Advanced Wireless Networks - 4G Technologies

818 QUALITY-OF-SERVICE MANAGEMENT

Data

Figure 21.1 Watermark embedding process.

Figure 21.2 Principle scheme of tracing watermarking for coder-channel quality assessment in multimedia communications.

data onto a channel, the mark undergoes the same alterations as suffered by the data. At the receiving side, the watermark is estimated and compared with the original. Since the alterations endured by the watermark are also likely to be suffered by the entire data, as they follow the same communication link, the watermark degradation can be used to estimate the overall alterations endured by the data.

The tracing watermarking procedure for coder-channel quality assessment is given in Figure 21.2. The watermark embedding is performed by resorting to the spread-spectrum technique proposed in Cox et al. [17] for still images and applied to video sequences in Hartung and Girod [18]. The watermark is a narrowband low-energy signal. It is then spread so that it has a larger bandwidth. Consequently, the watermark energy contribution for each host frequency bin is negligible, which makes the watermark imperceptible.

In the application addressed here, a system embedded into the data stream that is able to trace the degradations introduced by the transmission system composed by the coderchannel cascade and not perceptually affecting the data themselves is described.

A set of uncorrelated pseudorandom noise (PN) matrices is multiplied by the reference watermark. Both the PN matrices and the embedded watermark are known at the receiving side. The watermark is the same for each video sequence frame, whereas the PN matrices are different for each frame. This insures that the spatial localization of the mark is different frame-by-frame so that the watermark visual persistency is negligible.

After the randomization of the watermark by the PN matrices, the embedding of the tracing marks is performed in the DCT domain. The watermark is embedded in the DCT middle-band frequencies of the whole image. After the inverse DCT (IDCT) has been performed on each frame, the whole sequence is coded by a video coder and, finally, transmitted.

At the receiving side, the video is first decoded; then, from the DCT of each received frame of the sequence, a matched filter extracts the spread watermark, which is finally despread using the known PN matrices. After having extracted the received watermark, it is matched to the reference one, which is known at the receiving side, and the mean-square error (MSE), between the original mark and the received one, is used as an index of the degradation affecting the received watermark.

21.1.1 System modeling

A two-dimensional video sequence will be represented as { fi [n1, n2], i = 1, 2, . . . M}. The sequence is composed of M frames fi [n1, n2] of N1 × N2 pixels. Let ω[k1, k2] be the employed watermark, having dimensions K1 × K2 and { pi [k1, k2], i = 1, 2, . . . M} the M PN matrices of dimensions K1 × K2. The watermark ω[k1, k2] is a visual pattern, like a logo, consisting of binary pixels.

The PN matrices { pi [k1, k2], i = 1, 2, . . . M} are employed to spread a narrowband signal (watermark) over a much larger bandwidth signal (host frame) in such a way that

the watermark energy is undetectable in any single frequency of the host image. This can be expressed as follows: ωi(s)[k1, k2] = ω[k1, k2] · pi [k1, k2], i = 1, 2, . . . , M, where

ωi(s)[k1, k2] is the spread version of the watermark to be embedded in the ith frame. Let Fi [k1, k2] = DCT{ fi [n1, n2]} be the DCT transform of the ith frame fi [n1, n2]. The spread watermark is embedded, in the DCT domain, in the middle-high frequency region S of

Fi [k1, k2].

The embedding is performed in the DCT domain according to the following rule:

where Fi(ω)[k1, k2] represents the DCT of the ith watermarked frame fi [n1, n2], and α is a scaling factor that determines the watermark strength. Parameter α must be chosen in such a way as to compromise between not degrading the picture on one side and being detectable by the tracing algorithm on another.

The ith watermarked frame is then obtained by performing the IDCT transform

822 QUALITY-OF-SERVICE MANAGEMENT

To expand support for applications with QoS requirements, the IEEE 802.11E task group [37] is proceeding to build the QoS enhancements of the 802.11 MAC. The techniques for providing prioritized and parameterized QoS data deliveries have been discussed in the task group. An enhanced DCF was proposed, where each traffic flow is assigned with a different backoff time whose value decreases with increasing traffic priority, to achieve the prioritized QoS data delivery. To guarantee the bounded delay requirements in the parameterized QoS data delivery, the hybrid coordination function (HCF) was proposed [42]. In the HCF, an STA can be guaranteed to issue the frame transmission even during the CP using the contention-free burst (CFB). The CFB can be considered a temporary CFP during which the transmissions of STAs are coordinated by the polling scheme as the PCF. Moreover, a multipolling mechanism (called contention-free multipoll, CF-multipoll) was proposed [41] to reduce the polling overhead that the traditional PCF suffers from. In the multipoll, the PC can poll more than one STA simultaneously using a single polling frame.

