Welcome to our site! EDAboard.com is an international Electronics Discussion Forum focused on EDA software, circuits, schematics, books, theory, papers, asic, pld, 8051, DSP, Network, RF, Analog Design, PCB, Service Manuals... and a whole lot more! To participate you need to register. Registration is free. Click here to register now.
You have three basic choices:
1. Parallel (also known as unrolled) structure, this is the one that you will see in most DSP text-books, with every filter delay implemented as explicit registers and one result sample is available every clock cycle.
2. MAC (multiply-accumulate) based, this one computes one tap per cycle and stores intermediate result in an accumulator, it would take 16 cycles to produce one output sample.
3. DA (distributed arithmetic) based, this method is similar to (2), in the sense that it requires 16 cycles to produce a result and it needs an accumulator, but no multiplyers are needed since their functionality is implemented in a look-up table (LUT).
Google for the above for terms and you will find a wealth of implementations that are well documented.
If your taps are symmetrical, you can save one multiplication by adding the taps (that have the same weighting factor) together two at a time and then multiplying. This will save you about 8 multiplies total.
The shortest way is choose what kind of your fir, and then calculate the coefficent by the FDATOOL which intergrated in Matlab, import them to any DSP tools for FPGA such as Xilinx SystemGenerator for DSP and build the system.
what kind of fpga are you using? if it is xilinx, you can use their core generator. It very handy and flexible. I used it to implemented 255 tap, 14bit coefficient, 12 bit input fir filter.
If you want to save resource, you can use chips with embedded MAC. It's fun. Good luck
This site uses cookies to help personalise content, tailor your experience and to keep you logged in if you register.
By continuing to use this site, you are consenting to our use of cookies.
