Moreno Coli

Real time pipelined system design through simulated annealing

Abstract This paper is concerned with automatic pipeline implementation of a program subject to some real time (RT) constraints; the program is described through a Control Data Flow Graph (CDFG). We have developed a mapping methodology which assigns to each instruction of CDFG a time step and a HW resource for its execution. We have defined the space Ω of all the possible feasible mappings, as well as an adjacency criterion on it and a cost function evaluating the quality of the mappings. We have minimized the cost function through a Simulated Annealing algorithm. The minimization process returns a mapping which satisfies all RT constraints, has minimal schedule length and minimal HW resource requirement. In order to show the capabilities of the proposed mapping methodology, we apply it to a graph with 50 nodes and several RT constraints: the obtained mapping gives a pipelined execution modality of the graph which satisfies all the given RT constraints.

Synthesis of pipelined systems for the contemporaneous execution of periodic and aperiodic tasks with hard real-time constraints

Summary form only given. We address the design of pipelined systems able to sustain the throughput of a given periodic task and, at the same time, to serve aperiodic requests associated with hard real-time constraints. The proposed method is based on the allocation of the global graph (periodic and aperiodic tasks), over-dimensioning the design of the system devoted to process the periodic task, deserving the unused parts of the resources to the management of aperiodic requests. A formal definition of such a mapping problem, together with the formalization of the searching space, is given. The searching space is structured in a way such as the minimization process moves toward a solution, which satisfies, if possible, all the real-time constraints and has minimal HW requirements. Once formulated as a minimization problem, the pipelined architecture and the corresponding scheduling are determined by means of a simulated annealing algorithm. A theorem is given to ensure that all the feasible mappings are reachable in the optimization process.

A new method for optimization of allocation and scheduling in real time applications

Performance improvements achievable through parallel processing are useful in real time (RT) environments. The paper describes a method to map (i.e. allocate and schedule) a program with some RT constraints into a parallel system. We formulate the mapping problem as a minimization problem, defining a new cost function whose minimization leads to the optimal mapping of the program into the parallel system. The searching space over which the minimization must be carried out is defined; this space encloses all the feasible allocation and scheduling modalities for the program in the parallel system. The minimization is carried out through a simulated annealing algorithm, so we define an adjacency criterion on the searching space. Some examples illustrating the capabilities of the proposed method are presented.

A new crossover operator for genetic algorithms

Starting from a mathematical reinterpretation of the classical crossover operator, a new type of crossover is introduced. The proposed new crossover operator gives better performances than the classical 1 point, 2 point or uniform crossover operators. A theoretical investigation of the behaviour of the new crossover is presented. Compared to the classical crossover operators, it allows better exploration of the searching space and gives better results. Some comparative results relative to the optimization of test functions taken from literature are given.

Massively parallel processing approach to fractal image compression with near-optimal coefficient quantization

Abstract In recent years Image Fractal Compression techniques (IFS) have gained more interest because of their capability to achieve high compression ratios while maintaining very good quality of the reconstructed image. The main drawback of such techniques is the very high computing time needed to determine the compressed code. In this work, after a brief description of IFS theory, we introduce the coefficient quantization problem, presenting two algorithms for its solution: the first one is based on Simulated Annealing while the second refers to a fast iterative algorithm. We discuss IFS parallel implementation at different level of granularity and we show that Massively Parallel Processing on SIMD machines is the best way to use all the large granularity parallelism offered by the problem. The results we present are achieved implementing the proposed algorithms for IFS compression and coefficient quantization on the MPP APE100/Quadrics machine.

Virtual cut-through implementation of the HB packet switching routing algorithm

In the last years, since the early 80s, wormhole and virtual cut through routing modalities have replaced packet switching and circuit switching schemes. We give the theoretical bases to implement hole based (HB) routing algorithm (M. Coli and P. Palazzari, 1995) by using the virtual cut through modality. After reviewing the theory of HB routing, we demonstrate some theorems which allow us to use such a theory, developed for the packet switching case, also for the virtual cut through modality (HB-VCT routing algorithm). The main features of HB-VCT are its full adaptivity, the high fault tolerance capabilities and the low HW requirement. Furthermore, low latencies and high bandwidths are achievable because the use of virtual channels can be avoided. Some simulated results are presented in order to show the good traffic balance capabilities and the fault tolerant behavior of HB-VCT. Simulation is also used to show the influence on throughput of traffic intensity and of a HB-VCT control parameter (the timeout).

Nonlinear system process prediction using neural networks

A novel iterative technique is proposed by the authors in order to construct a discrete-time nonlinear dynamical system predictor from experimental input-output pairs. The proposed technique reduces the memory amount required to construct the predictor taking into account the intrinsic redundancy of the input-output pairs of the experimental data due to underlying physical laws. The paper deals with SISO systems, anyway it is easy to extend the results to the MIMO case. The technique is very well suited to work in conjunction with the cerebellar model arithmetic computer (CMAC) or radial basis functions (RBF). A convergence discussion on the proposed algorithm is provided. Finally computer simulations verify the stated theory.

Global execution time minimization by allocating tasks in parallel systems

We have studied the allocation of directed acyclic graphs (DAGs) into a given parallel machine (PM); this is an NP-complete problem. Previous papers presented allocation algorithms all making many rough simplifications so that the achieved allocations are too far from the optimum and do not minimize the actual execution time of the program. We analyzed the impact of the precedence relations on the execution time of DAG into PM issuing a new cost function (f/sub PR/) which takes into account both the PM topology and the precedence relations. f/sub PR/ is minimized through a genetic algorithm and, in order to speed up its convergence, we developed a heuristic criterion, based on the critical path idea, for the choice of the starting population. The best results achievable using our cost function have been illustrated by comparing the actual execution times of the allocation given by the minimization of f/sub PR/ with the ones obtained using the allocations given by the minimization of two cost functions described in literature.<<ETX>>

The toroidal neural networks

In this paper we present the toroidal neural networks (TNN), a new class of neural network derived from discrete time-cellular neural networks (DT-CNN). TNN are characterized by 2D toroidal topology with local connections, by binary outputs and by a simple equation describing the dynamic of neuron states; binary outputs are obtained comparing initial and final states. Due to the expression of state dynamic, TNN learning has a very appealing geometric interpretation: a transformation, specified by means of a training input sequence, is represented through a polyhedron in the TNN weight space. Along with the definition and theory of TNN, we present a learning algorithm which, for a given transformation expressed by means of a training sequence, gives the set of TNN weights (if existing) which exactly implement the transformation: such a set of weights is a point belonging to the polyhedron representing the training sequence. Furthermore, the algorithm gives the exact minimal spatial locality characterizing the problem; in order to reduce the number of TNN weights, a heuristic is used to try to move neuron connectivity from the spatial to the temporal dimension.

A new statistical model of non linear noisy oscillator

In this paper we propose a model of noisy oscillator to describe the effects of white noise sources on amplitude and phase noise spectrum that can be applied to linear and non-linear structures. This work proposes an extension of previous works to take into account deeper considerations about Analytical Signal and Averaging methodologies to extract a new model for oscillator dynamics. The Noisy Oscillator model has shown an excellent agreement to literature works, and results obtained with the proposed model have been compared to simulations performed with SpectreRF in Cadence 4.4.3 on a LC oscillator, in order to provide model validation.