CW Decode from HO-68 using MultimonNG - Updated

(This is a follow-up to my previous here.)

The author of MultimonNG suggested that in addition to altering -d and -g, I should also use a bandpass filter before passing it to MultimonNG as the code does no filtering of itself. I applied a bandpass filter of around 400Hz wide. This filter is wider than I would have liked to have used, but the signal drifts due to, probably, instabilities of my RTLSDR receiver and errors on the Doppler corrections - there are ways to automatically correct this, but I've not had time to implement them yet.

If I now decode the data after applying the filter I get the following ASCII:

            W X W A A A T T T A U   A T E E T T T T T T T T T T T T T T T T T T T T T T T T T T X W X W E E
B J 1 S A X W X W A A A T T T A U 4 T B E E T T T T T T T T T T T T T T T T T T T T T T T T T T X W X W
B J 1 S A X W X W A A A T T T A U 4 T N T E T T T T T T T T T T T T T T T T T T T T T T T T T T X W X W
B J 1 S A X W X W A A A T T T A U 4 A V T E T T T T T T T T T T T T T T T T T T T T T T T T T T X W X W
B J 1 S A X W X W A A U T 5 H H 5

B J 1 S A X W X W A A A T T T A U 4 A V E E T T T T T T T T T T T T T T T T T T T T T T T T T T X W X W
B J 1 S A X W X W A A A T T T A U 4 T B E E T T T T T T T T T T T T T T T T T T T T T T T T T T X W X W E
B J 1 S A X W X W A A A T T T A U 4 A V T E T T T T T T T T T T T T T T T T T T T T T T T T T T X W X W EIE
B J 1 S A X W X W A A A T T T A U 4 T D T E D T N N T T T N E

Which is significantly better than without the filter (see previous post) and better than dl-fldigi.

Next I investigated altering the -d and -g parameters. To do this I wrote a Python script to run multimonNG that counted the number of occurrences of the string "T T T T T" were in the output. I looped over values for -d from 0 to 202, and -g from 0 to 198. The out put can be see in the graphic below (sadly I forgot to label the axes before I created it) - the y-axis is -d and the x-axis is -g:

From this is can be seen that for this data the value of -d is not too important, but the value of -g is more critical. However looking at the output for the best combination of -d and -g shows no significant difference from the results above. It is also questionable if the number of "T T T T T" in the output is really a good metric to measure decoding quality - but for the moment it will suffice!

In conclusion - make sure you appropriately filter your input to MultimonNG and you should get good results. Over the coming weeks I'll be trying with data collected from other sources which won't be so nice and clean.