Post by thread:[PIC] PIC24 Filtering code (FIR)?

If you've ever tried to do DSP type functions on a PIC24, you'll find that the standard libraries don't include any sort of filtering functions- there are filtering functions for the dsPICs, but those functions are using the DSP type features of that family- the code doesn't port. I understand it isn't optimal, but has anyone found decent fixed-point FIR (finite impulse response) code for the PIC24? I can generate the FIR coeffecients. I've done a moving average filter [rectangular impulse response], because it is very easy- but I'm just not happy with the results. I could dive into 'Numerical Recipes in C,' but I'm hoping that someone has done/found some slightly more optimized code (for example, using the __builtin_xxx() type functions available in C30/XC16). Regards, Matt Matt Bennett Just outside of Austin, TX 30.51,-97.91 The views I express are my own, not that of my employer, a large multinational corporation that you are familiar with

On Tue, May 22, 2012 at 8:22 AM, Matt Bennett <spam_OUTmattpiclistTakeThisOuThazmat.com> wrote: > If you've ever tried to do DSP type functions on a PIC24, you'll find that > the standard libraries don't include any sort of filtering functions- > there are filtering functions for the dsPICs, but those functions are > using the DSP type features of that family- the code doesn't port. > > I understand it isn't optimal, but has anyone found decent fixed-point FIR > (finite impulse response) code for the PIC24? I can generate the FIR > coeffecients. Â I've done a moving average filter [rectangular impulse > response], because it is very easy- but I'm just not happy with the > results. I could dive into 'Numerical Recipes in C,' but I'm hoping that > someone has done/found some slightly more optimized code (for example, > using the __builtin_xxx() type functions available in C30/XC16). > > Regards, > > Matt Matt, C30 produces pretty good assembly code. A simple FIR filter shouldn't have to be more complicated than rolling the x history and doing the MAC operations in standard C statements. If you need intense optimization to get it done efficiently then I would have to argue that you really need the dsPIC. The PIC24 running full speed should be able to do a LOT of FIR MACs per second, all without any special DSP instructions. I have derived and written FIR and IIR code if you need professional level help. -- Martin K.

The __builtin_xxx() functions are just C shortcuts for the processor's instructions. The builtins that map to DSP instructions won't work in PIC24 devices. Isaac Em 22/5/2012 09:22, Matt Bennett escreveu: {Quote hidden}> If you've ever tried to do DSP type functions on a PIC24, you'll find that > the standard libraries don't include any sort of filtering functions- > there are filtering functions for the dsPICs, but those functions are > using the DSP type features of that family- the code doesn't port. > > I understand it isn't optimal, but has anyone found decent fixed-point FIR > (finite impulse response) code for the PIC24? I can generate the FIR > coeffecients. I've done a moving average filter [rectangular impulse > response], because it is very easy- but I'm just not happy with the > results. I could dive into 'Numerical Recipes in C,' but I'm hoping that > someone has done/found some slightly more optimized code (for example, > using the __builtin_xxx() type functions available in C30/XC16). > > Regards, > > Matt > > Matt Bennett > Just outside of Austin, TX > 30.51,-97.91 > > The views I express are my own, not that of my employer, a large > multinational corporation that you are familiar with.

On Tue, May 22, 2012 7:42 am, M.L. wrote: {Quote hidden}> On Tue, May 22, 2012 at 8:22 AM, Matt Bennett <.....mattpiclistKILLspam@spam@hazmat.com> > wrote: >> If you've ever tried to do DSP type functions on a PIC24, you'll find >> that >> the standard libraries don't include any sort of filtering functions- >> there are filtering functions for the dsPICs, but those functions are >> using the DSP type features of that family- the code doesn't port. >> >> I understand it isn't optimal, but has anyone found decent fixed-point >> FIR >> (finite impulse response) code for the PIC24? I can generate the FIR >> coeffecients. I've done a moving average filter [rectangular impulse >> response], because it is very easy- but I'm just not happy with the >> results. I could dive into 'Numerical Recipes in C,' but I'm hoping that >> someone has done/found some slightly more optimized code (for example, >> using the __builtin_xxx() type functions available in C30/XC16). >> >> Regards, >> >> Matt > > Matt, > C30 produces pretty good assembly code. A simple FIR filter shouldn't > have to be more complicated than rolling the x history and doing the > MAC operations in standard C statements. If you need intense > optimization to get it done efficiently then I would have to argue > that you really need the dsPIC. The PIC24 running full speed should be > able to do a LOT of FIR MACs per second, all without any special DSP > instructions. > > I have derived and written FIR and IIR code if you need professional > level help. I know I *can* do it- I was just hoping someone had already done it, tested it, and shared the code. A dsPIC is definitely a better choice, but sometimes you just have to play the cards you're dealt... the data I want to filter is pretty slow- only coming in at about 8Hz, and I think I'll put the corner of the low-pass <= 1Hz. Eventually, it will need to be optimized for low power (low cycle count), but I'm really trying to do a proof of concept now. This PIC24 is doing plenty of other stuff, I'm trying to wedge some additional functionality in there. Matt Matt Bennett Just outside of Austin, TX 30.51,-97.91 The views I express are my own, not that of my employer, a large multinational corporation that you are familiar with

> > C30 produces pretty good assembly code. A simple FIR filter shouldn't > > have to be more complicated than rolling the x history and doing the > > MAC operations in standard C statements. If you need intense > > optimization to get it done efficiently then I would have to argue > > that you really need the dsPIC. The PIC24 running full speed should be > > able to do a LOT of FIR MACs per second, all without any special DSP > > instructions. > > > > I have derived and written FIR and IIR code if you need professional > > level help. > > I know I *can* do it- I was just hoping someone had already done it, tested it, and > shared the code. A dsPIC is definitely a better choice, but sometimes you just have > to play the cards you're dealt... the data I want to filter is pretty slow- only > coming in at about 8Hz, and I think I'll put the corner of the low-pass <= 1Hz. > Eventually, it will need to be optimized for low power (low cycle count), but I'm > really trying to do a proof of concept now. This PIC24 is doing plenty of other > stuff, I'm trying to wedge some additional functionality in there. Look in the libraries available for the PIC32, IIRC there are IIR and FIR filters there, and should have source code. -- Scanned by iCritical.

On Tue, May 22, 2012 8:41 am, alan.b.pearceKILLspamstfc.ac.uk wrote: > Look in the libraries available for the PIC32, IIRC there are IIR and FIR > filters there, and should have source code. Yes, there are FIR functions for the PIC32 in C32- but they are really just wrappers around mips_fir16() and mips_fir16_setup(), which are only in libdsp.a- not source. I suspect that the source code is actually MIPS assembly. No joy there. Matt Matt Bennett Just outside of Austin, TX 30.51,-97.91 The views I express are my own, not that of my employer, a large multinational corporation that you are familiar with