Qinglin Zhao＊, Fangxin Xu, Shangguang Wang

1 The Faculty of Information Technology, Macau University of Science and Technology, Avenida Wei Long, Taipa, Macau, China

2 State Key Laboratory of Networking and Switching Technology, Beijing University of Posts and Telecommunications, Beijing, China* The corresponding author, email： zqlict@hotmail.com

I. INTRODUCTION

Today, wireless portable devices such as smartphones and notebooks are ubiquitous.While people leverage these devices to access Internet and collectively share data [1], the simultaneous data transmissions of these devices often lead to serious physical-layer interferences. These interferences will signi ficantly lower system performance. Under these interferences, how to coordinate data transmission has received a great deal of attention [2][3].

On the other hand, MAC-layer protocols often also lead to poor system performance due to serious collisions between data transmissions. IEEE 802.11 MAC protocol, carrier sense multiple access with collision avoidance(CSMA/CA) [1], has generally been built in these devices. In 802.11 networks, a sender infers a collision occurrence from the absence of an ACK feedback, only after it completes the entire frame transmission. To detect the collision early, CSMA/CN (collision notification)[5], which is compatible with conventional 802.11 protocols, has recently been proposed and has attracted a great deal of attention such as [6][7].

In CSMA/CN, when a receiver infers that the receiving frame gets corrupted, it will send its signature (i.e., a unique pseudo-noise sequence) as a CN to the sender. On the other hand, the sender, while transmitting the frame,constantly executes correlation (between the receiver’s signature and the arriving signal) to detect whether a CN arrives. Once detecting an arrival of the CN (i.e., when the correlation result is larger than a threshold), the sender aborts the ongoing transmission immediately,thereby avoiding continuing to transmit the erroneous frame.

In this paper, we propose a novel design called CSMA/CN+to enhance CSMA/CN in low SINR. The enhancement is at the cost of slightly increasing the software-implementation complexity.

In CSMA/CN, the sender, depending on the detection result, will determine whether to continue or abort an ongoing transmission at any time. Consequently, the CN detection performance (namely, the successful detection probability and the false-alarm probability of the CN) has a significant impact on the efficiency of CSMA/CN. By “false alarm”,we mean that the receiver does not send back a CN but the sender detects it mistakenly. In[8], we point out： the false-alarm probability of the CN is a dominating factor that influences the system performance significantly,and therefore we should keep this probability below a threshold in order to achieve high performance. The false-alarm probability defalse-alarm probability is a decreasing function of . To keep the false-alarm probability below a threshold, we should keep unchanged.Therefore, in high SINR, we should reduce L, while in low SINR, we should increase L.Assume that in high SINR, a signature with fixed length (e.g., 20 bytes suggested in [8])may fulfill the desired false-alarm requirement. However, in low SINR, the false-alarm requirement might be violated signi ficantly (as illustrated in Fig. 2 in [8]), leading to disastrous effects (for example, it might force senders to abort almost all ongoing transmissions).This motivates us to reduce the false-alarm in low SINR, while maintaining as high performance as CSMA/CN in high SINR.

In this paper, we propose CSMA/CN+ to overcome this drawback of CSMA/CN. In CSMA/CN+, we introduce an additional signature. The receiver, adapting to channel conditions and self-signal suppression capability,determines whether to send back zero, one,or two signatures (which are called a CN+noti fication) to the sender. In high SINR, the receiver transmits one signature; and therefore CSMA/CN+ reduces to CSMA/CN, achieving as high performance as CSMA/CN (as shown in Figure 7). In low SINR, the receiver transmits two signatures; and therefore the sender infers a CN occurrence only after detecting the arrivals of the two signatures (instead of detecting the arrival of one signature as in CSMA/CN), thereby reducing the false-alarm probability. In very low SINR, the receiver does not transmit signatures; and therefore CSMA/CN+ excludes the false alarm completely. In this way, we alleviate the adverse impact of the false alarm and therefore improve the system performance. Our design is applicable for CSMA/CN-supporting wireless networks, e.g., wireless LANs and wireless multi-hop networks (note that the authors in[5] presented their design mainly focusing on a two-link network, which is the fundamental component of multi-hop networks; our design is applicable for such a two-link network as well). Our improvement is at the cost of slightly increasing the software-implementation complexity (the overhead incurred is explained in Section 2.6). In this paper, we first present the design of CSMA/CN+. We then apply the design in a wireless LAN, and theoretically analyze the detection performance of the noti fication and the saturation throughput (where each sender always has frames to transmit). Extensive simulations verify that CSMA/CN+ can remarkably improve the system throughput of CSMA/CN and our analysis is very accurate.

