How to use AI in your Ham Shack

Artificial Intelligence is a powerful tool that can help you in your everyday life, even in you ham shack. In this video, I dive into how AI tools like ChatGPT or Claude.ai can revolutionize your amateur radio experience. From building antennas like a 40-meter dipole or an end-fed half-wave for 40 to 10 meters, to troubleshooting schematics of radios, AI proves to be an amazing companion in the shack.

 

Always double-check the results from any AI. Depending on the LLM used, the results may be great or lackluster.

 

Same could be said about advice from fellow hams LOL

 

you've posted a video.
Possibly you could have your AI summarize with a printed synopsis ?

 

Or flat out wrong. I catch Perplexity making math mistakes.

But it is still highly useful. I use the heck out of it. I especially like that it will quickly look up characteristics from datasheets. Saves me a ton of time. It also is very good at doing dB calcs...

 

I had lunch with a friend this week. He has a PhD in math from MIT. He quoted an equally credentialed friend of his: "AI is the best bullsh*t artist there is, because it reads the most bullsh*t."

Of course AI will improve. The question is whether it will improve in the direction of accuracy, or towards even better bullsh*t. The biases and desires of its creators and programmers inevitably creep in. Our economic model pushes those biases and desires towards getting masses of people to buy products, believe ideologies and channel their emotions into mass actions. So I'm not very optimistic. Technical information may be spared somewhat, but not completely. So use AI, sure, but don't trust. Verify.

 

I definitely agree you need to verify, but I would say that about any source. LLMs are trained on huge amounts of data, mostly from online, and we know how accurate information online can be. It's the old saying: garbage in, garbage out.
But using these LLMs can save a huge amount of time versus having to search for the answers yourself. Exciting to think we are just at the tip of the iceberg as far as what we can do with AI.

 

I let people know to assume the same with any answers I give.

And the first several tries of R&D which I was involved with that eventually led to valuable patents were worse than lackluster.

 

I just used Perplexity due to reading this thread. Interesting. I'll have to check this out further. I just hope SKYNET stays away.

 

I have tried asking it for aerial dimensions and that sort of stuff. It's fine, but there's nothing to get excited about.
I see radio as a very human thing, but I'm old, fat, ugly, and uncool, so it's very possible I'm a stuck in the mud old man.

Chat had this to say

AI can be a powerful tool for radio amateurs (also known as "ham radio operators") in a variety of ways, enhancing both their technical capabilities and the overall experience of the hobby. Here are some key ways AI can assist:

1. Signal Processing and Noise Reduction
AI can help radio amateurs improve the clarity of signals by applying advanced signal processing techniques. AI-driven software can analyze received signals in real time, identifying and reducing noise, interference, and distortion. This can be particularly useful in crowded frequency bands or during poor propagation conditions.
​

• Noise filtering: AI algorithms can help remove background noise, making weak signals more intelligible.​
• Automatic gain control (AGC): AI can dynamically adjust gain to optimize signal strength and quality.​

2. Digital Modes and Decoding
AI can be applied to decode digital transmission modes like FT8, PSK31, RTTY, and others. For example, AI-driven software can automatically detect, decode, and log transmissions from other operators, making it easier for beginners to engage with complex modes without needing deep technical knowledge.
​

• Automated decoding: AI algorithms can decode and log digital signals, even in challenging conditions.​
• Signal identification: AI can identify the mode being used and switch the software accordingly.​

3. Propagation Prediction
One of the most challenging aspects of ham radio is predicting radio wave propagation. AI can improve propagation prediction models by analyzing vast amounts of historical data and current environmental factors to predict optimal frequencies, time windows, and conditions for long-distance communication.
​

• AI-based propagation forecasting: AI models can predict solar activity, ionospheric conditions, and geomagnetic storms to suggest the best times and frequencies for communication.​
• Machine learning models: AI can continuously learn from past propagation data to improve accuracy over time.​

4. Frequency and Band Monitoring
AI can monitor frequency bands in real time, identifying open frequencies, analyzing band congestion, and providing recommendations for optimal operating frequencies. This can be particularly helpful during contests or crowded conditions.
​

