Breaking Through the Full-Duplex Wi-Fi Capacity Gain
In this work we identify a seminal design guideline that prevents current Full-Duplex (FD) MAC protocols to scale the FD capacity gain (i.e. the half-duplex throughput) in single-cell Wi-Fi networks. Under such guideline (referred to as :), a MAC protocol attempts to initiate up to two simultaneous transmissions in the FD bandwidth. Since in single-cell Wi-Fi networks MAC performance is bounded by the PHY layer capacity, this implies gains strictly less than over half-duplex at the MAC layer. To face this limitation, we argue for the : design guideline. Under :, FD MAC protocols ‘see’ the FD bandwidth through orthogonal narrow-channel PHY layers. Based on theoretical results and software defined radio experiments, we show the : design can leverage the Wi-Fi capacity gain more than at and below the MAC layer. This translates the denser modulation scheme incurred by channel narrowing and the increase in the spatial reuse offer enabled by channel orthogonality. With these results, we believe our design guideline can inspire a new generation of Wi-Fi MAC protocols that fully embody and scale the FD capacity gain.
Recent works have demonstrated the feasibility of Self-Interference Cancellation (SIC) techniques, turning Full-Duplex (FD) radios into a reality e.g. . Such radios are capable of receiving and transmitting simultaneously within the same frequency band, achieving a gain of the half-duplex link capacity in theory (i.e. the FD gain). An important question raised by that achievement is whether it is possible to design a Medium Access Control (MAC) protocol that accomplishes the goal of scaling the FD gain in a wireless network. A possible way to accomplish that consists in relying on the wide area implied by multi-cell deployments to activate multiple concurrent links . However, by surveying the MAC literature e.g. [3, 4, 5, 6], one can find out it is hard to accomplish that scalability goal within a single-cell Wi-Fi compliant Wireless Local Area Network (WLAN), since the contention overheads and the lack of spatial reuse can shrink the FD gain to as the network grows .
To tackle the limitation of current FD MAC protocols, we go a step further and identify a common design strategy we refer to as the : MAC design guideline. With the : design, an FD MAC protocol ‘sees’ the whole FD bandwidth through a single PHYsical layer. To maximize FD gains with such design, MAC protocols attempt to minimize the difference between the start time of two concurrent transmissions in the channel. This leads to gains bounded by the capacity of two nodes freely transmitting to each other in the channel. In fact, in a single-cell WLAN, the MAC throughput is bounded by the PHY layer capacity.Thus, doubling such capacity with FD radios may limit the maximum capacity gain achieved at the FD MAC layer to a value strictly less than the half-duplex throughput. This suggests one needs to improve the capacity below the MAC layer more than to give room for MAC protocols that actually approaches the FD gain.
In this paper we report novel results that break through the capacity gain leveraged by FD radios in single-cell WLANs. We accomplish this by arguing for an alternative FD MAC design guideline we refer to as :. Under that, the MAC layer arranges the FD bandwidth into PHY layers. Each PHY layer is assigned to a portion of spectrum that is narrower than the available FD bandwidth and orthogonal to the other PHY’s spectrum portions. Similar design have been studied before from the perspective of MAC and/or radio architecture e.g. [8, 9, 10, 11, 12]. These works highlight the advantages of parallel narrow channels on a single radio but under the half-duplex constraint. To fully realize the FD gain over a wireless bandwidth allocated to concurrent narrow channels, one has to refer to the same kind of wide-band SIC design (e.g. ) assumed by current state-of-the-art : FD MAC proposals. We refer to such advance to report unprecedented contributions towards the FD gain scalability in WLANs.
Our first contribution is to show that, contrary to the popular assumptions and beliefs, it is possible to attain more-than-doubled capacity gains within an FD bandwidth i.e. below the MAC layer. Indeed, narrowing a channel relaxes receive sensitivity requirements enabling denser modulation schemes [13, Table 18–14]. Thus, spectrum usage improves. For instance, instead of occupying a MHz channel with two (FD) transmissions, one can split it into two MHz orthogonal FD channels and activate four concurrent transmissions. This yields gains of over a MHz half-duplex link even considering guard-bands. We demonstrate this theoretically and through a proof-of-concept study with USRP platforms.
