Modeling the Effect of Interferences among N Collocated Heterogeneous Wireless Networks

Transcription

1 Modeling the Eect o Intererences among N Collocated Heterogeneous Wireless Networs Bilal Khan and Jong-Su Ahn Department o Computer Science and Engineering, Donggu University, Seoul, Korea bilalhan83@hotmailcom and jahn@dguedu Abstract-- With the advent o wireless networs, the usage o mobile devices has been rapidly exploded due to their cable-ree convenience The ubiquity and dense population o mobile devices have led several heterogeneous wireless networs to be redundantly deployed as an underlying inrastructure in a given area, allowing mobile users to choose their preerred wireless networs These co-placed dierent networs, however, tend to interere with each other and thereore suer rom severe perormance degradation To estimate this deterioration, this paper proposes an analytic perormance model which precisely evaluates both the inter-networ intererence and intra-networ collision eect on a victim networ when involved networs run the same baco algorithm Dierently rom a legacy hidden station model, our model requires a set o Marov chains to separately abstract behaviors o both interered and interering networs Simulation validates that our proposed model predicts the intererence impairments more accurately than the hidden station model and even the conventional intererence model that ignores the baco behavior in wireless networs Keywords- intererences, IEEE 802, perormance model o wireless networs I INTRODUCTION As a tremendous number o dierent mobile devices have been emerged, various mobile wireless networs have actively built to provide seamless and ubiquitous connections regardless o where users travel around To maximize the perormance o wireless networs under dierent topological constraints, urthermore, a variety o communication protocols have been actively standardized IEEE 8026, IEEE 802, and IEEE protocols also nown as WiMAX, WiFi, and Zigbee, or instance, have been redundantly installed to support the Internet access service over the area o cities, buildings, and rooms respectively Even though these collocated networs give the reedom o selecting a networ or aster data delivery and lower cost, however, these networs tend to hamper each other s communications especially when they occupy the same bandwidth WLAN and WPAN sharing 24GHz ISM (Industrial, Scientiic and Medical) band, or example, would incur requent intererences namely inter-networ collisions, resulting in severely degraded perormance Siora et al, report that a pacet error rate in IEEE networs increases by more than 90% when they are installed near to IEEE 802 networs [] Pollin et al [2] also demonstrate that the perormance o IEEE 802 networs go down by up-to 60% when it coexists with IEEE networs This intererence is expected to be exacerbated in uture as FCC (Federal Communications Commission) has designated its newly devised IEEE 802y and 8026h [3][4] standards to share 36GHz band This severe perormance deterioration is mainly attributed to two actors such as the intrinsic laziness o BEB (Binary Exponential Baco) algorithm adopted by almost all IEEE 802 variants or collision avoidance and the lac o mechanisms to discriminate collision-driven ailure rom intererence-driven ailure At irst, BEB algorithm unwieldy spends the time to reach the appropriate contention window when heavy congestion lasts or a long period since it always starts rom the small contention window regardless o the current networ status It would experience some number o timeouts to settle to a baco timeout suitable or the number o currently contending stations Secondly, BEB algorithm blindly doubles its contention window due to the lac o any explicit eedbac on the outgoing transmission status Without the explicit eedbac, precisely it cannot dierentiate collisions signaling intranetwor intererence rom corruptions caused by internetwor intererence Under heavy intererences, it is a better way or victim networs to raise their signal strength rather than expand their contention window to prevent throughput rom steeply alling down Research has been actively conducted to accurately and separately measure either collision or intererence eect on perormance One typical research proposes a perormance model which aims at calculating perormance downall o 802h when it operates in the same cell as 8026y[5] The drawbac o this approach, however, is to assume that both WLAN and WMAN run the ixed contention window unlie the standards [3][4] which stipulate to employ BEB algorithm To inluence the behaviors o BEB algorithm on the eect o intererence, this paper proposes a perormance model which extends a legacy hidden station model [6] This extension is required since the inter-networ intererence is asymmetric while intra-collision by hidden stations is symmetric In other words, in inter-networ collision, superior networs can continue to send rames ignoring transmissions o nearby victim networs due to their asymmetric transmission power whereas hidden stations symmetrically /204 - Copyright is with IFIP 67