In this section, we consider how to efficiently serve real-time traffic in the IEEE 802.11 WLAN by using the multipolling mechanism. The multipolling mechanism can increase the channel utilization and is robust in mobile and error-prone environments. The mechanism can be used in the PCF and the HCF. Moreover, a polling schedule is provided to guarantee the bounded delay requirements of real-time flows.

21.2.1 Contention-based multipolling

In the PCF or the HCF, each STA in the polling list takes a polling frame when polled. This polling scheme is called SinglePoll. The number of polling frames for a polling list can be reduced if a multipolling mechanism is used. Here, we discuss an efficient multipolling mechanism that has the advantages of high channel utilization and low implementation overhead. This multipolling mechanism, referred to as contention period multipoll (CPMultipoll), incorporates the DCF access scheme into the polling scheme.

In the DCF, any contending STA for the channel will select a backoff time in units of slot times and execute the backoff procedure as follows: if the channel is sensed idle for a DIFS period, an STA starts the transmission immediately. Otherwise, an STA should defer until the channel becomes idle for a DIFS period and then the backoff time is decreased. If the channel is idle for a slot time, an STA decreases the backoff time by one or else freezes the backoff time. When the backoff time becomes zero, an STA begins the frame transmission. If a collision occurs, the STA duplicates the backoff time used in the last transmission and executes the backoff procedure again. In the DCF, the virtual carrier sensing using the Network Allocation Vector (NAV) is performed at the MAC sublayer. The information on the duration of a frame exchange sequence for one STA is included in the frame header (duration field) and is announced to other STAs. Other contending STAs will wait for the completion of the current frame exchange by updating their NAVs according to the announced duration information. For a communication the RTS (request to send) and CTS (clear to send) frames are exchanged before the data frame transmission. Other STAs defer their channel access by setting their NAVs according to the duration field in the RTS, the RTS, or the data frame. The exchange of RTS/CTS frames can also avoid the hidden terminal problem [35].

The basic idea of CP-Multipoll is to transform the polling order into the contending order which indicates the order of winning the channel contention. Different backoff

time values are assigned to the flows in the polling group and the corresponding STAs execute the backoff procedures after receiving the CP-Multipoll frame. The contending order of these STAs is the same as the ascending order of the assigned backoff time values. Therefore, to maintain the polling order in a polling group, we can assign the backoff time value incrementally according to the expected polling order.

The CP-Multipoll has the following advantages: a polled STA can hold the channel access flexibly depending on the size of local buffered data, so it becomes easy to deal with the data burst; if a polled STA makes no response to the CP-Multipoll, other STAs in the same polling group will detect the channel idle right away and advance the starting of channel contention. Therefore, the CP-Multipoll can decrease the waste of channel space and can afford any polling error.

21.2.2 Polling efficiency

We consider the environment with overlapping BSSs. In the overlapping BSS, we distinguish the STAs associated with one BSS from the STAs associated with neighboring BSSs by using the terms ‘internal STAs’ and ‘external STAs’, respectively. Moreover, a STA that will cause an internal collision is called a nonbehaved STA; otherwise, the STA is called a behaved one.

The following terminology is used in the performance analysis:

frame num, maximum number of data frames allowed to be transmitted in a TXOP (transmission opportunity);

lt ype, number of bits in a ‘type’ frame;

ttype, transmission time on the channel for a ‘type’ frame;

single-poll, the single polling frame in the single poll (i.e. CF-Poll frame);

n, poll size (i.e. the number of poll records in a multipolling frame);

n-poll, the multipolling frame with n poll records;

ERRtype, probability that a ‘type’ frame is dropped due to bit errors;

α, probability that an STA has no data to send when polled;

β, probability that an STA becomes a nonbehaved STA;

h, number of PCs operating in the overlapping space;

InitBT, initial backoff time;

E, polling efficiency.

The following assumptions will be used in the analysis:

(1)The PC (point coordinator) always performs the initial backoff before sending any polling frame.

(2)The data frame is transmitted without any acknowledgment. When an STA is polled, the STA either sends a half of frame num data frames on average or sends a null data frame if there is no data to send.

Figure 21.5 The state diagram of the multipolling mechanism. (Reproduced by permission of IEEE [43].)

(3)There is an equal probability BER for a bit error to occur due to the channel noise

(interference, fading or multipath). Hence, ERRtype can be expressed as 1 − (1 −

BER)ltype .

(4)An STA becomes a nonbehaved one if the STA fails to receive the CTS frame from the PC. That is, β = ERRCTS.