Our second contribution is to scale the novel FD gain at the MAC layer. We characterize the ideal condition for an : FD Wi-Fi MAC protocol and show its performance improves more than twice under the : guideline. This happens because channel orthogonality multiplies FD opportunities by increasing the spatial reuse offer. We believe these results instigate further research towards a solid FD IEEE 802.11 stack.
Ii System Model and Background
We consider the design directives that a Wi-Fi compliant FD MAC protocol should follow to scale the FD gain. In this sense we focus on models to assess capacity upper-bounds at and below the MAC layer in a single-cell infrastructure IEEE 802.11 WLAN. For the MAC protocol study, the cell is composed of one Access Point (AP) and STAtions (STA). STAs perform the standard CSMA/CA to initiate a transmission to the AP (uplink). The AP is assumed to always have a frame enqueued to its current transmitting STA. Then, the AP can establish an FD (down)link to the STA upon processing its incoming header. As we discuss in section IV, this corresponds to an ideal condition the capacity upper-bound of an FD Wi-Fi MAC protocol can be derived from.
For each MAC proposal we assume saturated traffic and ideal channel conditions . These assumptions ensure we assess ‘the most each MAC protocol can do’ when provided with best conditions. Note, however, any MAC protocol under the design guideline we are about to present might actually perform better in noisy environments. This happens because the narrow Wi-Fi channels we rely on are less prone to noise, as we discuss in the section III-B. Also, we assume each compared MAC and PHY model suffers from the same level of negligible self-interference residue. Again, a successful (de)modulation process might be less demanding in terms of SIC requirements if performed over narrower channels instead of wide channels .
Ii-a FD MAC WLAN Terminology
The ultimate goal of any FD MAC protocol is to take advantage of FD opportunities within a given wireless channel to maximize capacity. It means the protocol attempts to activate two overlapping transmissions to maximize channel utilization so throughput. In Wi-Fi compliant WLANs, the Primary Transmitter (PT) is the first node to start transmitting a data frame after winning a typical CSMA/CA contention round. The node PT transmits to is called Primary Receiver (PR). During the primary transmission, the FD MAC protocol may start a secondary transmission in the channel. In this case the sender and receiver are called Secondary Transmitter (ST) and Secondary Receiver (SR), respectively.
Basically, the FD opportunities can be classified into either symmetric or asymmetric dual-links . In symmetric dual-links, PT and PR coincide with SR and ST, respectively (i.e. [PTSR][PRST], where the direction of each arrow denotes the destination of a transmission). In asymmetric dual-links, there must be a third node involved in the secondary communication. Such node is either a SR or a ST. In the former case, the PR coincides with the ST i.e. PT[PRST]SR. Otherwise the PT coincides with SR, i.e. PR[PT=SR]ST. Note the two possible asymmetric dual-links are not different views of the same scenario since in one case an already receiving node starts transmitting while in the other an already transmitting node starts receiving.
Ii-B Medium Access Control Challenges with Dual-links
The performance of an FD MAC protocol results from a balance between how effectively it exploits dual-links and the cost it takes towards that. Concerning asymmetric dual-links, the main challenge consists in assuring the secondary transmission does not collide with some possible ongoing primary transmission. Collisions may happen whenever the receiver node of a primary (secondary) transmission is within the interference range of a secondary (primary) transmission. In case of symmetric dual-links, the challenge consists in identifying a pair of nodes that have frames to each other. To maximize FD gains regardless of the type of dual-link, any FD Wi-Fi MAC protocol attempts to minimize . Particularly for our scenario, is the time at which a STA starts a primary transmission after winning a CSMA/CA contention round and is the time at which the AP starts the corresponding secondary transmission.
