Weather Balloon Data System Upgrade
Can the developed world’s digital make sense of stormy developing world analog?
Weather is a global business. The Indian government approached a small U.S. weather systems integrator to update a piece of their national weather infrastructure. I was contracted to create a system capable of digitizing Indian weather balloons’ analog data. Prior to this, India’s radio probe used a ground-based paper strip recorder and required human post-processing of the data. While the Indian government elected to continue using the existing radio probe, they desired to replace their antenna system and strip recorder with an auto-tracking antenna and automated digital system.
The challenges were significant. Transmission times varied among units and with atmospheric temperature. No unique demarcations existed between transmitted data tones. Pressure tones were derived from a mechanical bellows that swept a needle across a pattern of resistors — requiring matching a progression of tones against a reference wherein some patterns repeated.
I developed the system from a PC/104 stack with desktop processor and timer counter card. The analog signal from the radio probe was conditioned at the antenna unit to become a pulsetrain presented to my hardware. Using the timer counter card and a statistically-based signal processing method, I constructed a system able to identify each of the four tones the radio probe transmitted as a reference, pressure, temperature, and humidity value.
I designed the software as a stack of filters to compensate for many of the challenges already identified. To compensate for signal drop outs I devised a number of mechanisms based in time and altitude calculations to estimate pressure for sake of re-acquiring the pressure tone pattern.
For ground station development and testing, I created a desktop application that allowed me to simulate the radio probe’s transmissions. My application was able to generate tone sequences and noise variations through a desktop PC’s sound card wired to the signal processing input.
The final system produced excellent results. Save for a single deployment issue, the system has required no significant work after entering the field.