eHam.net - Amateur Radio (Ham Radio) Community

Call Search
     

New to Ham Radio?
My Profile

Community
Articles
Forums
News
Reviews
Friends Remembered
Strays
Survey Question

Operating
Contesting
DX Cluster Spots
Propagation

Resources
Calendar
Classifieds
Ham Exams
Ham Links
List Archives
News Articles
Product Reviews
QSL Managers

Site Info
eHam Help (FAQ)
Support the site
The eHam Team
Advertising Info
Vision Statement
About eHam.net



[Articles Home]  [Add Article]  

CW Decoding Using Neural Networks

Mauri (AG1LE) on July 14, 2012
Website: http://ag1le.blogspot.com/2012/05/morse-code-decoding-with-self.html
View comments about this article!

Member Comments:
This article has expired. No more comments may be added.
 
CW Decoding Using Neural Networks  
by K1CJS on July 14, 2012 Mail this to a friend!
Maybe it's just my computer, but....

Where is the article???
 
RE: CW Decoding Using Neural Networks  
by AG1LE on July 14, 2012 Mail this to a friend!
I am wondering the same...article content is missing, but the link to my blog is visible.

73 Mauri. AG1LE
 
CW Decoding Using Neural Networks  
by K0RGR on July 14, 2012 Mail this to a friend!
I think the link to the blog is the article.

Very interesting - it will be great if this leads to a more accurate method of computer CW recognition. There are programs out there now that do a great job on machine-sent code under fair conditions, but a program that could unscramble some of the hand sent code on the air would benefit Techs and no-coders quite a bit, while giving those of us who know code more people to work.

 
CW Decoding Using Neural Networks  
by TTOMAS59 on July 14, 2012 Mail this to a friend!
Neat information. I've always noticed that some letters are just easier to master than others and different sequencing of letters vary in difficulty also.

Tim

 
CW Decoding Using Neural Networks  
by K8QV on July 14, 2012 Mail this to a friend!
Sure, this is a neat computer project as theory, but...

Actually doing CW with a computer is like heating up a frozen pizza and pretending to be a chef. What's the point? You've completely bypassed the fun of discovering the proper seasoning, finding fresh ingredients, and creating a memorable meal. You've made something that can technically be called food....

CW is the art of listening and sending properly, often under trying conditions. It's brain activity, not computer programming. It's an activity that should bring a sense of accomplishment and pleasure. It has so much more to offer than merely decoding, encoding and displaying information on a screen. Seriously, why bother with CW if you're just going to turn it into yet another automatic digital mode?
 
RE: CW Decoding Using Neural Networks  
by AG1LE on July 14, 2012 Mail this to a friend!
> It has so much more to offer than merely decoding, encoding and
> displaying information on a screen. Seriously, why bother with CW if
> you're just going to turn it into yet another automatic digital mode?

We use computers to extend human capabilities in many other areas. Why not with CW ? CW is a pretty unique mode that is used not only in radio communications but also as a method for quadriplegic people to be able control / communicate via computers. We need software that adapts to humans and is able to cope with our imperfections.

Neural network type of software could also provide improved new CW learning features for humans. If you have a very patient software that listens as you learn to send proper CW and shows graphically what characters need more practice (improper rhytm or timing) wouldn't that be quite useful ?

I fully appreciate the fact that Morse code / CW has a unique role in ham radio hobby - it requires a lot of effort to learn and thousands of hours to really master. I don't see how computers / improved software would take anything away from those who have put the effort into learning it.


I did recently some experiments with -12.2 dB SNR signals using modified Morlet wavelets (see http://ag1le.blogspot.com/2012/06/morse-code-detection-using-modified.html). This is an example where software could potentially extend human capabilities and enable for example automatic monitoring of NCDXF/IARU beacons at very low SNR - would a human have patience to listen 24x7 these beacon stations and report propagation?

