Call-Level Design
Within the framework core each DSP component such as Acoustic Echo Cancellation, Tone Generator, Tone Detector, etc., should be made into a modular block. These blocks can then be joined together to form a data processing pipeline (see Figure 2). An inbound endpoint connection would consist of the inbound pipeline and an outbound pipeline.
To complete a call connection within the VPF, two endpoints need to be connected together. Figure 3 shows the setup for an endpoint IP call.
This provides a very powerful and flexible method of configuration call connections. N-way calls can be set up by connecting the endpoints together, and forking can be configured just by disabling the inbound pipeline of an endpoint.
A number of blocks can be strung together within a pipeline to perform the necessary soft DSP for the incoming data. Below are example configurations of the pipeline for IP and PCM inbound and outbound. Each block can be turned on or off as needed. The following pipeline configurations provide the flexibility to support many scenarios.
The Sample Rate Converter component is one example of how a flexible configuration can reduce CPU usage of the pipeline. Narrowband data requires less processing than wideband data within each component; therefore, to minimize CPU usage, as many blocks as possible should be run in narrowband mode. The following examples show how placement of the SRC block can be used to reduce the CPU usage of the system.
