I think I have found a way
to eliminate interference from the DCF77 pseudonoise modulation
in 74.55 kHz spectrograms.
As explained before, the problem
is that the phase-modulating PRN sequence carries the timecode of
DCF77, sending an inverted sequence with every "1" bit.
Thus what would otherwise be a comb spectrum of discrete 1 Hz lines is
being spread out, contaminating the space between the lines with noise-like
The idea is now to
unspread the lines, based on the known time bits, which are also
communicated by prolonged second gaps (0.2 s for a "1" bit). Multiplying
the 74.5 kHz stream by the 1, 1, -1... data will fold all
the interference energy back to discrete 1 Hz multiples.
These can then be notched out. Finally the despeading has to be undone
with the same data, restoring the desired spectral features (amateur
signals or Loran lines), but without the interference.
Example results and zipped
MathCad and data files are in
and a comparison of
unprocessed and processed spectrograms from the night November 7 /
(spectrogram as received),
(spectrogram with DCF77 sidebands removed).
Unfortunately there was no
signal from Bob or Dex during the night of the recording. All you can see are a
few Loran lines, eg. on 74556.6076, 74550.5868, 74549.9400 Hz. I want
to try this again, but not today as the big antenna which is needed for 74 kHz
is currently retracted due to heavy winds.
For anyone interested in
the postprocessing, here are the steps in detail:
SpecLab, VAC and SndInput, I took a narrowband IQ recording,
with decimation of 768 down to 15.625 Hz samplerate. There
are two audio channels, one centered on the desired band around 74.550 kHz, and
the other on 77.5 kHz for reference. In the example, the overnight
recording started 13-11-07 16:38, and ended next morning
6:24. The 6 MB .tmp file was preceded by a dummy .bmp
header to be able to read it with the old MathCad
2. In the MathCad script, the
second dips were identified and the bits extracted from therir duration. The
first 15 bits which carry encrypted weather information are not included in the
phase modulation, and have to be substituted by a fixed
111111111100000 sequence. In principle, the time telegrams are completely
predictable (except possible leap seconds), so with accurate time information
one could do without the 77.5 kHz channel and the bit
3. The time domain data
is split into overlapping 2048 sample chunks, allowing to generate
spectrograms at 7.63 mHz resolution. For each chunk, a windowed FFT is
generated, leading to the unprocessed spectrogram (sp_2e.png)
4. In time domain,
the samples during each second with a preceding "1" bit are multiplied
by -1, despreading the interference.
5. These data chunks
are also Fourier transformed, producing spectra with discrete 1 Hz peaks.
6. The bins on and
near integer Hz frequencies are nulled, notching out the
7. The data is brought back
to time domain with an inverse FFT.
8. The polarity inversion
of the "1" periods is undone, applying the same procedure as in step 4
9. Finally data is Fourier
transformed again, producing the clean spectrogram result
Sent: Tuesday, October 01, 2013 7:35
Subject: Re: LF: 74.550kHz Sep
Here are clearer shots of the DCF77
sidebands from the morning:
- the RX antenna is resonant around 75 kHz,
which emphasizes the PRN sidebands below the DCF77 carrier. The fifth and
sixth lobe are still visible. In reality, the upper sidebands are slightly
stronger. This is probably due to an offset (or a 75 kHz notch) in the
transmitter antenna matching, which happens to help us now.
- the Swiss time signal HBG on 75 kHz is no
longer on air.
- there is an RTTY signal at 73.6 kHz which
could be CFH.
- the 1 Hz lines are surrounded by a somewhat
regular fine structure, consisting of 16.6 mHz spaced sub-lines. This is
probably due to parts of the BCD timecode and weather
information data which are repeating or similar in consecutive
Sent: Monday, September 30, 2013 11:26
Subject: Re: LF: 74.550kHz Sep
Hi Bob, LF,
These are presumably artifacts from DCF77 which is
only about 160 km from here. In addition to the well known
AM timecode, it also carries pseudorandom phase modulation,
which has been proposed in the 80ies to provide higher resolution timing
(albeit orders of magnitude worse than Loran or GPS). The resulting sidebands
extend a couple of kHz on either side of the carrier, with pronounced minima
around multiples of the chip rate 77500/120 = 645.833 Hz, see
Attached is a spectrogram which was taken
tonight on the resonant antenna. Between statics, you can still
see the fourth sideband lobe which is centered near 74.6 kHz. The
spectral gaps are on
with small and sharp central lines, presumably
caused by slight inbalances or nonlinearities in the
By these criteria, if you have the choice I
would recommend to operate somewhere near these gaps, but not exactly in
their middle, and also preferably not exactly on integer Hz