Data Truncation

[Simon Bonner]

Hi all,

I've had a short discussion with John B off list regarding modeling signal strength and he (correctly) pointed out that I should move on-list (despite my fears that I'm about to ask some silly questions). 

The gist is that I am trying to model signal strength for some stationary tags that were monitored for a period of 8 weeks in the area of Old Cut last spring. I was surprised by how poorly the Friis function predicted the signal strength given that the displacement and relative orientation of the tag and antenna are known, and John has suggested that this may be due to signal truncation: "signal strength of detections is bounded below by around -80 dB relative to max". In our data the threshold seems to be near -73 db relative to max. Of the almost 2 million detections that occurred more 1% have signal strength less than -70, but none have signal strength less than -74.

I'd like to understand the process a little more. My immediate questions are: 1) is the threshold hard (all detections with signal strength lower than the threshold are discarded) or soft (detections are missed with some increasing probability as the signal strength decreases) and 2) how is the max value determined? 

I noticed in my data that there is a constant difference between sig and dbm of 50 suggesting that the max is set at -50 dbm. Is this correct?

Thanks in advance!

Simon

[john brzustowski]

Hi (again) Simon,

On Wed, Nov 30, 2016, at 12:08, Simon Bonner wrote:

> The gist is that I am trying to model signal strength for some
> stationary tags that were monitored for a period of 8 weeks in the
> area of Old Cut last spring. I was surprised by how poorly the Friis
> function predicted the signal strength given that the displacement
> and relative orientation of the tag and antenna are known, and John
> has suggested that this may be due to signal truncation: "signal
> strength of detections is bounded below by around -80 dB relative to
> max". In our data the threshold seems to be near -73 db relative to
> max. Of the almost 2 million detections that occurred more 1% have
> signal strength less than -70, but none have signal strength less
> than -74.
>
> I'd like to understand the process a little more. My immediate
> questions are: 1) is the threshold hard (all detections with signal
> strength lower than the threshold are discarded) or soft (detections
> are missed with some increasing probability as the signal strength
> decreases)

It's hard, at least nominally. An estimate of signal strength is made
for each putative pulse and compared to an estimate of "noise" signal
strength in a surrounding window, and if the difference is at least 6
dB (by default; this can be changed in the receiver's deployment.txt
file), the pulse is recorded. One issue is that presence of other
pulses in the "noise" window can lead to incorrect rejections.

Performance of the pulse detector has not been evaluated in any
systematic way, and needs revisiting. Source code is here:

https://github.com/jbrzusto/vamp-plugins/tree/2014/lotek

(the 2014 branch is what's installed on most SGs).

> and 2) how is the max value determined? I noticed in my data that
> there is a constant difference between sig and dbm of 50 suggesting
> that the max is set at -50 dbm. Is this correct?

The max comes from calibration of the funcubedongle against a good
quality radio source (USRP). This has also not received sufficient
attention; the -50 dBm value came from a quick test. I don't know
what kind of variability to expect across funcubedongles.

John B.