What we have here is a small .NET application that does data transformation using FFT and some other small mathematics. We need it ported to Android device and optimized so it runs faster then figures given in attached table.
Usage of native compiling for fft on android should be considered.
You will be given samples that should be used as an input, source code for the application and required outputs.
Results on android application should be bit-exact to a reference given. In fact, due to rounding of number some errors might be allowed, but error rate should be less then 1 bit in 10000 (0.01%).
Application should be delivered as an APK file containing all the resources (input files).
When run, application should process all of the files, and report a processing log, results of processing to the screen and e-mail address (to simplify testing on multiple devices).
If application can't be run faster, good explanation for it should be given.
Application uses 1d in-place fft with 8192 window size, Hanning window.
Open source solutions should be avoided, but in case that open source is good provider for fft for example, it could be discussed.
Algorithm that is in the original application is pretty straightforward; reads data, does framing, windowing, fft processing, then something simple with code, and outputs the data to another file. It is integral part of an larger algorithm that should be moved to android platform.