Improved calendar disc scheduler for LTE advanced networks with HARQ

Abstract

Long Term Evolution Advanced (LTE-A) is a broadband wireless technology that supports variety of services with different data rate. In order to achieve this the evolved Node B (eNB) uses different features provided in the 3GPP standards. Features like Carrier Aggregation (CA), Multiple input and Multiple output (MIMO) and Hybrid Automatic Repeat Request (HARQ) help to increase the throughput and spectral efficiency. In this paper, a novel two-level calendar disc algorithm with HARQ is introduced at the eNB for effectively scheduling real time and non-real time traffic with different service types. The algorithm also uses a burst profile management module that analyzes the current user profile and notifies the scheduler about the need to change in profile based on power boosting. The Calendar Disc Scheduler (CDS) is improved by adding HARQ retransmission index as a parameter in calculating the metric weight. The scheduler was tested for both adaptive and non-adaptive methods of HARQ in both synchronous and asynchronous modes. The proposed improved CDS scheduler was simulated with LTESim simulator and compared with calendar disc algorithm without HARQ improvements. Results show that the proposed scheduling method provides increased performance in terms of goodput, delay and spectral efficiency.

Keywords

Long term evolution Advanced (LTE-A)Calendar Disc Scheduling (CDS)Burst profile Management (BPM)Hybrid automatic repeat request (HARQ)

1. Introduction

Mobile telecommunication systems have gone through considerable evolution over the past few decades. With the high availability of high-end mobile devices and with introduction of new different applications there has been a significant increase in the data traffic. This indirectly increases the demand of high bandwidth services which has encouraged the Third Generation Partnership Project (3GPP) to come up with the proposal of Long Term Evolution advanced (LTE-A). LTE-A is a major improvement from the LTE network with added features and also backward compatible [3]. LTE-A is a fourth generation cellular system which can be operated in different spectrum allocations and it make use of Orthogonal Frequency Division Multiple Access (OFDMA) in the downlink and Single Carrier Frequency Division Multiple Access (SC-FDMA) in the uplink.

OFDMA has the capability to reduce the fading effects by which the system capacity can be increased. The total bandwidth of the LTE system consists of many subcarriers that are allocated to the users depending on their Quality of Service (QoS) demands. In LTE-A the downlink transmission is done through Resource Blocks (RB’s). A RB is equal to a slot that comprises of 7 symbols in time axis and 12 subcarriers in the frequency axis [1,2,23]. RBs are assigned in pairs to the users and each pair of RB is referred to as scheduling block or subframe which last of 1 ms. This is referred as Transmission Time Interval (TTI) of LTE. In uplink the users must be assigned two consecutive RBs to make up for that pair while in downlink users can be assigned any arbitrary two RBs to make up a scheduling block. In recent times, there have been multiple researches done to address problems of radio resource allocation in LTE-A downlink and have made useful contributions. Few such contributions include the sectioning of resource allocation scheme into scheduling and allocation modules [6]. Many scheduling methods had been proposed in literature for improving the system performance [4,5,7–12,14,15,17,20–22]. These methods depend on the Channel Quality Information (CQI) information reported by the user. A set of standardized QoS Class Identifiers (QCI) and their characteristics have been provided in the 3GPP specifications.

3GPP has grouped various services offered by LTE using QoS requirements and these groups are identified by different markers which are referred as QCIs. The QCI table provides values for priority handling, packet loss rate, and acceptable delay budget for each QCI label. Since the QCI information is reported by the end user, due to high possibility of change in channel conditions, it can lead to erroneous CQI reporting. This in turn can lead to improper selection of modulation order and transmit power to the user. Moreover LTE-A provides the data reliability using retransmission methods like Hybrid Automatic Repeat Request (HARQ) at the media access control layer and Automatic Repeat Request (ARQ) at the radio link control layer [13].

