Blind signal separation by Independent Component Analysis(ICA) is very attractive tool because of its potential applications. ICA can be used in speech recognition, telecommunications and medical signal processing. One of the algorithms for ICA is Maximum Entropy based on information theory. It is very concise and efficient algorithm for hardware implementation. In this paper, I propose the chip architecture using analog CMOS technique to implement the algorithm and present the Hspice simulation results.
The main issues for hardware implementation are offset problem, non-linearity and dynamic range in analog circuits. I analyze these problems and show the limitations of the algorithm and analog CMOS technique.
I design and simulate the 2 × 2 network using speechlike signals. The network is modular and expansible to N × N network.