If AvgD and AvgT denote the total number of bits in the data frames successfully sent from the polled STAs and the average complete time in time units for a poll, respectively, then the polling efficiency is defined as E = AvgD/AvgT. This represents the average uplink data rate during a poll.

In the single poll, a polled STA contributes data frames if the STA successfully receives a CF-Poll frame and has pending data frames to be successfully transmitted. The polled STA may suffer frame error due to interference from external STAs. Therefore,

AvgD = (1 − α) · frame num/2 · ldata− frame

The polled STA will give a response to the PC after an SIFS period for a successful CF-Poll. If a polled STA does not respond to the PC after a PIFS period for a failed CF-Poll, the PC takes over the channel control and may send the next CF-Poll frame. In the HCF, the RTS/CTS frames should be exchanged before the data transmission to prevent the interference from other STAs. Since the HCF is being substituted for the PCF in the 802.11E, we consider the single poll of the HCF in our analysis. Therefore,

AvgT = InitBT + (tsingle-poll + PIFS) · ERRsingle-poll + [tsingle-poll + SIFS

+ tRTS + tCTS + 2SIFS + (1 − α) frame num/2 · (tdata frame + SIFS) (21.6)

+α(tnull frame + SIFS)] · (1 − ERRsingle-poll)

Next, we use a state diagram to represent the situation after sending a multipolling frame with poll size n. In Figure 21.5, the state m : fail(0 ≤ m ≤ n) represents the event that there are m STAs in the polling group which failed to receive the multipolling frame. Let Pn,m denote the probability of the state m : fail. Let Dn,m and Tn,m denote the amount of uplink data frames in bits and the total time duration in time units under the state m : fail, respectively. Generally, we have the following values:

For the CF-Multipoll, the external STAs in the overlapping BSS will have their TXOPs overlap the ones allocated to the internal STAs. We assume that the interference from external STAs has been included in the parameter BER. Hence,

Dn,m = (n − m) · (1 − α) · frame num/2 · ldata frame · (1 − ERRdata frame)

Also, each successive TXOP starts an SIFS period after the predecessor’s TXOP limit expires in the CF-Multipoll. Note that the time spent by a polled STA is fixed regardless of the number of pending data frames. Hence,

In the CP-Multipoll, there are two processing phases: one is for the normal multipoll and the other is for the error recovery. In the normal phase, there are (n − m) STAs successfully receiving the multipolling frame under the state m : fail. Among these STAs, (1 − β)(n − m) STAs are behaved STAs and β(n − m) STAs are nonbehaved ones. Each nonbehaved STA is assumed to destroy one data frame transmitted by a behaved one. In the recovery phase, m STAs failed to receive the multipolling frame and (n − m) nonbehaved STAs will be served individually using the CP-Multipoll with poll size one. The analysis of this phase is similar to the one in the single poll. Hence, Dn,m has the following value:

The length of the time of the normal phase is dominated by the transmission time of those (1 − β)(n − m) behaved STAs. The total backoff time consumed in the normal phase is h × n + 1, regardless of the value m. Hence,

+α · tnull frame)Tnrecovery,m = [m + β(n − m)] · [(t1−poll

+2 Slot) · ERR1−poll + (t1−poll + 2 Slot + tRTS + tCTS + 2 SIFS + (1 − α)

frame num/2 · (tdata frame + SIFS) + α · tnull frame)

(1 − ERR1−poll)].

The superpoll can poll a group of STAs together as the CF-Multipoll and the CP-Multipoll. In the superpoll, the polled STA will attach the poll records of those polled ones whose polling orders are after it to its current transmitted data frame. This scheme can be considered

826 QUALITY-OF-SERVICE MANAGEMENT

Table 21.1 System and poll-related parameters. (Reproduced by permission of IEEE [43])

as one with replicated poll records. To keep the correct polling order, each polled STA in the polling group should monitor the channel and check whether the previous STA has finished sending a data frame. If a polled STA fails to receive its poll record, the next following STA in the polling group will wait for a timeout before its own frame transmission. However, the situation where some STAs cannot listen to other STAs’ channel activities is not considered. This may cause inconsistent setting of timers among polled STAs and may cause internal collisions.

21.2.2.1 Performance example

The parameter settings related to system and the polling schemes are listed in Table 21.1 [43]. The performance improvement of CP-Multipoll in percentage over other schemes is shown in Figure 21.6.

21.3 DYNAMIC SCHEDULING ON RLC/MAC LAYER

In wireless networks, the fading characteristics of the wireless physical channel may introduce location-dependent channel errors. A scheduling algorithm, referred to as channelcondition independent packet fair queueing (CIF-Q), is presented in Ng et al. [44] to solve the problem of location-dependent channel errors in wireless networks by suspending