CHAPTER 1 INTRODUCTION
2.3 Paper [20]
2.3.2 Limitations
Shi-Hua leads simulations with his mechanism and the results show a large improvement of the fairness of a network. Meanwhile, its protocol has some limita-tions.
Firstly, the capture effect in mixed downlink/uplink transmissions is not considered. As we will see in this paper, during a downlink transmission the AP may be able to capture the uplink transmissions of the stations in its BSS. This capture effect induces a STC fairness that has to be considered in the computation of the contention window sizes.
Secondly, two stations in the same BSS may not have the same under-standing of the capture effect and hidden node relationships of the whole network.
Indeed, when stations receive the effects relationships from the other stations, they record the information into two arrays. The number hidden array (NHA), records the numbers of hidden stations send from other stations and the number hidden capture array (NHCA) records the numbers of simultaneously hidden and captured stations sent from other stations. Then, the target of stations is to obtain the matrices HM and CM defined as follows:
• Hidden matrix (HM) stores the hidden information of the stations. The size of HM is N × N, where N is the number of total stations in the BSS. The element (i,j) of HM represents the hidden relation of station i and station j:
HM(i,j)=1, if station i and station j are hidden to each other
HM(i,j)=0, if station i and station j are not hidden to each other
Therefore, HM must be a symmetric matrix
• Capture matrix (CM) stores the capture information of the stations. The size of CM is N × N, same as HM. The element (i,j) of CM represents the capture relation of station i and station j:
CM(i,j)=1, if station i captures station j CM(i,j)=0, if station i do not capture station j
We denote the matrix, which are guessed by stations, as guessed hidden matrix (GHM) and guessed capture matrix (GCM), respectively. A station computes these matrices following an algorithm which only uses the value of NHA and NHCA.
For example, if the inputs are:
N HA = [2 2 1 1] and
doi:10.6342/NTU201701832
As shown in the example, these matrices are not unique and a station will get one pair of the GHM and GCM by the algorithm, which may be different from the actual HM and CM. In addition, two stations of the same network, which receive the same NHA and NHCA, may not compute the same HM and CM. To conclude, due to this centralized approach, the stations from the same network may not have the same understanding of the capture effect and hidden node relations among stations.
Therefore, their computation of the optimum CW may be wrong and the desired STC fairness or throughput may not be reached.
In this paper, we will present a complete new mechanism which avoids the drawbacks of [20]s. Our fully distributed mechanism first aims to provide the same and complete understanding of the capture effect and hidden node relationships to all the stations from the same network in uplink and downlink transmissions. Once the detection part is complete, each station will adjust its CW according to these relationships with the intention of maximizing the total throughput of the network .
GSR WITH CAPTURE EFFECT ONLY
In this paper, we consider a IEEE 802.11af networks in infrastructure mode.
Stations are randomly located in the transmission range of the AP and have always something to transmit. Mixed downlink and uplink transmissions are considered in which all stations transmit the same payload length at the same data rate. For downlink transmissions, we assume that a transmission of the AP is addressed to the different stations of the network with the same probability. In this chapter, we consider network without hidden node problem. ie. every station can hear each other.
3.1 Different level of capture effect
The capture effect between stations and between stations and AP is differ-ent. For uplink transmissions, if a device A captures another device B, then device A will always capture device B. However, for a collision between an AP in downlink transmission and a station in uplink transmission, the capture effected is more com-plex, as shown in Fig. 1. In Fig. 1 b) the AP and station B attempt a transmission at the same time. At the recipient of APs transmission, here station A, the signal from the AP is stronger than Bs signal so station A can decode its signal and answer with an ACK to the AP. In this case, the AP captures the transmission of station B. But, in Fig. 1 c), the AP and station B still attempts a transmission at the same time, at the difference that the transmission of the AP is addressed to a different station called station C. At the recipient of APs transmission, here station C, the signal from the AP is not enough stronger than Bs signal so station A cannot decode its signal and do not answer with an ACK to the AP. At the same moment, the AP, which is in
doi:10.6342/NTU201701832
3.2. OUR CAPTURE EFFECT DETECTION MECHANISM 25
transmitter mode, does not hear the transmission of B. There is no emission of any ACK inducing no capture effect and the both transmissions failed. In conclusion of this last category, there is no way a station can capture the AP, so we do not have to consider this case, but when the AP captures a station we have to know which station is the receiver of its transmission.
Figure 8: Topology of mixed downlink/uplink transmissions for the two different cases of capture effect.
