1. TDM PWs Yaakov (J) Stein Chief Scientist RAD Data Communications

2. TDMoIP Slide 2 Outline 1) Pseudowires 2) Emulating TDM 3) TDMoIP encapsulation formats 4) TDM signaling transport 5) Timing recovery 6) Packet loss and mis-ordering

3. TDMoIP Slide 3 Pseudowires Pseudowire (PW): A mechanism that emulates the essential attributes of a native service while transporting over a packet switched network (PSN)

4. TDMoIP Slide 4 The old model (X.200, OSI) Once upon a time networks were exclusively described by the OSI model However few networks actually work only that way highly inflexible (always need more layers!) some features only in one place (security, mux) missing features (OAM) doesn’t help to design transport networks PHYSICAL LINK NETWORK TRANSPORT SESSION PRESENTATION APPLICATION

5. TDMoIP Slide 5 Simple telephony counter-example this type of scenario important to carriers, and thus to ITU-T not captured by ISO layering model there can be an arbitrary large number of intervening layers all intermediate layers fulfill the same function -- transport voice channel E1 (TDM) E3 (PDH) STM1 (SDH) OC3 (OTN) voice channel E1 (TDM) E3 (PDH) STM1 (SDH) OC3 (OTN) OSI physical layer OSI application layer ?

6. TDMoIP Slide 6 The new model (G.805) A more general and applicable model for transport networks Layer network and trail Layering and partition Basic network modes Interworking Diagrammatic technique References: G.805 generic G.705 PDH G.732 ATM G.806 CO networks G.781 timing G.8010 Ethernet G.809 CL networks G.783 SDH G.8110 MPLS

7. TDMoIP Slide 7 Layer networks Layering Network may be decomposed (vertically) into layer networks Client-server relationship between adjacent layer networks Layer network Basic topological component for information transfer Link in layer network supported by network below Layer network provides link connection to layer above Layers are completely independent Trail Transport entity in layer network Contains client payload and OAM Partitioning Network may be decomposed (horizontally) into subnetworks connected by links Recursively, each subnetwork is similarly decomposed Peer-peer relationship between adjacent subnetworks

8. TDMoIP Slide 8 Network Modes Many native network types (technologies) for each mode –CS: TDM, PDH, SDH, OTN –CO: ATM, FR, MPLS, TCP/IP, SCTP/IP –CL: UDP/IP, IPX, Ethernet, CLNP Can layer any mode over any mode –BUT some layerings may involve performance loss –CL over CO over CS is EASY –CO over CL, or CS over CO is harder –CS over CL is very hard Circuit Switched (CS) Packet Switched (PSN) Connection Oriented (CO) Connectionless (CL)

9. TDMoIP Slide 9 Network interworking (tunneling) Network interworking may be provided by tunneling (edge to edge) Service Interworking requires more complex mechanisms network Native Service Native Service edge Native Service A Native Service B network

10. TDMoIP Slide 10 PseudoWire Emulation Edge to Edge PWE 3 Provider Edge (PE) Customer Edge (CE) Customer Edge (CE) Customer Edge (CE) native service Provider’s PSN PseudoWires (PWs) Customer Edge (CE) Customer Edge (CE) Provider Edge (PE) native service Pseudowire (PW): mechanism that emulates essential attributes of a native service while transporting over a PSN

11. TDMoIP Slide 11 Emulating TDM From PSTN to PSN

12. TDMoIP Slide 12 Classic Telephony Circuit switched ensures signal integrity Very High Reliability (“five nines”) Low Delay and no noticeable echo Timing information transported over the network Mature Signaling Protocols (over 3000 features) T1/E1 CO SWITCH analog lines SONET/SDH NETWORK PBXPBX extensions Access Network T1/E1 or AAL1/2 PBXPBX Core (Backbone) Network Synchronous Non-packet network CO SWITCH

13. TDMoIP Slide 13 A few G.XXX sayings … G.114 (One-way transmission time) –delay < 150 ms acceptable –150 ms < delay < 400 ms conditionally acceptable –delay > 400 ms unacceptable –G.126/G.131 echo control may be needed G.823/G.824 (timing) –primary vs. secondary clocks –jitter masks –wander masks G.826 (error performance) –BER better than 2 * 10 -4 –strict limitation on errored seconds

