1. A Cross-Layer Scheduling Algorithm With QoS Support in Wireless Networks
Qingwen Liu, Student Member, IEEE
Xin Wang, Member, IEEE,
Georgios B. Giannakis, Fellow, IEEE
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 55, NO. 3, MAY 2006
Presented by Jason, Li-Yi Lin
2019/1/15
NTU, IM, OPLab

2. Outline Introduction System Architecture Scheduler Design Simulations
Conclusion and Future Directions
2019/1/15
NTU, IM, OPLab

3. Introduction
Scheduling plays an important role in providing QoS support to multimedia communications in various kinds of wireless networks.
However many wireless standards define only QoS architecture and signaling, but do not specify the scheduling algorithm that will ultimately provide QoS support
Introduce a priority-based scheduler at the MAC layer for multiple connections with diverse QoS requirements, where each connection employs adaptive modulation and coding (AMC) scheme at the PHY layer.
2019/1/15
NTU, IM, OPLab

4. Outline Introduction System Architecture Scheduler Design Simulations
- A. Network Configuration
- B. QoS Architecture at the MAC
- C. AMC Design at the PHY
Scheduler Design
Simulations
Conclusion and Future Directions
2019/1/15
NTU, IM, OPLab

5. System Architecture – Network Configuration (1/6)
2019/1/15
NTU, IM, OPLab

6. System Architecture – Network Configuration (2/6)
All connections communicate with the BS using TDM / TDMA
At the PHY, multiple transmission modes are available to each user, with each mode representing a pair of a specific modulation format and a forward error control (FEC) code
The AMC selector determines the modulation-coding pair, whose index is sent back to the transmitter through a feedback channel.
2019/1/15
NTU, IM, OPLab

7. System Architecture – Network Configuration (3/6)BS SS
2019/1/15
NTU, IM, OPLab

8. System Architecture – Network Configuration (4/6)
Transmission modes (TM):
2019/1/15
NTU, IM, OPLab

9. System Architecture – Network Configuration (5/6)
At the MAC, each packet contains a fixed number of bits , which include packet header, payload, and CRC bits.
At the PHY, each frame contains a fixed number of symbols , and the frame duration (in seconds) is constant.
With TDM, each frame is divided into time slots.
Let each time slot contain a fixed number of symbols.
The time slots contain control information and pilots; the time slots convey data.
2019/1/15
NTU, IM, OPLab

10. System Architecture – Network Configuration (6/6)
MAC
PHY
symbols
2019/1/15
NTU, IM, OPLab

11. System Architecture – QoS Architecture at the MAC (1/2)
Unsolicited grant service (UGS)
- this service provides guarantees on throughput, latency, and jitter to the necessary levels as TDM services
- the QoS metrics are the packet error rate (PER) and the service rate
Real-time polling service (rtPS)
- provides guarantees on throughput and latency, but with greater tolerance on latency relative to UGS
- the QoS metrics are the PER and the maximum delay (or the maximum delay for a given outage probability)
2019/1/15
NTU, IM, OPLab

12. System Architecture – QoS Architecture at the MAC (2/2)
Nonreal-time polling service (nrtPS)
- provides guarantees only in terms of throughput
- the QoS metrics are the PER and the minimum reserved rate
Best effort (BE) service
- provides no guarantees on delay or throughput
- although no delay and rate is specified for BE connections, a prescribed PER should be maintained over wireless channels
2019/1/15
NTU, IM, OPLab

13. System Architecture – AMC Design at the PHY (1/3)
Efficient bandwidth utilization for a prescribed PER performance at the PHY can be accomplished with AMC schemes
Each connection with rtPS, nrtPS, and BE services relies on AMC at the PHY.
The objective of AMC is to maximize the data rate by adjusting transmission modes to channel variations while maintaining a prescribed PER
2019/1/15
NTU, IM, OPLab

14. System Architecture – AMC Design at the PHY (2/3)
Let denote the total number of transmission modes available ( for TM)
The boundary points are denoted as
To avoid deep-channel fades, no data are sent when which corresponds to the mode with rate bits/symbol
2019/1/15
NTU, IM, OPLab

15. System Architecture – AMC Design at the PHY (3/3)
Approximate the PER expression in AWGN channels as
Set the region boundary for the transmission mode to be the minimum SNR required to guarantee
2019/1/15
NTU, IM, OPLab

16. Outline Introduction System Architecture Scheduler Design Simulations
Conclusion and Future Directions
2019/1/15
NTU, IM, OPLab

17. Scheduler Design (1/8) - Scheduling UGS Connections
The transmission mode could be selected in the initial service access phase to meet the average PER requirement
Then, the transmission mode is fixed during the whole service time
Denote the total time slots allocated to UGS connections as per frame. The residual time slots are
2019/1/15
NTU, IM, OPLab

18. Scheduler Design (2/8) - Scheduling rtPS, nrtPS and BE connections
Each connection adopts AMC at the PHY
Given a prescribed PER , the SNR thresholds for connection are determined as described above by setting
The possible transmission rate (capacity) can be expressed as
where
2019/1/15
NTU, IM, OPLab

19. Scheduler Design (3/8) - Scheduling rtPS, nrtPS and BE connections
At the MAC, the scheduler simply allocates all time slots per frame to the connection
where is the PRF for connection at time
If multiple connections have the same value , the scheduler will randomly select one of them with even opportunity
2019/1/15
NTU, IM, OPLab

