Re: [802.19] Explanation to page 13 figure(s)
Hi Junyi,
The neighborhood definition is one level regardless of distance between networks. This interference threshold value is noise level (No+expected NF) + interference margin (IM). In all methods and in all "true" cases the results is compared to this value: above it = neighbor, below it = no neighbor.
If I modify your alternative 1, then it is correct:
(1) Check all possible interfering links from any of 20 notes +1BS in one network, to any of 20 notes+1BS in another network, if any one of links generate interference larger than interference threshold value, then decide the two networks are neighbor, otherwise if none of them may generate interference larger than the interference threshold value, then the two networks are not neighbor
These kind of decisions are generated 5000 times at each separation distance. In a transition range the distribution is changing from most of the decisions "no neighbor" to most of the decisions "neighbor" when the distance is becoming smaller.
In statistical method the 90% confidence value is also compared to the interference threshold value: above it = neighbor, below it = no neighbor. 90% means that in 90% cases, when CDF has been generated, the level is below the 90% confidence value.
BR,
Jari
-----Original Message-----
From: ext Junyi Wang [mailto:junyi.wang@xxxxxxxxxx]
Sent: 1. syyskuuta 2011 4:11
To: Junell Jari (Nokia-NRC/Helsinki); STDS-802-19@xxxxxxxxxxxxxxxxx
Subject: RE: [802.19] Explanation to page 13 figure(s)
Hi Jari
Thanks so much for your clear explanations.
Would you please clarify how you decided whether two networks are neighbor or not if 20 nodes inside each of two networks. Basically I have two choices as follows for this question, if yours is different from any of them, please also provide your option
(1) Check all possible interfering links from any of 20 notes +1BS in one network, to any of 20 notes+1BS in another network, if any one of links generate interference larger than 90% confidence value, then decide the two networks are neighbor, otherwise if none of them may generate interference larger than 90% confidence value, then the two networks are not neighbor
(2) Sum up interferences from all notes in one networks, if the sum power is larger than the 90% confidence value, you decided two networks are neighbors. Otherwise
Would you please also clarify how did you get the 90% confidence value. Do you keep this 90% confidence value fixed for any different distances of two networks, if not, how could this 90% confidence value be too low at certain distance? As far as I know, this 90% confidence value is calculated from statistic estimation at one distance case, and is used for that distance case, how is possible to be too low.
B.R.
Junyi
-----Original Message-----
From: Jari Junell [mailto:jari.junell@xxxxxxxxx]
Sent: Wednesday, August 31, 2011 5:00 PM
To: STDS-802-19@xxxxxxxxxxxxxxxxx
Subject: [802.19] Explanation to page 13 figure(s)
Hi,
The question was that why there is an abrupt change in missed detection curve of statistical method at a certain network separation?
Explanation:
- At each distance a CDF is calculated and 90% confidence value is chosen to represent the highest interference level from one network to another one. This value increases monotonically when the distance between the networks decreases.
- There has been defined a threshold interference value above which two networks are neighbors and below which they are not neighbors.
- For statistical method (as well as for other two methods) there is an abrupt change in neighborhood: before a certain distance decision made by a method gives no neighbor result and after that always a neighbor result.
- This result at each distance is compared to 5000 "true" TVBD location cases. For each case a decision is made that are the networks neighbors or not. At long separation distances all cases give always a result "no neighbor" and at quite short distances all cases give always a result " neighbor" and between those decision distribution changes depending on separation in such a way that when coming closer the number of "no neighbor" cases is decreasing.
Now to figure in page 13:
- because there was only one randomly located TVBD in both networks in each case when generating CDF, a 90% confidence value is clearly too low at certain distances compared to all cases when there are 20 randomly located TVBDs in both networks and the shortest distance is looked for. That's why the missed detection rate is approaching 100%. Now at a certain distance there is an abrupt change in neighborhood decision of statistical method. That causes an abrupt change in missed detection rate. Without that the missed detection rate would have reached 100% and stayed there (no abrupt change in statistical method).
BR,
Jari