INTERNATIONAL TELECOMMUNICATION UNION COM 11-R 205-E February 2000 Original: English TELECOMMUNICATION STANDARDIZATION SECTOR STUDY PERIOD 1997 - 2000 Question: 16/11 Texte disponible seulement en Text available only in Texto disponible solamente en E STUDY GROUP 11 - REPORT R 205 SOURCE*: STUDY GROUP 11 TITLE: IMPLEMENTORS' GUIDE (12/99) FOR RECOMMENDATION Q.706 (03/93) ________ Summary This Q.706 Implementors' Guide shows the MSU queueing delay and its standard deviation for the extreme error case of signal unit error rate (not MSU error rate) of ~0.004. Graphs are provided for several MSU sizes: 120 bits (i.e. 15 octets, the Figures 5 and 6/Q.706 value), 25, 50, 100, 150 and the limiting case of 279 octets. This Guide provides formulas for the Basic Error Correction (BEC) queueing delay in the presence of disturbance which yield the correct delay when the link loading is zero. To cater for long messages and high link loadings, the signalling loop delay is no longer assumed constant, and is normalized to that of a link with zero loading. Some typographical errors in the PCR formulas have been corrected, and two technical errors which affected moderately the standard deviation of queueing delay in the presence of disturbance have been corrected. Annex B of Q.706 is expanded to include a brief derivation of the BEC formulas. Annex B.2 contains the calculation for the effect of variable loop delay on the PCR formulas. This Guide aligns Q.706 with ITU-T Recommendation E.733, used for planning SS No. 7 networks. * Contact: TSB SG11\REPORT\R205E.WW9 (102326) Tel: Fax: +44 22 730 5866 +44 22 730 5853 09.02.00 -2COM 11-R 205-E CONTENTS Page 1 Introduction ................................................................................................................. 3 1.1 References ................................................................................................................... 3 1.2 Acknowledgements ..................................................................................................... 3 1.3 Background ................................................................................................................. 3 1.4 Scope of the guide....................................................................................................... 4 1.5 Contacts ...................................................................................................................... 4 1.6 Document history........................................................................................................ 4 2 Changes to Q.706 Section 4.2 ..................................................................................... 4 §4.2 §4.2.1 §4.2.2 §4.2.3 §4.2.4 §4.2.5 Queueing delays .......................................................................................................... General ........................................................................................................................ Basic error correction queueing delays ....................................................................... Preventive cyclic retransmission queueing delays...................................................... Formulas ..................................................................................................................... Examples..................................................................................................................... 4 4 6 7 9 10 3 Replace Figure 5/Q.706 .............................................................................................. 13 4 Replace Figure 6/Q.706 .............................................................................................. 19 5 Replace Figure 8/Q.706 .............................................................................................. 25 6 Replace Table 6/Q.706 ............................................................................................... 26 7 Replace Table 7/Q.706 ............................................................................................... 27 8 Replace Figure 13/Q.706 ............................................................................................ 28 9 Replace Figure 15/Q.706 ............................................................................................ 29 10 Replace Figure 16/Q.706 ............................................................................................ 30 11 In Section 5.2.5, insert a new Table 10bis .................................................................. 31 12 Replacement Annex B of Q.706 ................................................................................. 31 B.1 B.1.1 B.1.2 B.1.3 Derivation of basic error correction (BEC) message signal unit (MSU) queueing delays........................................................................................................... Preliminary definitions and formulas ......................................................................... BEC queueing delay in the absence of disturbance .................................................... BEC queueing delay in the presence of disturbance................................................... 31 31 34 36 B.2 PCR Loop delay .......................................................................................................... 38 B.3 Calculation of the kth moments of the MSU emission time for Tod ............................ 39 B.4 Approximate calculation of the 95% - values of Tod .................................................. 