Abstract
Evolvable Hardware (EHW) is a new method for designing electronic circuits. However, there are several problems to solve for making high performance systems. One is the limited scalability of the ordinary approach. To reduce this problem, a novel digital signal classification architecture has been developed that allows for incremental evolution. This is based on initially evolving subcircuits for each category to be detected.
The architecture is applied for classifying sensor data in a prosthetic hand controller. By applying the proposed method, the best performance achieved is substantially better than that obtained by more ordinary approaches. Analysis of the best circuit shows the importance of having an architecture containing some gates with random connections. Results of the analysis can also be used to collect better training data from users in the future.