73 Mauri AG1LE
 
CW Decoding Using Neural Networks  
by W4PC on July 14, 2012 Mail this to a friend!
This a great stuff.. :)
 
RE: CW Decoding Using Neural Networks  
by W4PC on July 14, 2012 Mail this to a friend!
K8QV

It's called 'fuzzy logic'. Been an AI topic for years :)
 
CW Decoding Using Neural Networks  
by VE6TL on July 14, 2012 Mail this to a friend!
I have been using Kohonen self organizing maps for many years in trying to make sense of seismic patterns as they relate to subsurface geology in the oil industry. This is a great way to visualize multidimensional signals in much fewer dimensions. When compared to a trained neural network (for example, with well control that pinpoints the geology at various locations), the self organized patterns may show similar correlations and allow for extrapolation of patterns into unknown areas. As one example, we can isolate stream channels buried a mile or more below the surface, which may contain hydrocarbons. I never thought of doing this with CW via a neural network approach. As one person commented, this kind of takes the fun out of it. Yet, nerds will be nerds and some people enjoy the exercise of applying what they know to solving a familiar problem in a novel way. We already have CW Skimmer and CW readers that solve the problem of converting audio signals into alphanumeric characters. Perhaps the SOM method will allow us to go beyond these tools into much noisier data. Its nice to know people are thinking about such things.

Jerry VE6TL
 
RE: CW Decoding Using Neural Networks  
by KE4ZHN on July 15, 2012 Mail this to a friend!
K8QV nailed it. Using a computer to decode CW is for folks too damned lazy to learn to use it the proper way..with your head and a key. You may as well be doing PSK if you use a computer to work CW. Takes all the fun and challenge out of it. I liken this to using a DX reflector to hunt DX....why not make it tougher and more of a challenge and hunt for it by tuning the bands? Anyone can shoot fish in a barrel. If you turn every amateur radio activity into a mindless computer program running by itself...you may as well go fishing and leave the rig off.
 
RE: CW Decoding Using Neural Networks  
by VE3EGA on July 15, 2012 Mail this to a friend!

Just a reality check to those out there that still assume that the code can be taught to anyone and as a natural progression of that, a person should then be able to decode it at speeds determined by a progressive learning level up to say 40wpm or beyond!

WRONG, WRONG, WRONG!!

It is so arrogant of many CW operators that have a unique ability (a gift) to send and receive at over 30 or 40WPM to look down upon those less gifted, as if they were not worthy to share the same bandwidth as them - and here they are right now in this thread mouthing off about how decoding CW is a natural progression and anyone who cannot learn it in their head (just as they can) is not trying hard enough!

I heard recently one self-indulgent old-timer / broadcast-station-owning ham arrogantly remark on how he would not work anyone who sent code at less than 35WPM (I guess I wont be working (callsign) 'Father and Mother' anytime soon!

Question: If you were to suffer a 'Stroke' would your CW ability be impaired? Of course it would!

Let me explain about how the human brain works:

It is quite possible for someone to learn the code as a combination letters and sounds yet still be unable to 'descramble' those same characters into meaningful language...

It's all to do with the neural cognitive language recognition and how the brain actually processes information!

I wont go into how neurons work here as it is beyond the scope of this thread, however I will leave you with some thoughts:

Some people speak many, many times faster than others - ask yourself why?

Some people take longer to answer a question - ask yourself why?

Put as a simple analogy: a Pentium 486 is well capable of several thousand MIPS but a PII is of course, much faster at processing the same information, yet both can still achieve the same end-result calculations in approximately the same amount of time!

It's all to do with inefficiencies - the human brain is exactly the same, how it orders and processes information is the key to how cognitive language is understood. Some people have more efficient pathways than others IN CERTAIN PROCESS RELATED ACTIVITIES!

So enjoy the GIFT!


Further reading:

http://www.technologyreview.com/featured-story/415823/intelligence-explained/

73





 
RE: CW Decoding Using Neural Networks  
by K8QV on July 15, 2012 Mail this to a friend!
"It is quite possible for someone to learn the code as a combination letters and sounds yet still be unable to 'descramble' those same characters into meaningful language... "

Thanks, Doc. We would have never thought of that.

Some people can learn to juggle while others can't. If you can't, you don't buy a juggling robot to do it for you - you take up golf or shuffleboard instead. Sheesh, does CW have to become yet another automated digital program? Why not just use one of the million keyboard software programs and not pretend you're using CW?
 
