Amezcua Peregrina

TI__Guru 69070 points

Miguel,

That's a 1024 pt complex FFT which is the max that the library handles.

You see 2048 floats in the input array because the input is complex (real + imaginary).

The key is the 'fftSize' parameter that is passed to arm_cfft_radix4_init_f32...

-Anthony

0 Anthony F. Seely over 9 years ago in reply to Miguel Amezcua Peregrina

TI__Guru 69070 points

Hi Miguel,

I really can't comment on your proposal without going through all the math myself; and I also can't tell how you intend to combine the results.

I can tell you that this is a different way than the DSPRelated article talks about - because there he is taking first the even and then the odd samples, running the FFT on both of them, and then combining the results of the FFTs to get a double length output vector. So all of the frequency bins of the 16 point DFT wind up getting computed, using 2 8 point FFT's plus some math.

He also shows a 24-point FFT (not power of 2) by doing 3 8 pt. and combining.

Also your 5200 length isn't the nicest, if you factor this into a power of 2 then its: 5200 = 16 * 325.
I don't think you would want to run 325 16-point FFTs and combine the results..

Can you change 5200 to say 5120 which is 1024 * 5? That would work nicely with the CMSIS DSP library constraints.

Again though - I would start working on this problem in some tool like Matlab first, to make sure I understood the algorithm and tradeoffs. Only after that - then would I try to code it for the MCU.
Just a lot faster that way... And if you go to matlab you can use the FFTW library implementation to compare your own decomposition against....

-Anthony

0 Miguel Amezcua Peregrina over 9 years ago in reply to Anthony F. Seely

Genius 4015 points

I was thinking on reducing the data to 5120. But then i need to do the orocessing 5 times right? Or the data lenght doesnt matter??

0 Anthony F. Seely over 9 years ago in reply to Miguel Amezcua Peregrina

TI__Guru 69070 points

Hi Miguel,

Well I just think that 5120 will simplify a bit decomposing the problem, say into 5 1024-pt FFTs.
But again I don't have the math down cold on this.

If you can make this compromise though I think that it removes one technical issue which is related to the oddball size of the 5200 pt DFT when trying to implement with an FFT algorithm.

0 Miguel Amezcua Peregrina over 9 years ago in reply to Anthony F. Seely

Genius 4015 points

Yes. I could use the pseudocode on the other post and then concatenate the results and obtain the max and everything. I don't need the complex magnitud since my signal is completely real so... no need for that

I will run some tests and let you know

thanks!

0 Miguel Amezcua Peregrina over 9 years ago in reply to Miguel Amezcua Peregrina

Genius 4015 points

Hi Anthony

Hope you're doing great

I already used the FFT function without problem

Now Im using the FIR example on my code. But I have this error every time I include the arm_math.h and all the other files from the include folder:

I have no idea of whats going on... (all the errors come from arm_math.h)

regards

0 Miguel Amezcua Peregrina over 9 years ago in reply to Anthony F. Seely

Genius 4015 points

Hi Anthony
I checked the compilation errors and all come from the type_def.h or the arm_math.h but I don't know whats going on... I was able to compile the fft example but when trying to use the libraries on my project the compilation gave me those errors...
any idea??
thanks!

0 Anthony F. Seely over 9 years ago in reply to Miguel Amezcua Peregrina

TI__Guru 69070 points

Miguel,

Did you make sure to set up your project with the required Predefined Symbols (-D compiler option).
[if not sure, please refer to the documentation file: file:///C:/ti/Hercules/Cortex-R4%20CMSIS%20DSP%20Library/1.0.0/Docs/html/index.html]

0 Miguel Amezcua Peregrina over 9 years ago in reply to Anthony F. Seely

Genius 4015 points

okok let me check that part =)

0 Miguel Amezcua Peregrina over 9 years ago in reply to Anthony F. Seely

Genius 4015 points

No Anthony. I tried but nothing....
I checked the file you told me but... I ended looking at the build options to look for that but nothing.
quick question.... Im working on mac right now.... could that be the problem???

thanks!

0 Anthony F. Seely over 9 years ago in reply to Miguel Amezcua Peregrina

TI__Guru 69070 points

Hi Miguel,

What do you mean exactly? That you don't have any predefined symbols, that you do have them correct, or that you cannot find the documentation file that I linked to?

Yes the file I linked to is a PC path and if you are on the MAC you need to find the equivalent path. It should be part of the hercules CMSIS DSP library.

Please do confirm you downloaded the Hercules CMSIS DSP library from TI.com and not ARM because what ARM distributes is only for Cortex M.

Hercules development on Mac isn't supported fully nor is it on Linux. CCS works on these platforms but a lot of the other installers will only work on windows (for example HET IDE, HalCoGen, and this CMSIS DSP library...) So I would recommend against MAC if you can help it.

-Anthony

0 Miguel Amezcua Peregrina over 9 years ago in reply to Anthony F. Seely

Genius 4015 points

ammm I already copied the files from the library into Mac in order to make easier for me to look for them. ammm I did find the document you told me. And I linked the path for the include to the .lib I needed.

I downloaded the library from TI.com

Let me try on the weekend to compile it on windows to see if its a problem with installation or something and let you know

Thanks!

0 Anthony F. Seely over 9 years ago in reply to Miguel Amezcua Peregrina

TI__Guru 69070 points

Yes but you need to add things to the project build options,
for example you need to add '-D CCS' to the compiler command line to build on CCS.

The documentation file tells you what symbols need to be defined.

0 Miguel Amezcua Peregrina over 9 years ago in reply to Anthony F. Seely

Genius 4015 points

I tried this:

and I tried this too

didnt work :/

0 Miguel Amezcua Peregrina over 9 years ago in reply to Anthony F. Seely

Genius 4015 points

ammmm how I do that part???
thanks

0 Miguel Amezcua Peregrina over 9 years ago in reply to Anthony F. Seely

Genius 4015 points

It already compiled. I saw where to put the CCS part you were saying.
I got some warnings because the types were already defined, on hal_stdtypes or somewhere else, but I don't think that could be a problem
Thanks Anthony

0 Anthony F. Seely over 9 years ago in reply to Miguel Amezcua Peregrina

TI__Guru 69070 points

I don't know, I haven't used CCS with a MAC before. I'd *assume* it looks the same in terms of where the settings are organizationally and maybe just the visual effects are different.

On the PC, here's what I'm talking about from a project I used with the CMSIS DSP library. It was for a floating point, big endian TMS570 so that's why I defined FPU_PRESENT as well as ARM_MATH_BIG_ENDIAN [in addition to CCS]

Arm-based microcontrollers

Arm-based microcontrollers forum

using Hercules DSP library