Xilinx System Generator: A summary of frequent errors during the Simulink – modelling stage

From the latency on the cordic block, that looks as if you've chosen to use a single processing element - that will produce results of limited accuracy. Very limited it appears :)

I tried using 10 elements for example, with [zeros (1,9) 1] for the "latency per processing element" - to get back to your 21 tick latency. Making sure I ran the simulation long enough (oops), I then got a result (using UFixed_16_11 inputs) of 0.501953125. Which is closer...

Does that help?

The cordic block is split up into a bunch of processing elements (PEs). Each PE is a bunch of logic operations. There is some latency associated with each PE already.

You can add additional registers at any point in the pipeline by putting non-zero values into the vector you have highlighted. The vector should have as many elements as you have requested PEs. In your case, you've requested one PE so it will only look at the first value of the vector, resulting in an additional latency of zero.

As regards notation - [] is standard Matlab for a vector (or matrix, depending on how it's used, but I'll not go into that here!). It's a bit like an array in other languages.

so

a=[5 6 7 8]

creates a 1x4 vector called a. Element 1 (which you can access in matlab by doing a(1) ) has the value 5. Element 4 has the value 8.

Note that Matlab starts its numbering from '1' not '0' like many other languages!