RE: CW Decoding Using Neural Networks  
by KG4NEL on July 15, 2012 Mail this to a friend!
No, but you might build a juggling robot if it leads you to solve another problem or adapt the technology for something else. I don't think Honda built Asimo because they wanted to replace our national soccer team, or IBM felt it would be all a better use of our time if our computers played our chess games for us.

Sheesh, not *everything* new is bad. :)
 
CW Decoding Using Neural Networks  
by K0RGR on July 16, 2012 Mail this to a friend!
Quite honestly, I think more computerized CW would benefit the hobby as a whole. We have hundreds of thousands of hams now who won't consider trying CW. Even in it's currently poor state, computer CW could still enable these guys to have CW QSOs with each other, and with other 'know CW' types as long as they keep the speed down.

If Technicians were smart, they'd all get set up for computer CW at 60 WPM. The computer CW decoders out there all work better at higher speeds, anyway, and most hams don't know the Morse character for ';'. But, I would hope that they would adopt a speed that most CW-savvy ops can copy in their heads. 18 WPM would probably be ideal - the speed W1AW uses for bulletins.

It's not hard to generate computer CW, either. There are even devices that will let you roll up a keyboard and send machine CW from your QRP picnic table setup. With that capability, the no-coders should at least stand a chance of being able to copy you.

If a computer program to copy hand-sent CW were perfected, that would simplify the whole thing.

I don't see CW going anywhere any time soon. We do see a lot of new licensees mastering CW. But, I think we're leaving the vast majority out in the cold, and CW could once again be a common ground for all hams if we could merely accept computers.

Why can't the universe back up and do everything my way? I don't really know the answer to that either, but being an elitist about your CW capability doesn't help the situation one iota. I'll match my code proficiency certificates against anybody's, and yes, Viriginia that was a 20 WPM test I passed at the FCC. I still operate more CW than most any other mode, I even operate a lot of mobile CW and I love it. I'd really like to be able to work all the Techs and no-coders that way, too.

One positive note is the number of Techs I hear about who are working the CW DX contests using computers. Since most of the CQ's and contest exchanges are computer sent, why not? Using a soundcard, they are probably closer to being on frequency than a lot of the other hams.
 
RE: CW Decoding Using Neural Networks  
by VE4EAR on July 16, 2012 Mail this to a friend!
This hobby is all about communication and communicating. If a few folks choose to not communicate with someone proficient in the skills of manual code sending and receiving, then it is their loss. They will miss the opportunity to talk with someone new and eager to learn.

This hobby has so many facets and everyone should be allowed to enjoy it using the modes and techniques that best suits them (within the law of course).

I started in this hobby 30 years ago and was one of the first no-code licenses in Canada. I have since realized the value of CW and have learned it, but I am not especially great at it. It is just another mode like PSK, RTTY, JT65, or any of the other digital modes.

Being able to send/receive 35+wpm is something to be proud of. Hell, sending/receiving 15 wpm is not too shabby and should be viewed as an accomplishment. If someone wants/or needs to use a computer to assist them so what. I say all new comers should have access to a CW decoder so they can actually see what some of these old kermudgeons are saying to each other. That way the will not waste their time trying to communicate with these dinosaurs. It is clear they have nothing useful to add to the hobby and are not worried about the future of the hobby or encouraging new growth.

Do the Kermudgeons also think we should get rid of grocery stores,and return to hunting and gathering for food? Any thing else is just too simple and you just can't take pleasure from simple things...Well at least they got that part right.


 
RE: CW Decoding Using Neural Networks  
by K9RJ on July 16, 2012 Mail this to a friend!
I am a long time cw op but not super fast. What I have wanted for many years is a great noise reduction capability that would allow me to copy those "in the noise" signals that can be extracted with certain digital modes. If the technology being discussed could do that I would love to use it when signals are weak. My current QTH has high noise, if an antidote is possible, bring it on!
 
CW Decoding Using Neural Networks  
by AD7II on July 17, 2012 Mail this to a friend!
A super article. It's hard to disagree with any of the comments, either the pro's or the con's. Morse is adequate for the Q codes we use as an international pidgin, but its shortage of case and punctuation limits it. I suppose that (dis)continuous wave could be used for sixteen or thirty-two bit characters if a computer was doing the work, so Morse isn't the only possible cw encoding. Trying to catch up with what the brain can do for standard Morse is fascinating - keep us informed.
 
