Bluetooth: Quality of Service

1. Bluetooth: Quality of Service Reference:“QoS based scheduling for incorporating variable rate coded voice in Bluetooth”; Chawla, S.; Saran, H.; Singh, M.; IEEE International Conference on Communications (ICC), 2001, pp. 1232 -1237 (BTQoS-1.pdf)

2. Introduction • Voice Activity Detection (VAD) • Detecting silence periods in speed • Reducing the coding rate during silence • Incorporation of variable rate coded voice over BT would necessitate a substantial change to the scheduling policies • BT intrinsically allocate 64Kbps to voice • 1. Adaptive TSCO scheduling • 2. For QoS of voice, EDD-based scheduling

3. Scheduling • For variable rate coded voice • 1. Adaptive TSCO • For voice channels with no error coding, TSCO is typically 6 TDD slots to account for 64 Kbps bandwidth • To adapt to the coding rate, the time period of scheduling SCO connections can be increased • For instance, TSCO = 16 would correspond to a bandwidth usage of 24 Kbps, which is approximately the bandwidth allocated to voice in GSM during talkspurt • In this scheme, we dynamically adjust TSCO depending on the activity of the call. Accordingly, TSCO toggles between two distinct values with change in activity of voice

4. Scheduling (cont)

5. Scheduling (cont) • 2. Voice over ACL • Use a QoS based scheduling scheme •  Latency based scheduling • If a connection has latency of ‘n’ slot, then each packet of that connection should be scheduled within ‘n’ slots of its arrival  scheduling based on deadline of packets • Use a greedy Earliest Due Deadline (EDD) scheme • EDD is known to give an optimal schedule in the sense that if it is possible to satisfy QoS requirements of every connection, they are satisfied • Necessitates admission control for the system

6. Scheduling (cont) • For a voice connection, end-to-end latency of less than 100 ms is acceptable as it is not noticeable to the human ear • A single hop should be much lesser than 100 ms • In order to avoid large end to end delay and jitter, try to avoid queuing of more than one voice packet in the system • Accordingly, choose max. tolerable scheduling delay of a voice connection to be the time gap between arrival of two packets • Thus, the latency of voice for 22.8 Kbps connection should be 18 time slots corresponds to a latency of 11 ms

7. Simulation Scenario • Using Network Simulator • Two state Markov chain model to simulate talkspurt and silence periods for voice • Voice traffic: in bursts of 20 ms length • Coding rate (similar to that in GSM) • 22.8 Kbps for talkspurt, 11.4 Kbps for silence • Latency for voice connections • 18 slots during activity, 36 slots during silence • Latency for data connections • 50 ~ 200 slots

8. Simulation Results • Fig. 3 • The scheduling delays for data packets in the case of adaptive TSCO are lesser on average than those in the case of fixed TSCO • Fig. 4 • Similar results as Fig. 3 • Fig. 5 • Both proposed scheduling schemes perform better than voice over SCO • Voice over ACL performs best

9. Simulation Results (cont) ?!@#$%^&*

10. Simulation Results (cont) ?!@#$%^&*

11. Simulation Results (cont) ?!@#$%^&*

12. Simulation Results (cont)

13. Simulation Results (cont)

14. Simulation Results (cont)

15. Conclusion • Variable rate coded voice in BT • 1. Adaptive TSCO • 2. Voice over ACL (EDD-based scheduling)