Ii-C Novel Classification for FD MAC Protocols
In this work we identify a new category for the design of FD MAC protocols. With this novel category, MAC protocols are classified according to the way they exploit the available wireless FD bandwidth. In this sense, we identify a seminal trend we refer to as the : MAC design guideline [3, 4, 5, 6]. Under the : guideline the MAC protocol ‘sees’ the FD bandwidth through a single PHY layer. Thus, the best-case of any : MAC protocol is bounded to the capacity of a dual-link. Moreover the resulting capacity is impaired because of the contention overheads.
Iii The : MAC Design Guideline
A reasonable way to overcome the performance limitation of : FD MAC protocols consists in, firstly, improving the capacity below the MAC layer. Toward that goal we advocate an alternative FD MAC design guideline we refer to as :. Under this novel guideline, a MAC protocol sees the FD bandwidth through PHY layers. Each PHY layer is assigned to a sub-channel that is narrower than the whole available FD bandwidth and orthogonal to the sub-channel of the other PHY layers. The value of is a trade-off figure of merit between the maximization of the number of concurrent transmissions and the minimization of the spectrum overhead needed to isolate channels through guard-bands. While a comprehensive understanding about the effects of varying makes a strong case for future research, along this work we propose a case study for to quantify the unique benefits our proposal brings for the design of FD MAC protocols.
Iii-a Increased Spatial Reuse Offer
The : design creates more FD opportunities than : by increasing spatial reuse offer, as shown in Fig. 1. In the : best-case scenario (Fig. LABEL:fig:1:1) a dual-link can increase throughput while avoiding that a hidden node (e.g. STA ) collides with an ongoing transmission (e.g. ). However, this sacrifices spatial reuse by interfering with all other STAs (dashed waved arrows) . By arranging the FD bandwidth into orthogonal narrower-channel PHY layers, the : best-case scenario overlaps additional dual-links in the same space. This is illustrated on Fig. LABEL:fig:1:N for , in which channel orthogonality (i.e. gray and black colors) also helps against collisions and enables one additional dual-link in the network.
Iii-B Improved Signal to Noise Ratio
Prior works [15, 16] show experimentally that halving a single Wi-Fi channel increases the total energy in the bandwidth, yielding an SNR gain of dB. We enhance these tests to check whether the SNR statement holds when the total active bandwidth remains the same but the number (then the width) of channels changes. In each test, we set Wi-Fi signals to the same parameters. However, one scenario considers a MHz-wide channel and the other considers two concurrent MHz-wide channels. To achieve such concurrency one can resample a MHz Wi-Fi signal by interpolating it to in the baseband. The resulting signal is duplicated and each copy is shifted to its specific half within a MHz band. In Fig. 2 we plot the Power Spectrum (PS) of the strongest signals as reported by a couple of single-antenna Ettus USRP B210 platform. We estimate the PS samples and their average based on the Matlab’s pwelch procedure. From the plots, one can see each narrow channel benefits from dB gain over the wider channel. In fact, although both narrow channels occupy the same MHz spectrum, they are employed independently. Thus, both the environmental and noise floors experienced within a channel does not account for the signal processing in the other.
Iii-C Capacity Model Below the MAC Layer
The SNR improvements resulting from channel narrowing can translate into higher capacity for a Wi-Fi bandwidth. Consider an AWGN Wi-Fi channel measuring (Hz) under a given (unitless). According to the Hartley-Shannon theorem, the maximum information that can be modulated and carried over a half-duplex bandwidth is Bits/Hz/s (Eq. 1). Assuming an FD radio and expressing the in dB (), one derives Eq. 2 for the capacity limit of FD MAC protocols under the : design.
With the : guideline, the FD bandwidth is equally divided among narrow channels. Considering , the dB gain induced by channel narrowing, the guard-band (Hz) and the FD capability assumed before, the total capacity achieved within is given by Eq. 3.
Iv FD Wi-Fi MAC Protocol Capacity Upper-Bound
In this section we characterize the ideal condition to derive the capacity upper-bound of a Wi-Fi compliant FD MAC protocol. Then, we present a model to assess such capacity under both the : and : MAC design guidelines.
