A VLSI parallel architecture of a piecewise linear neural network for nonlinear channel equalization

This paper proposes a systolic architecture based on a multilayer neural network (MNN) to solve the problem of the nonlinear channel equalization. This architecture is based on a piecewise linear multilayer neural network (PL-MNN) algorithm derived from a recursive version (PL-RNN). In place of a sigmoid function, both algorithms use a canonical piecewise linear function, which makes the MNN more suitable for a digital VLSI implementation. The PL-MNN algorithm is more suitable for a VLSI pipelined implementation. The pipeline technique is applied to obtain a high throughput circuit which can be used in a high speed adaptive channel equalization. A performance study on both linear and nonlinear channels is presented. A comparison of results obtained with two conventional methods (LMS and RLS) and the PL-RNN algorithm is presented, and a performance evaluation of the systolic architecture is carried out for a 0.5 /spl mu/m CMOS technology.

[1]  Elias S. Manolakos,et al.  A VLSI array architecture for the on-line training of recurrent neural networks , 1991, [1991] Conference Record of the Twenty-Fifth Asilomar Conference on Signals, Systems & Computers.

[2]  Dominique Godard,et al.  Channel equalization using a Kalman filter for fast data transmission , 1974 .

[3]  Tomas Nordström,et al.  Highly parallel computers for artificial neural networks , 1995 .

[4]  Jaekyun Moon,et al.  Data storage channel equalization using neural networks , 1997, IEEE Trans. Neural Networks.

[5]  Xiao Liu,et al.  A piecewise linear recurrent neural network structure and its dynamics , 1998, Proceedings of the 1998 IEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP '98 (Cat. No.98CH36181).

[6]  Daniel Massicotte.  A systolic VLSI implementation of Kalman-filter-based algorithms for signal reconstruction , 1998, Proceedings of the 1998 IEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP '98 (Cat. No.98CH36181).

[7]  Vijay K. Madisetti.  VLSI Digital Signal Processors: An Introduction to Rapid Prototyping and Design Synthesis , 1995 .

[8]  Bhaskar D. Rao,et al.  Fast adaptive digital equalization by recurrent neural networks , 1997, IEEE Trans. Signal Process..

[9]  Daniel Massicotte,et al.  Neural-network-based method of correction in a nonlinear dynamic measuring system , 1999, IMTC/99. Proceedings of the 16th IEEE Instrumentation and Measurement Technology Conference (Cat. No.99CH36309).

[10]  Elias S. Manolakos,et al.  Using recurrent neural networks for adaptive communication channel equalization , 1994, IEEE Trans. Neural Networks.

[11]  K. M. Curtis,et al.  Piecewise linear approximation applied to nonlinear function of a neural network , 1997 .

[12]  J. G. Proakis.  Equalization techniques for high-density magnetic recording , 1998 .

[13]  Steve McLaughlin.  Shedding light on the future of SP for optical recording , 1998 .