Prof. Dr. P. Tran-Gia VoIP over Wireless Opportunities and Challenges Universität Würzburg Lehrstuhl für verteilte Systeme
H.323 RTP Codec Voice-over-IP over Wireless (VoIPoW) UDP IMS G723.1 SIP G729 HSDPA WiMAX Skype G711 WiFi IP UMTS AMR LTE QoS QoE Play-out buffer Header Compression PTT Packet Loss Concealment
Overview > Why Voice-over-IP over Wireless? Motivation and Advantages Current problems and challenges > VoIP transmission in different radio access technologies IEEE802.11 WLAN IEEE802.16 WiMAX UMTS > An outlook to the future/ current research > Conclusion
Why VoIP over Wireless? flexibility of IP spectral efficiency future VoIPoW VoIPoW today circuit-switched voice QoS
VoIP and VoIPoW > Aspects of VoIP signaling and connection management (SIP, H.323, Skype, IMS ) transport protocol (RTP) voice codec play-out buffer packet loss concealment (FEC, ) jitter, packet loss, delay > Additional aspects of VoIPoW mobility management (MobileIP, VHO, IMS, ) properties of radio transmission high bit error rate time-variant channel limited, expensive bandwidth different radio access technologies must fit to each other
Problems, Challenges, Solutions > Desired: high spectral efficiency problem: small packets, large header RTP/UDP/IP header avoidable through header compression problem: MAC layer overhead frame aggregation problem: no delivery of erroneous packets voice codec could deal with rare bit errors MAC/UDP require correct packets higher SIR, more robust transmission, more retransmissions > Desired: low packet loss, delay, and jitter problems: retransmissions random access/medium access scheduling time-variant channel quality solutions: play-out buffer, adaptive codec, packet loss concealment
VoIP over Wireless LAN > IEEE802.11abg random access on up- and downlink no service differentiation bad spectral efficiency alternative: polling with PCF (point coordination function) > IEEE802.11e service differentiation dedicated resource allocation with HCCA (Hybrid Control Function Controlled Channel Access) > Header compression is possible but not used > Future challenges admission control adaptive contention parameters
VoIP over IEEE802.11g/e with Header Compression maximum number of clients 90 80 70 60 50 40 30 20 IEEE802.11g without service differentiation 54 Mbps 36 Mbps 24 Mbps 18 Mbps 12 Mbps 9 Mbps 6 Mbps dashed: without header compression solid: using header compression maximum number of clients 55 50 45 40 35 30 25 20 15 10 IEEE802.11e with service differentiation 54 Mbps 36 Mbps 24 Mbps 18 Mbps 12 Mbps 9 Mbps 6 Mbps dashed: without header compression solid: using header compression 10 10 15 20 25 30 35 40 frame size in milliseconds > Contention parameters for VoIP support decrease VoIP capacity adaptive contention parameters > Small benefits from header compression 5 10 15 20 25 30 35 40 frame size in milliseconds
VoIP over WiMAX (IEEE 802.16) > Possible scheduling services in WiMAX UGS (Unsolicited Grant Service) essentially a dedicated channel no support for silence suppression on uplink rt-ps (real-time Polling Service) regular dedicated bandwidth request opportunities support for silence suppression on uplink BE (Best Effort) not intended for VoIP contention based bandwidth requests collision free data transmission introduces delay and jitter > Problem: Services intended for VoIP (UGS, rt-ps) require detailed traffic characteristics and provide detailed QoS VoIP e.g. Skype transmitted over BE
