Frequency dependent regularization of a mixed-order Ambisonics encoding system using psychoacoustically motivated metrics

In mixed-order Ambisonics (MOA) encoding the right choice of the regularization parameter λ is crucial to the reproduction quality of the system. The choice of λ is a trade-off between low microphone noise amplification and high directivity. Commonly used performance metrics are not suited to determine an optimum λ(f)-curve due to their limited psychoacoustical relevance. Therefore, it is proposed here to use two perceptually motivated metrics that consider the reproduced signals at the listener's ear to quantify the reproduction error and the audibility of the total microphone noise. Regularization was then optimized towards previously defined requirements on both of these metrics separately. The results showed that at frequencies below 500Hz both requirements can be met. Above this frequency one or both of the requirements have to be relaxed in order to find a suitable λ(f).