14. TDMoIP Slide 14 TDM PWs TDMoIP replaces CS core with a PSN The access networks and their protocols remain ! TDM Pseudowire Can G.xxx compliance be maintained? T1/E1/T3/E3 analog lines PBXPBX extensions Access Network PBXPBX T1/E1 or AAL1/2 Packet Switched Network Asynchronous network No timing information transfer

15. TDMoIP Slide 15 Network Comparison TDM Circuit switched Guaranteed BW Low overhead Minimal delay Constant arrival rate Timing transport No information loss PSN Connection oriented / connectionless Shared BW High overhead Delay (introduced by forwarding) Packet delay variation (and bursts) No physical layer clock Packet loss (congestion, errors)

16. TDMoIP Slide 16 TDMoIP Protocol Processing Steps in TDMoIP The synchronous bit stream is segmented The TDM segments are adapted TDMoIP control word is prepended PSN (IP/MPLS) headers are prepended (encapsulation) Packets are transported over PSN to destination PSN headers are utilized and stripped Control word is checked, utilized and stripped TDM stream is reconstituted (using adaptation) and played out TDM frames TDM frames IP Packets PSN IP Packets

17. TDMoIP Slide 17 TDMoIP vs. VoIP Two ways to integrate TDM services into PSNs VoIP Revolution - complete (forklift) CPE replacement New signaling protocols (translation needed) New functionality (e.g. video-phone, presence) TDMoIP Evolution - CPE unchanged, IWF added at edge No change to signaling protocols (network IW) No new functionality Migration path

18. TDMoIP Slide 18 TDMoIP encapsulation formats

19. TDMoIP Slide 19 TDMoIP layering structure PSN / multiplexing Optional RTP header TDMoIP Control Word higher layers Adapted TDM payload

20. TDMoIP Slide 20 PW Multiplexing to reduce resources in core network PWs are sent inside PSN tunnels we often wish to send several PWs in same tunnel to demux we use a PW label for application muxing, IANA has assigned to TDMoIP UDP port number 0x085E (2142) in IP networks we use UDP source port number as bundle ID in MPLS networks we use an inner label for L2TPv3 we could use L2TP multiplexing

21. TDMoIP Slide 21 Packet Components PSN headers ensure packet transported to destination RTP header contains timestamp that may help in timing recovery Control Word enables detection of out-of-order and lost packets indicates critical alarm conditions TDM payload may be adapted to assist in timing recovery and recovery from packet loss to ensure proper transfer of TDM signaling to provide an efficiency vs. latency trade-off

22. TDMoIP Slide 22 TDM over IP and MPLS IP header (20 bytes) UDP header (PW label) (8 bytes) Optional RTP header (12 bytes) TDMoIP Control Word (4bytes) TDM payload PSN label PW label control word TDM payload

23. TDMoIP Slide 23 TDMoIP Control Word PID (4b) special uses flags (4 b) –L bit (Local failure) –R bit (Remote failure) FRG (2 bits) indicates fragmentation (only for special uses) Length (6 b) used when packet may be padded Sequence Number (16 b) used to detect packet loss / misordering PID flags FRG Length Sequence Number

24. TDMoIP Slide 24 TDM Payload What needs to be transported from end to end? Voice (telephony quality, low delay, echo-less) Tones (for dialing, PIN, etc.) Fax and modem transmissions Signaling (there are 1000s of PSTN features!) Timing T1/E1 frame SYNC TS1 TS2 TS3 signaling bits … … TSn (1 byte) “timeslots”

25. TDMoIP Slide 25 Why not N bytes? Why don’t we simply encapsulate N bytes frame? IP N TDM octets UDP RTP? because a single lost packet would cause service interruption need constant N (else don’t know how many TDM bytes were lost) need to conceal lost packet by proper amount of AIS “all ones” TDM synchronization would be lost SAToP is good for well-engineered networks essentially no packet loss very low PDV (see below)

