The stream output operates on instantaneous samples from the continuous input stream without data memory movements or intermediate storage. This approach prevents selective sample discarding while maintaining uninterrupted stream flow and continuous throughput performance. The figure below shows the transpose-form implementation of the convolution and correlation operations.
- The stream output length is independent of the compute mode (VALID).
- When comparing the output with reference data, ensure that the first M samples are discarded from the stream output, where M is an integer that can be computed based on configuration parameters.
This discarding of M samples is necessary because the transpose-form implementation of convolution and correlation produces initial transient samples that do not represent valid results until the computation reaches steady state.