Current-mode techniques for the implementation of continuous- and discrete-time cellular neural networks

Abstract

This paper presents a unified, comprehensive approach to the design of continuous-time (CT) and discrete-time (DT) cellular neural networks (CNN) using CMOS current-mode analog techniques. The net input signals are currents instead of voltages as presented in previous approaches, thus avoiding the need for current-to-voltage dedicated interfaces in image processing tasks with photosensor devices. Outputs may be either currents or voltages. Cell design relies on exploitation of current mirror properties for the efficient implementation of both linear and nonlinear analog operators. These cells are simpler and easier to design than those found in previously reported CT and DT-CNN devices. Basic design issues are covered, together with discussions on the influence of nonidealities and advanced circuit design issues as well as design for manufacturability considerations associated with statistical analysis. Three prototypes have been designed for l.6-pm n-well CMOS technologies. One is discrete-time and can be reconfigured via local logic for noise removal, feature extraction (borders and edges), shadow detection, hole filling, and connected component detection (CCD) on a rectangular grid with unity neighborhood radius. The other two prototypes are continuous-time and fixed template: one for CCD and other for noise removal. Experimental results are given illustrating performance of these prototypes.

The Speckle Noise Reduction and the Boundary Enhancement on Medical Ultrasound Images using the Cellular Neural Networks

“…The CNN local processing architecture is well adapted to vision algorithms and facilitates VLSI implementation (14)- (17) . A cellular neural networks is a two-dimensional structure, a lattice of cells, composed by identical analogical non-linear processors.…”

“…Cellular neural networks is as simple analog circuit, it has ability to real time information processing. And, in these days, it is applied to image processing field (14)- (17) . In this study, we aimed speckle noise reduction with edge boundary strengthening using CNN that the template is decided by IIRNN(Neural Network using IIR filter).…”

An analog scheme for fixed-point computation-Part II: Applications

“…The above implementation uses an operational transconductance amplifier (OTA)-C integrator; thus the dynamic range of is limited to the linear range of the OTA, which is typically a few tens of millivolts in CMOS implementations. The dynamic range (equivalently the word length) of the computational element can be improved by using current-mode integrators, which remove the need for wide-range linear convertors [25]. Another related problem is to solve the system of equations where is a stochastic matrix.…”

On global exponential stability of standard and full‐range CNNs

“…Advantages in the VLSI implementation of CNN chips with a large number of neurons have been obtained by using the ISR model of CNNs [2,3]: where m 0 is a constant and the diagonal mapping…”

“…This has led to the introduction of the so-called improved signal range (ISR) model of CNNs [2]. In the limit where the slope of the limiter device tends to 655 of S-CNNs, with or without delay, has been extensively investigated in the literature, see e.g.…”