• Frequency scanning: AI can scan large portions of the frequency spectrum to detect active transmissions and open spaces.​
• Band occupancy analysis: AI can identify underutilized portions of the spectrum and recommend frequencies with less interference.​

5. Logbook Management and Automation
Many radio amateurs keep detailed logs of their contacts (QSO), which can be time-consuming. AI can automate much of this process, making it easier to track contacts, manage log entries, and even generate reports for awards or certificates.
​

• Automatic logging: AI can automatically fill in details for each contact, including time, frequency, mode, and location, by integrating with digital interfaces or decoding software.​
• Awards and contest tracking: AI can help track progress toward specific awards (like DXCC or Worked All States) and help with contest logging and scoring.​

6. Antennas and Propagation Optimization
AI can help radio amateurs design, test, and optimize antenna systems. By analyzing variables like frequency, terrain, and local conditions, AI can assist in predicting the performance of different antenna configurations.
​

• AI-based antenna modeling: AI tools can optimize antenna designs for specific frequencies or operating conditions.​
• Virtual experimentation: Using machine learning, AI can simulate and predict how antenna changes will affect performance, saving time and effort compared to physical trial and error.​

7. Remote Operation and Automation
AI can make remote operation more accessible and efficient. With the help of AI, ham radio stations can be set up for automated operation, where AI can take over routine tasks like adjusting antenna positions, managing transmissions, and monitoring signal conditions.
​

• Automated station control: AI can control aspects of the radio station remotely, adjusting settings based on real-time conditions.​
• Remote assistance: AI can assist operators in troubleshooting issues with their equipment, offer maintenance suggestions, and optimize settings remotely.​

8. Training and Mentoring
For newcomers to the hobby, AI can serve as an educational tool. By offering interactive lessons, quizzes, and virtual mentorship, AI can help new operators understand radio theory, operating techniques, and regulations.
​

• AI-based tutoring: AI can provide training on topics like Morse code, digital modes, and radio theory through interactive learning modules.​
• On-demand assistance: AI can offer quick answers to questions, helping new operators troubleshoot problems or learn new skills.​

9. AI-Assisted Contesting
During contests, AI can be a huge help by automating parts of the contest process, such as logging and score calculation. It can also help strategize contest tactics, suggesting optimal frequencies, power settings, and operating times.
​

• Real-time logging: AI can automatically track contest exchanges and scores, providing feedback on performance during the event.​
• Strategic suggestions: AI can recommend when to switch bands, how to time transmissions, or how to maximize points during a contest.​

10. Machine Learning for Experimentation
Ham radio enthusiasts often experiment with new technologies, modulations, and communication methods. AI can assist in analyzing experimental data and optimizing communication techniques.
​

• Data analysis: AI can analyze experimental results from different transmission modes, modulation schemes, or antenna configurations.​
• Predictive modeling: AI can model the behavior of novel technologies or communication techniques based on limited data, offering new avenues for experimentation.​

Examples of AI Tools for Radio Amateurs:​

• WSPR (Weak Signal Propagation Reporter): AI-powered software can be used to predict optimal times and frequencies for weak-signal communication.​
• FLDigi: A software suite that supports digital modes can integrate AI features for automatic signal decoding and transmission optimization.​
• SmartSDR (FlexRadio Systems): Advanced SDR (Software Defined Radio) systems that integrate AI for signal analysis, noise reduction, and spectrum management.​
• Machine Learning-based Propagation Software: Software that uses machine learning models to predict optimal propagation paths and frequencies based on real-time data.​

In summary, AI can enhance the radio amateur experience by improving signal quality, automating complex tasks, assisting in experimentation, and providing valuable insights into propagation and station optimization. As AI technology continues to evolve, its potential to improve and transform the ham radio world will only grow.​

 

I like to run those sorts of things across different models if I'm not sure. Voting system, run it across several models and see where they agree.

 

I don't know how to apply AI to most of those issues. It's beyond an individual effort.

All the AI apps that I have access to are Large Language Models. They can answer questions and carry on a conversation. They consult the database on which they were trained as well as some degree of internet searching. Perplexity seems to emphasize search results. It does its own searches in response to your query and consults the results to form a response, but I am not aware of what search "engine" it uses - perhaps its own, perhaps not. I can't tell from the results I've seen.

