If you are as old as I am (61), you’re probably skeptical about anything labelled artificial intelligence, or AI. Over the course of my engineering career, many claims have been made about AI, and few really panned out. This includes stuff like expert systems and neural networks.
Having said that, though, researchers continued to work in this field, and they have made some gigantic strides. We don’t really call it AI anymore, but products like the Amazon Echo Dot, that can decipher what your saying to it, and services like Netflix, that suggest movies and TV shows that it thinks you might like to watch, are using AI techniques. The amazing thing is that these products and services are actually lear
With that in mind, when I saw Machine Learning: The New AI by Ethem Alpaydin, I checked it out. It’s a very well-written book. It explains the concepts without any deep mathematics or code listings. I think that in itself is a remarkable achievement.
Reading this book has, of course, has gotten me to thinking about how we will use machine learning in amateur radio. A couple of years ago, Mauri, AG1LE, started a Kaggle competition to use machine learning to copy Morse Code. He’s also continued working on this, and some of his work can be seen on his Google+ page.
I’m thinking that maybe I can use Amazon’s Alexa technology to control my radio. How cool would it be to say, “Alexa, QSY to 7035 kHz, mode CW?” A little more ambitious project might be to collect data on troubles for a particular radio, or maybe antenna, and then use that database to diagnose problems.
I think the possibilities are endless. What do you think that we could use machine learning for in amateur radio?
Cognitive radio comes to mind. This is where a software-defined radio system learns how to receive/decode transmissions of all sorts.
Or imagine two or more random stations that wish to find a mutual frequency and mode with good propagation, and uses adaptive cognitive techniques.
Get Alexa a callsign, and train it to do CW, and copy cudgel fists. Alexa could be a CW Elmer, found on 40m. Maybe a contester too. Alexa might even thank you.
Ed KC8SBV
I’m not sure about getting Alexa to copy Morse Code–although that doesn’t seem to be out of the question. What might be more possible is developing an app (skill? talent?) that would send Morse Code practice.
I just got my Echo Dot and connected it up to the network. Unfortunately, it messes up the operation of my Bluetooth mouse. It makes its response slow and jerky. I’m going to have to send it back if there’s no fix for that.
I am bit late to comment on this post but just would like to make aware of an aspect of Machine Learning that can really be very useful to Ham Radio. It is also my side project that I would, hopefully, someday complete. There is something called as Independent Component Analysis or ICA [1] in Machine Learning, Long explanations aside, ICA can be used to separate out different streams of data that are independent of each other but are combined somehow in the data you get. For Ham Radio, it can be potentially used to separate out noise from the actual voice that we get in the noisy communication. This is also more commonly illustrated by the Cocktail Party example [2] where different microphones receive sounds from different sources and outputs combined sound. ICA is then used to separate individual sounds. A small unit made of Raspberry Pi or FPGA that can take the noisy input from the radio and output only the voice stream can work wonders in Ham Radio. Someday soon….
1. https://www.cs.helsinki.fi/u/ahyvarin/whatisica.shtml
2. http://research.ics.aalto.fi/ica/cocktail/cocktail_en.cgi
This sounds like a great project! Please keep me up to date on your progress.
Wouldn’t it be great if the AI could automatically detect and switch to the proper mode to decide the transmission. From noise to CW or Psk or SSB with all the best filtering. Then with the correct transceiver, you could begin a transmission and an automatic RC / TX box would flip to TX without burning up the SDR Receiver. MFJ 7308B ( not sure this is the right model number.).