Advanced Simulation Methods for Switched Networks

[摘要] This thesis addresses several difficult simulation problems that have arisen in the rapid development of analogue sampled-data systems with the drive to more circuit complexity and demands by designers for more sophisticated simulation tools the current techniques and related software packages were not capable of providing satisfactory solutions. Nonideal sensitivity analysis is found to be very useful in filter design and optimisation. A method for generating derivatives of the extended state transition matrix and the excitation response matrix is proposed. Most nonidealities of interest, such as gain and bandwidth product of the opamp, input and output impedances of the opamp, switch resistance and nodal parasitics can be exactly investigated. Group delay sensitivity and overall sum of sensitivities evaluation capabilities, offer accurate measures for choosing better circuit realisation in terms of low sensitivity properties. Noise analysis of sampled-data systems is always believed to be very time consuming. An efficient and systematic noise analysis technique is presented. The use of adjoint techniques, spectrum analysis together with highly efficient software implementation makes the noise analysis of practical circuits feasible on a modest workstation. It is also suggested that various accurate noise models are used in order to obtain sensible results. A polynomial interpolation method is used for semi-symbolic analysis of large nonideal switched linear networks to which other symbolic approaches cannot apply. Measures for retaining the accuracy of the interpolation method are discussed and the results are very encouraging. The application of symbolic analysis to evaluate the noise behaviour is described. The comparison of theoretical computation costs is introduced, from which the potential efficiency of the symbolic method can be estimated. Nonlinear time domain analysis of sampled-data networks is usually performed by a general purposed circuit simulator. Unfortunately, this is very extravagant in computer time. An effective strategy is proposed from which the dominant nonlinearities present in sampled-data circuits can be efficiently analysed. The use of a piecewise linear description of the nonlinear characteristic allows many efficient linear techniques to be employed. Oversampling techniques relax the precision requirement of analogue circuits and provide high resolution of analogue to digital conversion. However the simulation of oversampled sigma-delta modulator is almost impossible by traditional methods. A generalisation of present periodic linear techniques to handle this problem is presented. A new definition of network state is proposed and a series of efficient computer implementation strategies are described. This has facilitated the linking of a high refined switched network simulator to standard commercial logic simulator to provide a good mixed-mode simulation.

[发布机构] University:University of Glasgow

[关键词] Computer engineering