Digital Signal Processing Reference
In-Depth Information
Fig. 2
An illustration of a speech compression system that is modeled using PSDF semantics.
packets, where the sequence size determines the extent of the associated graph
iteration.
Body graph iterations can even be defined to correspond to individual actor
invocations. This can be achieved by defining an individual actor as the body
graph of a parameterized dataflow specification, or by simply defining the notion
of iteration for an arbitrary body graph to correspond to the
next actor firing
in the
graph execution. Thus, when modeling applications with parameterized dataflow,
designers have significant flexibility to control the windows of execution that define
the boundaries at which graph parameters can be changed.
A combination of cooperating body, init, and subinit graphs is referred to as
a
PSDF specification
. PSDF specifications can be abstracted as PSDF actors in
higher level PSDF graphs, and in this way, PSDF specifications can be integrated
hierarchically.
Figure
2
illustrates a PSDF specification for a speech compression system. This
header packet from a stream of speech data, and configures
L
, which is a parameter
that represents the length of the next speech instance to process. The
s1
and
s2
actors are input interfaces that inject successive samples of the current speech
instance into the dataflow graph. The actor
s2
zero-pads each speech instance to
a length
R
(
R
L
) so that the resulting length is divisible by
N
, which is the
speech segment size. The
An
(“analyze”) actor performs linear prediction on speech
segments, and produces corresponding auto-regressive (AR) coefficients (in blocks
of
M
samples), and residual error signals (in blocks of
N
samples) on its output
edges. The actors
q1
and
q2
represent quantizers, and complete the modeling of the
transmitter component of the body graph.
≥