The rest of this paper is organized as follows. In Section II, we elaborate the design of CSMA/CN+. In Section III, we theoretically analyze the detection performance and the saturation throughput of CSMA/CN+. In Section IV, we verify the effectiveness of CSMA/CN+and the accuracy of our analysis Finally, Sec-tion V concludes this paper.

II. CSMA/CN+

In this section, we present the proposed CSMA/CN+.

2.1 Overview of CSMA/CN+

CSMA/CN+ is an enhancement of CSMA/CN.

Like CMSA/CN, the sender S utilizes two antennas： one for normal data transmission and another dedicated to listening for the notification. The receiver R utilizes a single antenna for transmission or reception. R has a signature with length LR(denoted by), which is known to S. Besides, CSMA/CN+also employs three additional PHY techniques adopted in CSMA/CN： physical-layer hints [9](which inform the MAC layer of how likely some bits of a frame are in error), cross-correlation (between the known pattern and the arriving signal), and self-signal suppression(that suppresses the self-signal, rather than cancels it perfectly).

The data transmitting and receiving process in CSMA/CN+ is similar to that in CSMA/CN, as illustrated in Figure 1. CSMA/CN+enhances CSMA/CN through reducing or excluding false alarms.

Below, we elaborate the design of CSMA/CN+.

2.2 CN+ parameter update at R and s

Algorithm 1： CN+ parameter update atandspeci fies the parameter update procedure at R and S, which is newly introduced in CSMA/CN+. We assume that R intriguers the activation/deactivation operations of CSMA/CN+.When CSMA/CN+ is deactivated, CSMA/CA runs; otherwise CSMA/CN+ runs.

Fig. 1 Basic operations of CSMA/CN+

?

?

?

?

2.3 Data frame transmission from S to R

Algorithm 2 specifies the data transmission process from S to R. WhenSwins the channel,it will transmit a data frame according to the

2.4 CN+ transmission from R to S

Algorithm 3 specifies the CN+ transmission process from R to S. Once inferring that the receiving frame gets corrupted,Raborts receiving the frame immediately, and executes the following actions according to the Flag

2.5 CN+ detection at S

Algorithm 4 speci fies the CN+ detection process at S. While transmitting a data frame, S will detect via correlation the noti fication from R and execute the following actions according one signature as in CSMA/CN (i.e., corr(LR)＞, which means that the cross-correlation＞), S will abort the ongoing transmission. In low SINR, by regarding the arrivals of two signatures as a noti fication, CSMA/CN+ may reduce the false-alarm probability signi ficantly, thereby improving the channel utilization.Here, corr(LR) and corr(L+) can be calculated by and, respectively.

where denotes thel-th symbol of the receiving signal after the self-signal suppression, from

2.6 Overhead and threshold setting

CSMA/CN+ improves the performance of CSMA/CN at the cost of slightly increasing the software-implementation complexity. First,our design has the same hardware requirement as the CSMA/CN. Second, in our design, the main overhead (introduced in the sender side and implied in Algorithms 1 and 4) is that the sender should execute an additional correlation operation with a time complexity of O(1)as in CSMA/CN, when two signatures are employed. The main overhead (introduced in the receiver side and implied in Algorithm 1)is that the receiver should determine whether to transmit a noti fication and how to set L+if a notification will be transmitted. The decision has a time-complexity of O(1) as the mainly involved variable L+has a closed form and can be evaluated in a constant time.

In our design, the receiver, adapting to channel conditions and self-signal suppression capability, prudently determines whether to send back zero, one, or two signatures to the sender. Two thresholds involving the decision are SINRth(which is the SINR threshold after self-signal suppression) and Lth(which is the threshold of notification length), as shown in Algorithm 1. In our design, the threshold settings are according to the theoretical model in [8]. We set SINRthto -10dB, because Fig. 9 in [8] shows that the system performance will deteriorate signi ficantly when SINR ＜ -10dB.We set Lthto the half of the average payload length in a data frame, because Fig. 7 in [8]shows that a longer notification length will remarkably delay the aborting time of an erroneous transmission.

2.7 Hidden and exposed node problems

In this section, we discuss the behavior of CSMA/CN+ when hidden and exposed node problems occur. We assume that all nodes (i.e.,A, B, C, and D) in Figure 2 support CSMA/CN+, and C always transmit frames to D.

Hidden node problem. Figure 2(a) illustrates this problem, where A and C are hidden nodes each other in terms of B. If A transmits a frame to B, C’s transmission will interfere with B’s reception. Therefore, B will send a notification consisting of its signature to A;then A aborts its transmission immediately, releasing the channel for possible transmissions of A’s neighbors, as done in CSMA/CN. In short, in this situation, CSMA/CN+ reduces to CSMA/CN.

Fig. 2 (a) Hidden node problem and (b) exposed node problem in CSMA/CN+

Exposed node problem. Figure 2(b) illustrates this problem, where B and C are exposed nodes each other, and thereby they cannot transmit frame simultaneously. Like [5],we also assume that the CMAP scheme [10]is used here that allows B and C to transmit in parallel according to an interference-map mechanism. When B transmits frames to A, at the same time it will also perform correlation operations to detect whether to receive a noti-fication from A. However, B is in a very low SINR state due to C’s interference. This might cause B to frequently trigger false alarms(which forces B to abort almost all ongoing transmissions) if only one signature with a short fixed length acts as a noti fication. Then,CSMA/CN+ can be used to determine whether to introduce an additional signature and how to set its length. In short, in this situation,CSMA/CN+ has great potential to improve the performance of CSMA/CN signi ficantly.

III. PERFORMANCE OF CSMA/CN+ IN A WIRELESS LAN

In this section, we model the saturation throughput of CSMA/CN+ in a wireless LAN.

3.1 Performance of CN+ detection

The performance of detecting a notification signi ficantly affects the system throughput in CSMA/CN+. Two performance metrics are associated with the detection performance：

· the false alarm probability, namely, the probability that R does not send a noti fication but S detects it mistakenly;

· the successful detection probability, namely, the probability that R sends a notification and S detects it successfully.

where Q(·) is the tail probability of the standard normal distribution, f(·) and g(·) are from [8].

Fig. 3 CSMA/CN+ in low SINR for a wireless LAN

3.2 Calculation of CN+ length

In [8], we show that to achieve a high throughput, we should ensure that the false

With CSMA/CN+, to achieve a high throughput in low SINR, we use two sig-

3.3 Saturation throughput of CSMA/CN+

For a complete analysis of the saturation throughput, pleases refer to [8].

IV. MODEL VERIFICATION

In this section, we show the performance of CN+ detection and the saturation throughput of CSMA/CN+.

4.1 Performance of CN+ detect

Figure 4 (a) and (b), respectively, plot the false-alarm probability and the successful detection probability when SINR increases from-15dB to 5 dB. From this figure, we have the following observations：

· When SINR ＞ -5dB, the CN+ detection performance of CSMA/CN+ (i.e., the probabilities of false alarm and successful detection) is almost the same as that of CSMA/CN. This indicates that in this regime, CSMA/CN+ should produce as high throughput as CSMA/CN, since the false alarm probability is the dominating factor that in fluences the system performance.

4.2 Saturation throughput of CSMA/CN+

· CSMA/CN+ achieves the highest throughput. The reason is that when the SINR is low, the condition of detecting two signatures significantly lowers the false-alarm probability (which makes normal transmissions successful with high probability),while the sender may still abort erroneous transmissions with high probability immediately.

· CSMA/CN achieves the lowest throughput which is almost 0. The reason is that the low SINR triggers a high false-alarm probability, which forces senders to abort almost all ongoing transmissions.

· CSMA/CA achieves a throughput which is between those of CSMA/CN and CSMA/CN+. The reason is that CSMA/CA neither triggers a false-alarm (so it is better than CSMA/CN) nor aborts erroneous transmissions early (so it is worse than CSMA/CN+).

The above observations manifest that CSMA/CN+ is very effective in improving the system performance of CSMA/CN. In addition, the simulation results well match with the corresponding theoretical results, indicating that the theoretical model is very accurate.

