0. 摘要
大部分多通道声源DOA估计算法遭受空域混叠问题。当频率超过混叠频率,会产生IPD卷绕问题。使得DOA估计产生较大误差。本文提出一种实时的算法解决IPD卷绕问题。算法分为两步,
In the first step, the obtained IPDs are replicated to all the possible 2π-periodic repetitions, thus performing the correct unwrapping while introducing errors inevitably. However, the replication can also emphasize the correct DOAs of the sources when the replicated matrix is transformed into a histogram.
The second step contains a post-processing method supporting our method more robust for a noisy environment.
1. 介绍
空间混叠在窄带声源定位领域是一大难题。空域混叠频率和麦克风间距和声源的入射角有关。避免 空域混叠的方法有两种,一是减小麦克风间距,一是相位解卷绕。
相位解卷绕算法可以分为三类:Firstly, sequential unwrapping algorithms。These methods process the phase samples sequentially along the frequency axis. Secondly, there are algorithms
which apply multi-stage schemes assisted by subband decomposition for suppressing aliasing components. Thirdly, in our previous work [20] we introduced a method that first estimates
the aliasing frequency based on the autocorrelation of the wrapped IPDs. IPDs are then unwrapped, and several post-processing steps are performed to compensate potential unwrapping errors.
2. 问题定义
传统的方法通过估计不同麦克风的Time Difference Of Arrival (TDOA, Δt)来进行声源定位。
第i频点的 IPD和TDOA的关系如下式:
远场场景,声波为平面波,IPD和DOA的关系如下式;
针对窄带:
针对宽带:
3. 本文提出方法
3.1. IPD matrix replication
3.2. Theoretical analysis