Traffic Differentiation in Dense Collision-free WLANs using CSMA/ECA

The ability to perform traffic differentiation is a promising feature of the current Medium Access Control (MAC) in Wireless Local Area Networks (WLANs). The Enhanced Distributed Channel Access (EDCA) protocol for WLANs proposes up to four Access Categories (AC) that can be mapped to different traffic priorities. High priority ACs are allowed to transmit more often than low priority ACs, providing a way of prioritising delay sensitive traffic like voice calls or video streaming. Further, EDCA also considers the intricacies related to the management of multiple queues, virtual collisions and traffic differentiation. Nevertheless, EDCA falls short in efficiency when performing in dense WLAN scenarios. Its collision-prone contention mechanism degrades the overall throughput to the point of starving low priority ACs, and produce priority inversions at high number of contenders. Carrier Sense Multiple Access with Enhanced Collision Avoidance (CSMA/ECA) is a compatible MAC protocol for WLANs which is also capable of providing traffic differentiation. Contrary to EDCA, CSMA/ECA uses a contention mechanism with a deterministic backoff technique which is capable of constructing collision-free schedules for many nodes with multiple active ACs, extending the network capacity without starving low priority ACs, as experienced in EDCA. This work analyses traffic differentiation with CSMA/ECA by describing the mechanisms used to construct collision-free schedules with multiple queues. Additionally, evaluates the performance under different traffic conditions and a growing number of contenders. (arXiv's abstract field is not large enough for the paper's abstract, please download the paper for the complete abstract.)

• Adaptation of CSMA/ECA to support multiple queues for collision-free traffic differentiation.
• Introduction of the Smart Backoff mechanism for eliminating Virtual Collisions (VC).
• First simulation results on throughput, collisions and delay for CSMA/ECA with four ACs.
• Implementation of a realistic non-saturation scenario, using well-known models for voice and video codecs.
• Evaluation of the coexistence and backwards compatibility with EDCA.
Results show that CSMA/ECA uses a more efficient collision avoidance mechanism, wastes less channel time recovering from failed transmissions, yields higher throughput, provides traffic differentiation, and creates the possibility of supporting more high priority flows for a higher number of contenders.
Equally important, results show that CSMA/ECA is able to coexist with EDCA without gravely impacting this type of nodes's throughput.
An overview of the traffic differentiation techniques used with EDCA will be provided in Section 2.
Then, we will present CSMA/ECA and its ability to provide traffic differentiation in Section 3. The performance evaluation is shown in Section 4, while we draw our conclusions in Section 5.

Related Work
Each node with a packet to transmit must join a contention for the channel. In CSMA/CA, nodes are deferred for a fixed period of idle-channel time and then for a random backoff period before attempting transmission. Because it only considers a single kind of traffic, the default contention parameters are the same for all contenders.
In this section we present the traffic differentiation capabilities of EDCA as well as other compatible contention parameters adjustment techniques proposed by the research community.

Enhanced Distributed Channel Access
EDCA provides traffic differentiation by defining three parameters for each of the four ACs. First, by adjusting the Transmission Opportunity (TXOP) an AC may transmit several packets without repeating the contention for the channel, thus achieving greater throughput than other ACs. The other two parameters are related to the contention process, namely the Contention Window (CW min and CW max , for minimum and maximum respectively) and the Arbitration Inter-Frame Spacing (AIFS). The contention windows limit the random backoff period, while the AIFS defines the fixed waiting period when the channel is idle. ACs with low contention windows and short AIFS will access the channel quicker, that is, have higher priority.
WLANs time is slotted. That is, it is composed of tiny empty slots of fixed duration σ e , collisions and successful slots (which contain collisions or a successful transmission, their duration denoted by σ c and σ s , respectively) 5 . DCF instructs backlogged stations to wait for a random number of empty slots (random backoff period) before attempting transmission. Transmissions always start at the beginning of a slot.
EDCA extends directly from DCF. In its place, EDCA declares up to four Access Categories (AC), each one an instance of DCF with different contention parameters that allow a statistical prioritisation  among them [13]. Traffic types, declared by the IEEE 802.1D standard [14] are then mapped to the four ACs in EDCA (MAC bridging). The mapping is shown in Table 1.  Table 2 shows the default CW, AIFSN and TXOP values specified for EDCA.
As it can be observed in Table 2, ACs BK and BE may only send one MAC Service Data Unit (MSDU) upon each attempt. Whereas VI and VO can allocate the channel for longer periods. The TXOP parameter offers resource fairness rather than throughput fairness, that is, all ACs of the same category will receive close to the same average channel time regardless of its data rate. Furthermore, because the CW and AIFSN values for VI and VO are smaller than the others, on average these ACs will access the channel quicker; thus providing priority in the contention.
While being effective in providing traffic differentiation and priority, in principle EDCA is unable to eliminate collisions. For instance, two ACs from different contenders may draw the same random backoff and consequently attempt transmission in the same time slot, causing a collision. Furthermore, if two or more ACs within a node experience a backoff expiration at the same instant, a Virtual Collision (VC) will occur. VC are resolved by granting the channel to the highest priority AC, while doubling the CW curr [AC] for the lower priority ACs; just as it is done in the event of a real collision.
It follows directly from above that collisions waste channel time and thus contribute to the throughput 6 The Short Inter-Frame Space (SIFS) is defined in [1]. It is equal to 10 or 16µs for 802.11 n and ac/ax respectively. degradation in WLANs. Moreover, the probability of collision increases as more contenders join the network, each one having four ACs attempting to gain access to the channel.