2 suer collision just lie covered stations since the transmission power o stations in a networ is assumed to be equivalent To appreciate this asymmetric aspect o N collocated heterogeneous wireless networs, the proposed model distinctly abstracts the behavior o each interacting networ to evaluate its interior interactions and at the same time the degree o intererence on other networs Note that the traditional hidden station model requires only one Marov chain regardless o the number o hidden stations As a result, or N collocated heterogeneous wireless networs our model requires N separate Marov chains to which the eect o intererence rom superior networs is added ns-2 simulations validate our proposed model with only less than 0% deviation that both the conventional model [5] and the hidden station model [6] underestimate by around 85% the eect o intererence in heterogeneous wireless networs on the achievable throughput o victim networs The remainder o this paper is organized as ollows Section II illustrates some related wor Section III presents our proposed model or N collocated networs Section IV describes the ns-2 module or dierentiating intererences rom collisions Section V explains the simulation results and compares them with the results obtained rom the analytic model Finally Section VI presents the conclusion and uture research issues II RELATED WORK As wireless mobile networs have become popular, huge literatures has been published about the impact o intererences among coexistent heterogeneous wireless networs To accurately evaluate the degree o intererences, various analytical models have been proposed One o them predicts the intererence eect on WLAN s throughput without explicitly assuming the presence o nearby wireless networs [7] It abstracts interactions o co-placed wireless networs by channel errors measured by bit-error-rate without analyzing behavioral details o the intervening networs It, urthermore, evaluates the perormance improvement by employing FEC (Forward Error Correction) codes as a way to overcome this intererence problem in WLAN In [8], authors introduce an analytic model or the perormance analysis o WLAN taing into account the NACK (Negative ACK) rame as a solution when transmissions ail due to channel errors Lie [7], the model in [8] is also unable to include the impact o intererence caused by the collocated networs A hybrid Marov chain model is introduced in [9] or the perormance evaluation o hybrid IEEE 802b and IEEE 802g networs Their targeted hybrid networs, however, are not a truly heterogeneous networ since transmission ailures in their hybrid networs are solely due to collisions whereas transmissions in heterogeneous networs are aborted by either collisions or intererences In addition to the lac o evaluating the eect o intererence in heterogeneous wireless networs, the applicability o their model is restricted to the two speciic types o networs Authors in [5] describe an advanced analytic model to address the issue o transmission ailure in collocated heterogeneous wireless networs In contrast to our model, however, their analytic model assumes the constant size o contention windows or all the participant heterogeneous networs or the convenience o analysis even though most real MAC protocols employ BEB Similar to [9], moreover, the analytic model in [5] alls short or the perormance evaluation o more than two heterogeneous networs III PERFORMANCE MODEL OF N COLLOCATED NETWORKS This section explains an analytic model or computing the throughput o a networ that shares the same requency band with N- other collocated networs under the assumption o all networs running the same BEB algorithm We believe that even though this assumption is not enough general to cover all types o wireless networs, it is still reasonable since the most common wireless networs such