CW Decoding Using Neural Networks  
by NN4RH on July 18, 2012 Mail this to a friend!
If you're going to use a machine to decode communications anyway, why not instead just use a digital data mode that is designed for that purpose?
 
RE: CW Decoding Using Neural Networks  
by AB8O on July 18, 2012 Mail this to a friend!
maybe the article is there but we can't decode it because we're not computers? ;)
 
CW Decoding Using Neural Networks  
by AA1RB on July 18, 2012 Mail this to a friend!
I'm now using an organic neural network which is still being trained. However, it's very old, slow , and error-prone.
 
CW Decoding Using Neural Networks  
by K8QV on July 19, 2012 Mail this to a friend!
Inspiring! It prompted me to think up lots of modern "improvements."

Like oil painting? Speed up that tedious process some have no talent for by buying a digital camera! Same thing only better!

Love to ride horses but they're too slow and they make your ass sore? Get a car! Same thing only better!

Want to play football but you're too uncoordinated? Buy the video game! Same thing only better!

With enough technology, we won't ever have to learn a skill or do anything for ourselves again!!!
 
RE: CW Decoding Using Neural Networks  
by AJ4MJ on July 20, 2012 Mail this to a friend!
"Using a computer to decode CW is for folks too damned lazy to learn to use it the proper way..with your head and a key."

Have you ever seen http://www.reversebeacon.net? That's an application of computer CW copying that benefits the entire ham community by providing a real-time view of propagation. Dozens of automated receivers throughout the world monitor entire bands and report every signal heard 24x7. That's simply not possible with hand copying.

73 de AJ4MJ
 
RE: CW Decoding Using Neural Networks  
by VE3EGA on July 21, 2012 Mail this to a friend!

...or people too lazy to follow the thread?
 
CW Decoding Using Neural Networks  
by W5TTW on July 26, 2012 Mail this to a friend!
" If you turn every amateur radio activity into a mindless computer program running by itself...you may as well go fishing and leave the rig off. "

Fishing sounds like a great idea! I have a new electronic fish finder that I'm anxious to try.
 
RE: CW Decoding Using Neural Networks  
by W3TTT on July 27, 2012 Mail this to a friend!
Correct. I am working with some "special needs" students. One student will *always* answer any question like this:

Firstword...aahhhhh...firstword...f..f..f..aahhhh...firstword and rest of answer.

Now why is this student like that? I dont know. But i suppose it is because his brain operates in that fashion.

Now how would he "hear" Morse code? I suppose he would hear it with the same internal brain stutter that he exhibits when he talks. In other words, he will never learn code.

Now, this is an extreme example, but the natural abilities of people will span - from the 35wpm easily person to the can't get it at even 5wpm person.

That is just the way people are.

Oh, by the way, the "student" was Albert Einstein.
 
CW Decoding Using Neural Networks  
by PA1Z on July 28, 2012 Mail this to a friend!
A thread was generated here, but I sincerely doubt that anyone who read the bloodspot above (including myself!) understood a single thing about what it was about. I don't see that it has anything to do with learning code, and I also doubt that the blog owner even understands it !

If anyone he here really understood it, please enlighten me ;-) !

Don't get me wrong; I am not against unusual and deep discussions, however i am really tired of scientific bla-bla on the net that has no respect for the backgrounds of intelligent readers. This blog posting is merely an attempt to impress other people, not to explain or educate them...
 
RE: CW Decoding Using Neural Networks  
by AG1LE on July 29, 2012 Mail this to a friend!

> I don't see that it has anything to do with learning code, and I also doubt that the blog owner even understands it !

Hi Theodore

My blog is about ham radio projects and experiments - this particular experiment was to find out how well a particular neural network algorithm called Self Organizing Maps (SOM) would work in learning Morse code characters purely from noisy data.

I am sorry if the experiment was not explained well enough. My intention was not to write a tutorial on SOM as there are several already available in the web. I did provide links to relevant material for readers to get familiar with SOM related concepts as I did not expect everybody to be familiar with the topic.

I was merely trying to demonstrate that SOM type neural network algorithms are able to cluster Morse code patterns, build a codebook automatically and learn to classify these patterns -- even if the original signal contains noise and jitter. I also used standard SOM visualization techniques to demonstrate these results.

