I was not getting a significant number of decodes near the threshold from
stations participating in the WSJT-X 2.3-devel testing. I had a number of
questions that I wanted to investigate for my own understanding, including:

Why was I not seeing any decodes below -42 when operating with the noise
blanker? Was this just lack of suitably weak signals, or some anomaly with
my setup? Would I be able to get decodes below that level?

What is the best overall noise blanker setting for me (most decodes without
too many false decodes) and how many decodes might that get that I miss
without the noise blanker?

To what extent might reported Doppler-spread measurements be less
accurate for decodes approaching the threshold?

This was NOT an attempt to determine the WSJT-X decoding threshold, but
was intended to help answer the above questions in my real world
receiving environment.

I wanted this to be a real world test, with signals injected into my
receiver by way of my usual receiving antenna. This would present all the
usual challenges of coping with my local noise and RFI, making it a
realistic real world test. Given that my daytime receiving conditions in
August are very similar to a good winter night in terms of atmospheric QRN
and local RFI, I chose to run these tests during the day. Over a period of
six days, more than 100 FST4W-1800A transmissions were made at very low
power to an outside antenna assuring "over the air" reception through my
receiving antenna.

The transmitting setup for these tests was WSJT-X audio and RF carrier
from a GPS locked HP-3325B synthesizer to a Mini-Circuits ZAD-8+ mixer.
Modulated signal output from the mixer was fed to a small outside antenna.
Signal power was adjusted with the HP-3325B.

The receiving setup was my normal 2200m setup, consisting of a 30 foot
"LNV" antenna, modified SoftRock Lite II receiver with GPS locked LO
and additional front end band pass filtering; HDSDR sending audio to
WSJT-X through a virtual cable.

In actual practice, local RFI became severe enough to prevent decodes on
my usual frequency, 137437, several times during the tests. For minor
situations I changed the WSJT-X Tx audio several Hertz. For major situations
involving more broadband QRM of significant strength to severely mess up
decoding, I changed the transmitter carrier by 500 or 600 Hz and the
receiver LO setting in HDSDR by a like amount. Had this not been done,
many more days would have been required to get this number of relatively
ucompromised test files.

Files related to this test:

FST4W-1800A_WeakSigTest_Notes.txt (test description and notes)
FST4W-1800A_WeakSigTest_Decodes.txt (list of all decodes)
200811.zip (wav files for August 11)
200812.zip (wav files for August 12)
200813.zip (wav files for August 13)
200814.zip (wav files for August 14)
200815.zip (wav files for August 15)
200816.zip (wav files for August 16)

These files are located at:

http://n1bug.com/FST4W_Test/FST4W-1800A_WeakSigTest_Notes.txt
http://n1bug.com/FST4W_Test/FST4W-1800A_WeakSigTest_Decodes.txt
http://n1bug.com/FST4W_Test/200811.zip
http://n1bug.com/FST4W_Test/200812.zip
http://n1bug.com/FST4W_Test/200813.zip
http://n1bug.com/FST4W_Test/200814.zip
http://n1bug.com/FST4W_Test/200815.zip
http://n1bug.com/FST4W_Test/200816.zip

What did I learn from this?

It appears I get decodes reported down to -44 with a rare -45, both with
and without the WSJT-X noise blanker. It only seemed like this wasn't
happening because there were no signals from others at those levels.

The best overall noise blanker setting for me appears to be 10%. With
signals near the threshold, this results in a significant number of
decodes that are not possible without use of the blanker. (NOTE: When
decoding files recorded at other stations I often find the best noise
blanker setting is something significantly higher or lower than 10%.)

There appear to be indications that reported Doppler-spread measurements
become less reliable near the threshold. This is not surprising and was,
in fact, exepected. Now I have a better estimate of where things begin
to become less meaningful.

It would be nice to have these decodes in a spreadsheet where they could
be sorted by various parameters. However, I am NOT proficient with
spreadsheets (I also have a gift for understatement) and have chosen to
preserve what little remains of my sanity instead of persuing that.

Paul, N1BUG