VoIP over Best-Effort Connections in Fixed WiMAX > Performance of VoIP connections over BE service > No background traffic, no packet loss, no header compression > 5MHz TDD > G723.1 Codec: 480bit every 30ms 10ms frame, QPSK, ½ code rate 5ms frame, 16QAM, ¾ code rate P{delay P(T <= 10 frames) P <= 100ms} 1 0.8 0.6 0.4 4 8 0.2 12 16 0 5 10 15 20 number of VoIP G723.1 connections P{delay <= 100ms} 1 0.95 0.9 0.85 4 8 0.8 12 16 0.75 5 20 10 15 20 25 30 35 number of VoIP G723.1 connections
VoIP over UMTS > Today: Typical: circuit-switched voice over dedicated channels using AMR codec (Adaptive Multi-Rate) VoIP transmission as normal data traffic on DCH/HSDPA typically no service differentiation > Future: IMS, special dedicated channels for VoIP Special support for VoIP over HSDPA/HSUPA? Scheduling disciplines VoIP in UTRA LTE enhanced VoIP capacity by enhanced transmission techniques? > CDMA2000 1x EV-DO Rev A similar to HSDPA/HSUPA special support for VoIP
Skype over UMTS > T. Hoßfeld, A. Binzenhöfer, M. Fiedler, K. Tutschku, Measurement and Analysis of Skype VoIP Traffic in 3G UMTS Systems, IPS-MoMe 2006 > ilbc codec: 108 Byte voice packet with every 60 ms 1 Uplink 1 Downlink CDF 0.8 0.6 0.4 0.2 DSL receiver with 1024 kbps UMTS sender with 64 kbps CDF 0.8 0.6 0.4 0.2 UMTS receiver with 384 kbps DSL sender with 128 kbps 0 0 100 200 300 400 packet interarrival time [ms] > considerable jitter > PESQ ~2.2 instead of ~3 in bottleneck LAN with 64 kbps 0 0 50 100 150 packet interarrival time [ms] > packet inter-arrival time deterministic > PESQ ~2.5 instead of ~3 in bottleneck LAN with 128 kbps
Outlook to the Future > Development of VoIPoW current codec optimized for circuit-switched data development of special codecs for VoIPoW differentiated packet dropping > Challenges and opportunities for VoIPoW adaptive modulation and coding channel-aware scheduling frequency-selective scheduling enhanced antenna techniques multi-hop networks heterogeneous networks
Skype: Adaptive Codec > ISAC codec with artificial time-variant packet loss > Packetization independent of packet loss > Variable bit rate by increasing packet size, i.e. more audio data 0.5 50 0.4 400 Packet loss packet loss receiver sender sender 0 0 0.5 1 1.5 2 2.5 0 3 Time [ms] x 10 6 Time between packets [ms] Packet loss 0.2 packet size packet loss 0 0 0 0.5 1 1.5 2 2.5 3 Time [ms] x 10 6 200 Packet sizes
Scenario: VoIP over HSDPA > G.711 codec: 64 kbps 160bytes per 20 ms > Performance of different schedulers Maximum CQI Scheduler optimizes throughput channel-aware starvation, unfairness Proportional Fair Scheduler optimizes throughput considering long-term throughput fairness channel-aware Round Robin optimal short-term time fairness channel-unaware FIFO First In First Out common buffer channel-unaware DEDF Scheduler Dynamic-Earliest-Deadline-First considers buffering time channel-aware optimizes delay CH-EDD Scheduler Channel-Dependent-Earliest-Due-Date considers buffering time channel-aware drops packet after deadline
VoIP over HSDPA packet dropping probability User MAX PF DEDF CH-EDD 13 User 13 1.81 % < 1 % < 1 % 2.09 % 16 6.75 % 1.90 % < 1 % 3.99 % 20 14.06 % 7.48 % < 1 % 7.76 % 16 User 20 User
Conclusion > Situation today circuit-switched voice is optimized for QoS and spectral efficiency little/no support for VoIP in cellular networks VoIPoW is VoIP over WLAN > Drivers for VoIPoW in cellular networks are all-ip infrastructure, IMS vertical handover possibilities of packet-switched radio transmission > VoIP over Wireless will replace circuit-switched voice in the future > Future challenges and opportunities enhanced packet-switched radio transmission multi-hop development of VoIP optimized codecs charging