For a single-rate filter, a 2.5 Gsps input sample rate also means a 2.5 Gsps output sample rate. Because the input stream is separated into two (even, odd) streams, the output stream must be split the same way.
Take a look at how y0 is computed:
If the data stream is split into two phases, it can be seen that the coefficients also have to be split into two phases.
Also take a look at how y2 is computed:
For the even output stream, the data and coefficient phases should match:
Even data phase sent through a filter built with the even phase coefficients
Odd data phase sent through a filter built with the odd phase coefficients
Take a look at how this is modified for the odd outputs:
In this case, the phases of the data and coefficients should be mixed:
Even data phase sent through a filter built with the odd phase coefficients
Odd data phase sent through a filter built with the even phase coefficients
There is a further difference between the two. In the odd output case, they (even data, odd coefficients) should discard one data at the beginning of the stream.
In the previous section, the balance between data transfer and compute performance of the AI Engine was obtained for a 1.25 Gsps data stream going through an eight tap filter. The balance is identical here. As we have eight different filters, we can process 4x 1.25 Gsps streams in parallel.
The data stream and the coefficients must be split into four phases and then recombined. In the following figures, the various colors correspond to a different phase for the data (blue) and the coefficients(red):
Output phase 0, will be split and recombined as follows:
Output phase 1, will be split and recombined as follows:
Output phase 2, will be split and recombined as follows:
Output phase 3, will be split and recombined as follows:
The Data and the Coefficients being split into N Phases (four in this case), the resulting architecture requires NPhases x NPhases (4x4 = 16) to be implemented.