In this paper, a new scheduling method is proposed and improved by incorporating HARQ priority parameters while adopting to user demand variations. This scheduling method is done by improving the method proposed in [16]. The simulation is carried out using LTESim [18], the simulation results show that addition of new Burst Profile Module (BPM) along with the HARQ priority optimization parameters greatly improved the goodput and link spectral efficiency while reducing the packet loss rate.

The reminder of the paper is organized as follows: Section 2 describes the system model and the HARQ feature in LTE-A. Section 3 describes the proposed improvement done on the algorithm to support HARQ feature. Simulation results and performance analysis are done in Section 4 and Conclusions are drawn in Section 5.

2. System model

An OFDMA system for single cell downlink is considered in this paper. A single microcell eNB is serving $I = {1, 2, \dots, i}$ number of users as depicted in Fig. 1. The system contains N subcarriers which depend on the entire bandwidth of the system. In LTE the radio interface is shared by all the users with common shared channel.

Fig. 1.

System model of LTE-A network.

The resource management is done by eNB in an efficient manner to accomplish the end users QoS requirements. The initial assumption is made such that 50% of the users support LTE carrier aggregation i.e. LTE-A users while the other 50% are standard LTE users. It is also assumed that the perfect channel quality information is periodically reported to eNB at each TTI. To put algorithm in more stress mode 10% of the users are always kept at the cell edge to trigger multiple packet re-transmissions.

2.1. HARQ classification in LTE-advanced

HARQ in LTE-advanced can be classified based on the timing and adaptation. The 4 types names Synchronous adaptive, Synchronous Non-Adaptive, Asynchronous adaptive and Asynchronous Non-adaptive (as given in Fig. 2) are explained as follows:

Fig. 2.

HARQ classifications.

Synchronous adaptive allows eNB to change the resource allocation and modulation information for retransmissions while maintaining the time synchronicity. In Synchronous non-adaptive mode the retransmission happens only at fixed timing intervals. The big benefit of this type of HARQ is that the control information needs to be transmitted along with the first sub-block only. But the major setback resource allocation cannot be adapted during retransmission according to the current channel conditions. An asynchronous adaptive HARQ mode provides full flexibility for retransmissions. The adaptation and timing are adjusted according to channel quality. The drawback of this method is that full control information needs to be sent with retransmissions. In asynchronous non adaptive HARQ resource allocation, MCS and MIMO formats are kept the same as initial retransmission. The drawback of this method is limited flexibility.

Fig. 3.

Flexibility vs overhead.

Figure 3 depicts the overhead vs flexibility for HARQ. The synchronous non-adaptive scheme occupies the low overhead as well it provides lowest flexibility, while the asynchronous adaptive provides greater flexibility with the expense of overhead.

3. Proposed improved calender disc scheduler

One of the biggest challenges of every radio based transmission is to achieve the highest possible effectiveness by the eNB. To reach this goal, the eNB use a number of features like carrier aggregation, modulation orders, HARQ, MIMO etc. In LTE-Advance network, users feedback about their channel quality to eNB, if the channel quality of the users changes within the TTI timings the MCS adopted by eNB for that user varies, which leads to loss of packets. To overcome this disadvantage a module is added to the input for the proposed scheduling algorithm. This is the first level of improvement for the proposed scheduling algorithm. Here a separate BPM module is used at eNB as shown in Fig. 4. This BPM module is responsible for increasing or decreasing the signal level of the user by boosting or reducing the transmission power of the eNB respectively.

Fig. 4.

Proposed BPM module as an input to scheduler.

The main job of the BPM module is to analyze the current profile of the user and notifies the medium access control (MAC) scheduler about the need to change the boosting level.

Generally, the current eNB transmit power of a user i can be written as $\begin{matrix} (1) & P_{t, i}^{n} = P_{t, i} + P_{boosting}^{n} \end{matrix}$

Here $P_{t, i}^{n}$ is the current transmit power of the ith user in the nth subcarrier, $P_{t, i}$ is the constant transmit power for the user i and $P_{boosting}^{n}$ is the current boosting level of power used i the nth subcarrier.