Iv-a Ideal Condition for Wi-Fi Compliant FD MAC protocols
To keep Wi-Fi compliance, a MAC protocol shall follow the CSMA/CA access method. In the context of FD radios, this means that at least the primary transmission initiates following a typical exponential back-off procedure. Since CSMA/CA is half-duplex by nature, some additional mechanism is required to admit a collision-free secondary transmission. The resulting time overhead to coordinate such a secondary transmission (i.e. ) is the key reason why MAC protocols’ performance falls well below the FD gains . Therefore, under an ‘ideal FD condition’, an Wi-Fi compliant MAC protocol maximizes the FD gain utilization by minimizing the time overhead .
A naive way of characterizing the ‘ideal FD condition’ is assuming i.e. . This implies that the same backoff number is shared without overheads by a pair of arbitrary nodes at the beginning of each time slot. This is a too strong assumption for our scenario because conflicts with the random uniform behavior of the CSMA/CA backoff procedure. A reasonable alternative for this consists in assuming that the PR always has a data frame enqueued to the PT. In our scenario this means that the minimum corresponds to the time interval the AP needs to start the secondary transmission just after processing the incoming primary transmission’s header . A prior work has shown an AP can manage to do that in real-time . The whole process is illustrated in Fig. 3. In the figure, an arbitrary STA starts a primary transmission to the AP at the time instant upon winning a CSMA/CA contention (not illustrated). After receiving and processing , the AP fetches a data frame and starts a secondary transmission to the corresponding STA at the time . This defines the minimum , which corresponds to in the figure. Note, however, that FD becomes profitable only at , the time at which useful data starts being transferred. To avoid collisions due to hidden terminals, both transmissions have to be finished simultaneously , then the maximum secondary payLoad (bytes) for the capacity upper-bound is dimensioned accordingly. The other parameters on the Fig. 3 are helpful for the capacity model, as we explain next.
Iv-B Capacity Limit Model
To compute the capacity limit of CSMA/CA under the ideal FD condition for each design guideline, we refer to the IEEE 802.11 capacity model proposed by Bianchi . The model is twofold. Firstly it computes the probabilities and that a CSMA/CA station transmits and collides, respectively. These probabilities are computed in the same way for our scenario, since the STAs contends for primary transmissions just as in half-duplex CSMA/CA. The second part of the model consists in a expression that computes the throughput for IEEE 802.11 WLANs regardless of the channel access mode. More precisely, the model computes the saturation capacity given both the payload carried per transmission and the time duration of each possible event in the channel.
To assess assuming an FD channel, we need firstly to characterize the possible events related to a primary transmission at the beginning of a time slot. In our case they correspond to same possible events of a CSMA/CA half-duplex channel, namely, ‘success’, ‘collision’ or ‘absent’ (empty slot). These events happen with probabilities , and and take , and absolute time units (e.g. s), respectively. Of these, is obtained straightforwardly from the standard waiting slot time . Moreover, only the first event carries an expected amount of useful payLoad, that we denote as .
Iv-B1 Probabilities of channel events
To compute , and , recall that each one of all STAs does transmit with probability and does not with probability . Thus, channel is idle with probability . A primary transmission succeeds if only a single STA transmits and the remainder STAs remain silent, what happens with probability . Since each of the STAs has the same chance to succeed . A collision happens if the channel is not idle and, at the same time, a primary transmission does not succeed i.e. .
Iv-B2 Duration and payload of a successful primary transmission
Let and be the PHY-MAC headers and payLoad sizes of a primary transmission, respectively. Similarly, and have equivalent meaning for a secondary transmission, as illustrated on Fig. 3. Also, let and be the time taken to transmit (or ) and under given control and data rates, respectively. Denoting as plus the total IEEE 802.11 standard time needed to acknowledge a data frame and as the propagation time of each frame, the overall duration of a successful primary transmission is given by Eq. 4. Note that also comprises i.e. the minimum Wi-Fi standard time interval all STAs must wait before assuming channel is idle again and restarting the CSMA/CA count-down.
