1. Kunxiao Zhou and Xiaohua Jia City University of Hong Kong
Channel Assignment for WLAN by Considering Overlapping Channels in SINR Model
Kunxiao Zhou and Xiaohua Jia
City University of Hong Kong

2. Outlines Introduction Related work Problem formulation
Proposed solution
Simulation results
Conclusion 2

3. Introduction Channel assignment
Wireless Local Area Networks (WLANs)
Access points (APs), clients
Basic service set (BSS)
Assign channels to BSSs
Partially overlapping channels (802.11b)
11 partially overlapping channels
3 non-overlapping (orthogonal) channels
SINR model
Accumulative nature of interference
Most of combinatorial optimization techniques based on protocol model are not applicable 3

4. Related work
A. Mishra, V. Shrivastava, S. Banerjee, and W. Arbaugh, “Partially overlapped channels not considered harmful,” SIGMetrics/Performance, 2006, pp. 63–74.
Interference of overlapping channels
Communication btwn two nodes by using overlapping channels
Y. Cui, W. Li, and X. Cheng, “Partially overlapping channel assignment based on “node orthogonality” for wireless networks,” Mini-Conference at IEEE INFOCOM, 2011.
interference model: two nodes interfering or not depends on channel distance and physical distance bwtn two nodes (i.e. “node orthogonality”)
A. Mishra, S. Banerjee, and W. Arbaugh, “Weighted coloring based channel assignment for wlans”, in ACM Sigmobile MC2R, 2005
ADJ-minmax and ADJ-sum
The first paper proposed a partially overlapping channel model and employed the model in two scenarios, namely wilreless LANs and multi-hop wireless mesh networks.
2. Cui et al. in [2] derive a novel interference model that considers both the adjacent channel separation and the physical distance of the two nodes. By defining “ node orthogonality” model, they propose an approximate algorithm MICA to minimize the total interference for throughput maximization.
3. ADJ-minmax: assign a (adjacent) channel to a node such that the maximal interference of nodes is minimized; similarly ADJ-sum… 4

5. Problem Formulation
Infrastructure-based IEEE WLAN, 5

6. Problem Formulation (cont’d)
W= {w1,…,wm}: set of APs; U= {u1,…, un}: set of clients;
F= {f1, f2, … , fk}: set of overlapping channels;
BSS: AP and its associated clients
G(U ∪ W, E) , (u, w) ∈ E: client u connects to AP w;
γ(v,u): overlapping degree btwn channels of node v and u.
Channel overlapping degree: 6

7. Problem Formulation (cont’d)
Data rate:
The throughput of a BSS :
Throughput of the system: 7

8. Problem Formulation (cont’d)
Total interference of u received from all APs:
Total interference of all clients in the system:
Maximize the system throughput is equivalent to minimizing system total interference 8

9. Algorithm Design Interference analysis:
Interference caused by AP v to client u (u S(v))
Interference by AP v to all clients in a BSS Bw,
We only consider downlink case,
Since in WLAN environment, the system is clustered into a set of BSSs. Each BSS consists of one AP and the clients served by this AP. In each BSS, only AP works all the time while all the clients work with the AP in a time sharing fashion. For a client u, the interference from a BSS is actually caused by the AP in that BSS (note that we consider only downlink traffic where AP is the transmitter). 9

10. Algorithm Design (cont’d)
Fig. 2, weighted interference graph GI(VI,EI)
1. Now, we introduce a directed weighted interference graph GI(VI,EI) , where VI is the set of BSSs VI={Bw1, Bw2,… Bwm} and EI is the set of edges representing interference between all BSSs. For an edge e = (Bwi , Bwj ), the weight of e is i(Bwi ;Bwj ). Graph GI (VI ;EI ) is a complete graph. Fig. 2 is the corresponding weighted interference graph of Fig. 1, where each BSS in Fig. 1 is represented by a vertex in Fig. 2 10

11. Algorithm Design (cont’d)
Total weights of all edges in GI(VI,EI) ,
The total weight of all edges in GI is equal to the total interference of the system.
Our problem is converted to minimizing the total weight of all edges in GI(VI,EI) 11