40 SG11\REPORT\R205E.WW9 (102326) 09.02.00 -3COM 11-R 205-E 1 Introduction 1.1 References [1] ITU-T Recommendation Q.703 (07/96) - Specifications of Signalling System No. 7 – Message Transfer Part, Signalling Link. [2] J.N. Daigle: Queueing Theory for Telecommunications, published by Addison – Wesley, 1992. [3] R.A. Skoog: Performance and Engineering of Common Channel Signaling Networks supporting ISDN, in Traffic Engineering for ISDN and Planning, published by Elsevier Science Publishers B.V. (North Holland), 1988. [4] L. Kleinrock: Queueing Systems, Volume 1, published by John Wiley & Sons, 1975. [5] D.R. Cox and H.D. Miller: The Theory of Stochastic Processes, published by Chapman and Hall Ltd., 1967. [6] ITU-T Recommendation E.733 (11/98) - Methods for dimensioning resources in Signalling System No. 7 networks. [7] ITU-T Recommendation G.821 - Error performance on an international digital connection forming part of an integrated services digital network, Vol. III. [8] Y. Watanabe and Y. Ikeda: Traffic Characteristics of PCR method for CCITT Signalling System No. 7, in Proceedings of the 10th International Teletraffic Congress 1993, Volume 1. 1.2 Acknowledgements This Implementors' Guide was prompted by a liaison from ITU-T Study Group 2, which highlighted the Q.706 BEC inconsistencies and errors. It provided the Laplace Transform of the MSU BEC queueing delay distribution in the absence of disturbance, and an outline of the BEC virtual service time derivation for the presence of disturbance. It also contained the mean and variance of the BEC queueing delays in the absence and in the presence of disturbance, and proposed replacement graphs for the Pu = 0.004 cases in Figures 5 and 6/Q.706. 1.3 Background This Q.706 Implementors' Guide modifies Figures 5/Q.706 and 6/Q.706 to show for the extreme error case the mean and standard deviation of queueing delays at the error rate when link changeover would occur. Link changeover occurs at a signal unit error rate (not MSU error rate) of ~0.004, and queueing delays with this SU error rate are significantly higher than those shown for an MSU error rate of 0.004. Graphs are provided for several MSU sizes: 120 bits (i.e. 15 octets, the Figures 5 and 6/Q.706 value), 25, 50, 100, 150 and the limiting case of 279 octets. This Guide provides formulas for the Basic Error Correction (BEC) queueing delay in the presence of disturbance which yield the correct delay when the link loading is zero. Those currently in Q.706 have a term dependent on the MSU error rate even at zero MSU loading. The new formulas are derived using the concept of "virtual service time" in the presence of disturbance applied to the BEC formulas for the absence of disturbance. Annex B of Q.706 is expanded to include a brief derivation of the BEC queueing delay in the absence of disturbance, it also outlines the derivation and application of the BEC virtual MSU service time in the presence of disturbance. The Annex now provides the Laplace-Stieltjes transform of the MSU queueing delay, which is used to derive the mean and variance of the queueing delay as functions of the first 3 moments of the MSU emission time on to the link. If the SG11\REPORT\R205E.WW9 (102326) 09.02.00 -4COM 11-R 205-E MSU length distribution is known, the transform of the MSU queueing delay can be inverted to obtain the queueing time distribution, which would replace the approximate negative exponential formula given. The assumption of constant signalling loop delay has been replaced. A part of the loop delay is due to the emission from the MSU receiving end of an acknowledgement and its absorption at the MSU sending end. This acknowledgement could be a FISU, or an MSU. The acknowledgement type depends upon the relative frequency of MSUs on the acknowledgement path, and the emission/absorption time for an MSU depends upon the MSU length. The time before the acknowledgement can be transmitted is the residual emission time of the signal unit before it, and this depends upon the link loading and MSU lengths. Section 4.2 has been split into separate subsections for BEC and Preventive Cyclic Retransmission to improve clarity, and some technical and typographical errors have been corrected here and in the rest of Q.706. This Guide aligns Q.706 with ITU-T Recommendation E.733, used for planning SS No. 7 networks. The Guide is intended to be an additional authoritative source of information for implementors, to be read in conjunction with the Recommendation itself. 1.4 Scope of the guide This guide records enhancements to the Recommendation in the following categories: • corrections of editorial errors; • corrections of technical errors; • extension to support ISDN User Part (ISUP) and Signalling Connection Control Part (SCCP) message lengths. 1.5 Contacts Editor Q.706 1.6 R.A. Adams Lucent Technologies (UK) Swindon Road Malmesbury, Wiltshire SN16 9NA United Kingdom Tel: +44 1666 83 2503 Fax: +44 1666 83 2712 E-mail: raadams1@lucent.com Document history Version Summary 12/99 New Implementors' Guide 2 Changes to Q.706 Section 4.2 §4.2 Queueing delays §4.2.1 General The MTP handles messages from different User Parts on a time-shared basis. With time-sharing, signalling delay occurs when it is necessary to process more than one message in a given interval of time. When this occurs, a queue is built up from which messages are transmitted in order of their times of arrival. SG11\REPORT\R205E.WW9 (102326) 09.02.00 -5COM 11-R 205-E There are two different types of queueing delays: queueing delay in the absence of disturbances and total queueing delay. §4.2.1.1 Assumptions for derivation of the formulas The queueing delay formulas are basically derived from the M/G/1 queue with priority assignment. The assumptions for the derivation of the formulas in the absence of disturbances are as follows: a) the interarrival time distribution is exponential (M); b) the service time distribution is general (G); c) the number of servers is one (1); d) the service priority refers to the transmission priority within level 2 (see 11.2/Q.703); however, service of link status