Hi All -
Its no good at this stage looking for small signals. All I'm trying
to do is prove the HARDWARE. At the moment I have a QSD type receiver
delivering a signal centred on 1kHz. This is bandpass filtered to
about 100Hz bandwidth, then I/Q sampled at 1kHz sampling rate to deliver 12
bit baseband values at 1kHz
At the moment all I'm doing with these is displaying them on a
vectorscope plot - like that shown in the attachment - in order to prove phase
coherency of the system. So only strong signals are acceptable at this
stage - decimation and narrowband filtering can come later - that is high
level software and easier to implement
And I've discovered something that needs explaining. Having just
been monitoring Droitwich on 198kHz for the last couple of hours, its phase
plot indicates drift of 300 degrees in that time. Expressed in terms of
a frequency offset that is 0.00012Hz, or 0.6 PPB at 198kHz.
Now :
My reference is a VE2ZAZ GPSDO which is long term spot on, but subject to
short term variations. I have to check if it remains better than 0.6PPB
over that period. The VE2ZAZ isn't my favourite solution for LF
working. But a local high standard frequency reference isn't an issue -
the Caesium tube can be turned on to solve that one!
The LO in the LF receiver is a DDS tuned by a rotary encoder.
The PIC maths was written from first principles using 64 x 64 bit
integer arithmetic, BUT the pre-stored -constants to start with were evaluated
via a spreadsheet - which I'm not convinced provides sufficient numerical
accuracy to simulate the maths resolution needed to get teh constantsin teh
first place. So that needs checking.
And, of course, although Droitwich is supposed to be a frequency
standard, it is only Rubidium controlled; set manually from "time to time"
against an on site Caesium source. 0.6 PPB is of the order of a Rb
source that hasn't been corrected for a while. Is Droitwich properly
maintained these days when very few people want it as a reference? can
we really assume its good enough?
I've now changed to monitoring MSF 60kHz so lets see what indicted drift
that shows - its reasonable to assume that really is good!
Checking the constants used in the arithmetic is the most difficult
task. My high level programming language only offer up to 64 bit
integers, so the multiplication used in the PIC cannot be simulated exactly in
a prog, meaning I need to work with double precision floats and cast to
integers as and when.
Andy G4JNT
