The Scanner subgraph performs a coarse-grained search of the MUSIC spectrum computed by the DOA subgraph, looking for regions of the spectrum that fall below a “null threshold”. The Scanner breaks apart the \(256\) spectrum bins into \(32\) contiguous groups of \(8\) bins each, and produces \(32\) output tags that are set to true if there exists a bin value in that group which falls below the given null threshold. To meet the 1 $\mu s$ throughput target, two AI Engine tiles are required, where each tile processes \(128\) of the \(256\) available bins. This algorithmic approach is outlined in the following diagram.
The following diagram shows the AI Engine physical array view for the Scanner subgraph. As noted above, the design requires \(2\) tiles and some additional storage for I/O buffers.