Title :
Analog CMOS implementation of nonholonomic ICA algorithm with automatic offset compensation
Author :
Cho, Ki-Seok ; Lee, Soo-Young
Author_Institution :
Dept. of Biosyst. & Brain Sci. Res. Center, Korea Adv. Inst. of Sci. & Technol., Daejeon, South Korea
Abstract :
Analog implementations of neural networks suffers from non-ideal characteristics such as multiplier offsets and limited dynamic ranges. The multiplier offsets in analog hardware are analyzed and compensated. The output SNRs are improved by about 15dB with proposed offset compensation techniques. Analog CMOS circuits under subthreshold operation are also developed to realize 12-channel nonholonomic ICA algorithm, which maintains proper scales of unmixing matrix values.
Keywords :
CMOS analogue integrated circuits; independent component analysis; matrix algebra; neural nets; analog CMOS implementation; automatic offset compensation; complementary metal-oxide-semiconductor; independent component analysis; multiplier offsets; neural networks; neuromorphic system; nonholonomic ICA algorithm; unmixing matrix values; Biological neural networks; CMOS analog integrated circuits; CMOS technology; Dynamic range; Hardware; Independent component analysis; Information theory; Maintenance engineering; Neuromorphics; Signal processing algorithms;
Conference_Titel :
Neural Networks and Signal Processing, 2003. Proceedings of the 2003 International Conference on
Conference_Location :
Nanjing
Print_ISBN :
0-7803-7702-8
DOI :
10.1109/ICNNSP.2003.1279265
