This week began with some bad news, as a more detailed analysis of the marina test results revealed that the wrong sequences were being sent, with much longer lengths than intended. This was tracked down to a bug in the control software and quickly fixed, but unfortunately it invalidates all of the test data. New tests will have to be scheduled as soon as possible.

On a happier note, a new 256-point FFT core was added to the project, the first piece of the actual receiver. It takes up only two multipliers and two BRAMs, with small enough latency that we can perform a number of transforms per symbol. Since we are interested only in using the magnitude of the transform, and in order to save resources (namely multipliers), I implemented the old trick of α∙max + β∙min (see e.g. Lyons, 1997, Sec. 10.2), with 1 and 3/8 as coefficients so that only four adders are used. The results were verified in MATLAB and produce results identical to those of the built-in FFT.