![]() In this kata you have to write a simple Morse code decoder. After you solve this kata, you may move to the next one. and don't miss the big game tonight.This kata is part of a series on the Morse code. You can read more about my experiments from by ham radio blog. With more training data and more compute capacity the race is on to build the ultimate Morse decoder. The model I used was actually designed to recognize handwritten text but it seemed to work just fine in learning and decoding noisy audio files. How can this artificial neural network learn to decode Morse code at 30 words per minute (WPM) in only 1 hr 51 minutes while it took me months to reach 15 WPM level back in the day? How can it recognize Morse characters from signals that are so noisy? Why do a few neural network nodes seem to figure out the decoding task better than the 100 billion neurons in my brain? I was actually quite surprised when my first crude attempt to train this new CNN-LSTM-CTC model with only 5.2 hours of audio material the model was able to decode even noisy test audio files with -6 dB signal-to-noise ratio. I took a different approach and started using "Deep Learning" AI and wrote a decoder that can learn Morse code from labeled audio files. Instead of adding more complex algorithms to track the speed, estimate noise level, tune the filter bandwidth, find the peak signal frequency etc. Over the last two weekends during the deep freeze in New England I started a new project to figure out a different way of building a Morse decoder. I have written software for multiple different Morse decoders over the years to help me to follow those high speed conversations, but these decoders all seem to have flaws - with the real world noise, interference and signal fluctuations they tend to decode garbage. CW is a tough mode to master but I have seen hams that can listen and understand high speed CW conversations effortlessly with a speed that I could never achieve. I learned CW when I started my ham radio hobby, and spend many hours practicing in the Signal Corps in the Finnish Military. This mode is a simple digital communication that humans can learn by practicing hundreds or hours. One of the popular modes of communication in ham circles is known as CW (continuous waves) also known as Morse Code. When another ham responds to your call from some 17,000 kilometers away to make a contact it surely feels like magic. ![]() While we have Internet and instant messaging it is somehow very satisfying to setup a small portable radio station on a remote island in Pacific and using less power than a flashlight to bounce radio signals from ionosphere to reach out to the other side of our planet. Sometimes I take the radios with me to some more exotic location and attempt to make contacts with fellow hams across the world. I have been a ham (amateur radio operator) for over 40 years and love to listen the HF radio bands. I have been interested in Machine Learning for several years and trying to figure out how to build a better Morse decoder using ML techniques. Well, for some of us this gives an opportunity to delve into the mysteries of Deep Learning and Artificial Intelligence. What can you do when a Polar Vortex lands over North America and it is just too cold to go outside?
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |