James Brian Piket

Partitioned echo canceller utilizing decimation echo location

In practice, digital echo cancellers require design trade-offs between the effective range, rate of convergence, and system resource (sizing and timing) requirements. While echo cancelling over an extended time range is often desirable, for a standard LMS approach it comes at the expense of increased system resource requirements and decreased convergence rates. In this paper, the authors investigate a Decimation Echo Location (DEL) technique which provides a novel approach for identifying the high energy regions of the echo impulse response model over an extended range. Accurate echo location provided by the DEL process enables a subsequent partitioning of an echo canceller. The Partitioned Echo Canceller (PEG) structure was developed to operate only on the identified regions of interest. In this manner, the echo canceller can be effective over an extended range without sacrificing system resource or performance requirements. DEL/PEC simulations have readily yielded a factor of five increase in algorithm efficiency compared to a conventional LMS approach, with additional refinements promising even greater gains. For a system with a predetermined modeling range requirement, DEL/PEC provides a solution with significantly reduced resource requirements. Similarly, for an application with a fixed system resource allotment, this technique greatly extends the effective modeling range capability.