Artificial Neural Networks: Quick Introduction
The basic components of a biological neuron are shown in Fig. 1. A slightly modified form is shown in Fig. 2 where the dendrites are shown as electrical conductors and the synapses are represented as weights or electrical resistors. The neuron cell is equivalent to an electronic amplifier. An op-amp is commonly used in hardware implementations. The artificial neuron shown in Fig. 2 can also be represented mathematically using basic circuit equations. The equations can be programmed into a computer or microcontroller. The equations are relatively simple for the linear approaches such as the ADALINE. However, implementing fast nonlinear neural networks in software can be a challenge.
An analog circuit for a nonlinear neuron is shown in Fig. 3. The neuron was designed by Jeff Alig for his MSEE Thesis (December 1997). It was used to double the tracking range of a phase-locked loop circuit. One aspect that is not shown in Figures 2 and 3 is neuron learning or adaptation.
Software complexity and development time is significantly increased with the nonlinear approaches although the benefit is more computational power and the ability to solve nonlinear problems. Neural nets are used to solve nonlinear problems that are encountered in control, signal, and image processing systems. The computational power of a neural net lies in the parallel architecture and the neuron's nonlinear I/O characteristic.