Improved DCT domain variable step-size Griffith's LMS algorithm based active noise control using observation noise cancellers for secondary path identification

Abstract

This paper proposes two improved DCT domain active noise control (ANC) systems using observation noise cancellers (NC) for the secondary path identification (SPI). This is achieved by identifying the main path by an adaptive filter which is based on input orthogonalisation by a DCT. Further for the proposed first ANC system, the available input orthogonal components are beneficially used in estimating the observation noise component required for the adaptive NC for the SPI. For the proposed second ANC system, the NC is based on an observation noise robust discrete cosine transform domain variable step-size Griffiths LMS (TVGLMS) algorithm. In both ANC systems, the NC enables improved SPI and this indirectly enables an improved ANC convergence performance. The second ANC system uses a noise robust variable step-size and gradient for its NC used in deriving the desired signal for SPI of an ANC system with TVGLMS noise canceller. The smoothed ensemble average square error (SEASE) is reduced by an additional 5 dB and its convergence rate is about 1.5 times faster over that of an ANC which uses only orthogonal components of its input. The performance of basic first ANC which uses input orthogonalisation only for its main path identification is very poor as its noise reduction is only -2.5 dB.