Re: [Discuss-gnuradio] Homing in on an RA detector chain

[Matt Ettus]

> Given the rather-long integration times characteristic of RA, would you
> suggest
>  that the integration be "split" between the decimating FIR filter, and
> a single-pole
>  IIR filter?

First, I see no reason to use IIR filters here, except to make things
feel like the good old days of RC pseudo-integrators.  The advantage of
IIR filters is fewer taps, but since you are decimating, you wouldn't
get that advantage anyway.  I would go with a straight boxcar FIR filter
to get an average.

The most efficient way to do this would be to make a block which
accumulates the incoming signal, and every so often outputs the sum and
clears the accumulator.  This execution efficiency is not necessary,
though, and comes at the expense of programmer time and effort.

Let's assume you want to get 8 outputs per second for an 8MS/s
datastream.  You would need a 1 million tap FIR filter.  Doing it that
way would be very slow because you would be thrashing your cache.  The
best thing to do would be to run a 1000 tap FIR filter with 1000:1
decimation.  All taps should be 1/1000.  Cascade another one with it,
and you have 1,000,000 point averages, coming 8 times a second.

> The interferers was my next little thought project.  Considered it on
>  the way home. It seems to me that the RA "receiver" can start up
> analysing a FFT for narrowband interferers (anything bins that's
> so-many sigma away from the standard deviation), and make those
> interferers the "notch" in a (multi?) notch FIR filter.  Is this
> feasible with a single FIR filter, or would multiple filters be
> required?  What I'm seeing right now is that with the feedhorn
> covered in metal, at 1420.405Mhz+/- 2Mhz, I'm seeing a strong (15dB
> out of the noise) narrowband spur at -500Khz, and a weaker one at
> about +1.8Mhz.

The filter itself is not hard, but we don't currently have any filter
design code which would do what you are looking for.  The easiest way to
do what you want is to design the filter in the frequency domain.  Put
ones everywhere and put zeros on the bins which are interfering.  You
would IFFT this and pass that to the filter block.

> It's possibly being picked up by the coax cable between my LNA and
> the DBS_RX. But my theory is that it's internal to the DBS_RX/USRP
> "stack".

Do you see this signal with a terminator on the DBSRX connector?  I
might expect to see some small spurs at integer multiples of 1 MHz, but
not anywhere else.

Matt