Hunting Hurricanes

The authors tell us about hunting hurricane using the Scanning Radar Altimeter based on the Linux system and analyzing the data with Yorick.
2ns PCI Waveform Digitizer, DOS, GageScope

The core data acquisition device in the SRA is the GageScope 8500-PCI waveform digitizer. It provides for up to 128KB of sequential samples taken every 2ns (nanoseconds). This permits us to digitize a 256-microsecond waveform. We actually digitize for 60 microseconds, beginning a few hundred nanoseconds before the radar pulse is transmitted and ending after enough time has expired to accommodate a signal return from our highest possible altitude. The pulse takes 2 microseconds to travel 1000 feet and return, so to accommodate a maximum altitude of 30,000 feet, we need to digitize at least 60 microseconds. Since a point is digitized every 2ns, there will be 30,000 points in each waveform. We don't read all 30,000 points out of the digitizer. We “track” the position of the returns and read out only 256 points centered around where we expect the return to come from. Since the ocean is basically flat, this technique works well.

The driver code provided by Gage for the 8500 supports DOS, Windows and Windows NT. It is extensive, to say the least. It contains several thousand lines of code solely to initialize most of the cards that Gage makes to an operational state. Apparently, much of the functionality of the card is loaded into programmable Logic Arrays from the DOS driver. The Gage driver code supports virtually every waveform digitizer made on several different OS platforms. They make extensive use of conditional compilation to select both the desired digitizer board and the desired operating system. They attempt to establish an isolating layer of driver code, so that a common set of driver calls appears to the users of their supplied library.

After looking at the driver start-up code, we though it might take more time to port the start-up code to Linux than we could afford. In order to avoid porting the long and complex start-up code, we elected to make the system dual boot Linux and DOS. This scenario has worked well, permitting us to get a DOS program going quickly which would configure the digitizer. After the digitizer is configured, the autoexec.bat DOS script loads Linux using loadlin, a DOS program which can load a Linux kernel from DOS. The DOS digitizer start-up code leaves the digitizer in a known functional state. The code required to use the digitizer is actually not very extensive and only requires accessing a few registers and memory locations on the Gage card. The folks at Gage were very helpful in getting it working.

Hardware Interrupts

Waveform data are extracted from the digitizer after a radar pulse event has occurred. One of the 16C65A microcontrollers controls all aspects of triggering the transmitter, actuating various gates, triggering the waveform digitizer and finally interrupting the Linux waveform digitizer interrupt handler.

Figure 9. RT-Linux Interrupt Jitter

The RT-Linux interrupt typically responds in 2 microseconds (on our 200MHz Pentium) with occasional jitter to several microseconds. When we did this same test with MS Windows a couple of years ago, we found the fastest response to be on the order of 50 microseconds (486-dx2 66MHz) with jitter well into tens of milliseconds. It is incredible just how responsive Linux is to interrupts. Figure 9 is a digital scope capture, where the top trace rising edge is the actual hardware interrupt signal on the ISA backplane. The bottom trace is a hardware signal generated by the interrupt code. It simply wrote a “1”, waited awhile, then wrote a “0” to the printer port. Each horizontal division is 2 microseconds. This demonstrates the typical latency of our RT-Linux system. Concurrent with this test, we ran a find command in another xterm so the system had something to do.

Microcontroller Board, 16c65a

Microcontrollers permit very hard and reliable real-time capabilities. They are well-suited to replacing arrays of chips and digital logic in many applications such as the SRA. We designed a special ISA interface board for the SRA which encompasses most of the special requirements of the radar and special interfaces.

The board is presently populated with four Microchip 16C65A microcontrollers. One microcontroller implements a real-time clock which automatically maintains time-of-day synchronization with our GPS receivers. It has a least significant fractional time bit of 200 nanoseconds and provides the SRA with up to 64 bits of accurate time information. This chip automatically captures the trigger time for each radar pulse. It and its neighbors can all be read and written by Linux.

A pair of microchips function together to convert the scan encoder pulse trains into radar trigger events. As our scan mirror rotates, a scan encoder measures the scan angle. At our scan rates, it produces a pair of 40KHz square-wave signals which are 90 degrees out of phase. One microchip is programmed to combine these two signals together and produce a single 80KHz signal, which is then counted to determine the position of the scanner. The second microchip is programmed to count the 80KHz signal and initiate a radar pulse at predefined angles. With its 200ns instruction time, this microchip directly controls all aspects of the transmitter and receiver electronics and also generates an interrupt to Linux once a waveform has been acquired by the waveform digitizer. For each SRA pulse, this microchip:

  • Protects the receiver front-end from damage.

  • Verifies that the receiver is protected.

  • Gates the digitizer on.

  • Gates the EIK amplifier on.

  • Delays exactly 200ns.

  • Triggers the transmitter modulator to generate an 8ns pulse and causes the real-time clock microchip to capture the present time.

  • Delays 200ns for the transmit pulse to be well clear.

  • Enables the receiver to receive return signals.

  • Interrupts the RT-Linux SRA module to extract the waveform data from the digitizer.