Figure 7 compares the normalized throughput between CSMN/CN+ and CSMA/CN, as SINR varies from -15dB to 5dB, where n=30 and the packet size = 1500 bytes. From this figure, we have the following observations.

Fig. 4 (a) Probability of false alarm, and (b) probability of successful detection for the CN+ noti fication

· In low SINR (say, below -6dB), CSMA/CN+ may achieve far higher throughput than CSMA/CN. The reason is explained as follows. In low SINR, using a single signaprobability significantly, making most normal transmissions successful in CSMA/CN+. Note that the throughput of CSMA/CN+ increases as SINR varies from -15dB to -9dB, but decreases as SINR varies from-9dB to -7dB. This is because when SINR increases to a threshold, the overhead (introduced by the additional signature) will exceed the gain of reducing the false-alarm probability.

Fig. 5 Saturation throughput vs. L+

Fig. 6 Comparison of saturation throughput among CSMA/CA, CSMA/CN, and CSMA/CN+

V. CONCLUSION

In this paper, we propose a novel design called CSMA/CN+ to enhance CSMA/CN in low SINR. The enhancement is at the cost of slightly increasing the software-implementation complexity. We then apply the design in a wireless LAN, and theoretically analyze the system throughput. Extensive simulations verify the effectiveness of the proposed design and the accuracy of the proposed theoretical model. Our design is applicable for CSMA/CN-supporting wireless networks. This study will greatly promote the practicality of CSMA/CN.

Fig. 7 Comparison of saturation throughput between CSMN/CN+ and CSMA/CN,as SINR varies from -15dB to 5dB

ACKNOWLEDGEMENTS

This work is supported by the Macao FDCTMOST grant 001/2015/AMJ, and Macao FDCT grants 056/2017/A2 and 005/2016/A1.

Reference

[1] G. Yang, S. He, Z. Shi, J. Chen, “Promoting Co-operation by the Social Incentive Mechanism in Mobile Crowdsensing”, IEEE Communications Magazine, vol. 55, no. 3, pp 86-92, March, 2017.

[2] W. Feng, Y. Wang, N. Ge, J. Lu, and J. Zhang,“Virtual MIMO in multi-cell distributed antenna systems: coordinated transmissions with largescale CSIT”, IEEE Journal on Selected Areas in Communications, vol. 31, no. 10, pp 2067-2081,October, 2013.

[3] W. Feng, Y. Wang, D. Lin, N. Ge, J. Lu, and S.Li, “When mmWave communications meet network densification: a scalable interference coordination perspective”, IEEE Journal on Selected Areas in Communications, Accepted for publication, April, 2017.

[4] ANSI/IEEE Std 802.11, Part 11: wireless LAN medium access control (MAC) and physical layer(PHY) speci fications, 1999 Edition (R2007).

[5] S. Sen, R. R. Choudhury, and S. Nelakuditi,“CSMA/CN: carrier sense multiple access with collision notification”, IEEE/ACM Transactions on Networking, vol. 20, no. 2, pp 544—556, April,2012.

[6] X. Ji, J. Wang, M. Liu, Y. Yan, P. Yang, and Y. Liu,“Hitchhike: Riding control on preambles”, Proceedings of IEEE INFOCOM, pp 2499—2507,May, 2014.

[7] S. Lv, X. Dong, Y. Lu, X. Du, X. Wang, Y. Dou, and X. Zhou, “3D pipeline contention: Asymmetric full duplex in wireless networks”, Proceedings of IEEE INFOCOM, pp. 790—798, May 2014.

[8] F. Xu, Q. Zhao, and Y. Zeng, “How Well Does CSMA/CN Work in WLANs?”, IEEE Transactions on Vehicular Technology, vol. 65, no. 9, pp.7662-7669, September, 2016.

[9] M. Vutukuru, H. Balakrishnan, and K. Jamieson,“Cross-layer wireless bit rate adaptation”, ACM SIGCOMM Computer Communication Review,vol. 39, no. 4, pp. 3—14, August, 2009.

[10] M. Vutukuru, K. Jamieson, and H. Balakrishnan,“Harnessing exposed terminals in wireless networks”, in Proc. 5th USENIX NSDI, pp. 59—72,April, 2008.