# [USRP-users] USRP captured data information

Raj Bhattacharjea rbhattacharjea at gmail.com
Sat Apr 18 16:19:20 EDT 2015

```With your mention of curve fitting and RSSI, I had a strong suspicion you
were doing some practical channel measurements and modeling. Here are some

What we call power in a signal is proportional to a time-average of the
square of the signal. See here:

https://en.wikipedia.org/wiki/Spectral_density#Power_spectral_density

I say proportional because to get actual real world units like Watts of RF
power at an arbitrary receiver antenna, there must be extra factors
included like pi and the intrinsic impedance of free-space, and you have to
and impedance.

To go from USRP measurements to something proportional to power, you can
form the complex baseband in Matlab using something like this: X = I+1j*Q,
assuming I and Q are the I and Q samples, respectively. Now abs(X).^2 must
be time-averaged over some window of time, which looks something like
conv(abs(X).^2,ones(1,N)/N,'valid'), where N is the number of samples over
which you want to time-average. That would give you a signal that is
proportional to the the average power in the original signal (over N-sample
windows). This is something akin to RSSI, and you could move around your
antennas and measure the power in various locations and do your curve
fitting and come up with path loss exponents and variances, etc.

You could reasonably also fix the unknown constant of proportionality in
the system through calibration: cable connect a TX to an RX, transmit your
waveform at a known level, measure the RX power, and record that as the
reference level. Then, divide all further measurements by that reference
value. Alternately, if you work in path loss, then the absolute power
levels drop out because you form the difference between transmitted and