Hello Markus, Jim, and all,
Markus wrote:
Yesterday I did basically the same experiment using an audio filter, and saw no difference at all in decoding
ability between full SSB bandwidth and a 250 Hz filter. I concluded (perhaps prematurely) that there is no blocking effect present.
I do confirm that there is a blocking effect in the WSPR decoder, but I
am not really sure about the details...
Today I did another test by injecting an audio carrier from the Speclab generator within
the WSPR search bandwidth. The "wanted signal" was F5WK signal at 1577 Hz with
-57 dB below ADC fullscale, displaying -15 dB SNR at the time. I found that at a 1450 Hz
carrier at -20 dBFS completely blocked decoding, even though the spectrogram traces of
the WSPR signal and the carrier both looked perfect. Reducing the blocking carrier to -30
dBFS seemed to reduce Michel's SNR to -20 dB.
In hindsight, and with the information from Jim, I assume that the out-of-band
interference levels in my first experiment simply have been too small to cause
an effect. DCF39, the strongest daytime carrier, is attenuated by the SSB IF
filter edge to about -34 dBFS at originally 3330 Hz, which is then
downconverted to 2830 Hz for the WSPR computer. So I now also believe that
decoding might possibly be prevented by a 25 dB larger signal anywhere in the
audio band.
I guess I will need to play more with this tomorrow.
... similar findings here, but they are difficult to sort out. In the
meantime I was able to improve the readability of Jim's and Michel's
signal by *carefully* narrowing the IF filter, to give the decoder less
out-of-band signal energy to chew on :
2152 -20 0.0 0.137500 0 M0BMU IO91 23
2152 -21 0.1 0.137577 0 F5WK JN18 37
2158 -23 -0.2 0.137500 0 M0BMU IO91 23
2158 -26 0.0 0.137577 0 F5WK JN18 37
< at this point I modified the filter settings (*) et voila : >
2210 -14 -0.3 0.137500 0 M0BMU IO91 23
2210 -12 0.0 0.137577 0 F5WK JN18 37
2216 -14 -0.1 0.137500 0 M0BMU IO91 23
2216 -11 0.1 0.137577 0 F5WK JN18 37
2222 -11 0.1 0.137577 0 F5WK JN18 37
But drawing any conclusion may be premature - some neighbour may simply
have switched off his local QRM generator.
My impression is that some processing stages within the WSPR decoder
lack dynamic range, or the FFT has significant leakage, or some part of
the calculation is done with short integer values (instead of double
precision floating point values). The decoder is open source, but
lacking any knowledge of the Fortran programming language, I won't try
to understand the algorithms.
(*) what I did with the TS850 was switch back to SSB filters in both
banks, but use the passband tuning to narrow the frequency range,
carefully watching on the broadband spectrogram. The high-pitched tone
from DCF and some other, "local" carriers are almost unaudible now, and
WSPR seems to be happy about that. Early today, on the club station,
the narrow preselector in the LF transverter has done a similar job.
73,
Wolf DL4YHF ...-.-
|