[Tinyos-devel] Lqi and CC2420

[Andreas Koepke]

Philip Levis:
> 
> On Mar 3, 2008, at 1:00 AM, Andreas Koepke wrote:
> 
>>>>
>>>>  Pragmatically: something that delivers a comparable performance.
>>> Phil can elaborate more because he has looked at RSSI and LQI in
>>> detail but it is not clear if one can get comparable performance using
>>> RSSI vs LQI.
>>> - om_p
>>
>> Well, his published results afaik lack a QQ-Plot using the packet 
>> delivery ratio PDR of RSSI vs. LQI. There may exist a cost function 
>> such that PDR = cost(RSSI) = cost'(LQI). Theoretically, RSSI and LQI 
>> must be strongly correlated in the absence interferers (both in system 
>> and out of system like microwave ovens). So I believe that there is 
>> such a function, and I would not be surprised if it is linear at least 
>> for the relevant interval. Excluding interferers is the difficult part 
>> here...
> 
> I'm not sure why a QQ plot would help. The two functions are very 
> different: while they fact that they measure the same thing means they 
> are correlated, that does not mean they have similar probability 
> distributions, as the LQI/RSSI plots show. Much of the data from the 
> published results is online, though, so someone can do a QQ plot if they 
> want.

I wanted to use PDR as the prob. function, but given your next 
explanation it would indeed be of limited use.

> There is no accurate function f() such that f(RSSI) = LQI. Two reasons. 
> First, the quantization of RSSI readings. LQI readings are essentially 
> lower-order bits on the SNR curve. A one dB change in RSSI can lead to a 
> 40+ unit change in LQI. Second, LQI represents SINR, while RSSI is 
> independent of the Interference and Noise.

Phew -- quite sensitive. So the linear interval would be very small...

> Basically, the CC2420 chip correlation indicator provides fine-grained 
> information on the part of the SINR curve where the slope is 
> significant. This is why the values show much larger temporal variation: 
> a tiny shift in SINR causes a change in LQI. Also, LQI, due to its 
> measurement, is more greatly affected by statistical sampling. Even with 
> a very stable RSSI and SINR, you'll see LQI variations.

Thanks Phil.

> Phil
>