As one can also see on Fig. 3, under the ideal FD condition, a dual-link comprises two data frame transmissions. Therefore, the total expected payload carried within the channel event ‘success’ is defined as . Note that , where is the amount of useful payload that the secondary transmission’s data rate could send during the time interval comprising fetching and transmitting (i.e. on Fig. 3).
Iv-B3 Duration of a collision
To detect a collision, the PT starts its timer just after pushing the last symbol header into the channel. As soon as PT detects an incoming symbol, it stops the timer and finishes receiving the whole incoming signal. If the received signal does not correspond to as expected or, alternatively, no signal is detected before the timer expires, then PT stops transmitting. The maximum timer estimation comprises , the header propagation time and the overhead on PR to start the secondary transmission appropriately. In  authors report an overhead of s to start an FD Wi-Fi like transmission in real-time.
Iv-B4 MAC guidelines saturation throughput
The FD CSMA/CA capacity formula (Eq. 5), comes from the ratio between the payload and the time duration associated to each possible event in the channel.
Each value in the ratio are weighted by the corresponding channel event probability. This formula stands for both design guidelines. The difference is that for the : design. Hence, under the ideal FD condition, each CSMA/CA round triggers two transmissions across the whole channel. With the : design, and each CSMA/CA round triggers narrow-channel transmissions under the same ideal condition. Also, all timing parameters are rescaled according to channel width just as the IEEE 802.11 standard mandates .
In this section we report results achieved by both : and : guidelines at and below the MAC layer. We also report the half-duplex performance for comparison purposes.
V-a Novel Capacity Limit Below the MAC Layer
In Fig. 4 we plot the capacity upper-bound for the : and the : guidelines across different SNRs (Eqs. 2 and 3, respectively). The total bandwidth is MHz so : corresponds to two MHz channels. Each MHz channel is separated by a guard-band KHz, what can be achieved by actual filters e.g. . We also plot the half-duplex capacity for comparison purposes (Eq. 1). As widely known, the gain of any FD radio is bounded by the half-duplex capacity. However, the SNR gains induced by channel narrowing breaks this currently prevalent gain even paying a KHz guard-band overhead. We verified the statement still holds for a up to about MHz.
To investigate whether such theoretical results preserves in practice, we propose a proof-of-concept study based on a pair of Ettus USRP B210 Software Defined Radio (SDR) platforms. Each radio is equipped with one antenna for transmission and one for reception. We compare a single MHz channel against two MHz channels. An ideal FD radio doubles capacity by entirely releasing the bandwidth for reception while transmitting. To mimic such behavior, we rely on an out-of-band FD test. Thus, in both scenarios, each radio has MHz channel dedicated for reception and another MHz for transmission, being these channels MHz away from each other. For each case, we set the highest modulation the IEEE 802.11 standard mandates under a Received Signal Strength Indication (RSSI) of dBm i.e. QPSK for MHz and -QAM for MHz [13, Table 18–14]. This yields data rates of and Mbps, respectively. We produce Wi-Fi signals based on the gr-ieee80211 GNURadio module  and measured all bytes transferred through saturated links. Since SDR experiments are dramatically affected by CPU load and FD doubles such processing demands, we assess the half-duplex link from the best FD link. For each experiment we gather as much sample as needed to calculate mean throughput with a confidence of and a relative error , following the statistical procedures of .
From the plots on Fig. 5 one can see our theoretical statement holds in practice. Our design improves over half-duplex about . For all cases the throughput is dramatically lower in comparison to theory mostly because of the large latency introduced by the (half-duplex) USB connection between USRP and PC. Finally, we recognize that the capacity of actual in-band FD radios is strictly less than half-duplex’s because of residual self-interference. However, our findings suggest that the gains claimed by single-band FD radio proposals might be underestimated. For instance, we believe the best currently reported result – in an MHz channel with an RSSI of dBm  – could be even better if performed over two MHz FD channels (with appropriate filters/guardbands) set to the densest Wi-Fi modulation scheme supported under dBm.