EDCA enhanced
Because ACs in EDCA perform contention independently of the others, each AC mimics a DCF station.
This explains why the collision probability in EDCA is higher than in DCF networks with the same number of saturated nodes. Furthermore, the contention parameters in EDCA work better in scenarios with low number of contenders, but often cause starvation of low priority ACs in crowded scenarios (see [9] and Section 4.2).
Great efforts have been directed towards parameter adjustments in EDCA, mostly to ensure QoS for high priority ACs while maintaining low delay and losses [15][16][17]. For example, by dynamically adjusting the AIFS for each AC it is possible to maintain traffic differentiation while avoiding the starvation of low priority ACs. This is especially relevant in WLANs where all ACs are required to have effective throughput, like in [16]. Further, by randomising the AIFS values it is possible to increase the channel utilisation in EDCA [17].
MAC parameter adjustment algorithms work as functions that select future values for contention or transmission parameters in each AC. Most consider changing the contention windows, mainly because these were the only contention parameters in DCF. Nevertheless, adjustment of AIFS, and/or TXOP are also possible. These can be classified as [18]: • Static or Adaptive: static algorithms define contention parameters for all ACs, which remain unchanged throughout the operation. An adaptive algorithm selects the best contention parameters for each AC depending on the detected flows. They also react to network congestion variations.  • If no other parameter update is feasible, the flow is rejected.
Algorithms may be combined, or focus on iteration in order to provide advantageous conditions for high priority traffic. Nevertheless, as proposals deviate too much from the IEEE 802.11 MAC standard, the chances of being accepted as an amendment decreases [22,23]. Moreover, performance evaluations should implement updated audio and video source models, using specifications of widely-used codecs in order to mimic realistic scenarios [24,25].
The way traffic differentiation is defined in IEEE 802.11e is through a static, completely distributed, and measurement-based algorithm, that is, EDCA. As its goal is to provide QoS to high priority ACs, low priority traffic is often deferred to the point of throughput starvation. Additionally, EDCA's random backoff mechanism is prone to an elevated number of real and virtual collisions, widening the effects of throughput starvation to higher priority ACs 7 .

Learning-MAC
Learning-MAC (L-MAC) and a survey of other collision-free MAC protocols for WiFi are presented in [29]. As its name suggests, L-MAC uses learning techniques to achieve collision-free schedules. It defines a learning strength parameter, β ∈ (0, 1). Each contender starts by picking a slot s for transmission from schedule n of length C at random with uniform probability. After transmission on slot s(n), the node conditions the selection of the same slot in the next cycle according to the result of the transmission. (1) and (2) extracted from [29] show the probability of selecting the same slot s(n) in cycle n + 1.

Success
(1) p s(n) (n + 1) = βp s(n) (n), for all j = s(n), j ∈ {1, . . . , C}. That is, if a station successfully transmitted in s(n), it will pick the same slot on the next schedule with probability one. Otherwise, it follows (2).
The selection of β implies a compromise between fairness and convergence speed, which the authors determined β = 0.95 to provide satisfactory results.
L-MAC converges to collision-free schedules in a few cycles. Further extensions to L-MAC introduced an Adaptative schedule length in order to increase the number of supported contenders in a collision-free schedule. This adaptive schedule length is doubled or halved depending on the presence of collisions or many empty slots per schedule, respectively. As ZC-MAC, L-MAC stations require common knowledge of the start/end of the schedule.

