A novel singular spectrum analysis-based multi-objective approach for optimal FIR filter design using artificial bee colony algorithm


LATİFOĞLU F.

NEURAL COMPUTING & APPLICATIONS, cilt.32, ss.13323-13341, 2020 (SCI İndekslerine Giren Dergi) identifier identifier

  • Cilt numarası: 32 Konu: 17
  • Basım Tarihi: 2020
  • Doi Numarası: 10.1007/s00521-019-04680-1
  • Dergi Adı: NEURAL COMPUTING & APPLICATIONS
  • Sayfa Sayıları: ss.13323-13341

Özet

Effective filter design plays an important role in signal processing applications. Multiple parameters must be considered to control the over-frequency response of the designed filter. In this study, a novel multi-objective approach is proposed for windowing finite impulse response (FIR) filter design. The windowing FIR filters are commonly used due to its linear phase property, frequency stability and easier implementation. However, windowing method can only control the cutting frequency of filter, and it suffers from the problem of insufficient control of the transition bandwidth, pass and stop band cutoff frequencies. Therefore, the window function was optimized using a novel multi-objective artificial bee colony (ABC) algorithm based on singular spectrum analysis (SSA) to eliminate these disadvantages of the windowing method. The proposed method was compared to three other multi-objective ABC variants. Novel SSA-based multi-objective approach yielded the best performance among four approaches. The proposed multi-objective approach that uses the SSA method has a significant advantage since it does not require user experience, it is not dependent on parameters, and there is no weight determination problem. Also, it does not have sorting and pooling stages that increase the cost of calculation. The obtained results were compared with the published literature studies. The SSA-based multi-objective approach offered better alternative to other literature techniques in terms of calculating the fitness function that deals with finding the most reasonable solution considering all error terms. Finally, the performance of the designed filter was tested on electroencephalography (EEG) signal. The EEG signal was decomposed successfully into subbands using proposed filter design approach. Based on numerical results of this study, the proposed filter provided the low-pass band and stop band ripple, and high stop band attenuation value of all, while having well enough performance.