12. Algorithm Design (cont’d)
Heuristic method
Select an initial channel for each BSS;
Choose the BSS has largest total weights of outgoing edges in GI(VI,EI)
Assign this BSS with the channel produces the small total edge weight
Repeatedly executing Algorithm 1 to improve the quality of channel assignment 12

13. Simulation Results (throughput)
1. In all these experiments except the last one, we compared the performance of five channel assignment algorithms.
2. Besides the current best-known methods ADJ-minmax and ADJ-sum, ”One-channel” means that all BSSs in the network
use the same channel. ”Greedy-all” and ”Greedy-orthogonal” denote our greedy method considering all and only orthogonal
(i.e. channel 1, 6 and 11) channels, respectively.
3. This subsection compares total network throughput with different methods differing in number of clients. First, from Fig. 3, we observe that the performance achieved by as-
signing overlapping channels is expected to be better (i.e. ”Greedy-all” is expected to achieve better performance than ”Greedy-orthogonal” and ”One-channel”). It means that by
using partially overlapping channels, we reduce the total interference, improve spatial channel re-use and thus get better performance.
4. ”Greedy-all” also outperforms ”ADJ-minmax” and ”ADJ-sum”. This is attributed to the more realistic interference model, which helps to obtain a more appropriate channel
assignment. By using SINR model, our greedy method directly minimize the total interference of the system which aims to improve each user’s data rate and achieve better performance. The other two methods do not mention number of clients assigned to an AP and physical distance between user and other APs, resulting a rough estimating of system interference. 13

14. Simulation Results (service ratio)
Service ratio: percentage of clients who are active. 14

15. Simulation Results (per-user throughput)
1. This subsection gives the per-user throughput comparison between the five methods with different numbers of clients. We only calculate the active users, i.e., users whose received
SINR are greater than the threshold.
2. From Fig. 5, we can get the following observations. Per-user throughput decreases with the increase of number of clients. With the increase of number of clients, the
clients that each AP serves become more. Since b/g uses packet-fairness for channel access, thus each user’s throughput will be reduced.
3. There are slight differences among the four methods except ”One-channel”. This phenomenon can be explained as follows. Since in our method, service ratio is greater than other
four methods. Our method will enable more active users with low data rates in the system, this will degrade the per-user throughput. 15

16. Simulation Results (channel utilization)
In this subsection, we show the utilization of each channel in Fig. 6. We aim to evaluate how channel overlapping degree affects the channel utilization.
In this experiment, ”Attenuation” means that we use Table. I as overlapping attenuation. ”Atte-linear” uses f1; 0:8; 0:6; 0:4:0:2; 0g as the channel overlapping
attenuation. ”Atte-square” and ”Atte-SRoot” denote the channel overlapping attenuation are corresponding to the square and square root of 1, 0.8, 0.6, 0.4, 0.2, and 0, respectively.
3. It can be seen that when we use the channel overlapping attenuation from [10], channel 1 and 11 are used more frequently. Because channels 1 and 11 are two borders of channel
sequence and they just cause interference in one direction, in our greedy method, these two channels are used mostly to reduce the total interference of the system.
4. Another observation we make is that when the channel overlapping degree is linearly decreased as the channel distance, only orthogonal channels are employed. In this scenario, we can not get a favor of overlapping channels.
5. Theorem .1: Supposing one BSS has 6 neighbors whose channels are f1; f2; f3; f4; f5 and f6 separately. The orthogonal channel distance is 5. Channel overlapping attenuation is
linear. Then only orthogonal channel (i.e. either channel f1 or f6) will be selected in our greedy method. 16

17. Conclusions
Studied channel assignment by considering partially overlapping channels in SINR model;
Converted the maximal throughput problem to minimizing total edge weight of the interference graph;
Proposed a heuristic method;
Channel overlapping attenuation coefficient is a major factor to affect the channel utilization. 17

18. Thanks!
Q & A 18