Descentralised collision-free traffic differentiation
These reservation-like protocols, namely, L-MAC and ZC-MAC could be adapted to support traffic differentiation by using multiple schedules. Semi-Random Backoff [27] is able to build collision-free schedules in one (where m is the maximum backoff stage of typical value m = 5) and use another random where CW(k) = 2 k CW min is the contention window at backoff stage k. Otherwise, the successful station will then pick a deterministic backoff for its next transmission, 3.1 Supporting many more contenders with Hysteresis, Fair Share and the Schedule Reset mechanism CSMA/ECA is also capable of allocating many contenders in a collision-free schedule by not reseting the backoff stage k after a successful transmission, as opposed to CSMA/CA. That is, a node at backoff stage k would select B d ← CW(k)/2 − 1 as its deterministic backoff after a successful transmission.
This extension to CSMA/ECA is called Hysteresis.
Hysteresis forces some contenders to have larger schedules than others, resulting in an unfair distribution of the network resources. This effect can be compensated by allowing nodes at backoff stage k to transmit 2 k frames, performing MPDU aggregation (AMPDU) and using Block Acknowledgement [13] upon each transmission attempt. We call this extension Fair Share and it ensures an even distribution of the available throughput among contenders. CSMA/ECA is able to outperform CSMA/CA, mainly due to the more efficient collision avoidance mechanism and the aggregation technique suggested by Fair Share.
CSMA/ECA instructs nodes not to reset their backoff stage after a successful transmission. This is done in order to increase the cycle length and provide a collision-free schedule for many contenders, which is desirable in dense scenarios. Nevertheless, having a big deterministic backoff increases the time between successful transmissions. Furthermore, if not operating in a scenario with many nodes the empty slots between transmissions are not longer negligible and degrade the overall throughput of the system. For instance, if nodes withdraw from the contention their previously used slots will now be empty. Contenders should be aware of this issue and pursue opportunities to reduce their deterministic  and the state of the observed slot; which equals to one when busy or zero when idle. After γ consecutive successful transmissions (sxTx), the bitmap is evaluated. If a change of schedule is possible, it is executed just after the next successful transmission.
It is possible to configure Schedule Reset in two modes, namely conservative and aggressive. These modes relate to the number of consecutive transmissions needed to evaluate the bitmap, that is, γ. A conservative SR contains the information of all users' transmissions, therefore no additional collisions are introduced as a consequence of the schedule change 9 . This implies a value of γ = 2 (m−k)+1 . On the other hand, setting γ = 1 triggers a bitmap evaluation after just two consecutive transmissions, rendering this choice of γ the aggressive mode.
Aggressive Schedule Reset coupled with an increase in the Stickiness after an effective schedule change has proven to be suitable for noisy scenarios in real hardware implementations of CSMA/ECA [7]. This work uses the same settings to provide the simulation results in Section 4. Stickiness is not a new concept to CSMA/ECA [11]. It simply instructs the contender to stick to the deterministic backoff even in the event of stickiness number of failed transmissions. This allows for a faster convergence towards a collision-free schedule. CSMA/ECA with a default level of stickiness equal to 1 has proven to provide the better combination of high throughput and low collisions, as shown in Figure 2.
To summarise, simulations results presented in Section 4 use Aggressive Schedule Reset and increase Providing priority is to ensure a more frequent access to some ACs over others. In CSMA/ECA this is only subject to the deterministic backoff. That is, an AC using a shorter B d would in turn access the channel more often than those using a larger one. To maintain compatibility with EDCA, CSMA/ECA considers the same four ACs.
Nevertheless, AIFS and TXOP are not fit for multiple CSMA/ECA queues. For instance, AIFS values are not required since differentiation is only provided by the deterministic backoff. The incorporation of different AIFS for each category would trigger Virtual Collisions that in turn may disrupt an existent collision-free schedule with real collisions. Figure 3 shows a VC in CSMA/ECA with two queues (indicated by the outline) consequence of using AIFS during a collision-free schedule. As the lower priority AC proceeds to select a random backoff, its next transmission may disrupt any ongoing collision-free operation.
TXOP in EDCA ensures that all traffic from the same category receives on average the same channel time. In contrast, CSMA/ECA's goal through Fair Share is to provide close to equal average throughput to same-priority ACs. The combination of Fair Share and Schedule Reset provides throughput fairness through aggregation. Further, it attempts to evenly distribute the channel time among AC increasing  BK  32  1024  5  15  511  BE  32  1024  5  15  511  VI  16  512  5  7  255  VO  8  256  5  3  127  Legacy  32  1024  5   the frequency of transmissions, permanently seeking opportunities to reduce the schedule. In order to provide a fair comparison with EDCA, Section 4 also shows simulation results for CSMA/ECA using the default TXOP.
As EDCA extends DCF into four ACs, similarly there is an instance of CSMA/ECA for each AC.
We will refer to CSMA/ECA with multiple ACs as CSMA/ECA QoS from here forward. Table 3 shows the CW, lowest and largest B d , and maximum backoff stage m.  Figure 4 indicates an VC in STA-1, which is resolved allowing AC1 and deferring AC2's transmission using a random backoff with a doubled CW. A future collision between STA-2's AC1 and AC2 from STA-1 is highlighted by the last outline.

