Adaptive Noise Cancellation with Computational-Intelligence-based Approach

A new intelligent noise filtering approach using Computational Intelligence (CI) is proposed for the problem of adaptive noise cancellation (ANC). Since the traditional linear filtering may not be good enough to handle with the noise complexity and time-varying statistic property, a self-constructing neuro-fuzzy system (SCNFS) is used as an adaptive filter to deal with the nonlinearity of noise. A hybrid machine learning algorithm with the methods of both random optimization algorithm (RO) and least square estimation (LSE) is introduced to enable the SCNFS with learning capability. The learning capability includes both the parameter learning and the structure learning. In the parameter learning phase, the premises and the consequents of the SCNFS are updated by RO and LSE, respectively. In the SCNFS structure learning, the system structure can be generated or rearranged using the proposed mechanism with rule-splitting and/or rule-expanding. To demonstrate the feasibility and the capability of the proposed approach, an example of adaptive speech noise cancellation is illustrated. With the experimental results, the SCNFS shows excellent filtering performance for noise cancellation. Key-Words: computational intelligence, neuro-fuzzy, learning, random optimization, least square estimation, adaptive noise cancellation.