IIR filtering algorithm for C28x+FPU

Chesterdam

Hello,

I've been implementing DSP algorithm for TMS320C28335. Algorithm utilizes IIR filters and I have used "C28x Fixed Point DSP Library v1.01" dsp library. However, the IIR is quite sensitive to coefficient quantization and the IIR filter output using fixed-point arithmetics is not optimal. We have FPU available for accelerating floating-point operations but alas, the "C28x Floating Point Unit DSP Library V1.30" does not have IIR filtering algorithms (it has FIR filtering though...).

Does TI provide any IIR filtering library for C28x+FPU?

Best regards,
Henrik

over 13 years ago

0 Edwin Krasser over 13 years ago

Expert 2420 points

Hi,

I had the same problem. No FPU IIR filter functions. Then I wrote it on my own, the fixed point functions are a good thing to start. Now, it's no problem any more ;-).

Best regards and good luck
Edwin Krasser

0 Igor Kordunsky over 13 years ago in reply to Edwin Krasser

Intellectual 835 points

Hi, Edwin,

It is a shame TI does not provide IIR assembler routine example for FPU.

Of cause it is possible to look on generated by C-compiler code, but it does not utilize all capabilities of FPU, besides someone needs to know in-depthl the architecture and assembler to re-create efficient assembler code from C compilation.

Would you be so kind to share your IIR routines with me and community?

Regards and thanks,

Igor

0 Edwin Krasser over 13 years ago in reply to Igor Kordunsky

Expert 2420 points

Hi!

Sharing might be a good idea for you, but I am not allowed to do that. But: The assembler optimizations are not as accelerating as in FIR filters. Hence, for the first tests I suggest to write the IIR second order section in C float. That should give a working solution for the first time.

Best regards and good luck,
Edwin Krasser

0 Igor Kordunsky over 13 years ago in reply to Edwin Krasser

Intellectual 835 points

Thank you, Edwin,

Of cause, I do have filters in C, working. I hoped assembler would give faster calculations.

I remember old days, C3x assembler IIR filter routine provided by TI works 15% faster than C-compiled.

I also did this exercise for ARM7TDMI: fixed point calculations iir 2nd order in assembler, and had got boost almost 50%, because C-compiler does not use multiply-and-add 32 bit words with 64 bit result.

I hoped something similar, like parallel assembly commands or clever arrangement of a,b coefficients and history buffer, or put these variables in a dual access RAM would benefit speed.

Can you tell did you get any performance boost for ii2 2nd order after you have converted C into asm?

Regards,

Igor

0 Edwin Krasser over 13 years ago in reply to Igor Kordunsky

Expert 2420 points

Hi!

I never did real hardcore assembler programming for c2000 processors. And I do not think that my asm functions as very much optimized, it is more a simple implementation. I started with the fixed point implementation (see iir5biq32.asm) and modified the integer operations to floating point.

iir5biq32:
            MOVL    XT,*--XAR6      ; T=dk(n-2)
            QMPYL   P,XT,*XAR7++    ; P= dk(n-2)*a2k
asm:
        MACF32 R7H,R3H,*XAR6,*XAR7++   ; R3H=x, R2H=dk(n-2)*a2k
        NOP ; to complete mac operation

And so on ...

Best regards and good luck

Edwin Krasser

0 Vishal_Coelho over 13 years ago in reply to Igor Kordunsky

TI__Mastermind 20850 points

Hi Guys,

I have a biquad implementation for the FPU that may be useful. I dont know how well ive optimized this code, might be able to squeeze a few more cycles out of it but its all i have at the moment....project runs on a 28069, but you can adapt it to any C28x+FPU

6011.f28069_FPU_IIR_Biquad_2p2z_121109.zip

0 Igor Kordunsky over 13 years ago in reply to Vishal_Coelho

Intellectual 835 points

Thank you, Vishal,

We will try it

Igor

0 Umesh Thirunavukkarasu over 12 years ago in reply to Igor Kordunsky

Intellectual 420 points

Hi Vishal,

I downloaded the code example for 28069. I am trying to implement the same in the 28335.. I want to know about the graph properties for this particular program??

0 Umesh Thirunavukkarasu over 12 years ago in reply to Vishal_Coelho

Intellectual 420 points

Hi,

Based on the project available in the above post i modified something and tried for 28335 processor, I am not getting any response.

Then i tried single step debug its getting hanged inside the Initsysctrl();

I have attached my project.

Kindly check and let me help to get the output.

Thanks & Regards,

Umesh T6082.IIR_2P2Z.rar

0 Vishal_Coelho over 12 years ago in reply to Umesh Thirunavukkarasu

TI__Mastermind 20850 points

Hi Umesh,

You need to assign the following sections to RAM in the linker command file:

firldb, iirfilt, sigIn, sigOut, coefffilt

firldb : > RAML4, PAGE = 1
iirfilt : > RAML5, PAGE = 1
coefffilt : > RAML5, PAGE = 1
sigIn : > RAML6, PAGE = 1
sigOut : > RAML7, PAGE = 1

0 Umesh Thirunavukkarasu over 12 years ago in reply to Vishal_Coelho

Intellectual 420 points

Hi Vishal,

Thanks for your valuable reply.. ya now i can able to see the variables values. If I put the processor in the Real Time Mode Its getting hanged after sometime and asking me to go for the Ruderetry.. Its entering into the Illegal ISR Routine.

In addition to that how plot the graph and can you please tell me the Dual Time Graph properties.

Thanks & Regards,

Umesh T.

0 Vishal_Coelho over 12 years ago in reply to Umesh Thirunavukkarasu

TI__Mastermind 20850 points

Hi Umesh,

I dont know much about the real time mode operation. I'd have to refer you to the CCS wiki on that.

http://processors.wiki.ti.com/index.php/C2000_Getting_Started_with_Code_Composer_Studio_v5#Enable_real-time_mode_to_monitor_the_watch_variables_as_they_are_changing

As for the graph, you need to plot fAdcInput and fBiquandOutput, 128 points in 32-bit floating point representation.

C2000™︎ microcontrollers

C2000 microcontrollers forum

IIR filtering algorithm for C28x+FPU