The minimum boosting level power budget can be written as $\begin{matrix} (2) & P_{boosting (min)}^{n, k} = P_{boosting}^{n, k} + P_{RSSI (min), i}^{k} - P_{RSSI, i} + γ_{correction}^{k} \end{matrix}$ After determining the power budget with minimum boosting level, the BPM module calculates the expected Received Signal Strength Indicator (RSSI) and Signal to Interference Noise Ratio (SINR) value for the user. The scheduler chooses the corresponding power level and modulation irrespective of the reported CQI value from the user. So though there is sudden change in the channel condition or erroneous CQI from the user the scheduler schedules the data effectively in the downlink transmission.

In improved CDS algorithm, two levels of scheduler are used. Here, separate disc scheduling metric is calculated for real time ( ${DS}_{RT}$ ) and non-real time services ( ${DS}_{NRT}$ ). After this metric is calculated for every packets, they are placed in a queue according to the type of service. The allocation is done first for the ${DS}_{RT}$ list and then followed by the ${DS}_{NRT}$ list by the level 1 scheduler considering the BPM of the individual user. The allocation is done by placing the traffic flow descriptor of the user on the component carriers forming calendar discs. The scheduler forms the calendar disc of component carriers as many as supported by the user. That is, one calendar disc for LTE support user and variable number of calendar disc for LTE-A support users. After creating the calendar discs the level 2 scheduler processes on all the calendar discs for effective utilization of spectrum. While calculating the scheduling metrics HARQ prioritization parameter is considered which gives priority to real time flows retransmissions.

The DS metric for real time flows is calculated as $\begin{matrix} (3) & {DS}_{RT} = E_{d}^{k} + (\frac{L_{i}^{k}}{r_{i (min) \times a}}) + (T_{slot} \times N_{c}) \end{matrix}$ where, $E_{d}$ is the expected departure time of a packet k, $L_{i}^{k}$ is the size of kth packet of user i, $r_{i (min)}$ is the minimum guaranteed bit rate for that service, $T_{slot}$ is the time slot duration which is equal to 0.5 ms and $N_{c}$ is the number of component carriers supported by the user and a is the HARQ prioritization parameter. For LTE user, the value of $N_{c}$ is always equal to one and for LTE-A user the value varies from 2 to 5 depending on the user supported component carriers.

The DS metric for non-real time users is calculated as $\begin{matrix} (4) & {DS}_{NRT} = E_{d}^{k} + (\frac{L_{i}^{k}}{r_{i (max) \times N_{b} \times a}}) + (T_{slot} \times N_{c}) \end{matrix}$ where $r_{i (max)}$ is the maximum bit rate for that service and $N_{b}$ is the number of bits per symbol, which is decided based on the channel condition of that user. The HARQ retransmission index, which is added as a part of the scheduler metric evaluation is used in minimizing the packet loss and delay in retransmission.

4. Performance analysis

The performance of the proposed CDS Scheduler is examined in terms of total goodput, fairness index, packet loss rate and delay. To explore the current behavior of the proposed CDS algorithm, the scheduler is simulated to execute the various HARQ modes described by the 3GPP.

Synchronous non-adaptive mode

Synchronous adaptive mode

Asynchronous non adaptive mode

Asynchronous adaptive mode

For validating performance improvement, the two level CDS scheduling algorithm is analyzed with and without BPM and HARQ improvements. In the existing system, the two-level CDS algorithm satisfies the QoS requirement of real time users. Here, the CDS algorithm performs efficiently when LTE (does not support carrier aggregation) and LTE-A (supports carrier aggregation) users are served simultaneously. The performance analysis is carried out using an open source simulator, LTESim. The modules present in the simulator are well explained in [19].

