NeuroPic

aus MariaTheresia, der freien Wissensdatenbank

Inhaltsverzeichnis

Introduction:

For our assignment in ANE we wanted to implement a neural network on a PIC microcontroller. The idea was to make an easy to use and versatile back propagation network on a microcontroller with an interface that can be used by novices to the area.

[edit]

How it works:

We implemented a MLP with back propagation and therefore we used the nn.c program from John Bullinaria[1] (http://www.cs.bham.ac.uk/~jxb/NN/nn.html) as base because it is written in C and therefore easy to convert into PICC which we used to program the microcontroller. We used the CCSC compiler version 3210 to compile the program and the Shane Tolmies microchipc PIC18Fx52 bootloader v1.5 [2] (http://www.microchipc.com/PIC18bootload/) to load the program on a PIC 18F452 microcontroller.

The interface application is written in java and needs JRE version 1.5 to be executed. The communication between the PIC and the java application works via RS232 which has to be available on PIC and the PC running the java application.

It is possible to set input values, target values and all parameters of the neural network through the java interface. The neural net outputs the learning progress for every 100 learning epochs and the final output values of the network.

[edit]