DESIGN OF ADAPTIVE TRANSVERSAL FILTER UNDER RANDOM DELAY VARIATIONS

Transversal filters consist of fundamental circuit elements such as adders, multipliers, and unit delay elements. The performance of these filters is affected by inaccuracies in these components, particularly limited precision in delay elements, which becomes significant in high-frequency semi-digital transversal filters. This letter proposes a cascaded delay element structure to mitigate precision errors. The main delay element is complemented by smaller cascaded delay elements, refining overall delay precision. To further enhance accuracy, a neural network (NN)-based adaptation scheme dynamically fine-tunes delay adjustments in real time. The proposed two-layer NN takes inputs from both the primary and cascaded delay elements and generates an optimized output for the next delay stage. The input layer neurons are randomly initialized, while the NN weights are iteratively updated using gradient descent to minimize errors. The neural network weights are determined during an initial factory‑calibration stage and remain fixed during all subsequent filter operation. Simulation results demonstrate that the cascaded delay structure, combined with NN adaptation, significantly reduces precision errors, enhancing semi-digital transversal filter performance for high-speed signal processing applications.