26. TDMoIP Slide 26 Why not one frame? Why don’t we simply encapsulate the T1/E1 frame? IP T1/E1 frame 24 or 32 bytes UDP RTP? because it is inefficient - however N frames is reasonable (structure-locking) because a single lost packet could cause service interruption and for CAS, signaling uses a superframe (16/24 frames) so superframe integrity must be respected too because we want to efficiently handle fractional T1/E1 because we want a latency vs. efficiency trade-off

27. TDMoIP Slide 27 TDM Structure handling of TDM depends on its structure unstructured TDM (TDM = arbitrary stream of bits) structured TDM … SYNC TS1 TS2 TS3 signaling bits … … TSn (1 byte) channelized (single byte timeslots) SYNCSYNC framed (8000 frames per second) SYNCSYNC SYNCSYNC multiframed frame … multiframe

28. TDMoIP Slide 28 TDM transport types Structure-agnostic transport (SAToP) for unstructured TDM even if there is structure, we ignore it simplest way of making payload OK if network is well-engineered Structure-aware transport (TDMoIP, CESoPSN) take TDM structure into account must decide which level of structure (frame, multiframe, …) can overcome PSN impairments (PDV, packet loss, etc)

29. TDMoIP Slide 29 Structure aware encapsulations Structure-locked encapsulation (CESoPSN) Structure-indicated encapsulation (TDMoIP – AAL1 mode) Structure-reassembled encapsulation (TDMoIP – AAL2 mode) headers TDM structure headers AAL1 subframe headers AAL2 minicell

30. TDMoIP Slide 30 Structure indication - AAL1 For robust emulation: adding a packet sequence number adding a pointer to the next superframe boundary only sending timeslots in use allowing multiple frames per packet for example UDP/IP T1/E1 frames (only timeslots in use) seqnum (with CRC) ptr TS1 TS2 TS5 TS7 7 @

31. TDMoIP Slide 31 Structure reassembly - AAL2 AAL1 is inefficient when timeslots are dynamically allocated each minicell consists of a header and buffered data minicell header contains: –CID (Channel IDentifier) –LI (Length Indicator) = length-1 –UUI (User-User Indication) counter + payload type ID TDM frame 124531122334455 12345hdr12345 12345 PSN hdrsCW TS1TS2TS3

32. TDMoIP Slide 32 TDM Signaling TDM Signaling

33. TDMoIP Slide 33 Signaling? signaling is used for network control –call setup/tear-down (including routing) –OAM –billing in TDM networks there may be different types: –Subscriber - CO –CO - CO –CO - CPE (e.g. PBX) there are four common PSTN signaling techniques: –Analog * (E&M, ground-start/loop-start, ring-voltage, etc) –In-band (dial-tone, ring-back, DTMF,etc) –CAS – Channel Associated Signaling –CCS – Common Channel Signaling * we needn’t discuss the analog techniques

34. TDMoIP Slide 34 In-band signaling in-band signaling is transferred in the audio (200-3600Hz) band for example: –call progress tones (dial tone, ring back) –DTMF tones, –FSK for caller identification, –MFR1 in North America or MFCR2 in Europe, audible tones in TDM time slot automatically forwarded this is not the case for VoIP! –speech compression may not pass (need tone relay) –VoIP protocols replace legacy signaling with its own SIP H.323 Megaco

35. TDMoIP Slide 35 CAS CAS is carried in the same T1/E1 as payload –but not in the audio –T1 uses robbed bits –E1 uses a dedicated time slot (usually TS16) Readily handled by TDMoIP (even for fractional T1/E1 links) VoIP systems need to –detect the CAS bits, –interpret them according to the appropriate protocol –transport them through PSN using a relay protocol –finally regenerate and recombine them at the far end

36. TDMoIP Slide 36 CCS Examples: ISDN PRI signaling, SS7 if occupy a TDM timeslot (trunk associated) then forwarded by TDMoIP (see HDLCoIP) if not trunk associated, then forwarded by signaling network or signaling gateway employed encapsulate (relay) the native signaling forward as additional traffic through the PSN