V-B Novel Capacity Limit at the MAC Layer
To check whether the PHY layer improvement scales at the MAC layer we assess the capacity upper-bound of the FD CSMA/CA under both : and : designs. The numerical results are computed in accordance with the section IV-B. We also report the half-duplex results under both the standard access modes namely, the -way (i.e. DATA followed by ACK) and the -way handshakes (Request-to-Send/Clear-to-Send, RTS/CTS-DATA-ACK). We assume a propagation time of s, a bandwidth of MHz and (i.e. two MHz channels for :). All other timing parameters are set according to the IEEE 802.11a best-effort traffic class.
We verify that the FD MAC protocols outperform the half-duplex CSMA/CA across different data rates and frame payload sizes. Due to space constraints, on Fig. 6 we only report results for data rate of Mbps in MHz channels. This implies in at least Mbps for MHz channels . Similarly, for these respective channel widths, we set control rates to Mbps and Mbps and MAC payload to bytes. Larger payloads dramatically damages -way half-duplex performance upon collisions, specially as network grows (Fig. 6). The -way handshake mitigates that by preceding data transmission with smaller RTS frames but the overall handshake slows all successful transmissions. In turn, with FD only a very small part of the primary transmission’s payload is exposed to collision. This happens with no penalty to successful transmissions. In addition to these abilities, the poor half-duplex performance over an increasing number of nodes causes the : FD CSMA/CA to be as higher as the half-duplex performance (as of nodes on Fig. 6). However, as one can also see on Fig. 6, such gains can be improved by conforming the FD CSMA/CA to the : design. Indeed, the MAC gain under the : design closely approaches the PHY layer improvement we report in this section, and keeps scaling over nodes. Moreover, the channel orthogonality exploited by the : design enables higher spatial reuse. Hence, the number of FD opportunities in the best-case increase from to . Although non-exhaustive, these results represent an unprecedented step towards the scalability of FD gains in single-cell WLANs.
Vi Conclusion and Future Work
In this work we study the capacity limits of single-cell FD WLANs. We inquire what prevents current Wi-Fi compliant FD MAC protocols to fully profit from the theoretical double of throughput leveraged by FD radios. In addition to the overheads at the MAC layer, we realize this is also explained by the capacity bound imposed below the MAC layer. Thus, we propose a design categorization based on which MAC protocols are classified according to the way they ‘see’ the FD bandwidth. In this sense, we identify current FD Wi-Fi MAC protocols are classified into what we refer to as the : design guideline, meaning they ‘see’ the FD bandwidth through a single PHY layer. With this, MAC performance is bounded by a pair of transmissions in the channel. Instead, under the : design guideline we advocate, MAC protocols ‘see’ the FD bandwidth through orthogonal narrow-channel PHY layers. Based on theoretical results and software defined radio experiments, we show it is possible to outperform the current assumed FD capacity gain at and below the MAC layer. To benefit from this novel more-than-doubling improvement, in future works we plan to design novel mechanisms that exploit the spatial reuse opportunities enabled by the : design guideline. Also, we intend to study the : design together the MIMO technology.
-  D. Bharadia, E. McMilin, and S. Katti, “Full duplex radios,” in Proc of the ACM SIGCOMM 2013, pp. 375–386, ACM, 2013.
-  X. Xie and X. Zhang, “Semi-synchronous channel access for full-duplex wireless networks,” in Network Protocols (ICNP), 2014 IEEE 22nd International Conference on, pp. 209–214, Oct 2014.
-  S. Goyal, P. Liu, O. Gurbuz, E. Erkip, and S. Panwar, “A distributed MAC protocol for full duplex radio,” in Signals, Systems and Computers, 2013 Asilomar Conference on, pp. 788–792, Nov 2013.
