Advanced digital filters without and with neural techniques

点击翻译按钮获取中文摘要

Filters are core sub-systems in an expanding spectrum of products in communications, consumer electronics, and others. Emerging applications have placed an increasing demand on the robustness, speed, and power of a filter to be accomplished within a shrinking product cycle. This research attempts to provide practical solutions to the associated problems. The robustness of a filter refers to its coefficient sensitivity, roundoff noise, and stability properties, and to its abilities to handle overflow. Different filter structures have different robustness properties. For a specific application, what will be the optimal structure to achieve desired robustness? Real-time functionalities and applications demand a filter to operate extremely fast and with low delay. Different filter structures have difference speed and delay properties. For a specific application, what is the best structure for real-time operations? Nowadays, portable applications impose an increasing demand on battery with a longer operating time, while overheating reduces processing speed and life. There exist different techniques for reducing the power consumption. For a specific application, what will be the best strategy for low-power solution? Once a filter structure is decided, we need to design its coefficients for applications. Recursive filters are more compact but their designs involve nonlinear problems that have no global solution. For a specific application, what will be the best design strategy? To answer the above questions, we propose to continue research on optimal structures and design algorithms for digital filters, Gabor filters, and adaptive neural/fuzzy-neural filters. Some of the research ideas are summarized here. In the search of an optimal filter structure, what is the relationship between the robustness of a filter structure and its passivity and stability properties? The most time-consuming and power-consuming element in a filter is multiplier. Is there a smart way to reduce or eliminate multipliers in a filter structure? If feasible, it would indeed provide a high-speed and low-power solution. What is the best way to solve the difficult nonlinear design problem of a recursive filter? How about a combination of techniques such as a combined global and local search?