4.1. Simulation scenario

A single cell scenario is considered with fixed eNB at the center of the cell and the users are uniformly distributed inside the cell. The users are moving with a speed of 30 kmph in random direction and when it reaches the cell boundary it will take the new random direction. At any instant of time, it is assumed that half of the users support LTE and other half support LTE with carrier aggregation. And also, it is assumed that at any point of time 40% of users receive video flows, 40% of users receive Voice over Internet Protocol (VoIP) flows and 20% of uses receive Best Effort (BE) flows. At the eNB, infinite buffer is assumed such that there is no packet loss because of the non-availability of the buffer space. The performance analysis is done by varying the number of users. The VoIP flows are generated with source rate of 8 kbps by G.729 audio coder, whereas, the video flows are produced with a source rate of 128 kbps by H.264 trace-based coder. BE traffic is generated by ideal source that always has packet to send. The delay bound is set to 0.1 s. The simulation parameters used for analysis are given in Table 1.

Table 1
Simulation parameters for the LTE-A network

Parameter Value

Bandwidth 10 MHz

Component Carrier Bands 0.7, 2.6 GHz

Frame structure FDD

Channel model 6-ray Typical Urban (TU)

Path loss 128.1 + 37.6 log r (dB)

Penetration loss 20 dB

eNB transmission power 43 dBm

Mobility model Random

Speed of UE 30 kmph

Maximum delay 0.1 s

Scheduling Algorithm CDS

Minimum and maximum boosting power −12 dB and 9 dB

Simulation duration 20 s

Modulation scheme QPSK, 16 QAM, 64 QAM

Traffic model Video – Trace based (H264)

VoIP – G.729 voice flows

BE- Infinite buffer

Parameter	Value
Bandwidth	10 MHz
Component Carrier Bands	0.7, 2.6 GHz
Frame structure	FDD
Channel model	6-ray Typical Urban (TU)
Path loss	128.1 + 37.6 log r (dB)
Penetration loss	20 dB
eNB transmission power	43 dBm
Mobility model	Random
Speed of UE	30 kmph
Maximum delay	0.1 s
Scheduling Algorithm	CDS
Minimum and maximum boosting power	−12 dB and 9 dB
Simulation duration	20 s
Modulation scheme	QPSK, 16 QAM, 64 QAM
Traffic model	Video – Trace based (H264)
VoIP – G.729 voice flows
BE- Infinite buffer

4.2. Results and discussion

The main objective of the improved calendar disc scheduling is to guarantee QoS for real time services, though the user is of LTE or LTE-A, while maintaining fairness among users. Total goodput is an important parameter that is measured in terms of the number of packets received successfully without errors by the users at a given duration. To test the behavior of the proposed algorithm when HARQ is enabled different types of HARQ modes are enabled and simulated.

Total Goodput for VoIP, video and BE services are shown in below Fig. 5. The Total goodput decreases as the number of users increases. This is because LTE system uses shared channel in the downlink. The proposed CDS algorithm shows better performance for real time flows when different HARQ modes are used. With Asynchronous adaptive mode, the total goodput is higher for the proposed algorithm. This is because this mode provides full flexibility for the retransmission. And also, the adaptation and timing are adjusted according to the channel quality.

Fig. 5.

Total goodput for a) VoIP b) video with multiple HARQ modes.

Fig. 6.

Delay for (a) VoIP and (b) video flows.

The delay experienced by the VoIP and video flows are shown in below Fig. 6. It is noted from the figure that proposed algorithm can be able to maintain lesser delay for real time services. By using the proposed algorithm, the delay experienced by the VoIP flows lies between 1 ms to 4 ms for the maximum number of users (100 users) used in the simulation. For video flows CDS algorithm provides lesser delay under 5 ms. Because in proposed algorithm, the allocation of resources starts from the DSRT list thereby priority is given to the real time services. Moreover, with the HARQ prioritization parameter the re transmission messages are given more priority than the actual packet transfer ensuring no delay for the retransmission packets.

