CLEAN-SC波束形成声源识别改进

CLEAN-SC波束形成声源识别方法计算速度快、成像干净清晰、结果准确度高,但当传统延迟求和算法在各声源处输出的主瓣严重融合时,亦无法准确分辨声源。造成该缺陷的原因为:主瓣严重融合时,CLEAN-SC所基于的延迟求和输出峰值所在聚焦点即为声源点的假设不成立。从源相干性角度,若某聚焦点处的延迟求和输出主要由某声源贡献时,该聚焦点可标示该声源,即基于该聚焦点的位置及强度信息可重构该声源在各传声器处产生声压的互谱矩阵。鉴于此,以CLEAN-SC识别的声源为初值迭代寻找正确的声源位置及强度,每次迭代中,最小化其余声源与某一声源的波束形成贡献的比值为每个声源选择标示点,根据标示点更新声源。仿真及试验均证明:所给方法比传统CLEAN-SC具有更高分辨率,使近距离低频率声源的准确识别变得可行。

Improved acoustic source identification based on CLEAN-SC beam forming

Acoustic source identification method based on CLEAN-SC beam forming has advantages of fast calculation speed, clear images and higher accuracy. However, it can’t recognize acoustic sources when main lobes output with the traditional delay-summing algorithm at various acoustic sources are seriously fused. The reason is that when main lobes are seriously fused, the assumption that the focus point of peak values from delay-summing output is the acoustic source point is not valid, while CLEAN-SC is based on this assumption. From the view point of source coherence, a focus point can be marked an acoustic source if delay-summing output at the focus point is mainly contributed by a certain acoustic source. The cross-spectral matrix of acoustic pressures at various microphones induced by the acoustic source can be reconstructed with the location and intensity information of this focus point. Here, the identified acoustic source based on CLEAN-SC was taken as the initial value to iterate and seek the correct acoustic source location and intensity. In each iteration, the ratio of contribution of each other source’s beam forming to that of a certain acoustic source’s one was minimized and used to select the mark point for each source, and sources were updated according to their mark points. Simulations and tests showed that compared with the traditional CLEAN-SC, the proposed method has higher resolution to make the correct identification of closely spaced and lower frequency acoustic sources become feasible.