Optimal Design of Digital Analysis Filters Based on PSO-BPNN for Aliasing Errors Cancellation in HFB DAC

• Published in: Engineering letters Vol. 33; no. 1; p. 159
• Main Authors: Wang, Yan, Liu, Shengjian, Yang, Xing, Zhang, Weiyuan, Yang, Jiansheng, Yang, Jiyao
• Format: Journal Article
• Language: English
• Published: Hong Kong International Association of Engineers 01.01.2025
• Subjects: Algorithms; Aliasing; Analogue filters; Approximation; Artificial neural networks; Back propagation networks; Complexity; Design; Design analysis; Design optimization; Digital to analog converters; Error analysis; Error reduction; Filter banks; Mixers; Neural networks; Neurons; Particle swarm optimization; Transfer functions
• ISSN: 1816-093X; 1816-0948

To cancel the aliasing errors introduced by the non-ideal characteristics of analog filters and mixers, this paper proposes a method that combines particle swarm optimization (PSO) with back propagation neural networks (BPNN) for designing digital analysis filters in hybrid filter bank digitaltoanalog converter (HFB DAC). A mathematical model for the HFB DAC is initially established to derive both its desired and practical transfer functions, facilitating the calculation of the estimation error between them. Next, the approximation error is derived from both the real and imaginary components of the estimation error. A BPNN method is then proposed to minimize the approximation error. To reduce the computational complexity of the traditional BPNN design, we also proposed a PSO algorithm for optimizing all the sub-filter orders, the number of the BPNN hidden layer neurons and the iterations of the BPNN, enabling that the given upper limit errors (upper limits of distortion and aliasing errors) is met. Finally, these optimized parameters are applied to the BPNN method, thereby deriving the optimal coefficients for digital analysis filters. Additionally, this paper derives the computational complexity of PSO-BPNN. Several design examples indicate that, comparing with the other four designs, our proposed PSO-BPNN design not only achieves better aliasing errors cancellation but also reduces all the sub-filter orders, the number of the BPNN hidden layer neurons and the iterations of the BPNN.