Vaisala RVP900

Chapter 5 _______________________________________________________ Plot-Assisted Setups
VAISALA______________________________________________________________________ 163
Begin with the FIR length that was chosen previously in the Pb command,
and use the N and W keys to set an initial bandwidth equal to the reciprocal
of the pulse width. The main lobes of the two plots should more-or-less
overlap. Experiment with changing the FIR length and bandwidth to
achieve a filter with the following properties.
- The filter width should be no greater than the burst spectral width. A
wider passband will reduce the SNR of the received signal because
out-of-band noise would be allowed to pass.
- The DC gain should be as small as possible, preferably less than
-65 dB (see discussion below).
- If there are conspicuous interference spikes at particular frequencies,
try to adjust the location of the filter’s zeros so that the interference is
maximally attenuated.
The filter should not pass any frequencies that do not actually contain
useful information from the original transmitted pulse. If anything, choose
a filter whose width is slightly narrower than the bursts spectral width.
Figure 28 on page 160 shows an example of a filter that is poorly matched
to the pulse. Although the filter has fairly good DC rejection, it passes
frequencies that are outside of the transmitter's broadcast range. These
frequencies contribute nothing but noise to the synthesized "I" and "Q"
data stream.
There are two procedures for optimizing the performance of the FIR filter:
- Manual Method—The process of arriving at a nearly optimal filter
requires a few minutes of hunting with the I, W, and N keys. Every
time you press any of these keys the RVP900 designs a new FIR filter
from scratch, and displays the results. Fortunately, the DSP chips are
fast enough that this can be done quickly and interactively. Even
though the user must still control two degrees of freedom (length and
bandwidth), the RVP900 internal design calculations are actually
performing several hundred iterative steps each time, which
preferentially select for the best filter. Because the FIR coefficients
are quantized in the filter chips themselves, the process of finding an
optimal filter becomes quite nonlinear.
- Automatic Method—Type the $ command and let the RVP900 do all
of the work (See description in Section 5.4.2 Available Subcommands
Within Ps on page 151).
The offset error of the IFDR A/D converter is at most 10mV, that is,
-27 dBm into its 50Ω input. If we wish to achieve 90 dB of dynamic range
below the converters +8dBm saturation level, then we expect usable "I"
and "Q" values to be obtainable from a (sub-LSB) input signal at -82 dBm.
This is 55 dB below the interference that would result from the worst-case
A/D offset. But a weak input signal at -82 dBm would still be damaged by