In a previous post, I asked whether electronically steered antennas (ESAs) would replace parabolic antennas in satellite ground stations. I read a few articles suggested by others and by Google search, used some common sense, produced a list of advantages of ESAs, and concluded that it was likely they would eventually replace parabolic antennas for many applications.
Many of the articles I found were written by companies selling products or services and I'm not an antenna expert -- more a curious journalist. ChatGPT has access to the entire Internet, and I wondered if it could have helped me improve what I wrote or convinced me to reach a different conclusion, so I queried it three times
Since I had listed the advantages of ESAs over parabolic antennas, I began by asking ChatGPT to list the advantages of electronically steered antennas over parabolic antennas for satellite ground stations.
In my post, I listed twelve bullet-point advantages. The ChatGPT answer was more verbose, beginning with a restatement of the question and listing and elaborating on seven advantages. The elaborated replies included all but one of my bullet points, spectral efficiency, and it included an advantage that I had overlooked, interference mitigation, and explained why that was so.
In spite of these glitches, ChatGPT reached nearly the same conclusion as I had, saying "It's worth noting that while electronically steered antennas offer numerous advantages, they also have some limitations, such as higher cost and complexity compared to parabolic antennas. However, ongoing advancements in technology are continuously addressing these challenges, making electronically steered antennas increasingly attractive for satellite ground station applications."
Since it had failed to mention spectral efficiency, I asked, "What about spectral efficiency?" and it replied with an apology for missing the point and went on to list and elaborate on five ways in which ESA made more efficient use of bandwidth.
Since ChatGPT generally agreed with me, I asked it to play devil's advocate and list the advantages of parabolic antennas over electronically steered antennas for satellite ground stations.
Perhaps I was biased by the experience of having written my earlier post, but I found myself questioning some of its cited advantages. For example, one was "wide coverage" but that would not apply to Thinkom's array of arrays. Another was "simplicity ... making them easier to manufacture, install and maintain." ChatGPT was contradicting its previous response in which it said ESAs were more reliable and required because parabolic antennas "have mechanical parts that can be subject to wear and require regular maintenance."
I also realized that my question was a bit off -- it focused on the antenna rather than the entire ground station -- so I asked it to "list the advantages of satellite ground stations that use parabolic antennas over satellite ground stations that use electronically steered antennas". One of the advantages it listed was suitability for high-frequency Ka and Ku bands, but both Thinkom and BlueHalo use those bands and Thinkom is targeting even higher frequencies. It also repeated the simplicity advantage.
Finally, I asked the title question from my previous post -- will electronically steered antennas replace parabolic antennas in satellite ground stations?
It listed points under three headings -- advantages of ESAs, limitations of ESAs, and application-specific considerations and concluded that:
"While there is potential for Electronically Steered Antennas to replace parabolic antennas in certain satellite ground station applications, it is not a straightforward or immediate transition. The adoption of ESAs will depend on factors such as technology advancements, cost reductions, performance improvements, and specific application requirements."
It reversed itself once more by listing improved reliability as an advantage of ESAs, but the conclusion was similar to the conclusion in my original post. It also revealed a shortcoming in my formulation of the question. Since I have been writing a lot about Internet service constellations, I overlooked other applications in my previous post and did not specify that I was interested in broadband Internet service. (A human editor would have known that I was writing for an Internet-related publication, would have been aware of my previous writing, and assumed I was focused on Internet applications).
It pointed out that parabolic antennas had an advantage for deep space communication, so I asked a more fully qualified question -- Will electronically steered antennas replace parabolic antennas in ground stations for LEO, MEO, and GEO Internet service constellations?
The reply seemed vaguer and non-committal this time, but was also similar to mine:
"Considering these factors, it is likely that a combination of ESAs and parabolic antennas will be employed in future ground stations for LEO, MEO, and GEO internet service constellations. The specific configuration and utilization of each technology will depend on various factors, including cost, performance requirements, deployment scenarios, and network architectures."
Conclusion
So, what is the role of ChatGPT in this sort of journalism? It served me as an editor or referee reviewing what I had written. I would have made two changes after getting feedback from ChatGPT -- I would have mentioned that I was focused on LEO, MEO, and GEO Internet service applications, and I would have included interference mitigation as one of the advantages of ESAs.
ChatGPT made a couple of misstatements and did not "know" about the possibility of an array of antennas like that of Thinkom with its wide elevation angle range. To its credit, I liked the way it apologized for overlooking spectral efficiency as an advantage for ESAs and it was an indefatigable and patient interviewee. It was also free (for the time being).
I consulted ChatGPT after drafting my article, but I could have used it as a research tool before writing. Had I done so, I might have been misled by some of its mistakes and I wouldn't have discovered Thinkom's innovation. Furthermore, its prose was not clear and concise -- I don't think it could pass a Turing Test on writing style. In this case, an old-fashioned Internet search engine was a far better pre-writing tool.
ChatGPT would be improved if it gave links to the sources of its assertions. Like autocomplete, it generates sentences by repeatedly appending the most probable next word in text documents found on the Internet, so the final string is novel. (That's not the way I generate sentences -- they follow from an idea). Might a list of the documents providing the most words be useful? (Google Bard adds citations, but I've not tried it out yet).
While ChatGPT helped me, I don't believe a ground station expert would have learned anything new by interacting with it and a beginner like a student writing a term paper would have been misled. The curious journalist was the sweet spot in this case, but this is version 3.5 of ChatGPT, which had a data-cutoff date before BlueHalo and Thinkom announced the products I've mentioned. I'll revisit it when I get access to ChatGPT4.
PS -- Let me know if you read the dialogs I linked to and notice something I missed,
