John B. Kenney

Finding maximal link disjoint paths in a multigraph

The problem of fast restoration of lost transport facilities between nodes of the growing high-capacity fiber network has recently emerged as a principle concern of both network operators and their customers. At the same time, there is increasing use of intelligent cross-connects in the telephone network. It should be possible to use the computing and connection capacity of cross-connects to recover from facility outages by enabling the cross-connects to automatically reroute traffic in the event of facility damage or loss. An algorithm by which a network node might be able to establish as many paths as required (up to the theoretical maximum) between itself and another node with which it has lost part or all of its directly connecting transmission facilities is presented. The algorithm has been simulated on a Sun 3/60 workstation and is able to find all four paths of the example network in under one second.<<ETX>>

The composite regressor algorithm

A relation between equation error and output-error adaptive algorithms is presented. It leads to a clarified view of output-error bias and stability. It is shown that a tradeoff between desirable stability and bias characteristics can be achieved in output error by manipulation of the adaptive gain. The composite regressor algorithm (CRA) is introduced as a means of affecting this tradeoff, independent of the gain. Output error is shown to a special case of CRA. A stability constraint for CRA is developed in a theorem, and the bias properties of this algorithm are discussed.<<ETX>>

A neural network solution for call routing with preferential call placement

A neural network solution to the problem of routing calls through a three-stage interconnection network is presented. The solution uses a Hopfield network with a binary threshold, rather than a sigmoidal function. An important feature of this solution is that the weights of the neural network are fixed for all time and are independent of the current state of the interconnection network. It is possible to implement various preferential call placement strategies through selection of external inputs to the neural network, again independent of the weights. The performance of the call placement strategy is guaranteed. The operation of the neural network solution is discussed in terms of a digital logic implementation that could be used to realize the same functionality.<<ETX>>

Bias analysis of a combined output error-equation error algorithm

An analysis is presented of the bias produced in the composite regressor algorithm. This study contains what the authors believe is the first analytical demonstration that there is bias in the output error algorithm when the algorithm is used with a small but nonvanishing adaptation step size. The analysis provides insight into general bias properties of infinite-impulse-response adaptive filtering algorithms.<<ETX>>

Queue-length distributions for two-stage CBR multiplexing

This paper is concerned with the queue length distribution at the output of an ATM multiplexer. More specifically, we derive an upper bound on the complementary queue-length distribution for a multiplex of several pre-aggregated cell streams arising from a multiplex of constant bit rate (CBR) cell streams. An upper bound on the second-stage complementary queuelength distribution is useful when sizing queues and determining cell delay bounds for any system using more than one level of CBR multiplexing. The upper bound is shown to be very tight throughout the domain and hence a very good approximation of the actual distribution. We compare our second-stage upper bound to that obtained by assuming all sources arrive at a single multiplexer (i.e., no pre-aggregation).

A simple shaping scheme for frame-based bandwidth allocation

Packet switching devices that rely on input buffering to prevent cell loss in the switch fabric must regulate the arrival pattern of data packets to the input links of the switch fabric. Frame-based bandwidth allocation can be used in cell-based switch fabrics to regulate the number of cells from a flow entering the fabric during a fixed-length frame of cell slots. We present a simple cell slot assignment algorithm for distributing each flows cell slot allocations within a frame. The algorithm controls the inter-cell spacing for each flow and bounds fabric queues to a small number independent of the frame length. The simplicity of the algorithm supports per-frame changes in the allocated rates.

Mean square error in an IIR adaptive echo canceller

An examination is made of the mean-square-error performance of an IIR (infinite impulse response) adaptive filter. The principal goal is to determine if IIR adaptive filters have an advantage over FIR (finite impulse response) adaptive filters in terms of either greater performance or less restrictive processing constraints. A stochastic analysis of the IIR algorithm, which results in a steady-state mean square error expression, is presented and that result is used as a tool to investigate performance in the echo canceller application. Simulations are used to verify the analytical result. The echo canceller study shows that the IIR algorithm offers approximately a 2:1 reduction in processing requirements compared to the FIR approach, without loss of performance.<<ETX>>