signal units and independent flags are not considered; e) the signalling link loop delay is the time to receive an acknowledgement to an MSU. It consists of a constant part, which is the sum of the signal unit MTP level 2 processing time in signalling terminals and twice the link propagation time, and a variable part. The variable part consists of the remaining emission time of the signal unit currently being sent in the return direction (which is the delay to start transmitting the acknowledgement) plus the emission/absorption time of the acknowledgement. For BEC, this acknowledgement could be an MSU (with probability ), or a FISU (with probability (1-)) where is the relative frequency of MSU arrivals. a a (1 a ). f) Tm Tf For an errored MSU in BEC, the variable part contains also the absorption time of the next (assumed correct) SU after the MSU in error, and the remaining emission time of the SU being sent in the forward direction before the MSU is retransmitted; for PCR, the FISU traffic is defined as a3, hence the probability of the acknowledgement being an MSU is (1 a 3 ).t f a 3 (1 a 3 ).t f , with a3 calculated assuming the mean value Tm for TMSU (t). For both BEC and PCR the link loading is assumed equal in both directions. For the formulas in the presence of disturbances, the assumptions are as follows: g) Bit errors are randomly distributed, this makes transmission errors of message signal units random; h) MSU transmission errors are statistically independent of each other; i) the additional delay caused by retransmission(s) of erroneous signal units is considered as a part of the virtual service time of the concerned signal unit. Furthermore, the formula of the proportion of messages delayed more than a given time is derived from the assumption that the probability density function of the queueing delay distribution may be exponentially decreasing where the delay time is relatively large. §4.2.1.2 Factors and parameters a) The notations and factors required for calculation of the queueing delays are as follows: Qa Mean queueing delay in the absence of disturbances Variance of queueing delay in the absence of disturbances a2 SG11\REPORT\R205E.WW9 (102326) 09.02.00 -6COM 11-R 205-E Qt Mean total queueing delay t2 Variance of total queueing delay Proportion of messages delayed more than T Traffic loading by message signal units (MSU) (excluding retransmission) Traffic loading by MSUs including retransmission Mean emission time of message signal units Emission time of fill-in signal units (FISUs) Signalling loop delay including processing time in signalling terminals at zero link load, excluding FISU absorption and emission times 2Tf and excluding remaining emission and absorption times Tf Signalling loop delay (variable) Error probability of message signal units Error probability of signal units (FISUs and MSUs) P(T) a aeff Tm Tf K TL Pu Psu k1 2nd moment of MSU emission time k2 3rd moment of MSU emission time k3 4th moment of MSU emission time Tm2 Tm3 Tm4 T M( 2 ) Tm2 T M( 3) Tm3 T M( 4 ) Tm4 NOTE - As a consequence of zero insertion at level 2 (see 3.2/Q.703), the length of the emitted signal unit will be increased by approximately 1.6 per cent on average. However, this increase has negligible effect on the calculation. b) The parameters used in the formulas are as follows: t f T f / Tm t L TL / Tm §4.2.2 Basic error correction queueing delays The functions, factors and parameters used are as follows: ( s) e st . dTM (t ) , 0 (s) is the Laplace-Stieltjes transform of the probability distribution function TM (t) of the MSU emission time. M ( s )(1 a). Tm 1 s.T .(1 e f ). , Tf s.Tm a.1 ( s ) M (s) is the Laplace-Stieltjes transform of the MSU BEC queueing delay in the absence of disturbance. SG11\REPORT\R205E.WW9 (102326) 09.02.00 -7COM 11-R 205-E Pu , 1 Pu P (1 Pu ) u , (1 Pu ) 2 N (1) N (2) N ( 3) Pu (1 4 Pu Pu2 ) , (1 Pu ) 3 a a (1 a ). Tm Tf TL(1) K 2(1 ) 1 a .T f 2 ak1 .Tm 2 2 k1Tm2 (1 )T f2 2.K ak1Tm (1 a)T f 4.Tm (1 )T f . ak1Tm (1 a )T f 4 K Tm (1 )T f 23 T f2 (1 a ) ak 2 Tm2 12 .(1 a ).T f ak1Tm 2 T L(3) K 3 6.Tm (1 )T f .k1Tm2 (1 )T f2 2k 2 Tm3 (1 )T f3 2.K ak 2 Tm2 (1 a)T f2 4.ak 2 Tm2 (1 a)T f2 .Tm (1 )T f 6 K 2 Tm (1 )T f 3.K 2 ak1Tm (1 a)T f 6 K Tm (1 )T f 2 k1Tm2 (1 )T f2 6.Tm (1 )T f 2 k1Tm2 (1 )T f2 .ak1Tm (1 a)T f 12 K Tm (1 )T f . ak1Tm (1 a)T f 12 .(1 a)T f3 a.k 3 .Tm3 32 .K .(1 a)T f ak1Tm 2 3.(1 )T f Tm . ak1Tm (1 a)T f 2 ak1Tm (1 a)T f .ak 2 Tm2 (1 a)T f2 T L( 2) K 2 2 Tm (1 )T f (1) Tvir Tm 1 N (1) N (1) TL(1) T N 3N 3T T .N 2 ( 2) Tvir TM(2) N (2) 2N (1) 1 2TmTL(1) N (2) N (1) TL(2) N (1) TL(1) . N (2) N (1) (3) Tvir a eff a (3) ( 2) 3N (1) 1 3TM( 2) TL(1) N (3) 2 N ( 2) N (1) 3TL( 2) Tm N ( 2) N (1) (1) 2 L (3) N (1) TL(3) N (1) 3TL( 2) TL(1) N ( 2) N (1) TL(1) (3) M m N 3 (3) 3N ( 2) 2 N (1) (1) Tvir Tm §4.2.3 Preventive cyclic retransmission queueing delays Assumptions: 1) the forced retransmission case is not considered; 2) after the error occurs, the retransmitted signal units of second priority are accepted at the receiving end until the sequence number of the last sent new signal unit is caught up by that of the last retransmitted signal unit. The functions, factors and parameters used in the formulas are as follows: a eff .k1 K 1 a eff a 3 exp a.