Data aggregation of mobile M2M traffic in relay enhanced LTE-A networks

4.1 RNs implementation and scheduling

The W _i,a (t) parameter indicating the quality of service (QoS) weight of bearer a of user i at time t of the weighted proportional fair (W-PF) as given in (1). The R _min,a is the bit rate budget (target data rate) and τ _max,a is the end-to-end delay budget (target delay) of QoS class of bearer a, R _avgi,a (t) is the average throughput of bearer a and τ _i,a (t) is the packet delay of bearer a of UE i, ρ _a (t) is a variable with value set to, e.g., 10 if τ _i,a (t) is above the threshold value of bearer a at time t (as the bearer packet delay has reached the threshold), otherwise equal to 1. The value 10 of ρ _a (t) raises the metric value of bearer a by 10 times and ensures immediate scheduling.

In (2), the parameters related to bearer delay are eliminated from (1) due to the fact that in uplink data packet transmission, the information about packet delay at the packet data convergence protocol (PDCP) buffer of the UE (or RN) is not available at the DeNB. The buffer status report (BSR) provides information only about the length of the buffer at the eNodeB. Although, the length of the buffer can be used to estimate the delay [31], but since the scheduling scheme in the CQA scheduler is already designed to prioritize the delay-sensitive GBR bearers by creating a separate GBR bearer list, so the delay estimation is not critical. Hence, the delay estimation is not considered for the scheduling decisions. Furthermore, the RN MAC scheduler is designed to allocate radio resources to the UEs and M2M devices connected to the RN at the Uu link. The round robin scheduling approach is adopted for such type of M2M traffic. All the active devices ready to transmit packets are scheduled in a round robin fashion.

4.2 Proposed M2M data aggregation and multiplexing

In the proposed approach, small data packets are aggregated at the PDCP layer of the RN in order to maximize the multiplexing gain without aggregating the additional headers such as those from the PDCP, radio link control (RLC) and MAC, as shown in Fig. 1. The PDCP layer exists in the UE, DeNB as well as in the RN; it is a part of LTE air interface control and user plane protocols. The major functionalities and services of PDCP layer for the user plane include header compression and decompression, user data transfer, delivery of upper layer packet data units (PDUs) in sequence as well as retransmission of the lost PDCP service data units (SDUs), etc. On the other hand, the control plane services include ciphering and integrity protection and transfer of control plane data. The proposed aggregation and multiplexing scheme can also be applied either at the lower layers of the RN Uu interface such as RLC and MAC, or in the PDCP of RN Un interface. However, the multiplexing gain will be reduced due to the multiplexing of additional headers along with the original IP packets. An aggregation buffer B is created at the PDCP layer of the RN which aggregates the incoming packets according to the size of available transport block − RN Un protocol overhead, see Fig. 2. The RN Un protocol overhead includes GPRS tunneling protocol (G T P=12 b y t e s), user datagram protocol (U D P/I P=28 b y t e s), and layer 2 (4 b y t e s). Later, the aggregated packet is sent to the RN GTP over the Un link. The additional overheads such as the GTP, UDP/IP, PDCP, and RLC are added. Then, from the physical (PHY) layer of the RN Un interface, the aggregated packets are sent to the DeNB.

The proposed approach of multiplexing significantly improves the PRB utilization [32]. However, there are certain constraints regarding latency requirements of high priority M2M traffic such as emergency alerting. Each packet waits until the size of the buffer B reaches the size of (TBS −Un protocol overhead). In a highly loaded scenario, the waiting time is not long due to a high arrival rate. However, in a low-load scenario, there is a comparatively longer delay due to the low arrival rate. Consequently, the performance of delay sensitive M2M applications such as e-healthcare and emergency alerting can be degraded. To tackle this issue, an expiry timer T _max is introduced. The timer is, e.g., set with a fixed value of 9 ms in the current implementation. This means that the buffer serves the aggregated packets after 9 ms at the latest. The value of the timer could also be adaptive, i.e., it can change its value adaptively according to priorities of the incoming packets. For this purpose, the algorithm must be fully aware of various priorities of M2M applications. The description of the proposed scheme is given in Algorithm 1.

When the aggregated packet is received at the DeNB PHY layer, it is sent to the DeNB GTP for de-multiplexing of the aggregated large IP packets. The original small IP packets are then sent to the application servers. The block diagram of the uplink data differentiation algorithm implemented at the DeNB is provided in Fig. 3 (left).