as Wii and Zigbee belong to this category For our model, these N networs are indexed in an increasing order o transmission power lie N, N 2,,N N Namely networ N radiates weaer signals than N- networs ranging rom N + to N N whereas it dominates the other remaining networs N, N 2,, N - that rerain themselves rom transmitting their rames whenever N occupies the channel Note that the subscript N and correspond to any positive integer and an integer lying between and N respectively Figure shows three Marov chains belonging to the strongest networ N N, to an intermediate networ N, and to the weaest networ N, respectively Marov chain in Figure (a), at irst, depicts a model or networ N N in which transmission is not aected by any other networ, thereore, the success and ailure o a transmission is only determined by the probability o collision, denoted as p c lie the one presented in [0] N- dotted boxes in Figure (a) represent transmissions o N- weaer networs that are vulnerable to the networ N N Transmissions o these N- weaer networs are interered when the transmission o networ N N alls inside one o the states in its corresponding box In other words, in a vulnerable period the transmission time o networ N N is overlapped with that o a weaer networ 68

3 τn-,n τ-, τn-2,n τ-2, τi,n τi, /W0 0,0 0, 0,2 pc/w i-,0 pc/wi i,0 i, i,2 pc/wi+ pc/wm m,0 m, m,2 m,wm-2 NN- τ,n 0,W0-2 0,W0 - i,wi-2 i,wi - m,wm- (-pe) (-pe) (-pe) (-pe) 0,0 0, 0,2 0,W0-2 0,W0 - i-,0 i,0 (-(-pe))/wi i,wi-2 i, i,2 i,wi - (-(-pe))/w (-(-pe))/wi+ (-(-pe))/w0 (-(-pe))/wm m,0 m, m,2 m,wm-2 m,wm- N- N-2 τ, (-pe) (-pe) (-pe) (-pe) 0,0 0, 0,2 0,W0-2 0,W0 - i-,0 i,0 (-(-pe))/wi i, i,2 i,wi-2 i,wi - (-(-pe))/w (-(-pe))/wi+ (-(-pe))/w0 m,0 m, m,2 m,wm-2 (-(-pe))/wm m,wm- NN-2 Ni Ni N N (a) V i (b) V i (c) Figure 2-Dimensional Marov Chains, (a) or N-th Networ N N (b) or -th Networ N (c) or -st Networ N Two symbols, τ i,n and V i around the dotted box N i in Figure (a) represent the probability with which networ N i is interered by networ N N and the vulnerable interval o networ N i In τ i,n, irst and second subscript represent the interered or victim networ and the interering or superior networ, respectively Vulnerable interval o networ N i is set to the transmission delay o one rame o networ N i which is assumed to be ixed in our model The vulnerable interval excludes transmission time o ACK since ACK rame is assumed to be not interered due to its comparatively higher transmission power, ie, 245dBm [3] In contrast, Figure (b) describes the Marov chain or networ N which is either superior or inerior to other collocated networs in terms o transmission power Lie Figure (a), - dotted boxes in Figure (b) represent vulnerable periods o weaer networs ranging rom N to N - The transmission o a weaer networ, or example N i, is interered with probability τ i, that networ N stays inside one o dotted boxes labeled N i when networ N i has already initiated its transmission Finally, Figure (c) shows the Marov chain or the weaest networ N Due to its low transmission power networ N does not pose any threat to the collocated networs A Collision and Intererence Probability o Networ N This subsection calculates p c and p e in Figure (b) representing the collision and intererence probability o networ N, respectively, as a unction o the number o collocated networs and number o stations in each networ Once p c and p e are calculated, () solves the probability o ailure p that is the complement o the probability o success p s Note that, in addition to collisions, intererences also contribute to the probability o ailure p, unlie the hidden station model [6] which ignores the eect o intererence p p ( p )( p ) () s e c p e is described in (2) where n j and τ,j represent the number o stations in networ N j and the probability o intererence caused by networ N j, respectively In other words, τ,j is the probability o transmission by a station in networ N j that is superior to networ N while the latter has already initiated a transmission The second term in (2) describes the probability that any networ superior to networ N does not initiate transmission when the latter has already occupied the channel These superior networs range rom N + to N N N j (2) p = - (-, j ) n j e j p c is described in (3) where n and ζ represent the number o stations in networ N and the probability o transmission o a station o networ N, respectively The second term in (3) describes the probability that n - stations in networ N do not transmit any rame, when one station o networ N has occupied the channel n - p c = - ( - ζ ) (3) Ater p is ound, b 0,0 is calculated rom (4) Here m, m and W speciy the maximum number o allowable retransmission, the maximum number o contention window s bacos, and the minimum contention window size For more details, please reer to [][2] b = 0,0 2(- p ) (-2p ) when mm' m m W (- p )(-(2p ) )+ (- p )(-2p ) 2 (- p ) (-2p ) m' m m' m ' mm ' W (- p )(-(2p ) )+ (- p )(-2p )+W 2 p (- p )(-2p ) when m>m' (4) 69