(1 ).t f .t f Tm 2 2 : traffic loading caused by fill-in signal units. (1 a 3 ).t f a 3 (1 a 3 ).t f SG11\REPORT\R205E.WW9 (102326) 09.02.00 -8COM 11-R 205-E t L(1) t L( 2) 1 a eff K 1 Tm 2 K Tm 2 a k .t f eff 1 2 K 2. . (1 ).T f k1 (1 )t 2f K . a eff k1 (1 a eff )t f T m (1 )t f . a eff k1 (1 a eff )t f 13 . a eff k 2 (1 a eff )t 2f 3 K K k 2 (1 )t 3f t L(3) . a eff k 2 (1 a eff )t 2f (1 )t f . a eff k 2 (1 a eff )t 2f T T m m 2 2 K K . (1 )t f 32 3. T . (1 a eff ).t f a eff .k1 Tm m K 3. . k1 (1 )t 2f 32 . k1 (1 )t 2f . (1 a eff ).t f a eff .k1 Tm 3. K . (1 )t f . (1 a eff ).t f a eff .k1 14 . (1 a eff )t 3f a eff .k 3 Tm a eff 1 a 3 a z 1 a a3 H 1 a.t L(1) H 2 a.k1 .t L(1) a 2 .t L( 2) H 3 a.k 2 .t L(1) 3.a 2 .k1 .t L( 2) a 3 .t L(3) F1 F2 F3 qa a.t L(1) 2 a.k1 .t L(1) 2 a.k 2 .t L(1) 2 a 2 .t L( 2) a 3 2 .k1 .t L( 2) a 3 .t L(3) 4 k1 ( a az ) a3t f 2 (1 a ) k2 ( a az ) a3t 2f ak1 sa qa 1 a 3(1 a ) ( a az ) k3 a3t 3f 3 ak1sa 2 ak2 qa ta 2 (1 a ) 4 (1 a ) Z1 = 2 Pu(1 H1) Z2 = 4k1 Pu(5k1 6H1 H2) Z3 = 8k2 Pu(19k2 27k1H1 9H2 H3) Y2 = sa 4k1 F2 2{qa(2 F1) 2F1} Y3 = ta 8k2 F3 3{sa(2 F1) qa(4k1 F2) 2F2 4k1F1} 12qaF1 1 a{2 Pu (1 at L(1) )} qd 2 q a at L(1) / 2 aZ 2 Y2 2(1 aZ1 ) SG11\REPORT\R205E.WW9 (102326) 09.02.00 -9COM 11-R 205-E sd aZ 2 aZ Y3 qd 3 1 aZ1 3(1 aZ1 ) qb q a 1 F1 1 a sb qc sc s a k1 F2 (1 a) 2 2{q a (1 F1 ) F1 } (1 a ) 2 ak1 q b (1 a) 2 q d 1 Pu (1 H1 ) 1 a s d k1 Pu (3k1 H 2 ) (1 a) Pv Pu .a. 2 2. q d Pu {q d (1 H 1 ) 2 H 1 } (1 a) q a 2 at L(1) / 2 a a 2 t L(1) 1 Pu 1 2a 1 2a 2 ak1 q c (1 a) 2 §4.2.4 Formulas The formulas of the mean and the variance of the queueing delays are described in Table 2. The proportion of messages delayed more than a given time Tx is: T Qx x P(Tx ) exp x x where Qx and x denote the mean and the standard deviation of queueing delay, respectively. This approximation is better suited in absence of disturbances. In the presence of disturbances the actual distribution may be deviated further. Relation between P(Tx) and Tx is shown in Figure 2. TABLE 2/Q.706 Queueing delay formula Error correction method Disturbance Absent BEC Present Absent PCR Present Qa T ( 2) 1 a . T f M 2 1 a Tm 2a T ( 2) T ( 3) 1 a 1 a M M 12 4 1 a Tm 3 1 a Tm Qt a eff T ( 2) 1 . T f vir (1) 2 1 a eff Tvir 2t ( 3) a eff T ( 2) Tvir 1 a eff 1 vir (1) (1) 12 4 1 a eff Tvir 3 1 a eff Tvir 2 T f2 2 T f2 Qa qa Tm Qt (1 Pu Pv ) qa Tm + Puqb + Pvqc SG11\REPORT\R205E.WW9 (102326) Variance 2 Mean Q a2 Tm2 t2 Tm2 s a q a2 (1 Pu Pv ) sa Pusb Pv sc Qt2 Tm2 09.02.00 - 10 COM 11-R 205-E P (TX ), probability of MSU delayed more than Tx 10 –1 10 –2 10 –3 10 –4 10 –5 10 –6 Qx Q x + 8 x Q x + 4 x Qx + 12 x Queueing delay time (Tx ) Qx x s T1156380-93/d02 Mean queueing delay (see Figure 3) Standard deviation (see Figure 4) FIGURE 2/Q.706 Probability of message signal unit delayed more than Tx FIGURE 2/Q.706...[D02] = 10.5 CM §4.2.5 Examples Assuming the traffic models given in Table 3, examples of queueing delays are calculated as listed in Table 4. In many cases, the bit error probability of signalling data links (terrestrial and satellite) is better than 10–7, at which rate the effects on queueing delays are minimal. NOTE - The values for models A and B in Table 3 were determined based on TUP messages. With the increase of the effective message length, using ISUP and TC, these values may be expected to be increased. Sheets 2 to 6 of Figures 5/Q.706 and 6/Q.706 show the effect of increased MSU length on BEC queueing delay. TABLE 3/Q.706 Traffic model Model A Message length (bits) 120 104 304 constant Percent 100 92 8 100 Mean message length (octets) 15 15 25-279 k1 1.0 1.2 1.0 k2 1.0 1.9 1.0 k3 1.0 3.8 1.0 SG11\REPORT\R205E.WW9 (102326) B C 09.02.00 - 11 COM 11-R 205-E TABLE 4/Q.706 List of examples Figure Error control Queueing delay Disturbance Model 3 BEC/PCR Mean Absent A and B 4 BEC/PCR Standard deviation Absent A and B 5 sheet 1 BEC Mean Present A 6 sheet 1 BEC Standard deviation Present A 7 PCR Mean Present A 8 PCR Standard deviation Present A 5 sheets 2-6 BEC Mean Present C 6 sheets 2-6 BEC Standard deviation Present C ms 2.5 2.0 1.2 R PC Mean queueing delay (Qa ) 1.0 sic Ba 1.5 TL 0 ms = 60 0.8 0.6 1.0 30 ms 0.4 s 30 m 0.5 0.2 0 0 0.1 0.2 0.3 0.4 a, traffic loading of MSU 0.5 Q a /Tm , mean queueing delay in multiples of T m Tm = 1.875 ms (120 bits and 64 kbit/s) Tf = 0.75 ms (48 bits and 64 kbit/s) 0 Erlangs 0.6 T1156390-93/d03 Model A Model B FIGURE 3/Q.706 Mean queueing delay of each channel of traffic in absence of disturbance FIGURE 3/Q.7...[D03] = 12.5 CM SG11\REPORT\R205E.WW9 (102326) 09.02.00 - 12 COM 11-R 205-E ms 2.5 T m = 1.875 ms (120 bits and 64 kbit/s) T f = 0.75 ms (48 bits and 64 kbit/s) Standard deviation ( ) Ba sic 2.0 1.5 TL = 600 ms 1.0 PCR 0.5 T L = 30 ms 0 0 0.1 0.2 0.3 0.4 a, traffic loading of MSU 0.5 0.6 Erlangs T1156400-93/d04 Model A Model B FIGURE 4/Q.706 Standard deviation of queueing delay of each channel of traffic in absence of disturbance SG11\REPORT\R205E.WW9 (102326) 09.02.00 - 13 COM 11-R 205-E 3 Replace Figure 5/Q.706 with: 18 16 14 12 Mean total queueing delay (Qt) ms 10 8 Psu =0.004 6 Pu = 0.001 4 2 Pu = 0 0 0 0.2 0.4 0.6 0.8 MSU channel load, Erlangs FIGURE 5/Q.706 (sheet 1 of 6) Mean total queueing delay of each channel of traffic Basic error correction method, 15 octet MSUs Tm= 1.875 ms (120 bits and 64 kbit/s) Tf = 0.75 ms (48 bits and 64 kbit/s) TL = 30 ms loop delay with zero link loading SG11\REPORT\R205E.WW9 (102326) 09.02.00 - 14 COM 11-R 205-E 20 Mean total queueing delay (Qt) ms 15 10 Psu =0.004 Pu = 0.001 5 Pu = 0 0 0 0.2 0.4 0.6 0.8 MSU channel load, Erlangs FIGURE 5/Q.706 (sheet 2 of 6) Mean total queueing delay of each channel of traffic Basic error correction method, 25 octet MSUs Tm= 3.125 ms (200 bits and 64 kbit/s) Tf = 0.75 ms (48 bits and 64 kbit/s) TL = 30 ms loop delay with zero link loading SG11\REPORT\R205E.WW9 (102326) 09.02.00 - 15 COM 11-R 205-E 30 25 Mean total queueing delay (Qt) ms 20 15 Psu =0.004 10 Pu = 0.001 5 Pu = 0 0 0 0.2 0.4 0.6 0.8 MSU channel load, Erlangs FIGURE 5/Q.706 (sheet 3 of 6) Mean total queueing delay of each channel of traffic Basic error correction method, 50 octet MSUs Tm= 6.25ms (400 bits and 64 kbit/s) Tf = 0.75 ms (48 bits and 64 kbit/s) TL = 30 ms loop delay with zero link loading SG11\REPORT\R205E.WW9 (102326) 09.02.00 - 16 COM 11-R 205-E 50 40 Mean total queueing delay (Qt) ms 30 20 Psu =0.004 Pu = 0.001 10 Pu = 0 0 0 0.2 0.4 0.6 0.8 MSU channel load, Erlangs FIGURE 5/Q.706 (sheet 4 of 6) Mean total queueing delay of each channel of traffic Basic error correction method, 100 octet MSUs Tm= 12.5 ms (800 bits and 64 kbit/s) Tf = 0.75 ms (48 bits and 64 kbit/s) TL = 30 ms loop delay with zero link loading SG11\REPORT\R205E.WW9 (102326) 09.02.00 - 17 COM 11-R 205-E 80 70 60 Mean total queueing delay (Qt) ms 50 40 30 Psu =0.004 20 Pu = 0.001 10 Pu = 0 0 0 0.2 0.4 0.6 0.8 MSU channel load, Erlangs FIGURE 5/Q.706 (sheet 5 of 6) Mean total queueing delay of each channel of traffic Basic error correction method, 150 octet MSUs Tm= 18.75 ms (1 200 bits and 64 kbit/s) Tf = 0.75 ms (48 bits and 64 kbit/s) TL = 30 ms loop delay with zero link loading SG11\REPORT\R205E.WW9 (102326) 09.02.00 - 18 COM 11-R 205-E 160 140 120 Mean total queueing delay (Qt) ms 100 80 60 Psu =0.004 40 Pu = 0.001 20 Pu = 0 0 0 0.2 0.4 0.6 0.8 MSU channel load, Erlangs FIGURE 5/Q.706 (sheet 6 of 6) Mean total queueing delay of each channel of traffic Basic error correction method, 279 octet MSUs Tm= 34.875 ms (2 232 bits and 64 kbit/s) Tf = 0.75 ms (48 bits and 64 kbit/s) TL = 30 ms loop delay with zero link loading SG11\REPORT\R205E.WW9 (102326) 09.02.00 - 19 COM 11-R 205-E 4 Replace Figure 6/Q.706 with: Standard deviation of queueing delay () ms 25 20 15 10 Psu = 0.004 Pu = 0.001 5 Pu = 0 0 0 0.2 0.4 0.6 0.8 a, channel load in Erlangs FIGURE 6/Q.706 (sheet 1 of 6) Standard deviation of queueing delay of each channel of traffic Basic error correction method, 15 octet MSUs Tm= 1.875 ms (120 bits and 64 kbit/s) Tf = 0.75 ms (48 bits and 64 kbit/s) TL = 30 ms loop delay with zero link loading SG11\REPORT\R205E.WW9 (102326) 09.02.00 - 20 COM 11-R 205-E 25 Standard deviation of queueing delay () ms 20 15 Psu = 0.004 10 Pu = 0.001 5 Pu = 0 0 0 0.2 0.4 0.6 0.8 a, channel load in Erlangs FIGURE 6/Q.706 (sheet 2 of 6) Standard deviation of queueing delay of each channel of traffic Basic error correction method, 25 octet MSUs Tm = 3.125 ms (200 bits and 64 kbit/s) Tf = 0.75 ms (48 bits and 64 kbit/s) TL = 30 ms loop delay with zero link loading SG11\REPORT\R205E.WW9 (102326) 09.02.00 - 21 COM 11-R 205-E 35 Standard deviation of queueing delay () ms 30 25 20 15 Psu = 0.004 Pu = 0.001 10 Pu = 0 5 0 0 0.2 0.4 0.6 0.8 a, channel load in Erlangs FIGURE 6/Q.706 (sheet 3 of 6) Standard deviation of queueing delay of each channel of traffic Basic error correction method, 50 octet MSUs Tm = 6.25 ms (400 bits and 64 kbit/s) Tf = 0.75 ms (48 bits and 64 kbit/s) TL = 30 ms loop delay with zero link loading SG11\REPORT\R205E.WW9 (102326) 09.02.00 - 22 COM 11-R 205-E 60 Standard deviation of queueing delay () ms 50 40 30 Psu = 0.004 20 Pu = 0.001 10 Pu = 0 0 0 0.2 0.4 0.6 0.8 a, channel load in Erlangs FIGURE 6/Q.706 (sheet 4 of 6) Standard deviation of queueing delay of each channel of traffic Basic error correction method, 100 octet MSUs Tm = 12.5 ms (800 bits and 64 kbit/s) Tf = 0.75 ms (48 bits and 64 kbit/s) TL = 30 ms loop delay with zero link loading SG11\REPORT\R205E.WW9 (102326) 09.02.00 - 23 COM 11-R 205-E 100 Standard deviation of queueing delay () ms 80 60 Psu = 0.004 40 Pu = 0.001 20 Pu = 0 0 0 0.2 0.4 0.6 0.8 a, channel load in Erlangs FIGURE 6/Q.706 (sheet 5 of 6) Standard deviation of queueing delay of each channel of traffic Basic error correction method, 150 octet MSUs Tm = 18.75 ms (1 200 bits and 64 kbit/s) Tf = 0.75 ms (48 bits and 64 kbit/s) TL = 30 ms loop delay with zero link loading SG11\REPORT\R205E.WW9 (102326) 09.02.00 - 24 COM 11-R 205-E Standard deviation of queueing delay () ms 200 150 100 Psu = 0.004 Pu = 0.001 50 Pu = 0 0 0 0.2 0.4 0.6 0.8 a, channel load in Erlangs FIGURE 6/Q.706 (sheet 6 of 6) Standard deviation of queueing delay of each channel of traffic Basic error correction method, 279 octet MSUs Tm = 34.875 ms (2 232 bits and 64 kbit/s) Tf = 0.75 ms (48 bits and 64 kbit/s) TL = 30 ms loop delay with zero link loading SG11\REPORT\R205E.WW9 (102326) 09.02.00 - 25 COM 11-R 205-E 5 Replace Figure 8/Q.706 with: 14 Tm = 1.875 ms (120 bits and 64 kbits/s) Tf = 0.75 ms (48 bits and 64 Kbits/s) 12 Pu=0.004 0.004 0.001 Standard deviation (sigma) ms 10 0.001 8 TL = 600 ms 6 TL = 30 ms 4 2 Pu = 0 0 0 0.1 0.2 0.3 0.4 0.5 a, traffic loading of MSU Erlangs FIGURE 8/Q.706 Standard deviation of queueing delay of each channel of traffic Preventive cyclic retransmission error correction method, 15 octet MSUs Tm = 1.875 ms (120 bits and 64 kbit/s) Tf = 0.75 ms (48 bits and 64 kbit/s) SG11\REPORT\R205E.WW9 (102326) 09.02.00 - 26 COM 11-R 205-E 6 Replace Table 6/Q.706 with: TABLE 6/Q.706 Outgoing link delay Tod with basic error correction method Outgoing link delay (ms) a [Erl] 0.2 TL [ms] 30 Disturbance No Yes 600 No Yes 0.4 30 No Yes 600 No Yes MSU – Length (Bytes) Value 15 23 50 140 279 Mean 2.7 4.0 8.2 21.5 39.6 95% 9.3 14.0 30.1 66.0 61.6 Mean 2.8 4.1 8.4 22.0 40.5 95% 10.8 15.5 31.7 69.9 69.0 Mean 2.7 4.0 8.2 21.5 39.6 95% 9.3 14.0 30.1 66.0 61.6 Mean 35.2 36.9 42.3 59.7 83.6 95% 277.7 283.6 303.5 357.5 391.1 Mean 3.5 5.2 10.8 27.6 46.9 95% 12.3 18.6 40.0 88.7 87.1 Mean 3.8 5.5 11.4 31.0 50.1 95% 15.0 21.4 43.6 98.2 104.8 Mean 3.5 5.2 10.8 27.6 46.9 95% 12.3 18.6 40.0 88.7 87.1 Mean 109.7 113.0 124.0 158.2 199.7 95% 580.0 591.3 629.3 738.7 834.6 a Traffic loading by MSUs (excluding retransmission). TL Signalling link loop delay at zero link load, with mean bit error probability < 10-5 in the presence of disturbance. In this table, the mean values and 95% points for all MSU lengths for 30 ms and 600 ms TL, for both a = 0.2 and 0.4, with disturbance present, have been replaced where appropriate. SG11\REPORT\R205E.WW9 (102326) 09.02.00 - 27 COM 11-R 205-E 7 Replace Table 7/Q.706 with: TABLE 7/Q.706 Outgoing link delay Tod with preventive cyclic retransmission