The total goodput of the VoIP, video and BE services, with and without HARQ parameter while scheduling are shown in the Fig. 7. From the figures it is confirmed that the proposed algorithm provides higher goodput when incorporating HARQ prioritization parameter.

Fig. 7.

Total goodput for (a) VoIP (b) video and (c) BE.

Apart from goodput and delay, the performance analysis is done for the proposed CDS algorithm for various parameters like Packet Loss Rate (PLR), Fairness Index (FI). The results are shown in the Table 2. The simulation is done individually with standard known algorithms like PF, EXP, MLWDF, LOG, EMBS, the proposed CDS algorithm, CDS algorithm with BPM module incorporated and CDS algorithm with BPM and HARQ incorporated. With proper burst profile management and with synchronous adaptive retransmission using HARQ, the proposed algorithm gives better performance in terms of goodput, delay, fairness index and packet loss rate.

Table 2

Performance comparison of CDS + BPM + HARQ algorithm

Algorithm	Video				VoIP				BE

	Goodput (Mbps)	Delay (s)	PLR (%)	FI	Goodput (Mbps)	Delay (s)	PLR (%)	FI	Goodput (Mbps)	FI
PF	2.4115	0.012	0.978	0.631	0.275	0.0021	0.0618	0.638	7.304	0.785
MLWDF	2.461	0.0099	1.151	0.715	0.287	0.0019	0.0566	0.813	8.226	0.783
EXP rule	6.46	0.0072	0.288	0.762	0.346	0.00188	0.0262	0.819	9.2162	0.782
LOG rule	3.2	0.0091	0.922	0.726	0.334	0.00189	0.0542	0.814	8.896	0.783
EMBS	30.7	0.0134	1.60	0.791	1.632	0.00316	0.150	0.776	31.6	0.840
CDS	29.9	0.0015	0.157	0.824	1.750	0.00256	0.0135	0.871	35.9	0.841
CDS + BPM	30.96	0.0015	0.131	0.824	1.847	0.00261	0.0088	0.871	36.23	0.841
CDS + BPM + HARQ (proposed)	31.11	0.0013	0.129	0.824	1.875	0.00261	0.0084	0.871	36.42	0.841

5. Conclusion

In this paper, a novel scheduling algorithm which supports the HARQ was introduced in the downlink transmission. The algorithm uses the scheduling metrics optimized for the HARQ along with the addition of a separate BPM module at the eNB which governs the modulation order and transmit power chosen for the particular user. By using this method there has been a 3% increase in the throughput for video users, 6% increase in throughput for VoIP users and 1.5% increase in the throughput for BE users with HARQ asynchronous adaptive mode. The addition of BPM module introduces some delay in system. There is 30% increase in the system processing delay because of the metric becoming more complex. The delay can be addressed by improving the system hardware.

References

3GPP, Technical Specification Group Radio Access Network – Physical Channel and Modulation (Release 8), 3GPP TS 36.211.

3GPP, Technical Specification Group Radio Access Network – Physical Layer Procedures, (Release 8), 3GPP TS 36.213.

3GPP Technical Report, Requirement for further advancements for E-UTRA (LTE-Advanced), 3GPP TR 36.913.

S.M.

Alavi,

Zhou and

W.W.

Gen, Efficient resource allocation algorithm for OFDMA systems with delay constraint, Computer Communications 36 (2013), 421–430. doi:10.1016/j.comcom.2012.11.004.

Brehm and

Prakash, Overload-state downlink resource allocation in LTE MAC layer, Wireless Networks 19 (2013), 913–931. doi:10.1007/s11276-012-0509-1.

Capozzi,

Piro,

L.A.

Grieco,

Boggia and

Camarda, Downlink packet scheduling in LTE cellular networks: Key design issues and a survey, IEEE Communications Surveys and Tutorials 15(2) (2013), 678–700. doi:10.1109/SURV.2012.060912.00100.