3.2 Our capture effect detection mechanism
In order to detect the capture effect relationships of the whole network, our mechanism proposes the following modifications of the original DCF:
• every station has a non-zero fixed contention window [1, CW].
• if a station attempts a transmission and receives an ACK addressed to another station instead of receiving a ACK addressed to itself, this station will transmit again its data at the end of the ACK as if its back-off counter is still equal to 0.
In the case of a transmissions collision in uplink transmissions only, the capture effect is illustrated in Fig. 9. Station A receives the right acknowledgement meaning that its transmission is successful but A does not know it has collide with B yet. From B’s perspective, it transmits but receives an ACK addressed to A and understands its has been captured by A. Based on our protocol, station B transmits again directly after the ACK addressed to A and its transmission is successful because
A has to choose a non-zero back-off value. Station A will in turn hear the retrans-mission of B and understand it has captured station B during its last transretrans-mission.
Figure 9: Capture event detection with the new mechanism in uplink transmissions.
The capture effect detection protocol is exactly the same when it occurs in a mixed uplink/downlink transmissions collision. In addition, if two stations collides without capture effect, there is no ACK or if there is just a successful transmission one back-off slot has to elapse to announce that no collision happened.
Based on this detection mechanism, devices will know the number of devices they captured NC and the number of devices they are captured by NBC . According to these two parameters, each station will select an optimum CW as their STC will be equal and the total throughput will be optimized. In the next section we will derive the optimum CW for every station with the Markov Chain.
3.3 Computation of the contention window .
3.3.1 Markov Model
Once the devices know the capture effect among each others, they have to choose a contention window based on their capture effect relationships. In order to simplify the protocol, devices do not increment their CW size as in original DCF. We establish a Markov chain in Fig. 20 of our protocol inspiring by [1]. and that reflects
doi:10.6342/NTU201701832
3.3. COMPUTATION OF THE CONTENTION WINDOW . 27
the main modification announced in part III. to calculate the CW for each devices.
Figure 10: Markov chain of the new protocol
The different states of our Markov chain are described below:
•
O
: The device attempts a transmission.•
O
T : The device keeps its back-off counter to 0 and transmits for the second time in this transmission attempt, directly after waiting DIFS after the last ACK received.•
O
W : The device waits a transmission time before counting down its back-off.The Markov Process of Fig. 20 for a station i is governed by the following transition probabilities:
P (k/k + 1) = 1 , k ∈ [0, CW − 1]
P (0T/0) = Pbci , P (0W/0) = Pci ,
P (k/0T) = P (k/0W) = 1/CW, k ∈ [1, CW ] P (k/0) = (1 − Pci+ Pbci))/CW, k ∈ [1, CW ]
(3.1)
Let bis = limt→∞P (station i is in state s) denotes the stationary distribution of the chain. The probability bi0 represents the probability that the back-off counter of station i reaches the value 0 in a random chosen time slot.
In the rest of our study, we will consider that for large networks:
CWi 1 ∀i ∈ {N } ⇒ bi0 = bj0 ∀i, j ∈ {N } (3.2) The different probabilities that appear in the Markov chain can now be defined and simplified considering (2) :
• Pci is the probability that station i captures 1 station knowing that station i collides with 1 station. where NCi is the total number of stations that station i captures,NAPi is the number of stations for which AP capture station i when it transmits to.
• Pbci is the probability that station i is captured by 1 station knowing that station i collides with 1 station:
Pbci = bi0∗ (
where NBCi is the total number of stations that capture station i.
We now want to obtain a closed-form solution for our new protocol Markov chain.