While this may not meet your criteria for a scientific publication I felt this was still newsworthy topic, especially as I implemented a SOM decoder in the latest FLDIGI software.

There has been many comments on this forum that pursuing a better Morse decoding software is nonsense - everybody should just learn Morse code like we did 25 or 50 years ago - using our ears and brains.

However, I feel that we need to use and improve the tools available. If ham radio as a somewhat technical hobby stagnates we will have difficulties getting the younger "facebook & twitter" generation interested and to participate - this generation is very fluent in using computers to communicate. Also, as the ham population gets older we will have more hams with not so good hearing who may need some assistance in copying Morse code.

What comes to my understanding of Morse code - I became ham in 1976 and learned CW in the army at around 15 - 18 WPM speed level. I admit that I was QRT for over 20 years and have not practiced CW actively since 1986. After receiving my current license in 2010 I have been using FLDIGI and it has helped me to start re-learning CW but I am not very fluent yet.

73 Mauri AG1LE



 
RE: CW Decoding Using Neural Networks  
by PA1Z on August 4, 2012 Mail this to a friend!
Hi Mauri,

Sorry if I come over too hard; I am also sure your intentions were sincere.
SOM's may be of interest in the way computers learn and organise material, with specific parallels to human learning.

However, the broad methodology you outline - for so far that I can follow even a bit of it - do not seem to take into account many other aspects of human learning process, such as the training of reflexes, the way muscles learn to connect with certain brain functions, and the way sound is processed amongst other items.

A computer system can learn to understand Morse code and organise it if I follow you. But I would expect that to be quite simple. The problem for humans is getting the brain, fingers, ears and muscles to react and coordinate, and I don't see how this relates to that, or gives any real insights? This is why I reacted the way I did.

In the end, as a conservatory trained musician, I learned that only patient and careful practice works for things like music and code. I don't think there is any real shortcut except one: everyone learns in their own way and finding out what works for you is the real 'shortcut'

Let me use myself as an example: I cannot use books to learn languages, I forget all that I read and it bores me. I only learn new languages by listening and trying to read foreign newspapers.

My partner on the other hand needs textbooks and wants to know all the verb forms before speaking a word. Each their own way…For code, it is the same I think.

Anyway, good luck with the endeavours and hope something good comes out of it !
73 The PA1Z
 
CW Decoding Using Neural Networks  
by KC9UBR on August 4, 2012 Mail this to a friend!
I'm new to ham radio. I've been malined because I only have a few look ups here. Seems to me though the same problem I see in the 1770 reenacting I do. Sooner or later you will run into the "old guy" that has been doing this for 20...30..50...100 years. And the response is almost always the same. "I had to research and learn and do this myself, I'll be damned if I help out this new guy, let him do the work I had to"
Same thing here. Sure learining code is cool and I hope to be able to do it sooner or later. But technalogy moves on. If the "old guys" don't keep up or help the new guys, they may be the ones setting around looked at a dead radio with no one to talk to.
I remember a person on u-tube bemoaning dx'ing, people not learning cw, no electronic skills...etc. Yeah he refused to offer help. "It's my way or no way", that's the way some see it. I only hope they wise up and pass along all the great weath of knowledge they have before they drop dead at the key. Very interesting article, to
 
CW Decoding Using Neural Networks  
by N3AIU on August 10, 2012 Mail this to a friend!

Some seem to be missing the point. Yes, in principle it can be used by codeless hams. Yes, if you know code well you can look down on people who use it.

I view this software differently. It's a good way to experiment on mimicking human behavior using AI stuff, especially in weak-signal and high-QRN environments.
 
RE: CW Decoding Using Neural Networks  
by S5M on August 17, 2012 Mail this to a friend!
PA1Z wrote: In the end, as a conservatory trained musician, I learned that only patient and careful practice works for things like music and code.

I was amazed in 2002 HST competition when top CW ops copied 90 wpm while I couldn't even follow group pauses. K1ZZ and myself asked Russian YL how to do it and the simple answer was: TRAINING. Any brain is capable of following single tone in 1:3:5 rhytm.

