I.1 The basic subsampling scheme

The simplest balanced scheme one can think of is based on straight subsampling The coding algorithm is based on a temporal subband decomposition. Consider for example the case of a PCM (Pulse Coded Modulation) encoded 8kHz audio signal decomposed into 3 PCM 2.7 kHz flows as shown in Figure 1.

   figure75
Figure 1: Hierarchical coding scheme

The temporal decomposition algorithm is carried out at the source done for each audio chunk (which typically includes 20ms, 40ms or 80ms of audio). At a destination, a receiver which receives all 3 flows can retrieve the original input signal. If one or two flows are missing, the destination uses the samples received to reconstruct an approximation of the original signal. An upsampling one-to-three flows is shown in Figure 2. Clearly, the larger is the number of received flows, the better the quality of the reconstructed signal.

   figure83
Figure 2: Reconstruction 1 to 3

Note that temporal decomposition handles signals sampled with different sampling rates. For example, a 48 kHz audio signal can be decomposed into three 16 kHz subflows, each of which can be decomposed into two 8 kHz subflows, which finally yields eighteen 2.7 kHz audio layers. Of course, temporal subband decomposition can be coupled with compression schemes.



Thierry Turletti
Mon Aug 11 11:02:05 MET DST 1997