37. TDMoIP Slide 37 HDLCoIP HDLCoIP intended to operate in port mode Data / control messages transparently transported Assume messages shorter than the MTU (no fragmentation) Only use when has potential to significantly compress BW Transmission : –monitor flags until frame detected –test FCS –if incorrect - discarded –if correct - perform unstuffing flags and FCS removed send frame

38. TDMoIP Slide 38 TDM Timing Recovery TDM Timing Recovery

39. TDMoIP Slide 39 TDM Jitter and Wander Jitter = short term timing variation * (i.e. fast jumps - frequency > 10 Hz) Jitter amplitude in UI pp Unit Interval pk-pk E1 : 1 UI pp = 1/2MHz = 488 ns Wander = long term timing variation * (i.e. slow moving- frequency < 10 Hz) Measure in MTIE(  ) or TDEV(  ) MTIE - max pk-pk error TDEV expected deviation Mask as function of  * compared to reference clock Note: requirements for E1 given in G.823 for T1 given in G.824

40. TDMoIP Slide 40 PSN - Delay and PDV PSNs do not carry timing clock recovery required for TDMoIP PSNs introduce delay and packet delay variation (PDV) Delay degrades perceived voice quality PDV makes clock recovery difficult PSN The arrival time is not constant!!! E1/T1 VOICE DATA E1/T1 VOICE DATA TDMoIP GW TDMoIP GW

41. TDMoIP Slide 41 Jitter Buffer Arriving TDMoIP packets written into jitter buffer Once buffer filled 1/2 can start reading from buffer Packets read from jitter buffer at constant rate How do we know the right rate? How do we guard against buffer overflow/underflow? PSN E1/T1 VOICE DATA E1/T1 VOICE DATA Jitter Buffer TDMoIP GW TDMoIP GW

42. TDMoIP Slide 42 Clock Recovery The packets are injected into network ingress at times T n For TDM the source packet rate R is constant T n = n / R The network delay D n can be considered to be the sum of typical delay d and random delay variation V n The packets are received at network egress at times t n t n = T n + D n = T n + d + V n By proper averaging/filtering  t n  = T n + d = n / R + d and the packet rate R has been recovered

43. TDMoIP Slide 43 What timing information ? We have three clues to the source clock rate TOA Jitter buffer level TOD (if in protocol) TOD PSN TDMoIP GW TDMoIP GW TOA level

44. TDMoIP Slide 44 FLL We can estimate the rate R by counting the number of arrivals N per unit time T the longer the averaging the better the estimate R = N / T Open loop frequency setting Better method is closed loop Measure reception rate F n = 1 / ( t n - t n-1 ) Correct present rate F according to filtered F n F = F +  TDMoIP GW F nn

45. TDMoIP Slide 45 PLL Phase difference between write (arrival) clock and read (present local) clock Number of packets written into the jitter buffer minus the number of packets read from the jitter buffer write events read events 360 o 270 o counter

46. TDMoIP Slide 46 Packet Loss and Misordering Packet Loss and Misordering

47. TDMoIP Slide 47 Reasons for packet loss In a perfect network all packets should reach their destination In real networks, some packets are lost Loss is caused by bit errors invalidating the data (detected by ECC) intentional dropping by forwarder because of congestion intentional dropping by forwarder due to policy ( e.g. (W) RED ) router

48. TDMoIP Slide 48 Handling of packet loss In order to maintain timing SOMETHING must be output towards the TDM interface when a packet is lost Packet Loss Concealment methods: fixed replay interpolation PSN

49. TDMoIP Slide 49 Voice Quality Comparison See draft-stein-pwe3-tdm-packetloss-00.txt

50. TDMoIP Slide 50 Mis-ordering In a perfect network all packets should arrive in proper order In real networks, some packets are delayed (or even duplicated!) Misordering is caused by parallel paths –aggravated by load balancing mechanisms Misordering can be handled by Reordering (from jitter buffer) Handling as packet loss and dropping later router 1 2 3 4 5 1 2 4 3 5 1 2 3 4 5