method Outgoing link delay (ms) a [Erl] 0.2 TL [ms] 30 Disturbance No Yes 600 No Yes 0.4 30 No Yes 600 No Yes MSU – Length (Bytes) Value 15 23 50 140 279 Mean 4.2 6.1 11.4 23.9 39.1 95% 12.6 18.9 38.3 73.4 60.0 Mean 4.2 6.2 11.9 25.0 42.1 95% 12.9 19.6 40.8 84.4 87.9 Mean 4.2 6.5 14.1 35.7 56.0 95% 12.7 19.4 42.2 93.2 86.2 Mean 5.2 7.6 (a) (a) (a) 95% 29.7 38.8 (a) (a) (a) Mean 5.0 7.6 15.6 32.4 45.5 95% 15.0 22.9 48.7 98.9 84.0 Mean 5.7 9.0 20.7 56.1 92.9 95% 25.8 42.1 107.2 318.2 444.5 Mean 5.0 7.7 16.7 41.8 63.9 95% 15.0 23.0 50.0 111.5 108.9 Mean 63.7 74.7 (a) (a) (a) 95% 386.6 443.1 (a) (a) (a) NOTE (a) - The signal unit error rate for the effective message traffic link load would cause the link to fail. SG11\REPORT\R205E.WW9 (102326) 09.02.00 - 28 COM 11-R 205-E 8 Replace Figure 13/Q.706 with: 100 90 80 Mean outgoing link delay (ms) 70 60 BEC: 50 TL = 30 ms a = 0.4 Erl Pb = 0/1 e – 5 PCR: TL = 30 ms a = 0.4 Erl Pb = 0 40 PCR: 30 TL = 30 ms a = 0.4 Erl PCR: TL = 30 ms a = 0.2 Erl Pb = 0 Pb = 1.0 e – 5 PCR: TL = 30 ms a = 0.2 Erl P b = 1.0 e – 5 20 10 BEC: TL = 30 ms a = 0.2 Erl Pb = 0/1 e – 5 0 15 23 50 140 MSU mean length (byte) BEC TL a Pb 279 T1133000-91/d13 Basic error correction method / PCR = preventive cyclic retransmission method Signalling loop propagation time including processing time in signalling terminal Traffic load by MSUs excluding retransmission Bit error probability FIGURE 13/Q.706 Mean outgoing link delay – TL = 30 ms SG11\REPORT\R205E.WW9 (102326) 09.02.00 - 29 COM 11-R 205-E 9 Replace Figure 15/Q.706 with: ERROR! OBJECTS CANNOT BE CREATED FROM EDITING FIELD CODES. SG11\REPORT\R205E.WW9 (102326) 09.02.00 - 30 COM 11-R 205-E 10 Replace Figure 16/Q.706 with: 1500 1400 1300 1200 1100 95% level values of outgoing link delay (ms) 1000 900 BEC: TL = 600 ms a = 0.4 Erl Pb = 1.0 e-5 PCR: TL = 600 ms a = 0.4 Erl Pb = 1.0 e-5 800 700 BEC: TL = 600 ms a = 0.2 Erl Pb = 1.0 e-5 PCR: TL = 600 ms a = 0.2 Erl Pb = 0 600 500 BEC: TL = 600 ms a = 0.4 Erl Pb = 0 400 PCR: TL = 600 ms a = 0.2 Erl Pb = 1.0 e-5 300 BEC: TL = 600 ms a = 0.2 Erl Pb = 0 200 PCR: TL = 600 ms a = 0.4 Erl Pb = 0 100 0 0 50 100 150 200 250 300 MSU mean length (octets) FIGURE 16/Q.706 95% level value of outgoing link delay – TL = 600 ms SG11\REPORT\R205E.WW9 (102326) 09.02.00 - 31 COM 11-R 205-E 11 In Section 5.2.5, insert a new Table 10bis TABLE 10bis/Q.706 Maximum PCR link loading and 95% level value for Tod for long MSUs with TL = 600 ms Bit error rate Maximum link load (Erlangs) 95% level value for Tod (ms) 279 octet MSUs 140 octet MSUs 279 octet MSUs 140 octet MSUs 1.0 10 -6 0.45 0.45 1 100 500 1.5 10-6 0.45 0.45 1 400 660 2.0 10 -6 0.32 0.45 300 750 3.0 10-6 0.23 0.45 210 1 150 3.5 10 0.21 0.45 200 1 300 ~3.8 10-6 < 0.2 < 0.15 ~200 ~100 ~5.0 10 < 0.17 < 0.1 ~180 ~90 -6 -6 Note that this table is calculated assuming constant length MSUs. For MSUs of variable length, the maximum link loading would be reduced. 12 Replacement Annex B of Q.706 (This annex forms an integral part of this Recommendation). B.1 Derivation of basic error correction (BEC) message signal unit (MSU) queueing delays B.1.1 Preliminary definitions and formulas This section is included to summarize the concepts used in deriving the BEC queueing delay. In the following, the construct Prob{event} is used to denote the probability of the event happening. Thus Prob{ X x} denotes the probability that the random variable X is less than the value x, etc. Laplace-Stieltjes transform and expectation of a continuous random variable (references [2] and [4]) The Laplace-Stieltjes transform of a probability distribution is defined as follows: if a continuous non-negative random variable X has the probability distribution function F(x), such that F ( x) Prob{X x} , then the Laplace-Stieltjes transform of F(x), ( s), is defined as: ( s) e sx . dF ( x) , where the lower limit of integration includes the origin of x. 0 The expectation of the random variable X is defined as E[ X ] x.dF ( x) , which is the mean value 0 of X. Note that: (s) E[e sX ] SG11\REPORT\R205E.WW9 (102326) 09.02.00 - 32 COM 11-R 205-E Moments of a continuous distribution (reference [4]) The nth moment of a probability distribution function F ( x) Prob{X x} is defined as: F ( n) d n( s) dsn x n . dF ( x ) ( 1)n 0 E[ X n ] s0 The mean value of X is thus F(1), and its variance is x F . dF ( x) F (1) 2 (2) F (1) 2 . 0 Moments of a discrete distribution (reference [4]) If Y is a non-negative integer random variable, with probability function p( y) Prob{Y y}, the kth moment of Y is defined as: Y (k ) p( y). y k E[Y k ] y 0 Generating function (references [2] and [4]) If Y is a non-negative integer random variable, with probability function p( y) Prob{Y y}, the generating function of Y, G(z), is defined as: G( z) p( y).z y E[ z Y ], where the expectation of zY is E[ z Y ] . y 0 If T denotes a random time variable, with probability distribution function F(t) such that F (t ) Prob{T t}, and if there is a Poisson process of rate , with Y denoting the number of events of this process to occur during time T, the generating function of Y, G(z), is: G( z ) E[ z ] Y E[ z Y | T t ].dF (t ) , where the E expression in the integral is the expectation of zY 0 conditional upon the time interval T having the value t. For a Poisson process of rate , the number of events in a time t is y with probability Prob{Y y| T t} E[ z Y | T t ] e t .(t ) y , hence: y! E[ z Y | T t , Y y]. Prob{Y y| T t} y 0 zy. y 0 e t .(t ) y , y! e .(1 z ).t Hence G( z) e .(1 z ).t .dF (t ) (.[1 z]) , where (s) is the Laplace-Stieltjes transform of F(t). 