TNF filter width

DrTeeth

I did read something about the width of a TNF on the panadaptor more closely matching the actual filtering width in the latest SSDR release.

I have just lost a contact on JT mode as this is not the case - a default 100Hz TNF actually blocks 300Hz of spectrum as seen in the WSJT-X waterfall.

I did notice that placing a default TNF in between the two RTTY tones quietened the whole signal.

Can the TNF marker be made to more accurately reflect the filtering bandwidth?

DrTeeth

I would refer people to this https://community.flexradio.com/flexradio/topics/tnf-vs-rtty-question where Steve N5AC commented.

Steve-N5AC

You should almost never use TNF with digital modes.  Most software for digital modes has filtering and decoding that already does required filtering and you're as likely to mess up the data as you are to remove something that is an issue.

DrTeeth

I use it to remove s9+++++ RTTY signals. You know how good and clean some EU signals can be, hi hi?

Surely they do have an adverse effect on the decoding process? I do like to listen to the audio even when using digital modes - heck, I cannot even work or go to sleep in silence ;-).

Steve-N5AC

I have a couple of things to say here -- first I do a lot of JT65 and in my experience the RTTY signals are not around 076, they are typically around 081ish.  Even working JT9 I find there are generally no RTTY signals below the JTx signals.  So my first recommendation would be to adjust the filter to cut them out of the passband.  But let's talk for a moment assuming that this is wrong and that they are in the middle of your operating region.

What you are looking for is dynamic range -- the ability to decode a signal many dB below the level of a larger signal.  You happen to be operating the radio with the best dynamic range in the HF world and the DAX output data from the radio contains all the dynamic range the radio is capable of.  You lose some (probably 10dB) running through Windows sound cards, but you shouldn't need this in most cases.

I've only looked briefly at the WSJT code and I don't know the signal-processing block diagram, but the author(s) are very smart folks.  If I was going to design something to do what JT65 does, I would use a long average FFT (the total duration of the sample set) to locate pilot tones (the tone that is returned to many times in the sequence).  Then I would place a PLL in the time-domain data to track this tone and an FFT or series of receivers for each of the expected tones spaced from the frequency captured in the PLL.  Each of these receivers would necessarily have to have a very sharp filter in order to hear up to 25dB in the noise.  These receivers would not allow anything from more than a few Hz away to affect the decoding provided the overall signal didn't overload the number space (a.k.a you have sufficient dynamic range). 

So intuitively, it may seem that a large signal in the passband might cause an issue, but as long as it is not on the same frequency as your WSJT signal, I don't expect it to have an effect.  I don't have proof of this other than it doesn't seem to be an issue when I operate and I have my speculation about my best guess at the design of JT65 that says it shouldn't make any difference.

DrTeeth

Once again Steve, thanks for explaining something with amazing clarity.

73