You can load a small document into the free version of Perplexity and ask it to read it, summarize it, or answer questions about its content. If you pay for the better version, you can upload larger documents. But I tried to get it to read entire books without success. I am sure it can do it, but the issue is revenue generation. How much do they bleed out to suck you in and how do they charge for its use.

My understanding is that you can incorporate many of these AI engines into a software product that you create and each deal is individually negotiated. I have product ideas that would be great for that, but I no longer have the drive to start companies to pursue such things. I won't be doing that. But those who do will create the next tier of wealth. Agriculture age, Industrial Age, Information Age, AI Age.

So I just ask Perplexity to convert units, calculate dB related things, find and read datasheets, do arithmetic for me, etc. It can do math - executing formulas that it finds. But it sometimes screws up. I just tell it that it is wrong and to do it again - fix its mistake. This almost always works. Sometimes it takes several tries. It can produce results that are contradictory. I insist that it resolve these internal conflicts and it has always succeeded. It appears to put more effort into it when you tell it to fix or confirm its results. That's cycle budgeting and I'm sure it can be manipulated to so that it uses more effort and produces better results. The easy way to get it to do that is to pay them $20 a month...

I find that ChatGPT is better than Perplexity at getting math correct. I've not caught it screwing up the math. Both show you what they are doing, step by step, so you can decide whether or not you trust the equations they are finding and you can usually spot results that are wonky.

I experimented with Perplexity to see how well it learns. I took a situation where it was wrong (my own background) and corrected it. It asked for corroboration. I directed it to documentary corroboration and it updated its version of my history. Then I let a day or two go by and tried it again in a new thread. It had reverted to the old data. Again I educated it and it responded well. But this time I instructed it to update its database with the information it had learned. It said it did so. A few days later I tested it and it had retained the new information! It had, indeed, learned. But you had to instruct it to do so! What I don't know is whether or not that new knowledge is confined only to conversations it has with me or if it is available to the world at large. I'll have to use a different IP address, different machine, and different internet identity to test that.

AI is useful now but I expect it to get much better. I would love to see someone do a startup making an LLM style AI product focused on Engineering and Physics. If you fed such a thing the entire library of engineering books from, say, 1900 up to last year and gave it the ability to search current material it should be dynamite. Is the market large enough?

I think the current crop of AI apps can be pretty effective if you watch out for errors. Treat it like a grad student intern - check its work and tell it when it is wrong and make it do it over until it gets it right. Tell it to update its database when it corrects its work.

Mike WY6K

 

I don't use AI in my shack and I never will. There are way to many flaws in its answers for me to trust it. As an example, a few months ago I wanted to know some more information about the Oliver 1650 farm tractor. I accidentally hit a '9' and therefore ended up asking it about the "Oliver 1659 farm tractor".

Well, those of you that know anything about Oliver farm tractors know that Oliver never manufactured a '1659'.
Claude.ai didn't know that. It came back with all sorts of baloney about the supposed "Oliver 1659".
Years manufactured, horsepower, engine type, number built, etc.

I then promptly responded with, "There seems to be some sort of error, I don't recall Oliver ever manufacturing a 1659."

Claude.ai responded with, "Yes, I see you are correct, Oliver never manufactured a 1659. It would seem that were inquiring about the Oliver 1650 possibly?"

Anyway, point is there are way to many errors in AI for me to believe it's ready to be used exclusively for people to rely on for accurate information.

As @W7UUU has pointed out, I've also found issues with dates and timelines.
Back to the Oliver 1650 queue, when claude.ai finally did give me information on the Oliver 1650, it told me that the tractor was produced in the "80s and 90s".
Wrong. Oliver was already part of White Farm Company by that point.
The tractor was actually manufactured between 1964-1969. I had to find that information on my own.

So, in the end I really don't see any point in using AI if I'm stuck verifying and searching for the information on my own.

Don't get me wrong, I'm not saying that AI is 100% useless, there are places for it. But I don't think that we should be relying on it to provide the most accurate and up to date responses.

Sorry for the long rant, just making my point.

 

AI has been working for Hams for quite some time already, many work the world while they do this or that, yet they never talked to a single person, but they`re in the log.