-  J. Y. Kim, O. Mashayekhi, H. Qu, M. Kazandjieva, and P. Levis, “Janus: A novel MAC protocol for full duplex radio,” in Stanford CSTR, 2013.
-  M. Duarte, A. Sabharwal, V. Aggarwal, R. Jana, K. Ramakrishnan, C. Rice, and N. Shankaranarayanan, “Design and characterization of a full-duplex multi-antenna system for WiFi networks,” Vehicular Technology, IEEE Transactions on, vol. PP, no. 99, 2013.
-  N. Singh, D. Gunawardena, A. Proutiere, B. Radunovic, H. V. Balan, and P. Key, “Efficient and fair MAC for wireless networks with self-interference cancellation,” in Int Symp on Modeling and Optimization in Mobile, Ad Hoc and Wireless Networks (WiOpt’11), pp. 94–101, 2011.
-  X. Xie and X. Zhang, “Does full-duplex double the capacity of wireless networks?,” in INFOCOM, 2014 Proceedings IEEE, pp. 253–261, April 2014.
-  S. Queiroz and R. Hexsel, “Translating full duplexity into capacity gains for the high priority traffic classes of IEEE 802.11,” in Applied Computing (SAC), 2015 ACM/SIGAPP 30th Symposium on, 2015.
-  S. Queiroz, “All-at-once or piece-by-piece: How to access wide channels in WLANs with channel width diversity?,” Communications Letters, IEEE, vol. 17, no. 11, pp. 2188–2191, 2013.
-  J. Fang, K. Tan, Y. Zhang, S. Chen, L. Shi, J. Zhang, Y. Zhang, and Z. Tan, “Fine-grained channel access in wireless LAN,” IEEE/ACM Trans. Netw., vol. 21, pp. 772–787, June 2013.
-  S. S. Hong, J. Mehlman, and K. Sachin, “Picasso: flexible RF and spectrum slicing,” in Proc ACM SIGCOMM’12, pp. 37–48, ACM, 2012.
-  K. Krishna, B. Radunovic, V. Balan, M. Buettener, S. Yerramalli, V. Navda, and R. Ramjee, “WiFi-NC: WiFi over narrow channels,” in Proc 9th USENIX, pp. 43–56, 2012.
-  IEEE, “IEEE standard for information technology– part 11: Wireless LAN MAC and PHY specifications amendment 5: Enhancements for higher throughput,” IEEE Std 802.11-2012, 2012.
-  G. Bianchi, “Performance analysis of the IEEE 802.11 distributed coordination function,” IEEE Journal on Selected Areas in Communications, vol. 18, no. 3, pp. 535–547, 2006.
-  M. Y. Arslan, K. Pelechrinis, I. Broustis, S. Singh, S. V. Krishnamurthy, S. Addepalli, and K. Papagiannaki, “ACORN: An auto-configuration framework for 802.11n WLANs,” IEEE/ACM Trans on Networking, vol. PP, no. 99, 2012.
-  R. Chandra, R. Mahajan, T. Moscibroda, R. Raghavendra, and P. Bahl, “A case for adapting channel width in wireless networks,” SIGCOMM Computer Communication Review, vol. 38, no. 4, pp. 135–146, 2008.
-  J. Mayank, J. I. Choi, T. Kim, D. Bharadia, S. Seth, K. Srinivasan, P. Levis, S. Katti, and P. Sinha, “Practical, real-time, full duplex wireless,” in Proc 17th Int Conf on Mobile Computing and Networking, pp. 301–312, ACM, 2011.
-  B. Bloessl, M. Segata, C. Sommer, and F. Dressler, “An IEEE 802.11a/g/p OFDM receiver for GNU radio,” in Proceedings of the Second Workshop on Software Radio Implementation Forum, SRIF ’13, pp. 9–16, ACM, 2013.
-  G. C. Ewing, K. Pawlikowski, and D. McNickle, “Akaroa2: Exploiting network computing by distributing stochastic simulation,” in Int Society for Computer Simulation, pp. 175–181, 1999.