The DeNB also serves the regular LTE-A traffic such as voice and video UEs. The DeNB takes into account the source of the incoming packets. If the packets are received from a standalone LTE-A user, then the PDCP of the DeNB forwards packets to the GTP for tunneling. After tunneling, the packets are routed towards the access gateway (aGW). On the other hand, if the packets are received from the RN, the data packets are sent to the GTP via IP and UDP for de-multiplexing. Later, the original standalone packets are sent to the aGW. Similarly, in downlink, the DeNB GTP recognizes the source and destination identities (IDs) of the incoming packets. If the destination ID of incoming packet belongs to an RN, then the packets belong to a user which is served by the RN. On the other hand, packets belong to a user which is directly served by the DeNB. The block diagram of the downlink data differentiation algorithm implemented at the DeNB is given in Fig. 3 (right).

6.1 Coverage enhancements

One of the major functionalities of the 3GPP LTE-A RNs includes coverage enhancements, especially for the UEs at the cell edge which usually experience poor channel conditions. The DeNB MAC scheduler allocates PRBs to each UE according to the received signal strength. The cell edge users usually experience poor signal strength which ultimately reduces the maximum utilization of the allocated PRBs. This inefficient utilization of PRBs significantly degrades the cell throughput. The RNs play a vital role in improving the QoS performance of the cell-edge UEs.

In order to evaluate the impact of relaying on the QoS performance of the cell-edge UEs, the scenarios are categorized into two groups namely direct-DeNB access and relay-based-DeNB access. In the first category, all the UEs communicate with DeNB directly. Whereas in the second category, an RN acts as an access point for the cell-edge UEs and later routes the UEs data to the DeNB. In both the categories, the UEs and RN are assumed to be stationary which results into non-varying channel conditions. The position of RN corresponds to MCS 16 with a TBS 1608 bits per TTI of 1 ms duration.

Each category is further subdivided into various subscenarios in which LTE-A video UEs are simulated. In the first subscenario, 10 video UEs are simulated. The number of the video UEs is incremented by 10 in the subsequent subscenarios. Figure 6 illustrates the simulation results of the average PRBs used with 95 % CI (confidence interval) in all the categories describes above. In the case of without relaying, the number of average PRBs used in the low-load case is less than 25 (total number of PRBs). However, increasing the number of UEs within cell also increases the average number of PRBs used. In the case of relaying, less number of average PRBs are used due to improved coverage which ultimately results in improved signal strength. Consequently, PRBs are used with the maximum capacity. However, in the high-load scenarios, e.g., 60 UEs, all the 25 PRBs are used which is the total available bandwidth in a 5-MHz system.

The E2E delay statistics of video traffic with 95 % CI is depicted in Fig. 7. In the low-load scenarios, the simulation results illustrate slightly higher delay in the case of relay-based-DeNB access due to relaying. However, in the high-load scenarios, the delay time in direct-DeNB access scenarios rises exponentially. This is due to the almost complete utilization of the available bandwidth which can be seen in Fig. 6. However, in the case of relaying, the delay time is still acceptable which is due to the fact that the maximum PRBs utilization due to improved coverage.

The simulation results of average traffic received with 95 % CI in both categories are depicted in Fig. 8. The better utilization of PRBs in the case of relaying gives increased average uplink traffic received at the PHY layer of the eNB. In all the subscenarios of both categories, the average traffic received is slighty better in the case of relaying. From the given simulation results, it is concluded that relaying can significantly improve the network QoS performance in terms of PRB utilization, E2E delay as well as average traffic received.

6.2 Efficient radio resource utilization

In this subsection, the performance of the proposed data traffic aggregation and multiplexing is evaluated. The scenarios are simulated according to three major groups. In the first group, M2M data packets are relayed in uplink without multiplexing. In the second group, the data packets from all the active M2M devices which are located in the proximity of the RN are aggregated at the Uu PDCP layer before being sent to the DeNB. However, only the periodic per-hop control scheme is used in which the large aggregated data packets are served when their size is equal to the available TBS − Un protocol overhead. In the third group, an expiry timer is introduced in order to limit the multiplexing delay especially in the low-load scenarios. In this case, the aggregated packet is served after T _max at the latest (periodic simple approach).

All the above mentioned groups are further sub-categorized into various subscenarios. In the first subscenario, 200 M2M devices are placed in the proximity of the RN. The number of M2M devices is incremented by 200 in the subsequent subscenariors. The position of the RN corresponds to an MCS of 16 and a TBS of 1608 bits per TTI with the allocation of 5 PRBs by DeNB. The performance of the proposed scheme is computed in terms of mean number of PRBs used and packet E2E delay in uplink for all the aforementioned scenarios.