20. Scheduler Design (4/8) - Scheduling rtPS, nrtPS and BE connections
The PRF for a rtPS connection at time is defined as
where is the rtPS-class coefficient and is the delay satisfaction indicator, which is defined as
with denoting the guard time region ahead of the deadline , and denoting the longest packet waiting time
2019/1/15
NTU, IM, OPLab

21. Scheduler Design (5/8) - Scheduling rtPS, nrtPS and BE connections
If , i.e., , the delay requirement is satisfied
If , i.e., , the packets of connection should be sent immediately to avoid packet drop due to delay outage
2019/1/15
NTU, IM, OPLab

22. Scheduler Design (6/8) - Scheduling rtPS, nrtPS and BE connections
Guarantee the minimum reserved rate to each nrtPS connection
If data of connection are always available in queue, the average transmission rate at time is estimated over a windows size as
2019/1/15
NTU, IM, OPLab

23. Scheduler Design (7/8) - Scheduling rtPS, nrtPS and BE connections
The PRF for an nrtPS connection at time is defined as
where is the nrtPS-class coefficient and
2019/1/15
NTU, IM, OPLab

24. Scheduler Design (8/8) - Scheduling rtPS, nrtPS and BE connections
The PRF for a BE connection at time is defined as
where is the BE-class coefficient
The role of and in (6), (9), and (11), respectively, is to provide different priorities for different QoS classes
2019/1/15
NTU, IM, OPLab

25. Outline Introduction System Architecture Scheduler Design Simulations
Conclusion and Future Directions
2019/1/15
NTU, IM, OPLab

26. Simulations (1/10) Assumptions:
A1: The wireless channel quality of each connection remains constant per frame, but is allowed to vary from frame to frame
A2: Perfect channel state information (CSI) is available at the receiver via training-based channel estimation.
The corresponding transmission mode selection is fed back to the transmitter without error and latency.
A3: Error detection based on CRC is perfect.
A4: if a packet is received incorrectly after error detection, we declare packet loss.
2019/1/15
NTU, IM, OPLab

27. Simulations (2/10) Channel Model:
The received SNR per frame is a random variable with a Gamma probability density function, i.e.,
where , is the Nakagami fading parameter
Finite-state Markov chain (FSMC) model
2019/1/15
NTU, IM, OPLab

28. Simulations (3/10) Parameter Setting: The frame length is ms
The packet length at the MAC is fixed to bits
For each rtPS connection , we assume that the arrival process to the queue is Bernoulli distributed with a given average rate and parameter
The instantaneous arriving rate at time can be expressed as
2019/1/15
NTU, IM, OPLab

29. Simulations (4/10) rtPS connection 1 rtPS connection 2 2019/1/15
NTU, IM, OPLab

30. Simulations (5/10) The guard time is set to ms
The delay outage probability
where window size ms
2019/1/15
NTU, IM, OPLab

31. Simulations (6/10)
For each nrtPS connection , we assume that the data are always available
nrtPS connection 3, 4
2019/1/15
NTU, IM, OPLab

32. Simulations (7/10)
For each BE connection , we assume that the data are always available
The rate performance of nrtPS and BE connections is also evaluated by the average service rate over a window size ms
The system is simulated over ms with bounds
BE connection 5, 6
2019/1/15
NTU, IM, OPLab

33. Simulations (8/10) 0.77% 0.24% 6.7 4.5 3.1 5.9 Average 2019/1/15
NTU, IM, OPLab

34. Simulations (9/10) 1.06% 0.29% 5.8 3.4 0.4 1.0 Average 2019/1/15
NTU, IM, OPLab

35. Simulations (10/10) 0.78% 0.18% 2.5 2.9 Average 2019/1/15
2019/1/15
NTU, IM, OPLab

36. Outline Introduction System Architecture Scheduler Design Simulations
Conclusion and Future Directions
2019/1/15
NTU, IM, OPLab

37. Conclusion and Future Directions (1/3)
Efficient bandwidth utilization is achieved through in
Delay bound is provided for rtPS connections and we can control the delay outage probability below the practically acceptable values by adjusting
Throughput is guaranteed for nrtPS connections if sufficient bandwidth is provided
2019/1/15
NTU, IM, OPLab

38. Conclusion and Future Directions (2/3)
Implementation complexity is low because the scheduler simply updates the priority of each connection per frame
Flexibility is provided because the scheduling does not depend on any traffic or channel model
Scalability is achieved.
2019/1/15
NTU, IM, OPLab

39. Conclusion and Future Directions (3/3)
The upper-bound and the delay guard time were set heuristically, their effects on performance are worthy of further research
Scheduling multiple connections each time may lead to better performance
The fairness issue for the users in the same service class is another topic
The effects of imperfect channel state information due to estimation error and feedback latency are also worth further study
2019/1/15
NTU, IM, OPLab

40. Thanks for your listening
2019/1/15
NTU, IM, OPLab

41. Conclusion and Future Directions
For rtPS
For nrtPS
For BE
2019/1/15
NTU, IM, OPLab

42. Finite-state Markov chain (FSMC) model
2019/1/15
NTU, IM, OPLab

43. Finite-state Markov chain (FSMC) model (cont’)
2019/1/15
NTU, IM, OPLab