1: \begin{abstract}

2: This paper investigates the problem of implementing

3: proportionate-type LMS family of algorithms in hardware for sparse

4: adaptive filtering applications especially the network echo

5: cancelation. We derive a re-formulated proportionate type algorithm

6: through algorithm-architecture co-design methodology that can be

7: pipelined and has an efficient architecture for hardware

8: implementation. We study the convergence, steady state and tracking

9: performances of these re-formulated algorithms for white, color and

10: speech inputs before implementing them in hardware. To the best of

11: our knowledge this is the first attempt to implement

12: proportionate-type  algorithms in hardware.  We show that Delayed

13: $\mu$-law Proportionate LMS (DMPLMS) algorithm for white input and

14: Delayed Wavelet MPLMS (DWMPLMS) for colored input are the robust

15: VLSI solutions for network echo cancelation where the sparsity of

16: the echo paths can vary with time. We implemented all the designs

17: considering $16$-bit fixed point representation in hardware,

18: synthesized the designs and synthesis results show that DMPLMS

19: algorithm with $\approx25\%$ increase in hardware over conventional

20: DLMS architecture, achieves $3X$ improvement in convergence rate for

21: white input and DWMPLMS algorithm with $\approx58\%$ increase in

22: hardware achieves $15X$ improvement in convergence rate for

23: correlated input conditions.

24: 

25: \end{abstract}