4 m V = b, j s, c s0 c0 X m+ p p - X X X m W - (2p ) - p p - p V 2 2 (V +) b 0,0 + +(V +) - when 0 V Wm' and 0 X m' = 2-2p p 2 - p - p 2W - 2 when V >Wm' (6) B Transmission Probability o Networ N and Intererence Probability o Its Superior Networs The probability o transmission ζ o a station in networ N is the sum o the state probabilities in the irst column in Figure (b) as shown in (5) Here, s the irst subscript in b s,0 indicates the number o bacos or the level o row in Figure (b) - p m m = b s,0 = b (5) 0,0 s0 - p (6) determines the probability τ,j with which networ N is interered by superior networ N j In other words τ,j is the sum o all the state probabilities o states contained within the box labeled with N within the Marov chain o networ N j where j ranges rom + to N Note that c the second subscript in b s,c, in (6) is the number o time slots in the vulnerable interval ranging rom 0 to V X is the minimum baco stage or which the contention window o superior networ N j is greater than the vulnerable interval V o networ N For example, i W <V W 2, then use X=2 in (6) C Throughput o Networ N Throughput TH o networ N is calculated in (7) where L is the payload size and p I, p S, p are the probabilities that there is no transmission in the considered raction o time, the probability o successul transmission o networ N and the probability, which is computed in (), that the transmission o networ N is unsuccessul due to either collision or intererence pl TH = s where p = - p - p pst s+ p T +piti I s Moreover T S, T and T I in (7) account or the time intervals or the channel being busy due to successul transmission, the time spent in unsuccessul transmission due to either collision or intererence and the time when the channel was idle, respectively (7) IV A MODULE FOR INTERFERENCES IN ns-2 To dierentiate inter-networ collision rom intra-networ collision, we add some number o steps as shown in Figure 2 since the current version o ns-2 [3] can only simulate the interaction o wireless networs o the same type According to the module we added when another rame denoted as the 2 nd rame arrives while a rame named as the st rame is being received, the receiver checs which part o the st rame is being received A rame is successully received when a receiver receives both PLCP and MPDU The execution o dierent steps can be summarized as below: An inter-networ collision is declared when a 2 nd rame arrives while MPDU o the irst rame is being received o the st rame This is due to the act PLCP o the st rame has been already received beore the arrival o the 2 nd rame When a 2 nd rame arrives with receiving power ratio, labeled as Power (PLCP 2 )/Power (PLCP ), higher than a certain threshold denoted as CP TH while PLCP o the st rame is being received [4] In this case, the st rame is dropped due to intererence and intererence is declared In contrast, an intra-networ collision is declared when a 2 nd rame arrives during the PLCP reception o the st rame and ails the PLCP reception power ratio chec In this case receiver ails to successully receive PLCP header and MPDU [5] and thereore both rames are discarded 2 nd Frame Arrives st Frame PLCP is being received No Yes Power(PLCP 2)/ Power(PLCP ) >CP TH Yes () Declare Intererence (2) Discard st Frame (3) Start Receiving 2 nd Frame No () Declare Collision (2) Discard both Frames END Figure 2 A Module or Telling Collision rom Intererence in ns-2 TABLE PARAMETERS USED IN SIMULATION AND MATHEMATICAL ANALYSIS Parameters Values Parameters Values Data Rate Mbps DIFS 50 µs Control Rate 2 Mbps SIFS 0 µs PHY Header 20 bits Slot Time (σ) 20 µs MAC Header 272 bits CW min 3 Transmission 0 dbm (WLAN) CW max 023 Power 30 dbm (WMAN) Retransmission Limit (m) 5 ACK Frame 2 bits Maximum Bacos (m ) 5 70

