That's the experience of Jim Wilson, an electronics engineer at the Environmental Technology Laboratory at the National Oceanic and Atmospheric Administration (NOAA) in Boulder, CO. When setting up a PC-based system for measuring fishery stocks in the oceans, he insisted on the reliability and realtime capabilities of the QNX operating system. For some plug-in I/O cards he found it necessary to write his own QNX drivers, but the fact that UEI supplies professionally developed QNX drivers for its PowerDAQ line of I/O hardware made those cards an easy choice.

Fish spotting with lasers

First, though, take a quick overview of Jim's project. Federal regulation of the offshore fishing industry has once again become national news, but unfortunately legislators often must make far-reaching decisions without the benefit of detailed information. When scientists try to determine how many fish of a given type are in a specific region, they typically rely on direct sampling, egg sampling, sonar (echo-sounding) surveys or aerial surveys. The surface-based techniques take considerable time and are expensive to conduct. And while aerial surveys can cover a larger area, the resulting data is generally less reliable.

That was the situation until Jim's group at NOAA started to take a concept developed for military applications and apply it to fishery surveys. The core of the system is LIDAR (Light Detection and Ranging), which operates in a fashion similar to RADAR except it uses an industrially safe laser instead of radio waves. One reason the laser is so useful is that approximately 96% of its incident optical power survives a round-trip propagation through an air/water interface and light is fast enough to make the round trip from aircraft to fish and back again before the aircraft position has changed significantly. Given this performance it’s practical to operate LIDAR from a light aircraft moving as fast as 100 m/sec. Seawater, however, rapidly attenuates light signals, so the technique is limited to measurements of fish in the upper ocean layers only as deep as 50m. Thus the system works best on upper-ocean fish types such as anchovy, sardine, mackerel and herring. Nonetheless, a typical survey flight can cover hundreds of kilometers in just a few hours. The cost, measured per survey kilometer is less than 10% of that for a ship survey, and the depth penetration is more than three times that of a visual survey.

Because the total equipment weighs only 150 kg, it's suitable for installation into a small aircraft as long as the plane can supply at least 1.5 kW to power the experiment. The complete system consists of three major sections. The first is the laser along with its associated beam-control optics. The LIDAR employs a green laser (doubled Nd:YAG, 532-nm) to generate 12-nsec pulses and emits them at a rate of 30 pulses/sec. The linearly polarized beam gets diverged with a lens to produce a swath that varies in size with daytime or nighttime operation.

Fig 1 – A small rack holds all the equipment in the airborne LIDAR, a QNX-based system that digitizes inputs from a telescope to help researchers locate various fish species during an efficient air survey. When the operator sees something of specific interest on the display, he can push a button to insert a marker into the datafile to make it easier to find that point during post-acquisition analysis.

Second come the receiver optics and detector. To collect the optical signal reflected from the fish, the system relies on a 17-cm refracting telescope. Before entering the telescope, the light first passes through a polarizing filter because experience shows that the cross-polarized component of the returned light produces the best contrast between fish and the scattering from small particles in the water. Further, an interference filter rejects background light. Finally, a photomultiplier tube generates a low-level voltage based on the light it receives from the telescope.

The third stage consists of signal-conditioning circuitry, digitization, display and storage into a datafile. The detector output goes into a logarithmic amplifier that compresses the signal’s dynamic range to match the capabilities of the digitizer card in the system. That card is a single-channel, 8-bit, gigasample/sec card from WaveEdge (formerly Sonix). By sampling at 1 GHz, the card allows the mapping of fish to a depth resolution close to 10 cm.

These data alone can't be adequately evaluated unless the researchers also have detailed data about the state of the experimental equipment and the environment at the time the data collection took place. For this housekeeping activity, Jim selected a PD2-MF-16-150/16H from United Electronic Industries. He uses eight differential 16-bit inputs on this 150-kHz multifunction board. One channel records the output of an infrared radiometer that measures the temperature of the ocean surface. Two more channels read the day and night gain-control voltage settings on the receiver. A fourth channel makes note of the polarization setting on the receiver. Two further channels record the pitch and roll angles of the instrumentation in the aircraft. The next input is tied to a shutter button in the cockpit, which the copilot pushes to close the laser-output shutter temporarily when passing over a ship or for any other reason.

The final input accepts a control signal from the equipment operator that inserts a marker into the file. This feature is convenient because the housekeeping card digitizes all of its inputs as fast as possible running the 150-kHz A/D at its peak rate, and it makes 30 such scans every second. The system then combines this status information with the raw data from the LIDAR and saves it in one file. A typical file that consists of 2000 "shots" of the ocean takes up 2.4 M byte. This large amount of data can make it difficult to sift through to find events of particular interest. Thus, as the instrument’s operator monitors system activity in real time on a PC screen, he watches for interesting events. When one arises, he can push a button to insert a special marker in the datafile to allow the person performing post-experiment analysis to go to that particular spot in the datafile.

