PCInt project history
The Post Correlator Integrator project started in October 1998 with a requirement analysis. It was followed by reviewing possible implementations of the unit based on Off The Shelf components. In December 1998 this resulted in completing the design specifications for the Post Correlator Integrator. The proposed solution had split the unit in two parts, a recording board and a processing board. Both parts could be developed independently.
In January 1999 the components for the recording boards were ordered. At that time it was decided to postpone the purchase of the processing board. The selected board was not yet deliverable because the required Digital Signal Processor were not yet available from Texas Instruments.
In February 1999 work was started on the Correlator board DSP software. This
was developed at Haystack observatory and provided to ASTRON as part of the
correlator board. It was expected that only functionality had to be added
to this software to allow use of the high speed output port of the correlator
board.
Despite the fact that Haystack claimed that the software was fully operational,
it turned out that it was not directly in the ASTRON application of the correlator
board. Changing into useful software turned out to be a real problem, it was
extremely difficult to change part of the code without effecting operation of
other parts.
- changed error and status report handling to CUCC
- modified the source code organisation to allow modifications
- removed several bugs from the code
This entire operation took over 3 months to complete. After that tasks were
added to the code to read data from the correlator chips, process the data
headers, perform data operations and send the data out of the high speed output
port.
At this point the entire task interaction is implemented only the actual
processing algorithms are to be added. These algorithms have to be taken from
the CUCC code, which does the processing right now. This action is of low priority
and postponed for now.
In June 1999 work was started on the development of the recording board software. Because this recording board is constructed of 2 modules, a C44 PMC module and a PMC SCSI controller, and internally based on a PCI bus. A lot of effort went in the implementation of the PCI protocol (configuration and access commands). Also the control and handling of the SCSI controller and tape device required a lot of development. By the end of the September the implementation is only a few days away from reaching the first milestone of writing data to tape.
In the second week of October the first milestone of writing data to tape was reached. This was followed by improving the architecture of the code and improving the data path for writing data to the SCSI controller and tape. In November a setup was made for the actual recording board application. This application integrates receiving correlator board data and writing it subsequently to tape. It required rewriting the routine for writing to tape and it also showed a few errors in the processor module used on the recording board. These errors were reported to Pentek the board manufacturer. As work around was created for this problem and by the end of December the second milestone was reached. By that time it was possible to write (test) data coming from a correlator board to a tape.
It was planned to order the Pentek DSP board (processing board) at the end of the third quarter, however during the year two similar board where introduced by two other companies. In October a comparision was made between 3 DSP boards to select the most suitable board to be used as the processing board in the project. It turned out that the original selected processing board from Pentek was best replaced by a board from Spectrum Signal Processing. This board fits into the original design of the PCInt unit but also makes it possible to simplify the design by placing the SCSI controller directly on the processing board. This simplification would mean that the recording board became redundant. But since a major part of the development on the recording board software is directly transferable to the processing board, development on this board continued in order to be able to use the board for testing purposes.
The selected processing board was ordered at the begining of December and was expected to be delivered in February 2000.
After the board was delivered, work started to incorporate the recording board and processing board software onto one board. Since the DSPs used are all code-compatible, this was not a huge task. Issues that had to be dealt with were issues like synchronization between the four processing nodes, interfacing to a new PCI-bus controller, making the DSPs boot via the VME bus.
By the end of August 2000, the DSPs could be booted via the VME bus in the test setup. Using the dzbtp software of A.Bos, it is possible to read a bootimage from disk, load it onto the DSPs and run the image. In order to move on to the production system, time had to be spent on the correlator board.
The correlator board DSP software in the production system correlator had to be upgraded in order to effect the following: support the PCInt (i.e. output via the highspeed serial link) and support for using the whole correlator. Up till the moment the correlator board DSPs became active, the usage of the correlator was limited to two boards per crate (whereas eight boards per crate are available). This has to do with the limited capability of the realtime system to read/calculate/write the headers for all boards within an acceptable timeslot (i.e. subsecond).
It took until beginning of 2001 before the correlator board software was capable of performing normal production correlation. This enabled the correlator to start operating at its full resolution.
Now the time had come to move away from the test setup and start on creating the production PCInt. It was felt that this was really a new phase in the development of the PCInt, after demonstrating that data could be read from the correlator board and written to tape. Also, it was deemed useful to postpone completing the documentation until the production capable PCInt became operational.