The authors have measured and modeled the temperature characteristics and material properties of quarter wavelength mirrors made using InP and InGaAsP. The center wavelength of the mirror moves at 0.110 nm/ degrees C,...
详细信息
The authors have measured and modeled the temperature characteristics and material properties of quarter wavelength mirrors made using InP and InGaAsP. The center wavelength of the mirror moves at 0.110 nm/ degrees C, in good agreement with the theory. This is an important result for the thermal design of such devices as VCSELs. This result can be used to find the shift in the resonant mode of the cavity or the change in penetration depth of the mirror with temperature. The authors have also characterized the reflectivity and bandwidth of this mirror as a function of the number of layers. Using selective wet chemical etchants, they removed one layer at a time and then measured the spectrum of the mirror. From this structural analysis, it was found that the center wavelength is much more sensitive than the bandwidth or peak reflectivity in detecting drift in layer thicknesses. This technique can be used to find the actual reflectivity for different numbers of layers without resorting to several expensive, time-consuming mirror growths. The measured reflectivity and bandwidth are in good agreement with theory.< >
Novel techniques based on classical tools such as rational approximation and harmonic analysis are developed to study the computational properties of neural networks. Using such techniques, one can characterize the cl...
详细信息
Novel techniques based on classical tools such as rational approximation and harmonic analysis are developed to study the computational properties of neural networks. Using such techniques, one can characterize the class of function whose complexity is almost the same among various models of neural networks with feedforward structures. As a consequence of this characterization, for example, it is proved that any depth-(d+1) network of sigmoidal units computing the parity function of n inputs must have Omega (dn/sup 1/d- in /) units, for any fixed in >0. This lower bound is almost tight since one can compute the parity function with O(dn/sup 1/d/) sigmoidal units in a depth-(d+1) network. The techniques also generalize to networks whose elements can be approximated by piecewise low degree rational functions.< >
In this paper we formulate and solve a control-oriented system identification problem for single-input, single-output, linear, shift-invariant, distributed parameter plants. In this problem the available apriori infor...
详细信息
In this paper we formulate and solve a control-oriented system identification problem for single-input, single-output, linear, shift-invariant, distributed parameter plants. In this problem the available apriori information is minimal, consisting only of worst-case/deterministic, time dependent, upper and lower bounds on the plant impulse response and the additive output noise. The available aposteriori information consists of a corrupt finite output time series obtained in response to a known, non-zero but otherwise arbitrary, applied input. A novel system identification method is presented for this problem. This method maps the available apriori and aposteriori information into an "uncertain model" of the plant. The resulting uncertain plant model is comprised of a nominal plant model, a bounded additive output noise, and a bounded additive model uncertainty. The upper bound on the model uncertainty is explicit, worst-case/deterministic in nature, and expressed in terms of both the l 1 and H ∞ system norms. Under the assumption that the available apriori information is "correct" for the underlying physical plant, the resulting uncertain plant model has the property that it not only "explains" the available aposteriori information, but will also explain all aposteniori information observed in the future. Because this property hinges on the correctness of the available apriori information, a method is also presented for developing confidence that the available apriori information is in fact correct.
Techniques based on classical tools such as rational approximation and harmonic analysis are developed to study the computational properties of neural networks. Using such techniques, one can characterize the class of...
详细信息
Techniques based on classical tools such as rational approximation and harmonic analysis are developed to study the computational properties of neural networks. Using such techniques, one can characterize the class of functions whose complexity is almost the same among various models of neural networks with feedforward structures. As a consequence of this characterization, for example, it is proved that any depth-(d+1) network of sigmoidal units computing the parity function of n inputs must have Omega (dn/sup 1/d- epsilon /) units, for any fixed epsilon >0. This lower bound is almost tight as one can compute the parity function with O(dn/sup 1/d/) sigmoidal units in a depth-(d+1) network. The techniques also generalize to networks whose elements can be approximated by piecewise low degree rational functions. These almost tight bounds are the first known complexity results on the size of neural networks computing Boolean functions with continuous-output elements and with depth more than two.< >
The authors show that the game of life algorithm, which is equivalent to a Turing machine, can be realized by cellular neural networks (CNNs). They also present a multipath CNN algorithm that demonstrates the capabili...
详细信息
The authors show that the game of life algorithm, which is equivalent to a Turing machine, can be realized by cellular neural networks (CNNs). They also present a multipath CNN algorithm that demonstrates the capabilities of analog/logic (analogic) software. Complex image processing tasks can be realized by programmable dual computing CNNs. A specific example of blob-counting in a gray-scale image is presented.< >
The authors introduce a MOS circuit for the integrated implementation of pulse-coded competitive learning. They describe an autoadaptive synapse circuit for a pulse-coded competitive learning rule. The specific focus ...
详细信息
The authors introduce a MOS circuit for the integrated implementation of pulse-coded competitive learning. They describe an autoadaptive synapse circuit for a pulse-coded competitive learning rule. The specific focus is upon an adaptive synapse cell which combines a capacitive analog storage element with subthreshold adaptation circuitry. The adaptation circuitry is designed to compensate for nonlinear device transconductance in the subthreshold operating region. The simulation results presented verify circuit operation in a 2-input-3-output competitive network. Accurate clustering of random training data was demonstrated.< >
A number of robust stability problems take the following form: A polynomial has real coefficients wvhich are multiaffine in real parameters that are confined to a box in parameter space. An efficient method is require...
详细信息
A number of robust stability problems take the following form: A polynomial has real coefficients wvhich are multiaffine in real parameters that are confined to a box in parameter space. An efficient method is required for checking the stability of this set of polynomials. We present two sufficient conditions in this paper. They involve: checking certain properties at the corners and edges of the parameter space box.
This article describes the first implementation of a hardware architecture that solves the problem of collision detection for an arbitrarily complex collection of arbitrarily complex objects in a visual simulation sys...
This article describes the first implementation of a hardware architecture that solves the problem of collision detection for an arbitrarily complex collection of arbitrarily complex objects in a visual simulation system in real time. Resolving all collisions between N moving objects in real-time requires O(N2) calculations are performed every frame time. One possibility is to have one central processor performing all the calculations. This quickly reaches a performance limit. Alternatively, N processors can each do O(N) calculations, since moving objects only require calculations to be performed relative to themselves. Moreover, only active objects need to do calculations. The hardware architecture described in this article uses this calculation partitioning to solve the collision detection problem. There are a number of collision detection chips per active object, which compare that active object's polygons against all other polygons within the pilot's 360-degrees viewing range. The number of collision detection chips per active object is proportional to the polygon complexity of an object-not the number of objects in the scenario. Further, we describe the model governing the collision detection hardware design and how the hardware is to be incorporated in the visual simulation system.
This paper presents a foundation for an integrated approach to the design of controls and diagnostics in reliable control systems. In this approach the control module and diagnostic module of the control system are de...
详细信息
暂无评论