The hunt for drivers

Clearly, requirements on the housekeeping A/D card aren't terribly severe and any number of cards could handle the job. The problem was, though, finding cards with the necessary driver support. "I started developing this system roughly eight years ago on a 386-based machine," Jim recalls. "Given all the tasks the system must perform without any hesitation, there was absolutely no processing power to spare. And because Windows could sometimes eat up CPU cycles just a brief moment before it could move on to a critical task, that situation was unacceptable for the application. That's why we had to go to a realtime operating system, and I chose QNX 4.x"

Now he had to get the peripherals running under QNX, and fortunately in this setup the only nonstandard peripherals consisted of the two data-acq cards. For the WaveEdge board he had to adapt an existing DOS driver written in C. Although that project took him only an afternoon, it's not the same as having a tested, verified driver from the hardware supplier on hand. Jim had looked at other vendors of gigahertz cards and notes that some of them support QNX unofficially. "They say, 'Well, here's a driver that worked for some other guys, and maybe it'll work for you – give it a try!' I'm looking for a higher level of support."

Jim also relates that his original configuration for the housekeeping portion of the system used a card from Data Translation. That firm likewise had no QNX drivers, and it turned out that the card was a bit expensive and overkill for the intended tasks. When he started searching for alternatives, he looked for a data-acq vendor with QNX support and found UEI. Further, he's completely satisfied with a cost-effective card in the PowerDAQ family, which handles all the tasks he requires at a price that fits in his budget.

System development with the PowerDAQ card proceeded smoothly, although Jim feels that he would have liked more-detailed documentation about the card and the software. However, he adds, a few quick calls to tech support solved the few minor questions that arose. The only snag came when he found a bug in the QNX driver, one that he discovered as being among the first users of the driver. It was a problem that tech support solved very easily: in one of the .H setup files a line had an incorrect variable value which inadvertently got transposed from a 0 to a 1. Once that was corrected (and was immediately fixed on the general release of the QNX driver), the card worked fine.

 As he enhances the system and wants to add other peripherals beyond the two DAQ cards, he is considering a switch to another OS, possibly even Windows. "We wanted multiple serial ports, but you don't just walk down to the local computer store and pick up a multiport serial board with QNX drivers off the shelf." His search did ultimately lead him to a RocketPort card from Comtrol, a firm that does support QNX. One of the major problems Jim faces now is that hardware is continually being developed that he would like to add to the LIDAR system, such as a CD burner. The problem stems from the fact that QNX has moved on to QNX 6.0 and the developer isn't supporting version 4.x any more. "We have a big decision to make," Jim reflects, "we need to either port everything over to QNX 6.0 or find a new OS".

And in the years since he first designed the system architecture, PCs have become tremendously more powerful in terms of computational ability, video display, bus speeds and transfer rates. "Because of this power, we don't have to pay as much attention to the realtime aspects as we did with hardware from just a few years ago," notes Jim. Even so, he plans to stick with UEI for data acquisition because of the experience he got from the firm's support department and the bang for the buck. He only wishes they had a gigahertz sampling card.

Algorithms being refined

Jim's group is also working on the software side. LIDAR has been in use for more than 30 years in oceanic environments, but it's only been in the last decade that advances in signal-processing algorithms and calibration experiments made it worthwhile to perform a detailed analysis of incoming data. However, much work is still required in algorithms to convert LIDAR data into quantitative information about stock abundance, species and size.

Fig 2 – A typical screen shot from an aerial survey shows only that there is a fair amount of plankton in this region. The X number is the Shot Number, and the Y axis represents depth. Water surface is 0m, so the range here is 5m above the surface to 50m below the surface. White means no signal returned, while cool colors represents a low signal, increasing to reds for higher signals. In this case, the violet layer between 5m and 10m shows a light layer of plankton, the darker blue sections show a denser plankton concentration. The blue below approximately 30m is noise due to a large amount of glacial silt in the Alaskan waters where this shot was taken.

For instance, today the system can't distinguish between a single large fish or a school of small fish. However, with the support of the ship sampling crews, researchers can make a reasonably good estimate of the type and density of the fish by combining both LIDAR data and surface-ship observations. Researchers must consider other clues available such as the habitat being surveyed, in particular the water column and latitude, time of year and the types of fish sampled by the ships. The researchers are also discovering algorithms and techniques that allow them to use characteristics of the return signal such as its strength, polarization and frequency spectrum to provide even more information about the size, density and possibly even the species of the fish being observed.

For more information on Fish LIDAR, contact the NOAA/ETL web site at http://www.etl.noaa.gov/technology/instruments/floe/