Location Engine CC2431

 Hi JP,

RSSI Location is a nice technology. However, since it is amplitude based and not phase based like GPS it will never be that accurate. To be succesfully with this technology you will need to invest most of your time on network deployment, characterisation and infrastructure to really understand the behaviour. First of all it is important to understand RSSI, what affects it value and how it is influenced by its RF environment. Please see some science and IEEE papers that are closely related to the topic that can be a great resource for you to understand the limitations and capabilities.

 
1. "Preformance of Ad-Hoc Wireless Network on 2.4Ghz Band in Real fields”
http://ieeexplore.ieee.org/iel5/30/4470011/04470027.pdf?isnumber=4470011&arnumber=4470027

2. “Localization in ZigBee-based Sensor Networks
http://www-md.e-technik.uni-rostock.de/veroeff/2007_10_05_-_WISP2007_WeightedCentroidLocalizationInZigBee.pdf

3. "Packet error Rate Analysis of ZigBee Under WLAN and Bluetooth interference”
http://ieeexplore.ieee.org/iel5/7693/4290009/04290024.pdf 

4. “RSSI is under appreciated”
http://www.eecs.harvard.edu/emnets/papers/levisEmnets06.pdf

Enjoy,  

LPWRocks  

So, for TI, the technology is dead ?

I am wondering something. If you have a slide rule, say in millimeters, you could only mesure (and make calculations) based on these resolution. But if I ask you to draw 2 lines spaced of 1.5 millimeter, you can do it (if you don't shake) and use sub-millimeter precision during the slidind of the intern bar. RF location is like you have fuzzy resolution, based on staticaly probable resolution. So I should be able to improve the precision if I am able to find technics to reduce the uncertainity, even if it is done after measurements!

e.g. : I used to work with ADC sampling precise analog signals. I often choose simple 12b ADC and with oversampling and accumulate-and-dump technics, I am a software-improved ADC of 16b (or more). So software could do miracles sometimes, and certainly helps when hardware is not enough.

Hardware is never perfect, never as fast as one wants, never as precise, never .... but technologies improve with time, precision are finer, speed explodes, ... so I just wants to know if TI will improve it soon (my current prospective project), or if not, that would means something like the death of the location estimation on 802.15.4. Anyway, UWB will be more accurate.

And last remark : RSSI is not perfect compared to other technologies, ok ! But it is really pratical, and as useful as other, and NO technology to date is better than RSS/RSSI without big practical drawbacks (i.e. : AOA (Ubisense, with TDOA), TDOA (GSM), ultrasound (MIT's cricket), ...).

Hi JP,

There is definately an opportunity to reduce uncertainty by algorighms on top of the basic trigonometry calculations done in the location engine. Our software does this to a certain extent, and I am sure that you could improve on that in many applications. Given a moving blind node, for example, taking several RSSI measurements over an appropriate period of time could enable you to compensate for multi-path fading.

Another option is to build more "environment awareness" into the trigonometry calculations. Especially for indoor locationing, the possibility of including the attenuation and reflections of walls into account will improve performance. This would require you to do the trig calculations on a central server rather than in hardware on the chip, but in many applications this is where the location information ends up anyway. 

Peder

Thanks so much ...

The first solution, averaging, has been done. But the second solution (mapping) is something I would add but I still interrested to keep the hardware LE as a faster estimator. Perhaps I could use both information : the distributed estimation and on top this fast clue a centralised, more high level refinements or correction made with RSSI of several signals.

What I would know is if TI would help. They seem to abandon both the LE and the CC243x family.

 Another option for a fast estimator in a centralized solution could be the Least Square Fitting algorithm in the Gnu Scientific Library (GSL). 

Maybe use Kalman filter with CC2431.

http://www.medialab.ch/archiv/pdf_studien_diplomarbeiten/da06/da2006-113_SensorNetworkSoccer.pdf 
http://bwrc.eecs.berkeley.edu/publications/2006/Publications/Theses/tufan_phd_thesis.pdf

Hobbyist

What I would need is to know the distribution at the output of the Location Engine, I assume that a Gaussian distribution is the best qualifying the precision on the input.

But since I could not have information from the DEV (Motorola R&D or TI), I am already trying to implement the solution described above ("accumulate and dump"), and if I have the time (as I know myself, it is a dream), I will try implementing a Kalman & Particle Filter (already planned that Hobbyist !! But that is a lot of work !). I regret only that nobody seems working on improving that hardware-supported feature. I found nobody trying to get 3D data, integrate this location data into fusion engine ...

And from one of the above paper, a conclusion was that average RSSI is better than LQI, so I guess from the fact that so few person write about the subject that noone is interested with the applications.

But, from experience, no other technologies are really better than RF RSS signal. The best one (resolution) could be the ultrasound Criket-style ranger, but it use a huge infrastructure (base nodes and mobile ones). That is not very handy.

So why not the CC2431 LE ?