by Kevin
"But It's OPTIMAL!"
During the summer of 2009, I was helping develop and maintain a software-defined radio called the "ATR-507" for GD. We had built and fielded the radio to provide Blue-Force Tracking services worldwide. Now, our customers had approached us about hosting an additional waveform.
The PRC-112 was a handheld AM radio used for voice communication since 1980's. It became a "Hook-112" with the addition of an external module that provided digital messaging by modulating the AM carrier tone with a digital (MSK) signal.
A first attempt at the receiver function had been made by a GD mathematician and an FPGA programmer. I couldn't even get their design to compile. With their firmware loaded, the receiver's maximum range was about 3 miles. It should have been line-of-sight to the horizon. Nor could it receive from closer than a half-mile.
The mathematician told me that I must be mistaken. (I would hear that a lot on this project.) He had a paper. The receiver design was Optimal! [1]
Back to Basics
Time to crack the textbook, in this case Korn's mighty doorstop, Digital Communications [2], wherein I read that while MSK is technically continuous-phase FSK, it can also be seen as differentially-encoded OQPSK. By demodulating it as such you get about 3 dB of processing gain over FSK. [3]
So, you know the drill: analysis, pages of diagrams, Matlab simulation, FPGA code, testbenches, troubleshooting, yaddah-yaddah.
Nobody gave me any performance requirements, so I chose my own: performance should be as good as it was for the existing COBRA waveform receiver. And it was. BER results were within a half-dB of the theoretical maximum for OQPSK.
The radio would emit a second or two of the unmodulated lower MSK tone at the start of each message. By demodulating it as an MSK signal, I retrieved a repeating pattern of complex data points that I could use as a message header. With over a second's worth of samples to correlate against, I could detect 100% of incoming messages with SNR's far into the negative 10's of dB.
(Drat Claude Shannon, anyway. I couldn't retrieve useful message data at those low SNR's, but still...)
"That's Impossible!"
Our team at GD wouldn't ship radios unless they passed a drive test, so I loaded my firmware in the receiver and connected it to an antenna on the roof. Then I wired up an AC converter, a DC power supply, the Hook-112 and a wire whip on the roof of my SUV and took off driving around Scottsdale.
Only one problem. Every time I pushed "transmit," the radio would die.
I drove myself ragged checking the voltages, the supply current limit, everything I could think of, until I had a strange thought: My setup worked perfectly in an air-conditioned lab, but now I was driving around Arizona in the summer. Could it be the heat?
"That's impossible," I was told, and told again, until finally I dragged the radio to the other building and waved it in the engineers' faces.
"Oh, that!"
A Successful Delivery
With the power problem fixed, I drove to a distant mountain ridge with line-of-site visibility to the GD facility and started transmitting while my buddy Gary watched the receiver.
I hit "transmit" 10 times in a row. All 10 messages were received. The new firmware was ready to ship.
Then it was on to the next thing!
Notes
[1] I wish I still had a reference for that paper. It amounted to FSK demodulation with a hard differential decoder -- not optimal in any sense.
[2] I do have a reference for the doorstop: Korn, Israel, Digital Communications, pub. Van Nostrand Reinhold, 1985.
[3] Because where MSK symbols are orthogonal, the BPSK symbols on the I and Q channels are antipodal, resulting in further separation between constellation points.
