There are two situations where changing the configuration can temporarily reduce the throughput of the core:
1. A Pipelined Streaming FFT is processing frames and the transform size (NFFT) is changed.
2. A Burst I/O core with bit reversed outputs is processing a frame, and the master supplies frame data in time to avoid the core automatically flushing the frame, and the transform size (NFFT) is changed.
Both the Pipelined Streaming architecture and the Burst I/O architectures (when bit reversed outputs are used) implement pipelining to achieve better throughput. In the case of the Pipelined Streaming architecture, it pipelines the loading, processing and unloading of entire frames (see This Figure ). In Burst I/O architectures when bit reversed outputs are used, the core implements a partial pipeline to overlap the loading on one frame with the unloading of another (see This Figure ).
However, a change to the transform size can only be applied when the pipeline is empty. Changing the transform size when the pipeline is not empty would result in data loss, so the core prevents this. When new configuration information is sent to the Configuration channel, and that information contains a change in transform size, the core does not load more frames until all frames already in the pipeline are processed and unloaded.
This is all handled automatically by the core, allowing you to send the configuration information at any time. However, throughput drops until the pipeline is fully flushed. This behavior only occurs if the transform size is to change. All other configuration options can be applied without waiting for the core pipeline to empty.