The first step is to express all the steady state probabilities as a function of bi0:
bi0T = Pbci ∗ bi0 (3.6)
bi = Pi ∗ bi (3.7)
doi:10.6342/NTU201701832
3.3. COMPUTATION OF THE CONTENTION WINDOW . 29
The other stationary probabilities for any k ∈ [0, CW − 1] follow by resorting to the state transition diagram: Knowing that we are working with large networks which involve big contention win-dows, the right hand term at the denominator can be neglected and the equation becomes:
bi0 ' 2
CWi+ 3 (3.11)
The last step is to extract from the Markov chain closed-form the probability of transmission attempt in random given time slot P0i of a station i. In our mechanism, every time a station is captured, this station is going to directly attempt a new transmission as shown in Fig. 9. Therefore, according equations (3),(5) and (10), the probability of transmission attempt is equal to:
P0i = 2
CWi+ 3 ∗ (1 + (NBCi + NAPi
N − 2) ∗ bi02 ∗ (1 − bi0)N −2) (3.12)
3.3.2 STC Fairness
The STC fairness in a network {N } is defined by:
PSi = PSj ∀i, j ∈ {N } (3.13)
where PSi is the probability of successful transmission of station i when station i attempt a transmission. For each station following our new protocol, this probability is equal to : We consider now a station neutral called n that has no capture effect relationships, ie. this station does not capture any stations (NCn = 0) and is not captured by any other stations (NBCn = 0). We assume this station has a fixed contention window called CWn that we will be able to modify later to achieve other enhancements as power efficiency or total STC maximization. To achieve the STC fairness we need :
PSi = PSn ∀i ∈ {N } (3.16)
By using the equations (12), (13) and (14) we obtain for every station : 1 − bi0
Considering the large networks assumption made in (2), we finally find that, to achieve STC fairness, the value of the contention window of the AP and a station i from the same network is :
CW = CW + 2 ∗ (Ni + Ni ) + b2 ∗ NAPi
c (3.19)
doi:10.6342/NTU201701832
3.3. COMPUTATION OF THE CONTENTION WINDOW . 31
CWAP = CWn+ b2 ∗
NCAP
X
h=1
NAPh
N − 2c (3.20)
3.3.3 Throughput optimization
We want to maximize the total throughput of the whole network by choosing a optimum value of CWn. In [1], Bianchi formulates the quantity that has to be maximized for a IEEE 802.11 network using the original DCF and without considering the capture effect. According to the different analytical results from part III and IV, we can modify his equation to make it correspond to our mechanism :
S =
PN i=1PSi 1 −QN
i=1(1 − P0i) ∗ 1
(1−QN
i=1(1−P0i)) QN
i=1(1−P0i)) + TCσ
(3.21)
where TCσ is the duration of a collision measured in slot time units σ. Based on our mechanism, each stations knows the capture effect relationships of the whole network and can find the optimum CWnwhich maximizes the quantity S and then the total throughput of the network.
GSR WITH BOTH CAPTURE EFFECT AND HIDDEN NODES
In this chapter, we will consider 802.11af networks with capture effect and hidden nodes.
We enable for all stations the RST/CTS frame exchange to reduce the cost of collisions.
Our main goals are the distributed detection of the capture effect relationships and the hidden node relationships by the whole network. We want all the stations to have the same understanding of the capture effect and hidden nodes relationships among stations.
4.1 Detection mechanism of the capture effect re-lationships in IEEE 802.11af networks with hidden nodes.
In chapter 3, we introduced a distributed detection mechanism of the capture effect relationships in IEEE 802.11af networks without hidden nodes and without the RTS/CTS frame exchange. Now, we take in consideration the hidden nodes and the previous approach cannot be considered. Indeed, when a station is captured its retransmission is not safe any more in an environment with hidden nodes. For instance, as illustrated in Fig. 11, if station B is captured by station A and station B is hidden to station C, then there is a probability that station C also transmits its RTS during Bs retransmission.
This is a reason why we need a safe space to be sure of a successful retransmission and we bring these new modifications of the DCF :
• every time a station access the channel, it reserves the channel for two consecutive
transmissions by setting the duration of its RTS to D1= 2*dataLength+CTS+RTS+5*SIFS+DIFS+1
• if the first transmission of a station is successful, its does not start to transmit its
doi:10.6342/NTU201701832 4.1. DETECTION MECHANISM OF THE CAPTURE EFFECT RELATIONSHIPS IN IEEE
802.11AF NETWORKS WITH HIDDEN NODES. 33
Figure 11: Previous mechanism applied to networks with hidden nodes
RTS for the second transmission but instead it waits for a timeout time T1 equal to RTS+SIFS+1 = 23 slots. After this timeout, the station directly transmits its data (no RTS/CTS exchange).
Figure 12: Creation of a safe space for retransmissions
Our idea is simple, every time a station attempts a transmission it will reserve the channel for two transmissions. Then, if the first transmission of a station is successful, the other stations will not try to transmit during its second transmission and that transmission is assured to be successful. The timeout T1, gives an available space for a station that has just been captured to transmit its RTS and receive the CTS from the AP.
