Abstract

In this work, we introduce, for the first time, the design of a compact neuromorphic architecture to efficiently support a filtered-x error-coded affine projection-like (FXECAP-L) algorithm that is based on affine projection (AP) algorithms for active noise cancellation (ANC) in an acoustic duct. To date, few practical ANC implementations have used AP algorithms because of their high computational complexity, despite providing fast convergence speeds. One of the main factors that increases their computational complexity is linked to the dimensions of the matrix used in the AP algorithm’s computations. Evidently, the largest dimensions of the matrix increase the convergence speed of the AP algorithms by paying a penalty in terms of area consumption. However, convergence speed is crucial in ANC applications since this factor determines the speed at which the noise is canceled. Recently, an FXECAP-L algorithm with evolving order has been proposed to dynamically reduce the dimensions of the matrix by maintaining the convergence speed of AP algorithms. Here, we propose a compact neuromorphic architecture with a dynamic routing mechanism to efficiently implement the evolutionary method of the FXECAP-L algorithm by creating a virtual matrix, whose dimensions can be modified over the filter processing. In this way, we avoid spending a large amount of memory to save the largest matrix elements. In addition, the inclusion of the dynamic routing mechanism in the proposed neuromorphic architecture has allowed us to guarantee low area consumption since the neuromorphic architecture is capable of simulating different adaptive structures without modifying its structure. Here, the neuromorphic architecture has been configured as the system identification and ANC controller for practical noise cancellation in an acoustic duct. Our results have demonstrated that the combination of the properties of the FXECAP-L algorithm and the implementation techniques generate a versatile signal processing development tool that can be used in practical real-time ANC applications.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call