Collisions and Virtual Collisions-free operation using Smart Backoff
Consider a complete schedule of length C = 2 m CW min , and m = 5. With CSMA/ECA and a single AC is possible to allocate a collision-free transmission slot for up to C/2 = 512 contenders (the highest B d +1 for AC Legacy in Table 3  What results from Algorithm 1 is a Smart Backoff counter that will not cause a VC on the next transmission attempts. 10 The maximum number of collision-free contenders in saturation is reduced when using the Schedule Reset Mechanism. This is due to the reduction of the average backoff stage of AC

Performance Evaluation
In order to test the traffic differentiation in CSMA/ECA QoS and its capability of outperform EDCA in terms of number of supported delay-sensitive flows and aggregate throughput, we have used a customised version of the COST simulator [30], which is available via [31]. If not expressed otherwise, each point in the presented figures is obtained from averaging twenty executions of duration equal to forty seconds. Further considerations: • PHY/MAC headers, and other unspecified parameters follow the IEEE 802.11ax (5 GHz) standard [32].
• All nodes can be assumed to be in communication range with each other.
• Transmission of several frames per attempt supposes AMPDU aggregation with compressed Block ACK [13].
• RTS/CTS mechanism is used, as transmitting multiple frames in a TXOP requires a protection mechanism in EDCA [4].
• Smart Backoff is used in CSMA/ECA QoS .
• Dynamic Stickiness defines a maximum stickiness = 2. The RTS/CTS message exchange between transmitter/received was originally intended to solve the hidden node problem in WLANs [13]. However, it also has advantages for a large number of contenders, as it reduces the collision duration, which compensates for the RTS/CTS overhead. Initially, a transmitter enters in contention in order to send a short Request to Send (RTS) message to the receiver. Consequently, the receiver performs contention to respond with a Clear to Send (CTS) message (which is received by all contenders), allocating the next TXOP to the transmitter. Upon reception of the CTS message, the transmitter is granted contention free access to the channel during TXOP. Collisions using the RTS/CTS mechanism are shorter that using Basic Access (BA) (in which collisions are normally assumed to occupy as much channel time as a successful transmission), given the short size of RTS and CTS packets.
Additionally, Table 4 provides information about relevant PHY and MAC parameters used in the simulator.
Apart from the assumptions presented above, the following provide details about the traffic source generators, channel conditions and overall scenarios to be evaluated. Then, simulation results for achieved throughput, number of collisions and time between successful transmissions are presented.

Traffic conditions
There are two main scenarios regarding traffic generation in a node. The saturated traffic condition refers to a node that always has a packet for transmission in its MAC queue. On the other hand, a non-saturated node empties its MAC queue and withdraw from the channel contention. These states do  [24].
Its improved compression tools makes it ideal for high quality video streaming. Video source modelling greatly depends on the video source, that is, action films after packetised produce very different frames than a static interview. This results in rate variability. As also tested in [24], an example Group of Images (GOP) representative of an action movie source is selected 11 . A GOP is composed of I, P and B frames, used to represent past, present and future in a video stream. For a given image quality (PSNR) and size (in pixels by pixels), Table 5 shows the average and standard deviation of the I, P and B frame sizes, alongside other video source characteristics.
• AC[BE] and AC[BK] sources: queues are saturated in all scenarios.
As the goal of the saturated scenario evaluation is to compare the efficiency of the contention mechanisms used by EDCA and CSMA/ECA QoS , all ACs use circular MAC queues, which are filled at startup 11 Due to its higher rate variability.