5 Throughput(Mbps) Throughput(Mbps) Throughput(Mbps) V SIMULATION STUDY This section perorms simulation using ns-2 to evaluate the eect o intererence when the arbitrary number o networs N is set to 2, equivalent to a situation where one WLAN and one WMAN competes or the same wireless channel Table lists the values assigned to PHY/MAC parameters o 802b that is assumed to be employed by both WLAN and WMAN It is worth mentioning that Par et al, in [5] also assumes that WLAN and WMAN run 802a or the convenience o analysis Furthermore, we use 400 bytes o payload size or both WLAN and WMAN or all experiments A Evaluating the Accuracy o the Proposed Model Figure 3 illustrates WLAN throughput versus the number o WLAN nodes when there is no WMAN station present in the vicinity At irst, Figure 3 proves that the throughput obtained using analytic model is dierent only by 5~0% rom that obtained using simulations or various number o WLAN stations Secondly when collocated with only one WMAN station WLAN suers severe reduction in its perormance as shown in Figure 3 This reduction in WLAN throughput is over 99% when there is less number o WLAN stations, eg, 2~0, but decreases when the number o stations in WLAN increases The slight improvement in WLAN perormance when there are many WLAN stations is due to the availability o a station to transmit in the idle time between the two consecutive WMAN transmissions No WMAN Station: Model One WMAN Station: Model No WMAN Station: Simulation One WMAN Station: Simulation Number o WLAN Stations Figure 3 WLAN Throughput in Absence and Presence o WMAN B Comparing the Proposed Model with the Conventional Model This sub-section compares our model with the conventional model proposed in [5] as shown in Figure 4 One WMAN station is present in the vicinity o WLAN with the number o stations varying Note that our proposed model assumes that WLAN and WMAN run BEB algorithm with the initial size o CW set to 3, whereas the model o Par et al, [5] sets CW constant 3 or both WLAN and WMAN Figure 4 veriies the validity o our proposed model ater the results obtained rom it is matched with those obtained rom the simulation We see in Figure 4 that conventional model overestimates WLAN throughput in the presence o one WMAN station The gap between the throughputs obtained rom the two models starts decreasing ater the number o WLAN stations reaches 5 Unlie the IEEE standards [3][4] that state the use o binary exponential baco or the contention resolution, [5] uses constant baco contention window in its proposed model, thereby deviating rom the results obtained rom the simulation which uses binary exponential baco Finally, [5] determines throughput using probabilistic approximations o the transmission attempts o WLAN stations and WMAN stations unlie our proposed approach that is based upon calculating the collision and intererence probability One WMAN Station: Conventional Model One WMAN Station: Proposed Model One WMAN Station: Simulation Number o WLAN Stations Figure 4 Comparison o Throughput between Proposed and Conventional Model C Dierentiating the Proposed Model rom the Hidden Station Model We argued earlier that the presence o one or more stations o heterogeneous type aects the perormance o WLAN dierently than the presence o a hidden station and thereore their aects shall be analyzed using separate models Figure 5 shows the comparison between the perormances o WLAN in the presence o a hidden station [6] to the perormance o WLAN when it is collocated with a WMAN station There are two points worth noticing First, WMAN severely aects the perormance o WLAN as compared to the eect o a hidden station Secondly, the model or the perormance evaluation o the eect o the hidden station cannot be applied to the situation when WLAN is aected by the presence o WMAN, thereby justiying the extension we made to the hidden station model One Hidden Station: Hidden Station Model One Hidden Station: Simulation One WMAN Station: Proposed Model One WMAN Station: Simulation Number o WLAN Stations Figure 5 Throughput Dierence between Proposed Model and Hidden Station Model 7

