Super Sample Rate - Coefficient & Data distribution - 2023.2 English

Vitis Libraries

Release Date
2023.2 English

The base mode of SSR is driven by the use of template parameter TP_SSR. The parameter TP_SSR allows a trade of performance for resource use in the form of tiles used.

When used, a number of TP_SSR input phases and a number of TP_SSR output paths will be created. An array of TP_SSR^2 FIR sub-graphs will be created to connect input phases and output paths.

Input data samples are distributed across the input phases in a round-robin, sample-by-sample mechanism where each input phase processes a fraction of the input samples, i.e.: TP_INPUT_WINDOW_VSIZE / TP_SSR. More details in: Super Sample Rate - Sample to Port Mapping.

Coefficients are distributed in such way that each output path consists of all the FIRs coefficients, but each FIR sub-graph in any given output path is only configured to operate on a fraction of the FIR length, i.e. operates on TP_FIR_LEN / TP_SSR number of coefficients.

As a result, each FIR sub-graph operates on a fraction of coefficients and a fraction of the data, giving an overall increased performance.

In addition, each FIR sub-graph can be further split into multiple FIR kernels with the use of cascade interface, which is driven by template parameter: TP_CASC_LEN.

For example, a FIR with TP_SSR=4 and TP_CASC_LEN=2 will create a kernel structure presented below, in Internal structure of FIR with TP_SSR=4 and TP_CASC_LEN=2.


Figure 28 Internal structure of FIR with TP_SSR=4 and TP_CASC_LEN=2