Multilayer feedforward neural network (MFNN) trained by the backpropagation (BP) algorithm is one of the most significant models in artificial neural networks. MFNNs have been used in many areas of signal and image processing due to high applicability. Although they have been implemented as analog, mixed analog-digital and fully digital VLSI circuits, it is still difficult to realize their hardware implementation with the BP learning function efficiently. This paper describes a special BP algorithm for the logic oriented neural network (LOGO-NN) which we have proposed as a sort of MFNN with quantized weights and multilevel threshold neurons. Both weights and neuron outputs are quantized to integer values in LOGO-NNs. Furthermore, the proposed BP algorithm can reduce high precise calculations. Therefore, it is expected that LOGO-NNs with BP learning can be more effectively implemented as digital type circuits than the common MFNNs with the classical BP. Finally, it is shown by simulations that the proposed BP algorithm for LOGO-NNs has good performance in terms of the convergence rate, convergence speed and generalization capability.

We consider a network consisting of phase locked loops coupled one another through frequency dividers. When the network structure is rotationally symmetric, spatially periodic simple patterns in terms of the phase of the PLLs are formed. The patterns determine the lock-in frequency of the network. The stability of the pattern is determined by the spatially distributed simple coupling weight patterns. Therefore, a signal with which the network synchronizes is indirectly selected by the weight patterns when several signals are simultaneously applied to the network. The selectivity plays an important role in an intelligent network model.

It is important to analyze a tracking or synchronizing process in Spread Spectrum (SS) receiving system. The most common SS tracking system considered here consists of pseudorandom (PN) generator, Lowpass Filter (LPE) and Voltage Controlled Oscillator (VCO). The SS receiver is to track or synchronize its local PN generator to the received PN waveform by VCO. The fundamental equation of the system is known by a second order nonlinear differential equation in terms of phase difference between local PN generator and received PN waveform. The differential equation is nonautonoumous due to PN function of time t with period T. Picking up the gain of VCO as the main parameter in the system we show that the system has bifurcation from the normal oscillation through subharmonic oscillation to finally chaos. In the final case, chaos is confirmed by investigating maximum Liapunov number and both stable and unstable manifolds.

Piecewise linear (PWL) circuit modules operating on sigma-delta (ΣΔ) modulated signals and nonlinear signal processors built of these modules are proposed. The proposed module library includes absolute circuits, min/max selectors and negative resistances. Their output signal-to-noise ratio is higher than 50dB when their oversampling ratio is 28. A nonlinear filter and a stochastic resonator are presented as applications of the PWL modules to ΣΔ domain signal processing. The filter is structured with 37% of logic gates consumed by an equivalent filter with a 5-bit parallel signal form.

This paper describes a digital delay-lock Loop (DLL) to which delta-sigma (Δ Σ) modulation technique is applied in order to reduce circuit elements. The DLL is evaluated in both transient and steady-state behavior by theoretical analysis, computer simulations and circuit experiments. Not deteriorated by the internally generated Δ Σ-modulation noise, the DLL shows good tracking performance in transient response and steady-state RMS jitter of phase error against additive white Gaussian noise. Using the proposed DLL most parts of receiving circuits are realized by digital integrated circuits. After realizing the circuit, power-line communication system with spread spectrum is possibly expected in a small size with low cost.

A digital circuit technique is proposed to process directly bit-stream signals from analog-to-digital converters based on sigma-delta modulation. Newly developed adder and multiplier are fundamental circuit modules for the processing. Using the fundamental modules and up/down counters, other circuit modules such as divider and square root circuits are also realized. The signal processors built of the modules have advantages over multi-bit Nyquist rate processors in circuit scale by the following two distinct features: First, single-bit/multi-bit converters are not needed at the inputs of the processors because the arithmetic modules directly process bit-stream signals. Secondly, the arithmetic modules consist of small number of logic gates. As an application of the technique to digital signal processing for communications, a QPSK demodulator is presented. The demodulator is structured with 40% of logic gates consumed by an equivalent multi-bit demodulator.

The threshold voltage variations for actual size MOSFETs obtained by capacitance measurement are compared with those obtained by the current measurement, and their differences are studied for the first time. It is found that the threshold voltage variations obtained by the capacitance measurement show the similar behavior to those current measurement and the absolute value is less than those obtained by the current measurement. The reason for the difference is partially explained by that the local channel dopant non-uniformity along the current path makes the threshold voltage variation obtained from current measurement larger. It is found that the flat-band voltage variations, which are obtained from the measured C-V curves, are small and not significant to the threshold voltage variation.

We propose binary-quantized and spatio-temporally discretized network models of linear diffusion systems and investigate their filtering effect on single-bit sigma-delta (ΣΔ) modulated signals. The network consists of only one kind of elements that add ΣΔ modulated signals and quantize the sum in the form of single-bit signal. A basic one-dimensional network is constructed first. Then, the network is extended into two dimensions. These networks have characteristics equivalent to those of linear diffusion systems in both time and frequency domains. In addition, network noise caused by the quantization in the elements contains low-level low-frequency components and high-level high-frequency components. Therefore, the proposed networks have possibility to be used as signal propagation and diffusion media of ΣΔ domain filters.

Associative memories composed of sparsely interconnected neural networks (SINNs) are suitable for analog hardware implementation. However, the sparsely interconnected structure also gives rise to a decrease in the capability of SINNs for associative memories. Although this problem can be solved by increasing the number of interconnections, the hardware cost goes up rapidly. Therefore, we propose associative memories consisting of multilayer perceptrons (MLPs) with 3-valued weights and SINNs. It is expected that such MLPs can be realized at a lower cost than increasing interconnections in SINNs and can give each neuron in SINNs the global information of an input pattern to improve the storage capacity. Finally, it is confirmed by simulations that our proposed associative memories have good performance.

We propose a method to search all the zeros of a complex function in a given compact region D Cn. The function f: Cn Cn to be considered is assumed to consist of polynomial and transcendental terms and to satisfy f (x) Rn for any x Rn. Using the properties of such a complex function, we can compute the number of zeros and determine the starting points of paths on the boundary of D, which attain all the zeros of f in D without encountering a singular point. A piecewiselinear approximation of the function on a triangulation is used for both computing the number of zeros and following the paths.