The single-layer backpropagation algorithm is a gradient-descent method that adjusts the connection weights of a single-layer perceptron to minimize the mean-square error at the output. It is similar to the standard l...
详细信息
The single-layer backpropagation algorithm is a gradient-descent method that adjusts the connection weights of a single-layer perceptron to minimize the mean-square error at the output. It is similar to the standard least mean square algorithm, except the output of the linear combiner contains a differentiable nonlinearity. In this paper, we present a statistical analysis of the mean weight behavior of the single-layer backpropagation algorithm for Gaussian input signals. It is based on a nonlinear system identification model of the desired response which is capable of generating an arbitrary hyperplane decision boundary. It is demonstrated that, although the weights grow unbounded, the algorithm, on average, quickly learns the correct hyperplane associated with the system identification model.
An algorithm is presented which computes optimal weights for arbitrary linear arrays. The application of this algorithm to in situ optimal reshading of arrays with failed elements is discussed. It is shown that optima...
详细信息
An algorithm is presented which computes optimal weights for arbitrary linear arrays. The application of this algorithm to in situ optimal reshading of arrays with failed elements is discussed. It is shown that optimal reshading can often regain the original sidelobe level by slightly increasing the mainlobe beamwidth. Three examples are presented to illustrate the algorithm's effectiveness. Hardware and software issues are discussed. Execution time for a 25-element array is typically between 1 and 2 min on an HP9836C microcomputer.
We investigate the problem of decoding digital data when soft decisions are constrained to take on values from a finite set. We propose a physically reasonable objective function for selecting the desired assignment o...
详细信息
We investigate the problem of decoding digital data when soft decisions are constrained to take on values from a finite set. We propose a physically reasonable objective function for selecting the desired assignment of metrics to the received analog signals, We develop a search algorithm for designing a table-look-up that is used by the decoder to select the appropriate intermediate metrics and show that an optimum solution exists, We provide a number of illuminating examples to elucidate our ideas and work out in detail some practical cases.
We describe the application of the PVL algorithm to the small-signal analysis of circuits, including sensitivity computations. The PVL algorithm is based on the efficient computation of the Pade approximation of the n...
详细信息
We describe the application of the PVL algorithm to the small-signal analysis of circuits, including sensitivity computations. The PVL algorithm is based on the efficient computation of the Pade approximation of the network transfer function via the Lanczos process, The numerical stability of the algorithm permits the computation of the Pade approximation to any accuracy over a certain frequency range, We extend the algorithm to compute sensitivities of network transfer functions, their poles, and their zeros, with respect to arbitrary circuit parameters, with minimal additional computational cost, We demonstrate the implementation of our algorithm on circuit examples.
Dynamic equations are developed in nonrecursive symbolic form for chain-structured robotic manipulators with compliant links. A program is developed in REDUCE to automate the symbolic expansion of these equations for ...
详细信息
Dynamic equations are developed in nonrecursive symbolic form for chain-structured robotic manipulators with compliant links. A program is developed in REDUCE to automate the symbolic expansion of these equations for any given chain-structured manipulator. The symbolic nonrecursive form of the dynamic model is particularly suitable for controller synthesis and real-time control implementations.
We develop a systolic design-rule checker (SDRC) for rectilinear geometries. This SDRC reports all width and spacing violations. It is expected to result in a significant speed up of the design-rule check phase of chi...
详细信息
We develop a systolic design-rule checker (SDRC) for rectilinear geometries. This SDRC reports all width and spacing violations. It is expected to result in a significant speed up of the design-rule check phase of chip design.
DQDB (distributed queue dual bus) is a media access control (MAC) protocol which is being standardized as an IEEE 802.6 MAN [11]. The DQDB MAC protocol offers two types of services to its users: isochronous and asynch...
详细信息
DQDB (distributed queue dual bus) is a media access control (MAC) protocol which is being standardized as an IEEE 802.6 MAN [11]. The DQDB MAC protocol offers two types of services to its users: isochronous and asynchronous. This paper thoroughly analyzes the asynchronous part of the DQDB MAC protocol with only one priority class. According to our methodology, DQDB behavior is first analyzed in normal conditions (i.e., when the offered load is lower [underload] or slightly higher [overload] than the medium capacity), and then in asymptotic conditions (i.e., when each DQDB MAN node is trying to seize all the medium capacity). For each condition, performance and fairness metrics have been devised. The results obtained clearly show that DQDB is unfair when the bandwidth balancing mechanism (BWB) is disabled, and that the unfairness depends, in particular, upon the medium capacity and bus length. While this unfairness does not meaningfully degrade a node's quality of service in underload, it does cause unpredictable behavior in overload. When the BWB is enabled, DQDB behavior changes significantly. In asymptotic conditions, after a transient time, completely fair behavior is reached. In normal conditions, even though DQDB behavior still remains unfair, some major drawbacks are solved in the steady state (e.g., distribution of packet loss among the nodes). Our analysis is then broadened to include higher medium capacities up to 1.2 Gb/s. Finally, the extension of the methodology used in this paper to the analysis of the broad class of resource-sharing algorithms is sketched.
We show that the use of nondedicated Global Positioning System (GPS) sensors to determine the attitude parameters of a vessel yields the same level of performance as the use of a dedicated multiantenna receiver, namel...
详细信息
We show that the use of nondedicated Global Positioning System (GPS) sensors to determine the attitude parameters of a vessel yields the same level of performance as the use of a dedicated multiantenna receiver, namely an agreement of the order of 0.1-degree (1sigma). The test platform is a survey launch operating at cruising speeds of 10 to 15 kt. The dedicated multiantenna receiver is a four-antenna Ashtech 3DF unit, while the nondedicated sensor array consists of three NovAtel GPSCard receivers. The approach used to resolve the relative carrier phase integer ambiguities between the antennas is discussed and the use of antenna baseline constraints is analyzed. A least-squares procedure which utilizes all the position information from the antennas for the estimation of the attitude parameters and their accuracy is presented. The attitude determination results from the two configurations tested are intercompared.
An iterative algorithm for the design of multichannel cosine-modulated quadrature mirror-image filter (QMF) banks with near-perfect reconstruction is proposed, The objective function is formulated as a quadratic funct...
详细信息
An iterative algorithm for the design of multichannel cosine-modulated quadrature mirror-image filter (QMF) banks with near-perfect reconstruction is proposed, The objective function is formulated as a quadratic function in each step whose minimum point can be obtained using a closed-form solution, This approach has high design efficiency and leads to filter banks with high stopband attenuation and low aliasing and amplitude distortions, The proposed approach is then extended to the design of multichannel cosine-modulated QMF banks with low reconstruction delays, which are often required, especially in real-time applications, Several design examples are included to demonstrate the proposed algorithms, and some comparisons are made with existing designs.
Discrete wavelet transform (DWT) is often used to approximate wavelet series transform (WST) and continuous wavelet transform (CWT), since it can be computed numerically. In this research, we first study the accuracy ...
详细信息
Discrete wavelet transform (DWT) is often used to approximate wavelet series transform (WST) and continuous wavelet transform (CWT), since it can be computed numerically. In this research, we first study the accuracy of the computed DWT coefficients obtained from the Shensa algorithm as an approximate of the WST coefficients. Based on the accuracy analysis, we then propose a procedure to design optimal FIR prefilters used in the Shensa algorithm to reduce the approximation error. Finally, numerical examples are presented to demonstrate the performance of the optimal FIR prefilters.
暂无评论