Hierachical Bandwidth Limitation: MSBTF vs. KLT

The MSBTF transform is independent of programme content and uses fixed transform matrices; the transformed signals are ordered according to their spatial importance assumed by Gerzon. However, the actual order of transformed signals may be incorrect for some excerpts in terms of the perceptual importance, which has been confirmed by means of some preliminary experiments conducted by the authors.

While in KLT, the transformed signals are statistically independent and coefficients in the encoding and decoding matrix are based on the signal content. The results of a subjective experiment showed that the signals of KLT eigenchannels were typically ordered according to their perceptual importance [P2]. It is expected that the signal-dependent KLT could achieve better audio quality than the signal-independent MSBTF as a means of preprocessing of bandwidth limitation, however this supposition requires experimental verification.

In this work, the perceptual effects of the hierarchical bandwidth limitation based on MSBTF and KLT techniques were studied. The main research questions were as follows:

1. How does hierarchical bandwidth limitation affect the basic audio quality of surround sound? How is it influenced by the nature of audio excerpts?

2. Which hierarchical encoding technique is better for bandwidth limitation: KLT (signal dependent) or MSBTF (signal independent)?

3. What kind of bandwidth allocation strategy is suitable for the hierarchical bandwidth limitation? In order to answer these questions, a formal listening test was carried out the method and results of which are presented in this paper.

In order to answer these questions, a formal listening test was carried out. Two different bandwidth allocation strategies were used. In the case of MSBTF1 and KLT1 algorithms, the highly ranked channels in the hierarchy were kept with full bandwidth and the channels with lower ranks were discarded completely (Profile 1). Removal of signals was considered as a case of "extreme bandwidth limitation". In the case of MSBTF2 and KLT2 methods, the bandwidth was step-down along with the decreasing ranks in the hierarchy (Profile 2). The results showed that (see Figure 1):

1. The signal dependent hierarchical encoding technique (KLT) is better than signal independent technique (MSBTF) for the purpose of bandwidth limitation.

2. With identical overall bandwidth, the low-frequency content of the low-ranked hierarchically encoded signals is preferred to be retained than high-frequency content of high-ranked signals.

Figure 1: Basic Audio Quality for different processes

This work is reported in detail in [P3].