DDR Buffers and PL URAM Buffers - 2025.1 English - XD100

A set of 586 radar pulses of length 2048 is stored in the pulse_i buffer in DDR. These are the zero-padded radar signatures collected from the GOTCHA data set [1] over azimuth angles \(38,39,40,41,\), and \(42\) for “Pass 1” on the “HH” polarization. This data set is input once to the engine for every radar pulse processed.

A set of 586 tuples $(A_x,A_y,A_z)$ representing the 3D coordinates of the antenna platform for each radar pulse captured is stored in the coord_i buffer in DDR. This buffer acts as a real-time parameter (RTP) vector for the BP engine that is transferred once to the AI Engine array prior to the processing of any radar pulses.

A set of 586 samples representing the range $R_0$ from antenna platform to SAR target image center is stored in the R0_range_i buffer in DDR. This buffer acts as a real-time parameter (RTP) vector for the BP engine that is transferred once to the AI Engine array prior to the processing of any radar pulses.

A single buffer image_buffer located in PL URAM stores the full SAR target image. The AIE array streams the input image from PL, updates the image for the current radar pulse in the AIE array, then streams the updated resultant image back to the PL URAM where it is stored in the same buffer.

The AI Engine array is enabled to run for NPULSE_USE radar pulses. The BP engine graph is designed to process a full radar pulse per graph iteration.

Asynchronous DMA transfers are used to send the pulse_i data from DDR to AI Engine array. A single DMA transfer is initiated for all radar pulses. Each radar pulse is processed by a single graph iteration.

After all radar pulses have been processed, the final target image may be uploaded from PL to DDR using a synchronous DMA transfer.