Channel errors
The inability to receive an ACK frame is handled as a collision, both in EDCA and CSMA/ECA QoS . This could happen due to channel imperfections preventing the receiver from decoding the transmissions. In order to simulate the effects of channel errors over the MAC protocol, we define the likelihood of a MPDU not being acknowledged, p e . It affects every MPDU independently. That is, for every transmission we draw a number from a random variable X ∼ U[0, 1], if the number drawn is lower than p e the frame will not be acknowledged. In the case of MDPU aggregation (AMPDU), it is considered a failed transmission only if all MPDUs in the AMPDU are independently affected by p e . A value of p e = 0.1 has been selected for the simulation of the non-saturated scenario, but a comparison with different values for p e is also provided. The saturation scenario is tested with a perfect channel.  As shown in the figure, CSMA/ECA QoS+FS is able to keep a steady overall throughput for a large number of contenders in saturated conditions. Moreover, as ACs aggregate frames proportionally to its 12 The JFI is an indicator of fairness regarding the ditribution of the available throughput in a system. As the throughput in WLANs is to be equally distributed among contenders, a JFI= 1 is expected. current schedule length, throughput fairness is achieved for high priority ACs. Collision-free operation is reached for N ≤ 12, as shown in Figure 5-b. This is lower than the maximum of N = 32 mentioned in Section 3.3 and is a consequence of Schedule Reset's γ = 1. For N ≤ 12, SR often fails to encounter further reduction opportunities, often succeeding keeping ACs with shorter schedules than the maximum.

CSMA/ECA performance evaluation
At higher N > 12, the aggressiveness of SR due to γ = 1 leads to schedule reductions that cause collisions.  Figure 6-a, the difference between selecting Half the current schedule and looking to reduce it to the Smaller available length are not significant in terms of average final backoff stage. Nevertheless, a reduction is observed when compared against not using SR. Looking at the average time between successful transmissions, the Half configuration provides better results given that a drastic reduction of the schedule increases the collision probability when using γ = 1. As this value of γ is required in order to increase the reduction attempts in non-saturated conditions with p e > 0, the Half configuration is used. That is, Schedule Reset will evaluate the bitmap and only perform a reduction to half the current deterministic backoff.
Smart Backoff prevents virtual collisions and consequent disruption of collision-free schedules. As shown in Figure 6-b, collision-free operation is only achieved with SB, and for N ≤ 14 during simulation time. 13 .
In non-saturation, CSMA/ECA QoS+FS in Figure 5 is able to construct collision-free schedules for short periods of time that allow AC[VO] and AC[VI] to saturate at a much higher number of contenders. Further, as shown in Figure 5-d the queueing delay of the highest priority AC[VO] is lower than other ACs. The value of p e = 0.1 is selected because it produces a moderate increase in the total number of failures observed in Figure 7, where a range of p e > 0 with a fixed N = 1 are tested. As nodes are supposed to be in communication range among each other, we avoid using higher p e values.   Figure 9 shows the percentage of empty, successful and failure slots observed during a simulation in saturated conditions. Despite clearly outperforming EDCA for high number of contenders, CSMA/ECA QoS+FS shows lower overall throughput for N ≤ 5 in Figure 8-b. This is due to Fair Share, which aggregates according to the current backoff stage 14 . As collisions are quickly eliminated with Smart Backoff, the level of aggregation produced by Fair Share is often lower than TXOP[AC], hence the lower throughput.

EDCA comparison and coexistence
Turning to the non-saturation scenario, Figure 10 shows the average aggregate throughput, fraction of failures, and time between successful transmissions as rows i = (1, 2, 3), using labels j = (a, b, c, d) to identify each AC as a column. Subfigures are referred as Figure 10.i.j.
In Figure 10

CSMA/ECA QoS+TXOP
Fair Share aggregates up to 32 frames in an AMPDU, nevertheless, the variable-size video frames proposed for the non-saturation scenario often sum up to more than the maximum TXOP limit defined for EDCA (see Table 2). Conversely, EDCA aggregates more packets at lower number of nodes. As Fair Share performs aggregation according to the AC's schedule length, at N ≤ 12 CSMA/ECA QoS+FS ACs reach collision-free operation with short schedules. 16 From the point of view of each AC.  Looking at the bottom of Figure 12, CSMA/ECA QoS+TXOP clearly outperforms EDCA in the nonsaturation scenario, besides, its average time between successful transmissions is practically equal to the one observed in Figure 10.3.
As Figure 11 in Section 4.3.1, the new Figure 13 shows a Mixed Scenario where 50% of nodes use EDCA, while the other 50% use CSMA/ECA QoS+TXOP . The figure shows that the interaction among nodes with different protocols is pretty much the same as when using CSMA/ECA QoS+FS .