We take a simple example, as illustrated in Fig. 13, where station A captures station B and station C is hidden to B. Station A and B send their RTS at the same time and AP
answers with a CTS addressed to A because A captures B:
• from As perspective : its RTS transmission is successful and so it can start transmit-ting
• from Bs perspective : it receives a CTS addressed to A so it understands it has been captured by A
• from Cs perspective : it receives a CTS with a duration time set to D1, so it sets its NAV to this duration time
Figure 13: Capture effect detection mechanism
At this time point, station B is the only one to know it is captured by A and we want it to pass the information to the other stations. To do that, we force station B to transmit again one time after the reception of the ACK to A. This protocol is similar to the previous mechanism but this time B transmits a RTS with the duration time set to one transmission.
The AP is going to decode this RTS and answer with a CTS. This CTS which has duration time set to D2=2*SIFS+dataLength will inform the whole network as follows:
• from As perspective : it decodes a CTS with a duration time set to D2 during its timeout T1. Then station A understands it has captured station B and sets its NAV to D2.
doi:10.6342/NTU201701832 4.2. DETECTION MECHANISM OF THE HIDDEN NODE RELATIONSHIPS IN IEEE
802.11AF NETWORKS. 35
• from Bs perspective : its RTS transmission is successful and so it can start transmit-ting.
• from Cs perspective : it receives a CTS addressed to station B with a duration time set to D2 while its NAV has already been set to a duration time D1. The value of D2 is bigger than the remaining time of D1, so station C sets a new NAV with a new duration time D2. This changement of NAV from CTS A/D1 to CTS B/D2 makes station C understand that station A captures station B.
With this new mechanism, adapted to networks with hidden nodes, every station has the same understanding of the captured effect relationships of the network. We now want the same result with the detection of the hidden nodes.
4.2 Detection mechanism of the hidden node re-lationships in IEEE 802.11af networks.
4.2.1 Detection of hidden nodes of a station by itself
A station can easily understand if it is hidden to another station or not by analyzing the header frames it receives on the channel. As illustrated in Fig. 14, if station B can decode the RTS and the header of data frames from station A, then station B concludes it is not hidden to A. On the contrary, as illustrated in Fig. 15, if station B cannot decode the RTS and the header of data frames from station A, but just decodes the CTS and ACK frame addressed to station A, then station B concludes it is hidden to A and vice-versa.
4.2.2 Detection of hidden nodes of a station by the whole network
As far, no works have been done to achieve that kind of distributed detection for the hidden node problem. By this kind of detection, we mean that the whole network under-stands by itself an hidden node relation. For instance, in Shi-huas thesis, a station detect only its own hidden nodes, but not the ones of other stations. To achieve this detection, we use the timing properties of IEEE 802.11af networks. We define two periods as illustrated in Fig. 16, where station B is hidden to station A while station C is not hidden to station A.
Figure 14: B no hidden
The first period is called hidden period (HP). When a station attempts a transmis-sion, it first sends a RTS frame and then waits a SIFS time for the CTS response. If a station iattempt a transmission and is hidden to station j therefore station j will sense the channel idle until the reception of the CTS frame for station i. The HP represent this whole period, equal to RTS+SIFS, the channel is idle for station j instead of being busy because of the hidden node relationship between i and j. In Fig. 16, In the IEEE 802.11af standard, a RTS transmission time with the lowest data rate is 408us and a SIFS time is 120us. As a consequence, an HP is 408us+120us=528us=22 backoff slots, where the backoff slot length is 24 us. The second period is called collision period (CP). If a station attempts a RTS frame transmission while one of its hidden station is already transmitting a RTS frame, a collision of RTS frame happens. If there is no capture effect, then the AP cannot decode any frames and the both RTS transmissions failed. We define CP as the possible collision period between two hidden stations, as illustrated in figure. Therefore, a CP is equal to and RTS transmission time and in the IEEE 802.11 standard, CP is 408us=17 backoff slots.
Between the end of these two periods, there is an interesting gap of 5 backoff slots.
Indeed, if station A first transmits its RTS and station B starts transmitting its RTS in the
doi:10.6342/NTU201701832 4.2. DETECTION MECHANISM OF THE HIDDEN NODE RELATIONSHIPS IN IEEE
802.11AF NETWORKS. 37
Figure 15: B hidden
NCP of A , the transmission of A is still successful. The transmission of B does not have any effect on As. A receives a CTS and transmits its data and waits for an ACK from the
NCP of A , the transmission of A is still successful. The transmission of B does not have any effect on As. A receives a CTS and transmits its data and waits for an ACK from the