Ferruccio Barsi

Fast base extension and precise scaling in RNS for look-up table implementations

Both base extension and scaling are fundamental operations in residue computing and several techniques have been proposed previously for their efficient implementation. Using look-up tables, the best result (log/sub 2/ n table took-up cycles, where n is the number of residue moduli in the system) has been obtained by using the Chinese remainder theorem (CRT) at the expenses of a redundant representation of the numbers and of an approximated scaling. The CRT approach is reconsidered and it is shown that the same average time performances (log/sub 2/ n lookup cycles) can be achieved without any redundancy and with a precise result for scaling. >

Efficient heuristics for data broadcasting on multiple channels

The problem of data broadcasting over multiple channels consists in partitioning data among channels, depending on data popularities, and then cyclically transmitting them over each channel so that the average waiting time of the clients is minimized. Such a problem is known to be polynomially time solvable for uniform length data items, while it is computationally intractable for non-uniform length data items. In this paper, two new heuristics are proposed which exploit a novel characterization of optimal solutions for the special case of two channels and data items of uniform lengths. Sub-optimal solutions for the most general case of an arbitrary number of channels and data items of non-uniform lengths are provided. The first heuristic, called Greedy+, combines the novel characterization with the known greedy approach, while the second heuristic, called Dlinear, combines the same characterization with the dynamic programming technique. Such heuristics have been tested on benchmarks whose popularities are characterized by Zipf distributions, as well as on a wider set of benchmarks. The experimental tests reveal that Dlinear finds optimal solutions almost always, requiring good running times. However, Greedy+ is faster and scales well when changes occur on the input parameters, but provides solutions which are close to the optimum.

Asynchronous Corona Training Protocols in Wireless Sensor and Actor Networks

Scalable energy-efficient training protocols are proposed for wireless networks consisting of sensors and a single actor, where the sensors are initially anonymous and unaware of their location. The protocols are based on an intuitive coordinate system imposed onto the deployment area, which partitions the sensors into clusters. The protocols are asynchronous, in the sense that the sensors wake up for the first time at random, then alternate between sleep and awake periods both of fixed length, and no explicit synchronization is performed between them and the actor. Theoretical properties are stated under which the training of all the sensors is possible. Moreover, both worst-case and average case analyses of the performance, as well as an experimental evaluation, are presented showing that the protocols are lightweight and flexible.

Efficient Location Training Protocols for Heterogeneous Sensor and Actor Networks

In this work, we consider a large-scale geographic area populated by tiny sensors and some more powerful devices called actors, authorized to organize the sensors in their vicinity into short-lived, actor-centric sensor networks. The tiny sensors run on miniature nonrechargeable batteries, are anonymous, and are unaware of their location. The sensors differ in their ability to dynamically alter their sleep times. Indeed, the periodic sensors have sleep periods of predefined lengths, established at fabrication time; by contrast, the free sensors can dynamically alter their sleep periods, under program control. The main contribution of this work is to propose an energy-efficient location training protocol for heterogeneous actor-centric sensor networks where the sensors acquire coarse-grain location awareness with respect to the actor in their vicinity. Our theoretical analysis, confirmed by experimental evaluation, shows that the proposed protocol outperforms the best previously known location training protocols in terms of the number of sleep/awake transitions, overall sensor awake time, and energy consumption.

A fully parallel algorithm for residue to binary conversion

Abstract A new method for converting numbers from a residue system to a binary notation is proposed which is based upon an original formulation of the Chinese Remainder Theorem. To prove its effectiveness in VLSI implementations a converter is presented and evaluated following a general VLSI model of computation. The proposed structure compares favourably with preceding results presented in the literature; in particular, the time which is required to perform conversion is O(log s ), with s representing the total number of input bits.

Mod m arithmetic in binary systems

Abstract Even if the most common approach to mod m arithmetic is based on look-up tables, the use of binary systems is valid in those situations, such as large-moduli residue arithmetic or conversion processes, where a memory approach is not viable. A general approach to the problem of performing mod m computations in binary systems is presented. The proposed solution proves useful in various applications, such as converting binary integers to residue notation and mod m addition or multiplication. Examples are given together with possible VLSI implementations.

Asynchronous training in wireless sensor networks

A scalable energy-efficient training protocol is proposed for massively-deployed sensor networks, where sensors are initially anonymous and unaware of their location. The protocol is based on an intuitive coordinate system imposed onto the deployment area which partitions the anonymous sensors into clusters. The protocol is asynchronous, in the sense that the sensors wake up for the first time at random, then alternate between sleep and awake periods both of fixed length, and no explicit synchronization is performed between them and the sink. Theoretical properties are stated under which the training of all the sensors is possible. Moreover, a worst-case analysis as well as an experimental evaluation of the performance is presented, showing that the protocol is lightweight and flexible.

Cheapest Paths in Multi-interface Networks

Multi-interface networks are characterized by the property that each node in the network might choose among several communication interfaces in order to establish desired connections. A connection between two neighboring nodes is established if they both activate a common interface. Each interface is associated with an activation cost. In this context we investigate on the so called Cheapest Path problem, i.e., given a node s of the network, for each other node j , we look for the cheapest set of interfaces that must be activated among the nodes in order to establish a path between s and j . Polynomial algorithms are provided.

A VLSI architecture for RNS with MI adders

Abstract Over several years, RNS applications were limited to addition, subtraction and multiplication with results expected within a predetermined range because of the absence of explicit information on number magnitude in the residue representation. Hybrid notations have been proposed to overcome this obstacle. In this paper, an architecture for adding and overflow checking is presented which is based upon Residue Number Systems with Magnitude Index (RNS with MI) and its area-time complexity is evaluated. It is shown that considerable execution time reduction may result for a wide class of applications at the cost of a slight increase of area cocupancy as compared with binary realizations.

Efficient binary schemes for training heterogeneous sensor and actor networks

Sensor networks are expected to evolve into long-lived, autonomous networked systems whose main mission is to provide in-situ users - called actors - with real-time information in support of specific goals supportive of their mission. The network is populated with a heterogeneous set of tiny sensors. The free sensors alternate between sleep and awake periods, under program control in response to computational and communication needs. The periodic sensors alternate between sleep periods and awake periods of predefined lengths, established at the fabrication time. The architectural model of an actor-centric network used in this work comprises in addition to the tiny sensors a set of mobile actors that organize and manage the sensors in their vicinity. We take the view that the sensors deployed are anonymous and unaware of their geographic location. Importantly, the sensors are not, a priori,organized into a network. It is, indeed, the interaction between the actors and the sensor population that organizes the sensors in a disk around each actor into a short-lived, mission-specific, network that exists for the purpose of serving the actor and that will be disbanded when the interaction terminates. The task of setting up this form of actor-centric network involves a training stage where the sensors acquire dynamic coordinates relative to the actor in their vicinity. The main contribution of this work is to propose an energy-efficient training protocol for actor-centric heterogeneous sensor networks. Our protocol outperforms all know training protocols in the number of sleep/awake transitions per sensor needed by the training process. Specifically, in the presence of κ coronas, no sensor will experience more than ⌈log κ⌉ sleep/awake transitions and awake periods.