Assessment of VoIP quality across ADSL downlinks

Transcription

1 Assessment of VoIP quality across ADSL downlinks Technische Universität Dresden Chair of Telecommunications Qin Dai

2 Contents 2 Introduction and motivation VoIP quality vs. ADSL downlinks VoIP challenges Delay in downlink Packet dropping Simulation and results Simulation 1 Simulation 2 Conclusions

3 Introduction and motivation 3 Background: VoIP application gains more and more interest both from ISPs and users. Goal of this study: guarantee VoIP quality when across ADSL downlinks Constructing the simulation model Assessing VoIP quality (in downlink direction) Approaches for improving VoIP quality

4 Contents 4 Introduction and motivation VoIP quality vs. ADSL downlinks VoIP challenges Delay in downlink Packet dropping Simulation and results Conclusions

5 VoIP challenges 5 Delay one way delay <= 150ms (ITU-T G.114) the only one biggest obstacle Jitter (delay variation) Loss Loss ratio (in packet) < 1~ 5% (according to different codec and PLC (packet loss concealment) function) Loss bursts impact more than random loss

6 System architecture 6 Web server VoIP phone DSLAM ADSL router Internet ATM ADSL link Gateway PC with other applications

7 Downlink model 7 Web server FTP server VoIP phone ADSL router DSLAM buffer (downlink, FCFS) ADSL downlink Main problem Large queueing delay in DSLAM Method limit the number of data packets can be placed in the DSLAM queue PC with Web brower and FTP VoIP phone

8 Approaches w 8 TCP packet dropping TCP: window-based protocol Enforce the TCP sender to close its congestion window size by introducing packet loss. Simple but effective Can result in poor TCP application behavior with high dropping rate Other approaches Can still maintain good TCP performance t

9 Contents 9 Introduction and motivation VoIP quality vs. ADSL downlinks Simulation and results Simulation 1 Simulation 2 Conclusions

10 Simulation model 1 network model 10 2 Web sessions C = 150 Mbit/s D = 7.5 ms + i * 0.1 ms (i=0...(num_srv-1)) num_srv Web server C = 150 Mbit/s D = 0 ms num_ftp FTP server C = 150 Mbit/s D = i * 10 ms (i=0...(num_ftp-1)) 1 FTP session TCP-Internet access point C = 150 Mbit/s D = 1, 10, 100 ms VoIP C = 150 Mbit/s D = 0 ms P Loss = 0.1 % C = 150 Mbit/s D = 0 ms num_voip VoIP phone (bidirectional) Discarding packet Ethernet: C = 10 Mbit/s D = 0 ms TCP application (Web and FTP Clients) DSLAM ADSL uplink C = 128 kbit/s D = 2 ms PC ADSL router DSLAM buffer (FCFS) ADSL downlink C = 1 Mbit/s D = 2 ms Queue with high priority for VoIP Start point of VoIP delay measurement End point of VoIP delay measurement VoIP num_voip VoIP phone (bidirectional) Delay budget: 50ms 1 VoIP session

11 Simulation 1 -- Performance metrics 11 VoIP quality Fraction of late VoIP packets No more than 1% Number of loss bursts in 10 minutes (burst length > 2,5,8) The number of loss burst (burst length>5) no more than 2 debatable TCP quality Web response time FTP throughput (one-way network delay about 1ms,10ms,100ms)

12 VoIP quality in downlinks 12 Fraction of late VoIP packets e-04 1e-05 Late VoIP packets Burst length > 2 Burst length > 5 Burst length > Number of loss bursts in 10 minutes 3 1e Delay budget / s High delay is necessary (more than 200ms) for guarantee acceptable loss ratio (1%)

13 TCP packet drop (delay budget 50ms) 13 Fraction of late VoIP packets Late VoIP packets Burst length > 2 Burst length > 5 Burst length > Number of loss bursts in 10 minutes FTP throughput / (MBit/s) FTP throughput 1 FTP throughput 10 FTP throughput 100 Web response time Web response time / s Distance between discarded packets (in packet) Distance between discarded packets (in packet) 0 High packet dropping rate necessary: more than 10% Low FTP throughput High Web response time (more than 8s)

14 Simulation model 2 MOS estimator 14 ns2 source Network model degraded signal G.729 Packet Data generator extraction G.729 PESQ PMOS value

15 VoIP quality estimation PESQ value 15 Perceptual evaluation of speech quality (PESQ): ITU-T Recommendation P.862 An objective method for speech quality assessment Compares an original signal with a degraded signal that is the results of passing through a communication system Highly correlated to the subjective scores: (random loss?)

16 Performance metrics and simulation setting 16 Performance metrics: Fraction of packet loss Fraction of BAD PESQ MOS value (PESQ MOS < 3.5) Number of loss bursts in 20 seconds (burst length > 2) Simulation setting: Voice sample: German, male, 20s Codec: G.729, 8kbit/s VoIP packet: 20ms/packet Competing traffic: 2 Web sessions simulation time: 400 times voice sample

17 PESQ value in 100 iterations 17 4 PESQ value PESQ MOS value Number of iteration original PMOS PMOS PMOS (random loss) Voice sample

18 (TCP) Packet dropping 18 VoIP packet loss / Bad PMOS ratio mean loss ratio mean bad MOS ratio (MOS < 3.5) number of burst (burst length > 2) Number of loss bursts in 20s Drop period / packet Packet drop (1/30) original Best (3.754) Worst (3.319)

19 PESQ values with algorithms 19 4 PESQ MOS value Number of iteration without algorithm Packet drop (1/30) Best method Mean Best Worst Bad ratio no algorithm Packet drop (1/30) Best method

20 Contents 20 Introduction and motivation VoIP quality vs. ADSL downlinks Simulation and results Conclusions

21 Conclusions 21 In ADSL downlink the VoIP packet can be seriously delayed in DSLAM by the packets from other applications. Some methods should be introduced to guarantee the VoIP quality across ADSL downlinks. Our proposed algorithms will improve the VoIP quality.

22 Thanks!