0 Derivation of ordinary M/G/1 system queue size generating function (references [2], [4] and [5]) For an M/G/1 system (i.e. a system with Poisson arrivals, general service time and one server) in which each arriving item's service time is an independent identically distributed random variable, the generating function for the queue size is calculated by considering the number in the queue left behind by a departing item (i.e. by an item completing service). The set of departure instants form an embedded Markov chain - that is, the state of the system can be determined at one departure SG11\REPORT\R205E.WW9 (102326) 09.02.00 - 33 COM 11-R 205-E instant from the state at the previous departure instant and the number of arrivals in the meantime. For any system that changes state by one step (up or down), the limiting distribution for the number of items in the system left behind by a departure is the same as the limiting distribution of the number found in the system by an arriving item. Since in our case the arrival process is Poissonian, and a Poisson arrival observes the system as a random observer, the statistics of the queue length measured at the set of departure instants are the same as those for arrival instants, and hence provide the solution for all points in time. If we denote the number of items left behind by the nth departing item as qn, the number of arriving items during the nth service by vn, and putting (q – 1)+ = max(q – 1,0), we have: q n1 (q n 1) v n1 , where n+1 and (qn – 1)+ are independent random variables. As n , q n 1 q n q, v n 1 v n v, and the generating function Q(z) for the number in the queue is given by: Q( z ) E[ z q ] E[ z ( q 1) v ] E[ z ( q 1) ] E[ z v ] P( z ) V ( z ) , with P(z) and V(z) the generating functions for (q 1) and v respectively. v represents the number of arrivals during a service time, hence from the formula for G(z) given previously, V ( z) ( .[1 z]), where ( s) is the Laplace-Stieltjes transform of the service time. Now, P( z ) E[ z ( q 1) ] E[ z (q1) x 0 1 | q x].Prob{q x} Prob{q 0} E[ z q | q x].Prob{q x} z x 1 1 1 Prob{q 0} E[ z q | q x].Prob{q x} Prob{q 0} z x 0 z Hence: 1 1 P( z ) (1 ) Prob{q 0} Q( z ) , so z z Q( z ) (1 z ).(1 a ). ( .[1 z ]) ( .[1 z ]) z Here Prob{q 0} 1 a , where a is the server utilization, and a . d ( s) ds s0 . Tm . Tm is the mean service time. Probability distribution function and transform of the residual time to the next event (references [5], [2] and [4]) Consider the following diagram, where an observation is made at an arbitrary time t. The time x to the next event (the residual time) has a probability distribution function Rt(x). The probability density function of Rt(x) is rt(x), which gives the probability that the next event occurs in the interval ( x, x x] as r t ( x). x . t t+x time axis y event Rt = PDF of time to next event observation event Suppose the kth event occurs at a time X1 X 2 ... X k , with { X1 , X 2 ,..., X k } independent positive random variables. Let X2, X3, ... be identically distributed with probability density functions f(x) SG11\REPORT\R205E.WW9 (102326) 09.02.00 - 34 COM 11-R 205-E and probability distribution functions F(x) such that f ( x) dF ( x) , and let X1 have the probability dx density function f1 (x) which might be the same as f(x). Let the probability density function of events be h(t), then: h(t ) lim Prob{event in (t , t t ]} t 0 t From the diagram, the next event is assumed to occur in the interval ( x, x x] after t, this might be the first event with probability f1 (t x). x , or there could have been an event at a time interval y before t, and the next event occurs in the interval ( x y, x y x] . Hence: t r t ( x). x f 1 (t x). x h(t y). f ( y x). dy. x 0 As t , h(t ) 1 , r t ( x) r ( x) , R t ( x ) R( x ) , f 1 (t x ) 0 , where is the mean event interarrival time. Hence, in equilibrium, r ( x) 1 f ( y x). dy 1 F ( x) 0 . The Laplace-Stieltjes transform of R(x) is: r ( s) 0 e s.x dR( x) e s.x r ( x). dx 0 e s. x 0 1 F ( x) . dx 1 ( s) . s where (s) is the Laplace-Stieltjes transform of F(x). B.1.2 BEC queueing delay in the absence of disturbance The basis of this calculation is the M/G/1 system with server multiple vacations, described in reference [2]. In the BEC method (reference [1]), fill - in signal units (FISUs) are transmitted continually if there are no MSUs to transmit. The first MSU to arrive waits until the FISU has finished emission, and then starts being emitted itself (i.e. it starts "service"). Denote as "type 1" MSUs those MSUs arriving whilst a FISU is being emitted, and "type 2" MSUs as the rest (type 2 MSUs arrive whilst an MSU is being emitted). In order to calculate the probability generating function for the number of queued MSUs, it is helpful to modify (temporarily) the order of MSU service: any type 2 MSU arriving whilst a type 1 MSU is being emitted is emitted before any waiting type 1 MSUs. This trick enables the calculation of the queue length, which is independent of the order of service (i.e. the order of MSU transmission). Thus, the start of emission of a type 1 MSU starts a "busy period", which lasts for the time of emission of this MSU and emission of all its associated type 2 MSU arrivals. The transmitter will transmit FISUs when the busy period associated with the last type 1 MSU to arrive during the previous FISU emission ends. Just one type 1 MSU is emitted in a busy period. Consider the queue size when an MSU has just been emitted. Since the MSU arrival process is Poissonian, and the emission times of successive MSUs are independent, the future evolution of the system depends only upon the MSU queue size immediately following an MSU departure and future MSU arrivals. In addition, the number of type 1 MSUs present is independent of the number of type 2 MSUs present, and if we write the total number of MSUs left behind by the departure of an arbitrary MSU as M, with M1 as the number of type 1 MSUs, and M2 the number of type 2 MSUs, we have: SG11\REPORT\R205E.WW9 (102326) 09.02.00 - 35 COM 11-R 205-E M = M1 + M2, and generating function N MSU ( z) E[ z M ] E[ z M1 M 2 ] E[ z M1 ] E[ z M 2 ] N1 ( z) N 2 ( z) . N2 (z) is analogous to Q(z) in the previous section. N1 (z) is given by the number of MSUs arriving during the residual emission time of a FISU, since the distribution of the number of type 1 MSUs arriving from the Poisson arrival process is the same as the distribution of the number of type 1 MSUs left behind by the departure of an arbitrary MSU. N 1 ( z) 1 F ( .