Design and Application
of High Complexity Mixed-Signal Integrated Neuro-Fuzzy Controllers
TESIS
DOCTORAL
Rafael
de Jesús Navas González
Advisor: Dr. Fernando Vidal Verdú
Departamento de Electrónica Universidad de Málaga
ABSTRACT:
The main contributions of this Thesis
focuse on the analog and mixed-signal integrated circuits design and the
application field of neuro-fuzzy systems, and its aim is preserving most
of the advantages of the analog implementations of fuzzy controllers, while
allowing a notorious increment of the system complexity.
Specifically, a new architecture
for neuro-fuzzy controllers is proposed to accomplish a spatial multiplexed
strategy of an analog core which implements the simplest fuzzy system in
each subregion of the input space. This core provides the output, for a
given input vector, and from the active rules set in each subregion of
input space, as rule base, which could be called the real rule set.
In order to provide an output for any input value, and fulfil the whole
control surface, such core is multiplexed among the whole knowledge base
which is composed by the virtual rule set. The latter is actually just
a set of programming parameters stored in digital and analog memories,
and the multiplexing is realized by means of digital techniques. Thus,
the signal path is entirely analog, while the internal control circuitry
is mixed, thus the analog performances are preserved. To validate this proposal, a
monolithic controller has been implemented in standard CMOS 0.7mm technology.
This
example controller has two inputs, one output and sixty-four rules, much
more than the reported pure analog monolithic fuzzy controllers, while
preserves most of their advantages. Experimental results from that example
implementation are presented to show its major features. Finally, to illustrate the capabilities
of the fuzzy controller chip in a real application, a DC motor soft-start
procedure has been proposed and developed. This is an original application,
not previously reported and it is complex enough to show the performances
of the controller when it is forced to work in a high-speed control loop.