6 WLAN Throughput(Mbps) D Analyzing the Eect o Contention Window o WMAN on the Perormance o WLAN The purpose o this experiment is to observe the ability o our proposed model in assessing the eect o tuning the initial size o CW o WMAN on the perormance o WLAN In addition, this experiment gives us a hint on tuning one o several parameters to improve the perormance o WLAN when it is collocated with a networ having higher transmission power In this experiment we set the total number o WLAN stations 0 with the number o WMAN stations varying rom one to two Figure 6 shows the perormance o WLAN when the size o its initial CW is 3 while that o WMAN is varied rom 3 to 023 We see that WLAN throughput is very low when the size o WMAN initial CW is small even when there is only one WMAN station It shows WMAN with small CW gives less transmission opportunity to WLAN As WMAN sets the size o initial CW to 63 or urther increases its CW, we observe improvement in the perormance o WLAN even when there are two WMAN Stations The throughput o WLAN reaches maximum when the initial size o the CW o WMAN is set to maximum, ie, 023 In general, the higher the size o WMAN initial contention window, the better the WLAN perorms WMAN Stations: Model 2 WMAN Stations: Model WMAN Stations: Simulation 2 WMAN Stations: Simulation WMAN Initial CW Figure 6 WLAN Throughput as a Function o Contention Window o WMAN VI CONCLUSIONS AND FUTURE RESEARCH Inherent drawbacs o the BEB algorithm coupled with the asymmetric nature o heterogeneous wireless networs are a main cause that degrades the perormance o WLAN This study proposes a perormance model that consists o a persystem Marov chain to analyze the eect o superior networs on the collocated victim networs by taing into account both the intra-networ and inter-networ collisions Simulation results veriy that the proposed model can evaluate the intererence eect accurately than a conventional model For our uture research, we will continue to expand our model to evaluate various variables aecting the degree o intererences such as transmission power, rames size and contention window etc We will also analyze the aect o already proposed solutions such as NACK with MAC header CRC (Cyclic Redundancy Chec) to estimate the perormance improvement o the victim networ ACKNOWLEDGMENT This research is supported by Basic Science Research Program through the National Research Foundation o Korea (NRF), unded by the Ministry o Education, Science and Technology [Grant No NRF202-RAA ] REFERENCES [] A Siora and VF Groza, Coexistence o IEEE with other systems in the 24GHz ISM band, Proceedings o the IEEE Instrumentation and Measurement Technology Conerence, IMTC 2005, pp [2] S Pollin, I Tan, B Hodge, C Chun and A Bahai, Harmul coexistence, between and 802: a measurement based study, Proceedings o the 3 rd International Conerence on Cognitive Radio Oriented Wireless Networs and Communications, CrownCom 2008 [3] IEEE Std 802y-2008(2008) IEEE Standard or Local and Metropolitan Area Networs Speciic Requirements, Par : Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Speciications Amendment 3: MHz Operation in USA IEEE, New Yor [4] IEEE Std 8026h (200) IEEE Standard or Local and Metropolitan Area Networs Part 6: Air Interace or Fixed Broadband Wireless Access Systems Amendment 2: Improved Coexistence Mechanisms or License-Exempt Operation IEEE, New Yor [5] E -C Par and M Rim, Fair Coexistence MAC Protocol or Contention-Based Heterogeneous Networs, The Computer Journal, 20 [6] F Y Hung, and I Marsic, Perormance Analysis o the IEEE 802 DCF in Presence o Hidden Stations, The International Journal o Computer and Telecommunications Networing, vol 7, no 4, pp , October 200 [7] J S Ahn, J H Yoon and K W Lee, Perromance and Energy Consumption Analysis o 802 with FEC Codes over Wireless Sensor Networs, Journal o Communications and Networs, vol 9, no 3, September 2007 [8] Q Pang, C M L Victor and S C Liew, Improvement o WLAN Contention Resolution by Loss Dierentiation, IEEE Transactions on wireless communications, Vol 5, No 2, December 2006 [9] S C Wang, Y M Chen, T H Lee and A Helmy, Perormance evaluations or hybrid IEEE 802b and 802g wireless networs, IEEE International Perormance, Computing, and Communications, IPCCC, 24 th April 2005, pp -8 [0] G Bianchi, Perormance Analysis o 802 Distributed Coordination Function, IEEE Journal on Selected Areas in Communication vol 8, no 3, March 2000 [] W Haitao, Y Peng, K Long, S Cheng and J Ma, Perormance o Reliable Transport Protocol over IEEE 802 Wireless LAN: Analysis and Enhancement, IEEE INFOCOM 2002, PP [2] B Khan, and J S Ahn, A Perormance Model or the Eect o Intererences among the Collocated Heterogeneous Wireless Networs, ICACT South Korea, 30~3 January 202, pp [3] The networ simulator ns-2, [4] A Kochut, A Vasan, A U Shanar and A Agrawala, Sniing out the correct physical layer capture model in 802b, Proceedings o the 2th IEEE International Conerence on Networ Protocols, ICNP 2004, pp , 5-8 October 2004, Berlin, Germany 72