ben-wpan/usrp

Antenna measurements
====================

The objective of antenna measurements is to determine how much energy the
antenna transfers at different frequencies. For this, we set up a sender,
a receiver, connect one to the antennas being tested, and the other to an
arbitrarily chosen lab antenna.

Since none of the items (sender, receiver, lab antenna) are calibrated,
we can only compare antennas but we cannot determine any absolute
characteristics.


Preparing a measurement run
---------------------------

Before measuring the characteristics of an antenne, we need to set up the
test environment and obtain a number of filtering parameters. The filters
are used to reduce the effect of noise on the measurements and to suppress
contamination from other sources.

1) Install transmitter and receiver. The transmitter is an atusb or atusd
   board, the receiver an USRP2+XCVR2450 with the antenna to test.

   (The same setup may also work with a USRP1 or UN210, and a RFX2400
   board.)

   Both should be spaced at least twenty times the wavelength, or 2.5 m
   apart. For test runs that can be compared with each other, antenna
   placement and orientation have to be exactly the same.

   The sender runs tools/atrf-txrx/atrf-txrx, the receiver runs utilities
   from gnuradio.

2) Obtain baseline performance values. For example, activate the sender
   with

   atrf-txrx -f 2455 -p 0.5 -T +0.5

   Emit a constant wave at 2455+0.5 MHz with a power of 0.5 dBm or 1.1 mW.

   Monitor the received signal with

   usrp2_fft.py -f 2455.5M -d 16

   Record the range in which the frequency peak falls. Variations of a few
   dB are to be expected.

3) Generate a series of sample for a specific setting.

   Example:

   The following script sets up the transmitter, lets it "warm up" for ten
   seconds, then takes 100 measurements, stored in files tmp00 through
   tmp99 in a directory $PWD/100/.

   In this setup, the receiver's gnuradio runs on a different host than
   the sender. Therefore we use ssh and pass the directory from $PWD.

    atrf-txrx -f 2455 -p 2.6 -T +0.5 \
      'sleep 10;
       for a in 0 1 2 3 4 5 6 7 8 9; do
         for b in 0 1 2 3 4 5 6 7 8 9; do
           ssh ws usrp2_rx_cfile.py -d 16 -f 2455.5M -g 46 -N 1124 \
	     '$PWD'/100/tmp$a$b
         done
       done'

   Each measurement obtains 1124 samples, 1024 samples for the FFT and
   100 samples to cut off (see below).

4) Determine the shape of the captured waves in the time domain, e.g.,
   with

   gnuplot
   gnuplot> plot "<./avg 1 <100/tmp00" with lines

   "avg" outputs the magnitude of the recorded wave, averaging over the
   specified number of sample.

   Some waves will probably show a peak in the first few samples. We need
   to cut off these peaks in the later processing steps. In this example,
   we will skip the first 100 samples.

   Besides the initial peak, the waves should be of comparable amplitude.

5) Verify the distribution in the frequency domain and determine the noise
   floor.

   gnuplot> plot "<./fft -s 100 -d <100/tmp00" with lines
                             ^
                          skip initial peak

   The spectrum should be U-shaped, with narrow peaks tens of dB above
   the noise floor near the beginning and the end. Note that the noise
   floor is curved and not perfectly flat.

   From this, we pick level of the noise floor. The value should be at or
   slightly below the highest peaks of the noise between the large peaks
   at the end of the spectrum.

   This noise floor value is used to filter uninteresting samples later
   on, removing a constant bias from the results.

   In this example, we'll use a noise floor value of -50 dB.

6) Determine the "interesting" frequency range. For this, we consider all
   the spectra of the measurements:

   gnuplot> plot "<for n in 100/tmp*; do ./fft -s 100 -d <$n;echo;done" \
     with lines

   There should be a thick noise band in the middle, with pronounced
   narrow peaks at the edges. If there are one or two signals on top of
   the noise band, some measurements have been compromised and need to be
   removed or redone. We will do this in the next step.

   When zooming into the left peak, the "bins" which contribute to the
   peaks can be identified. The range should be chosen with some
   tolerance, since the frequency may shift a bit during the measurement
   process.

   By not considering bins far from the peak, less noise is included in
   the final result, complementing the filtering by noise threshold from
   step 4). Restricting the bins also eliminates the second peak at the
   end of the spectrum.

   In this example, we'll use a range from 0 to 20.

7) Obtain the peaks from all measurements

   gnuplot> plot "<for n in 100/tmp*; do ./fft -s 100 5 15 50 <$n;done" \
     with lines
                                                   ^  ^  ^ ^
                                                   |  |  | |
                                   skip, from step 4  |  | threshold, 5)
                                             lowest bin  highest bin

   This should yield a jagged more or less horizontal line with values
   differing by not more than 1-2 dB. If there are any large outliers,
   they have been contaminated and should be dropped.

8) The final result for one measurement run can be obtained as follows:

   for n in 100/tmp*; do ./fft -s 100 0 200 50 <$n;done | ./range -v 2

   In this example, "range" eliminates all outliers more than 2 dB from
   the average and reports this.

   The output are three numbers: the average (after eliminating
   outliers), the minimum, and the maximum.