Team :
Gopalakrishna , Varun
Khanikar, Prakash

TITLE : Adaptation Behaviour of Pipelined Adaptive Digital Filters

ABSTRACT

Adaptive digital filters are difficult to pipeline due to the presence
of long feedback loops, careful calibration of step size and depth of
pipelining.  The LMS algorithm using the stochastic gradient approach is
implemented with recursive weight adaptation.  However with the RLS
algorithm , the resulting rate of convergence is typically an order of
magnitude faster than the LMS algorithm.  We implement the exponentially
weighted RLS algorithm which converges in the mean square sense in about
2M iterations, where M is the number of taps in the transversal filter. 

Pipelined implementation of these adaptive filters yield higher
throughput, higher sample rates and low power designs.The relaxed
lookahead transformation techniques are implemented to pipeline the
adaptive filters with no increase in hardware.  Stability analysis for
these filters are performed (with pole zero analysis) on a preliminary
basis with different stages of pipelining.  The above simulations are
run in Matlab.  With extensive literature search we also intend to
present a systolic array architecture for the RLS algorithm to reduce
the overhead and issues with computation complexity. 
Concurrencies/parallelism available in the computation of both these
implementations need to be investigated and exploited with the
pipelining and parallel processing techniques.