S.S.

Fouziya and

Nakeeran, Study of downlink scheduling algorithms in LTE networks, Journals of Networks 9(12) (2014), 3381–3391.

S.S.

Fouziya and

Nakkeeran, Performance analysis of service based scheduler in LTE OFDMA system, Wireless Personal Communications 83(2) (2015), 841–854. doi:10.1007/s11277-015-2429-8.

S.S.

Fouziya and

Nakkeeran, Performance evaluation of downlink packet scheduling for real time traffic in LTE system, International Journal of Information and Communication Technology 10(3) (2017), 263–275. doi:10.1504/IJICT.2017.083261.

10.

S.S.

Fouziya and

Nakkeeran, Performance evaluation of downlink packet scheduling for real time traffic in LTE system, International Journal of Information and Communication Technology 10 (2017), 263–275. doi:10.1504/IJICT.2017.083261.

11.

S.S.H.S.

Fouziya and

Nakkeeran, A simple three level downlink packet scheduling to improve throughput in LTE network, Communication Express 4(9) (2015), 293–298. doi:10.1587/comex.4.293.

12.

I.G.

Fraimis and

S.A.

Kotsopoulos, Queue-aware resource allocation for multi-cell OFDMA system with QoS provisioning, Wireless Personal Communication 71 (2013), 3033–3044. doi:10.1007/s11277-012-0988-5.

13.

Galaviz,

D.H.

Covarrubias,

A.G.

Andrade and

Villarreal, A resource block organization strategy for scheduling in carrier aggregated systems, EURASIP Journal on Wireless Communications and Networking 2012(1) (2012), 1–17. doi:10.1186/1687-1499-2012-1.

14.

Ivesic,

Skorin-Kapov and

Matijasevic, Cross-layer QoE-driven admission control and resource allocation for adaptive multimedia services in LTE, Journal of Network and Computer Application 46 (2014), 336–351. doi:10.1016/j.jnca.2014.09.010.

15.

W.K.

Lai and

C.-L.

Tang, QoS-aware downlink packet scheduling for LTE networks, Computer Networks 57 (2013), 1689–1698. doi:10.1016/j.comnet.2013.02.017.

16.

Pande and

Piro, Optimal resource allocation scheme for LTE-A systems with carrier aggregation, IEEE International Conference on Advanced Networks and Telecommunication Systems (ANTS) (2014).

17.

Pande and

Piro, Optimal resource allocation scheme for LTE-A systems with carrier aggregation, in: IEEE International Conference on Advanced Networks and Telecommunication Systems (ANTS), 2014.

18.

Piro, LTE-Sim – The LTE Simulator [online] http://telematics.poliba.it/LTE-Sim, (accessed December 2016).

19.

Piro,

L.A.

Grieco,

Boggia,

Capozzi and

Camarda, Simulating LTE cellular systems: An open source framework, IEEE Transactions on Vehicular Technology 60(2) (2010), 498–513. doi:10.1109/TVT.2010.2091660.

20.

Sadiq,

S.J.

Baek and

de Veciana, Delay-optimal opportunistic scheduling and approximations: The log rule, IEEE Trans. Networking 19(2) (2011), 406–418.

21.

Sajid Mushtaq,

Fowler,

Mellouk and

Augustin, QoE/QoS-aware LTE downlink scheduler for VoIP with power saving, Journal of Network and Computer Applications 51 (2015), 29–46. doi:10.1016/j.jnca.2014.02.001.

22.

Song,

G.-T.

Gil and

D.-H.

Kim, Packet-scheduling algorithm by the ratio of transmit power to the transmission bits in 3GPP LTE downlink, EURASIP Journal on Wireless Communications and Networking (2010), 1–8.

23.

Zyren, Overview of 3GPP Long Term Evolution physical layer, 2007, freescale semiconductor.