The paper describes a scalable extension of the
AVC coder. The assumption is to introduce possibly
minor modifications of the bitstream semantics and
syntax as well as to avoid as much as possible the
technologies that are not present in the existing structure
of the AVC codec. The coder combines spatial with
temporal scalability. The coder consists of two motioncompensated
sub-coders that encode a video sequence
and produce two bitstreams corresponding to two
different levels of spatial and temporal resolution. Each
of the sub-coders has its own prediction loop with
independent motion estimation. The system employs
adaptive interpolation. The interpolation-dependent
modes are carefully embedded into the mode hierarchy
of the AVC coder thus obtaining the codes that
correspond to the mode probabilities.