Principle of Data Format Conversion The data is

All Read pointers are synchronous. Delay is given by Write pointers offset

Basic sequence Every 4 nsec : One block of 16 samples is written at

A few assertions It takes the same time (16 x 4 nsec) for the

The geometrical delay It is the time spent by one sample in the RAM

Blanking When one Read Pointer reaches the end of one sector, it jumps back

Overwriting When the Write Pointer has completed a full turn, it starts overwriting the

Next comes an animation that may facilitate the understanding. For this the RAM space

Antenna X Sector F Delay X Write pointer Sector 0 Sector 1 Antenna Y

Antenna X Sector F Write pointer is 16 x faster Sector 0 Sector 1

Antenna X Sector F Sector 0 Sector 1 Antenna Y

Antenna X Sector F Antenna Y Sector 0 Delay X Sector 1 Delay Y

Slides: 104

Download presentation

Principle of Data Format Conversion The data is recorded at natural speed (4 Gsp/sec) in a RAM and played back to 16 different correlators that operate at 1/16 th the speed (250 Msp/sec). The 16 correlations are performed on consecutive signal segments and are added together later. (Thanks to the signal for being stationary) The RAM is circularly organized so the segments are better called sectors.

All Read pointers are synchronous. Delay is given by Write pointers offset

Basic sequence Every 4 nsec : One block of 16 samples is written at the WP address. The WP address is incremented by 1 One block of 16 samples is read at one of the 16 RP addresses. The sector number is increased by 1. The downloaded block is pushed into the corresponding correlator via a 16 -sample parallel-to-serial converter.

A few assertions It takes the same time (16 x 4 nsec) for the correlators to load and process 16 samples from the P/S converter as for the Read Pointer to poll the 16 sectors. It takes the same time for the Write Pointer to cover the full RAM space as for one Read Pointer to read one sector.

The geometrical delay It is the time spent by one sample in the RAM between its writing and its processing. For convenience all the Sector 0 players (one per antenna) synchronously start at RAM address 0 at the beginning of each 32 pps event. The write pointer of each antenna is offset by the corresponding delay value. One RAM location represents 16 samples, or 4 nsec of signal. 12 microseconds (4 kilometres) use 3000 RAM locations.

Blanking When one Read Pointer reaches the end of one sector, it jumps back to the beginning. The correlator shift register is still full with old samples that would not be correlated with the new ones. Integration must be blanked while the correlator pipe gets purged. The related loss of time is (correlator length/sector length) e. g. 1 k/128 k

Overwriting When the Write Pointer has completed a full turn, it starts overwriting the old data. It is important that the data on one sector should not be overwritten before the full sector has ended being read. This does not occur as long as the geometrical delay does not exceed 1/16 th the reading time of one sector. Next is a timing diagram where all this can be seen

Next comes an animation that may facilitate the understanding. For this the RAM space has been deployed linearly although its nature is rather circular. One cursor hit represents 16 cycles of 4 nanoseconds.

Antenna X Sector F Delay X Write pointer Sector 0 Sector 1 Antenna Y Delay Y Write pointer

Antenna X Sector F Write pointer is 16 x faster Sector 0 Sector 1 Antenna Y

Antenna X Sector F Sector 0 Sector 1 Antenna Y

Antenna X Sector F Antenna Y Sector 0 Delay X Sector 1 Delay Y

Antenna X Sector F Sector 0 Sector 1 Antenna Y