Sine function implementation in FPGA??

Question

Hi,I am trying to implement a DSP algorithm in a FPGA. My algorithm hassine and cosine functions in it. Can somebody help me in implementingsine and cosine functions in MATLAB fixed point (using fixed pointtoolbox) or VHDL.Thanks,SD

Symon · Accepted Answer

Dear SD,Try  this in the search box :-sine generation group:Isn't the internet great?Cheers, Syms.

Ray Andraka · Answer

How many bits precision do you need?  How many angles, or what is theangular resolution you need?  How about the angular range if less than+/-pi?  There are several ways of obtaining the sine.  For lower angularresolution, a look up table works fine if you have the on-chip memoryavailable.  There are also several approximations (one that I've used isa two step linear approximation with a correction that uses two fairlysmall look-ups).  You can also compute the sign algorithmically using aCORDIC rotation at the expense of more latency.  If your design needs tomultiply some signal by a sine, say a mixer and local oscillator, thenCORDIC might be advantageous as it can accomplish both the sine and themultiply in one operation.  I have papers on my website on CORDIC and aXilinx XCELL article I wrote on using CORDIC as a mixer in a digitalradio app.  For behavioral VHDL, you can use the sine and cosinefunctions in  are computed using CORDIC in thatpackage, by the way.SD wrote:----Ray...

Kevin Neilson · Answer

If a BRAM lookup isn't precise enough, a method I've seen used is a 5- or6-term Maclaurin series using the Horner method.  You need embeddedmultipliers for this though.  You can use one multiplier, or one for eachterm for a fully parallel implementation.  It all depends how fast you wantto do it.  The CORDIC probably uses a lot less hardware.  If you have extraembedded multipliers, though, you can be lazy.

john jakson · Answer

In the past I used sin(x) == sin(x[msb]+x[lsb]) and the expansion gives me 3 or 4 multiplies and small table lookups with a final PLA to correct a few odd cases. The table lookups for a course sin,cos (x[msb]) can be very small.

Also visit Ray Andraka's site, I'm sure there's something there on sin/cos with some interesting cordic functions.

You could also stuff a blockram with a pretty good size table too.

regards

johnjakson_usa_com

glen herrmannsfeldt · Answer

news:...I have a National data book from the days when ROMs needed +12volts with a SIN lookup rom set.  There is the course ROM,and then more ROMs and an adder.  I believe the fine ROMsare linear interpolation, such that only an adder is needed.As for the original problem, it would be easier to say withmore description of what the real problem is.  The bestsolution might depend on details not yet given.-- glen

Sine function implementation in FPGA??

Join the Discussion

Didn't find your answer?