Signal Processing and Information Technology. Second International Joint Conference, SPIT 2012, Dubai, UAE, September 20-21, 2012, Revised Selected Papers

Research Article

Effect of Finite Wordlength on the Performance of an Adaptive Network

Download63 downloads
  • @INPROCEEDINGS{10.1007/978-3-319-11629-7_17,
        author={Wael Bazzi and Amir Rastegarnia and Azam Khalili},
        title={Effect of Finite Wordlength on the Performance of an Adaptive Network},
        proceedings={Signal Processing and Information Technology. Second International Joint Conference, SPIT 2012, Dubai, UAE, September 20-21, 2012, Revised Selected Papers},
        proceedings_a={SPIT},
        year={2014},
        month={11},
        keywords={adaptive networks distributed estimation least mean-square (LMS) quantization},
        doi={10.1007/978-3-319-11629-7_17}
    }
    
  • Wael Bazzi
    Amir Rastegarnia
    Azam Khalili
    Year: 2014
    Effect of Finite Wordlength on the Performance of an Adaptive Network
    SPIT
    Springer
    DOI: 10.1007/978-3-319-11629-7_17
Wael Bazzi1,*, Amir Rastegarnia2,*, Azam Khalili2,*
  • 1: American University in Dubai
  • 2: Malayer University
*Contact email: wbazzi@aud.edu, a_rastegar@ieee.org, a.khalili@ieee.org

Abstract

In this paper we consider the performance of incremental least mean square (ILMS) adaptive network when it is implemented in finite-precision arithmetic. We show that unlike the infinite-precision case, the steady-state curve, described in terms of mean square deviation (MSD) is not always a monotonic increasing function of step-size parameter. More precisely, when the quantization level is small, reducing the step-size may increase the steady-state MSD.