Silicon brains are even better with digital signals. I had CW robot making 200+ contest QSO in ARRL 1991 as part of my PhD research in DSP. K6STI and myself agreed NOT to launch commercial product in order to preserve the hobby. K6STI memorable quote was: I don't want robot doing sex for me! Nothing significant appeared until VE3NEA CW Skimmer breaktrough. I refreshed robot under Windows C# as I can operate in Expert category of CQ WW contest now. However, I hate to miss 10 m openings in this Sun cycle peak. I also developed RTTY Skimmer for 44 kHz bandwidth.

Mauri, great AI research along your professional work. Let us keep hamradio as the best scientific hobby of 21st century. Otherwise it might die with us although there would always be a small minority interested in natural phenomenum of ionosphere.

73 de Mario, S56A, N1YU, hamradio & CW since 1962
 
RE: CW Decoding Using Neural Networks  
by S5M on August 17, 2012 Mail this to a friend!
BTW our work on SDR, RTTY Skimmer aand CW robot with YT7PWR is publicly available at:

http://code.google.com/p/cw-expert/source/list

73 de Mario, S56A, N1YU
 
RE: CW Decoding Using Neural Networks  
by WD8KNI on August 25, 2012 Mail this to a friend!
I for one get very tired of hearing about the old guy who looks down at the new guy. Since 1972 I have NEVER seen this type of activity. I find that when I hear something like this it is spewed by a very unstable person. Someone who would rather "drink the cool aid" rather than have real experiences. Simply an excuse and a type of behavior I will not associate with. I for one learned CW the hard way, found out what I was doing wrong, and relearned again. I think several iterations occurred along the way.
The old guy syndrum is nothing but a crutch for laziness, it is not a reality in any sense of real life.
I will slow down to work anyone at any time. I will not however send slower than the person I am speaking to is sending. When working at higher speeds I always very my speed to find out if the other guy is using a decode program. If I find they are I quickly end the QSO. Simply said if I wanted to talk to a computer I would be on the green keys, or PSK31.
I have also NEVER found anyone who could not learn CW,if they wanted to. I have found hundreds who were learning the incorrect way. Learning CW is 100% like learning a new language. You didn't learn your native language, or a secondary language overnight, CW will be no different.
Ok now time to get off my soapbox. The old guy syndrum is akin to saying I can not learn physics when you have never read a book regarding the subject. Stop blaming the old guy. Nothing is accomplished without spending time to perfect something. No mechanic ever learned to rebuild an engine without actually doing it. Fred
 
RE: CW Decoding Using Neural Networks  
by WD8KNI on August 25, 2012 Mail this to a friend!
I for one get very tired of hearing about the old guy who looks down at the new guy. Since 1972 I have NEVER seen this type of activity. I find that when I hear something like this it is spewed by a very unstable person. Someone who would rather "drink the cool aid" rather than have real experiences. Simply an excuse and a type of behavior I will not associate with. I for one learned CW the hard way, found out what I was doing wrong, and relearned again. I think several iterations occurred along the way.
The old guy syndrum is nothing but a crutch for laziness, it is not a reality in any sense of real life.
I will slow down to work anyone at any time. I will not however send slower than the person I am speaking to is sending. When working at higher speeds I always very my speed to find out if the other guy is using a decode program. If I find they are I quickly end the QSO. Simply said if I wanted to talk to a computer I would be on the green keys, or PSK31.
I have also NEVER found anyone who could not learn CW,if they wanted to. I have found hundreds who were learning the incorrect way. Learning CW is 100% like learning a new language. You didn't learn your native language, or a secondary language overnight, CW will be no different.
Ok now time to get off my soapbox. The old guy syndrum is akin to saying I can not learn physics when you have never read a book regarding the subject. Stop blaming the old guy. Nothing is accomplished without spending time to perfect something. No mechanic ever learned to rebuild an engine without actually doing it. Fred
 
Email Subscription
You are not subscribed to discussions on this article.

Subscribe!
My Subscriptions
Subscriptions Help

Other Recent Articles
Army MARS at the ARRL Convention:
Skywarn Warriors: Radio Buffs Work Front Lines for National Weather Service
Amateur Radio Operators Descend on Civic Center for Annual Hamfest:
Ham Radio Users Could Be Vital Resource In Emergency:
The First Social Network: Chewing the Rag with India's Ham Radio Operators: