Forum Discussion

Altera_Forum's avatar
Altera_Forum
Icon for Honored Contributor rankHonored Contributor
18 years ago

vhdl problem for iir filter

hi: guys

i have a problem makes me crazy.

how to use vhdl codes to implement iir filter

y(n)= 2cos(w)*y(n-1) - y(n-2)

and the initial conditions is y(-1)=sin(w),y(-2)=sin(2w).

which means i need output this filter every cycle calculation like following y(0),y(1)....

y(0)=2cos(w)*y(-1)-y(-2);

y(1)=2cos(w)*y(0)-y(-1);

...

i use a mac and a rounding operation to do it.

but its not work like i describe above.

how should i do. Am i need plus some selfcycle computation on testbench or on original filter codes. And how to plus it into this codes.

thanks

4 Replies

  • Altera_Forum's avatar
    Altera_Forum
    Icon for Honored Contributor rankHonored Contributor
    18 years ago

    Hello,

    I can't indentify a filter, cause there is no signal input. If you are actually trying to design an oscillator, then you should probably choose a structure less sensitive to rounding errors or use an NCO instead.

    But apart from this consideration, the shown IIR structure can surely be synthesized in VHDL. I understand, that cos(w) is a constant that either can be calculated at compile time or loaded as a parameter (which is big difference for FPGA implementation), the same with the said initial values. Next you have to define a numeric format (probably signed) and a resolution. Then the algorithm can be build from a multiplier and a substractor megafunction or infered from HDL. Two registers are needed to store previous y(n).

    Basic operation can be evaluated also in a spreadsheet tool, e. g. Excel to study rounding error effects with different numerial resolution.

    Regards,

    Frank
  • Altera_Forum's avatar
    Altera_Forum
    Icon for Honored Contributor rankHonored Contributor
    18 years ago

    to Frank:

    thank you pay attentation to my problem.

    yes, as you said i am doing a digital oscillator utilized a iir structure.

    you said i need two registers to store previous y(n). i show u my codes as attached, may be when u see my codes you can understand my problem totally.

    Please check out my codes, and give me some advise agian.

    Thank you
  • Altera_Forum's avatar
    Altera_Forum
    Icon for Honored Contributor rankHonored Contributor
    18 years ago

    sorry to Frank

    the forum can not attached the .rar files, so please show me your email address, i send it to you directly.