Discussion
After the analysis, it is clear that the number of contenders, channel and traffic conditions play a main role in the performance of both MAC protocols.
A perfect channel, RTS/CTS, and low number of contenders are ideal conditions for EDCA in saturation. Nevertheless, CSMA/ECA QoS+TXOP ACs converge into collision-free schedules with different lengths. Despite Schedule Reset's efforts to reduce the schedule length, ACs rapidly reach collisionfree operation and no further reduction is possible without introducing new collisions, producing the throughput oscillations observed in Figure 12 at N ≤ 10. This issue is normally solved with Fair Share.
Interestingly, using TXOP aggregation instead of Fair Share produces higher throughput for low number of nodes (despite the irregular throughput distribution), and as TXOP [VO] transmissions are shorter the overall delay of lower priority AC's transmissions is reduced when compared against Fair Share.
As scenarios become crowded with more contenders, CSMA/ECA QoS+TXOP consistently outperforms EDCA (see Figure 12). Further, it shows lower fraction of failed transmissions for a considerably higher number of contenders. Failed transmissions and non-saturated sources keep changing the structure of Schedule Reset's bitmap, providing more opportunities to reduce the deterministic backoff. Finally, a priority inversion is observed at the bottom row of Figure 12, where EDCA AC[VI] shows lower average time between successful transmissions than AC[VO], which is almost starved due to the tight contention parameters.
CSMA/ECA QoS+TXOP results suggest it is better than EDCA for crowded scenarios, specially if: • Traffic differentiation is to be ensured for high number of contenders.
• Transmissions from low priority ACs are not to be starved.
• To prevent AC priority inversions.
From the point of view of EDCA nodes in the mixed scenario, the deterministic backoff used by the other 50% of CSMA/ECA QoS+FS nodes during collision-free periods produce an increase in the number of empty slots. More empty slots imply lower probability of collisions. This means that sharing the network with CSMA/ECA QoS nodes reduces the collision probability for EDCA nodes. Therefore, the number of successful transmissions from low priority ACs is expected to be higher than in the EDCA-

Conclusions
EDCA is able of providing effective traffic differentiation in WLANs. It does so instantiating DCF for each of its four supported MAC queues, or Access Categories (AC), and defining different contention and transmission parameters that allow an statistical differentiation among them.
Results highlight EDCA's problems at serving many contenders with multiple ACs. Specifically, its contention mechanism being based on a random backoff is in principle unable to eliminate collisions that degrade the overall performance of the network. Strict differentiation techniques like AIFS, and the additional transmission deferrals due to Virtual Collisions starve low-priority ACs in terms of throughput.
Further, apart from low priority AC starvation, high priority AC inversion is observed with high number of contenders. CSMA/ECA QoS is able to construct collision-free periods that provide an overall throughput increase, while still providing Contention Window-based traffic differentiation for many more contenders. That is, CSMA/ECA QoS is able to bring traffic differentiation to crowded networks without killing the throughput of low priority ACs, as EDCA does. Further, because both protocols use similar contention parameters, CSMA/ECA QoS can coexist with EDCA nodes in the same network and still enjoy higher throughput and traffic differentiation.
Authors are even more confident that CSMA/ECA QoS can be a suitable replacement for EDCA for high number of contenders because: • Suppose a simple modification to the existing backoff mechanism of EDCA, and therefore DCF.
Suggesting that implementation on real hardware may only require customisation of existing EDCA firmware code, as done with DCF in [7].
• Is able to support many more high priority flows for a higher number of contenders, making it suitable for the crowded scenarios envisioned for upcoming standard amendments, like 802.11 ax [3,32].
• Coexistence with EDCA nodes in the same network do not impose costly degradation on the performance. In fact, reduces the collision probability of EDCA nodes allowing them to achieve higher throughput than in an homogeneous network.
Even-though our proposal is backwards compatible with EDCA, authors strongly believe in MAC protocol reconfigurability, as done with Wireless MAC Processors using MAClets [22,35]. We envision WLANs scenarios where backwards compatibility is no longer an issue because users download the MAC protocol from the AP (which can be selected according to different considerations, like: number of users, QoS, privacy, delay, among others). Finally, the most important lesson learned from this work is that there really is no "One-Fits-All" MAC protocol for all WLAN scenarios. We believe that reconfigurability using Software Defined Network-like strategies are the path to follow for future WiFi scenarios.