The simulation results for the mean number of PRBs used and packet E2E delay with 95 % CI are illustrated in Figs. 9 and 10, respectively. The values of the upper and lower bound of the confidence intervals are very small in most of the scenarios as depicted in Figs. 9 and 10. The simulation results in Fig. 9 clearly show the efficient utilization of PRBs in uplink with the proposed M2M data aggregation. For instance, in the “no multiplexing” scenario with 400 devices, the arrival rate at the RN Uu PHY layer is 262.4 bits per TTI which almost utilize 1 PRB. However, in the case of multiplexing, only half of the PRBs are used to serve the 400 devices. Similarly, without multiplexing the RN serves nearly 2400 devices with 5 PRBs in uplink, which actually depicts the limiting case of no multiplexing approach. Moreover, the system utilization, ρ can be determined in terms of arrival and service rate. The arrival rate is given as, λ= (N×656) b/s where N is the number of active devices and 656 bits is the amount of data per device received at the RN Un PHY layer. Moreover, the service rate is μ=μ _PRB ×5 (P R B s) b/s. Thus, the system utilization, ρ and the maximum number of served devices, N _max can be determined according to \(\rho = \frac {N\times 656}{321600}\) and \(N_{max}= \frac {\rho \times 321600}{656} \), respectively. However, the present system parameter settings serve approximately N _max =2400 devices, which also resembles with Fig. 9.

However, in case of multiplexing, the number of devices served by the RN nearly doubles. This is due to the fact that in the case of “no multiplexing”, each data packet contains an additional Un air interface overhead of GTP, UDP, IP and layer 2. The additional overhead causes an extra PRB usage and overall reduces the PRB utilization efficiency. Moreover, in low-load scenarios, the average number of PRBs used is slightly higher in the case of “multiplexing with timer”. This is due to the fact that the RN serves the traffic at the latest after 9 ms and thus the PRB is not necessarily used with its maximum capacity due to the low arrival rate. However, in high-load scenarios, the timer has almost no impact and nearly equal numbers of PRBs are used with and without timer.

Figure 10 depicts the simulation results of M2M mean packet E2E delay in all three aforementioned scenarios. The results show that the value of packet E2E delay is higher in the case of multiplexing for low loads. This increase in delay time is due to less arriving packets especially in the low-load scenarios. The buffer aggregates packets until its size + the additional RN Un overheads is equal to the offered load. However, the use of an expiry timer limits the delay by serving the aggregated packets at the latest after 9 ms. However, in high-load scenarios (e.g., 2000 devices) the E2E delay is slightly higher compared to the case of no multiplexing. This is due to high arrival rate and the buffer aggregates the incoming packets to make a large aggregated packet within less time. Moreover, the value of E2E delay is very large in fully loaded scenarios as depicted in Fig. 10, when the RN utilizes all five PRBs with the maximum capacity.

6.3 Impact of timer expiry, T _max

In this subsection, we investigate the impact of timer expiry, T _max on mean PRB utilization and E2E delay by varying input traffic. However, we only consider low-load scenarios due to the fact that the timer expiry has almost no affect on mean PRB utilization and E2E delay in high-load scenarios in the multiplexing process, as discussed earlier in Figs. 9 and 10. In general, the mean value of used PRBs decreases and E2E delay increases for larger values of the timer expiry. Additionally, this trade-off is heavily dependent on the arrival rate, as discussed earlier.

In this work, 10 scenarios are simulated in order to evaluate the impact of timer expiry on the multiplexing process. In the first scenario, 100 M2M devices are considered to generate input traffic. The number of devices in incremented by 100 in the subsequent scenarios. The above scenarios are simulated for the timer expiry values of 4, 9, and 14 ms. The simulation results of mean number of used PRBs and E2E delay are compared for the given values of maximum waiting time. Figures 11 and 12 compare the simulation results of mean PRB utilization and E2E delay, respectively. Figure 11 shows that less PRBs are used in the case of larger values of T _max , when the arrival rate is kept constant. For instance, in the first scenario with N=100, average numbers of PRBs used are 0.48, 0.27, and 0.20 for the timer expiry values of 4, 9, and 14 ms, respectively. It shows that for the larger values of maximum waiting time, less number of PRBs are used on average, as it allows more packets to be multiplexed before timer expiry. Resultantly, it increases the multiplexing gain. Additionally, it is further noted that larger values of T _max has almost no affect on PRBs used in the case of high loads. For instance, when N=1000, the average values of PRBs used are 1.219 and 1.213 for the maximum waiting time of 9 and 14 ms, respectively.

Similarly, Fig. 12 compares the simulation results of the mean E2E delay for the given timer expiry values. As discussed in the beginning, the larger values of waiting time, T _max increase the mean packet E2E delay, particulaly in the low-load scenarios. However, the effect reduces with the larger values of N. For instance, in the low-load scenario when N=100, the given values of maximum waiting time introduce a mean E2E delay of 0.0045, 0.0070, and 0.0095 ms, respectively. Since the effect of timer on the multiplexing process decreases due to the increasing arrival rate, the mean E2E delay values are reduced to 0.00368, 0.00416, and 0.00419, respectively for N=1000. Furthermore, it is noted that the values of mean E2E delay for T _max of 10 and 15 ms are almost similar. Moreover, the impact of timer expiry completely vanishes when input traffic load is increased beyond 1000, see Figs. 9 and 10.