[1 z]) .[1 z]. T f , where F ( s) is the Laplace-Stieltjes transform of the FISU emission time probability distribution, with Tf the FISU emission time. F ( s) e st . (t Tf ). dt e sTf , so: 0 N 1 ( z) .[1 z ].T f 1 e .[1 z ]. T f Hence: .[1 z ].T N MSU ( z) f (1 z).(1 a ). ( .[1 z]) 1 e ( .[1 z]) z .[1 z]. T f The size of the queue, and hence the generating function for it, does not depend on the order of service, so it is the same for first come, first transmitted MSUs. Hence, denoting the transform of the total time spent in the system by an MSU by (s) (this is the transform of the queueing plus emission time - i.e. the transform of the sojourn time of the MSU), N MSU ( z) ( .[1 z]) Hence: ( s) (1 a ). Tm ( s) s .T .(1 e f ). Tf s. Tm a .1 ( s) , since a . Tm Now, the sojourn time S equals the queueing time Q of the MSU plus its emission time TMSU on to the link, and its queueing time is independent of its emission time. Hence: ( s) E[e s.S ] E[e s (Q TMSU ) ] E[e sQ ] E[e sTMSU ] M (s) (s), where M (s) is the Laplace-Stieltjes transform of the MSU queueing delay time. Hence: M ( s )(1 a). Writing M (s) g ( s) , h( s ) Tm 1 s.T .(1 e f ). Tf s.Tm a.1 ( s ) M ( s).h( s) g ( s) , the various moments of M (s) can be determined by differentiating the latter equation sufficiently and taking limits as s 0 . The first moment is the mean queueing delay in the absence of disturbance, Qa d M ( s ) ds T ( 2) 1 a . T f M 1 a Tm s 0 2 , where TM(2) k1 Tm2 is the second moment of the MSU emission time. The variance of the queueing delay in the absence of disturbance can be derived from the mean and the second moment of M (s) to be: 2a 2 T ( 2) T ( 3) 1 a 1 a M M , 12 4 1 a Tm 3 1 a Tm T f2 where TM( 3) k 2 Tm3 is the third moment of the MSU emission time. SG11\REPORT\R205E.WW9 (102326) 09.02.00 - 36 COM 11-R 205-E B.1.3 B.1.3.1 BEC queueing delay in the presence of disturbance Virtual emission time of an MSU, and its moments (reference [3]) In the BEC method, MSUs that have been transmitted are held in the retransmission buffer until acknowledged. An error in a received MSU causes the receiver to request retransmission, and this MSU plus any subsequently transmitted MSUs are then re-sent in order. Hence the delay caused by the error in the MSU can be viewed as an increase in its emission (service) time. If the MSU error rate is Pu, the signalling loop delay is TL and the number of retransmissions of the MSU is N, then the virtual service (emission) time of the MSU is: Tvir TMSU N .TMSU TL (1) ... TL ( N ) , with N different and independent values of TL. For randomly distributed MSU errors, the probability that the number N of retransmissions is n is: Prob{N n} Pun (1 Pu ) . The kth moment of N is thus N (1) N ( 2) N (3) N (k ) nk Pun (1 Pu ) , and hence: n 0 Pu , 1 Pu Pu (1 Pu ) (1 Pu ) 2 , Pu (1 4 Pu Pu2 ) (1 Pu ) 3 The Laplace-Stieltjes transform of the distribution of TL can be derived as follows: The acknowledgement to an MSU is carried in the backward sequence number of either a FISU or an MSU, the choice depends upon the relative frequency of MSUs and FISUs on the link in the return direction. A decision to send a negative acknowledgement (NACK) is made after receiving the next SU with correct cyclic redundancy check (CRC) bits after receiving an errored SU. Before the NACK is sent (in the next SU to be emitted in the return direction), the SU's backward sequence number must be inserted, and its CRC bits computed and inserted. The CRC bits can be computed and inserted during SU emission. After the current SU has been emitted, the SU carrying the NACK is sent. At the NACK receiving end, the CRC bits must be recomputed and the SU absorbed completely before it is accepted as a NACK. Hence, the overall signalling loop delay for an errored SU consists of the propagation time of the MSU, plus the absorption time of the next (assumed error-free) SU, plus its processing time at level 2 at the receiving end, plus the time to finish emission of the SU before the NACK, plus the NACK SU emission time, plus the propagation delay on the link, plus the MTP level 2 processing time of the NACK, plus the remaining emission time of the SU currently being transmitted in the forward direction. The erroneous MSU can start retransmission after the now currently emitting SU. In the above, errors in the next SU after the errored SU in the forward direction, and errors in the return direction, extend the loop delay time, but this is a second order effect and is ignored here. From the above, T L* ( s ) E e sTL E e f E e sSU r f m SG11\REPORT\R205E.WW9 (102326) s K 2(1 )T f TMSU (1) TMSU ( 2) 2.SU r . 1 e sT f . 1 ( s ) m f m f T f m Tm s. f m , E e sK . E 2 e sSU .E 2 e sSU r , E e sSU (1 ).e sT f .( s ) 09.02.00 - 37 COM 11-R 205-E hence: sK (1 )e TL* ( s ) e f 1 a a , m Tf Tm sT f 1 e sT f 1 ( s ) f m .( s ) s 2 2 , SUr is the residual time to transmit an SU before the NACK, which can be a FISU or MSU. TMSU is the time to emit and absorb the NACK if it is in an MSU. Now, the residual time for SU emission is independent of the time to absorb the NACK, so each component of the expectation is independent, and hence the total expectation is the product of the expectations of the components. The first 3 moments of TL are: TL(1) K 2(1 ) 1 a .T f 2 ak1 .Tm 2 2 k1Tm2 (1 )T f2 2.K ak1Tm (1 a)T f 4.Tm (1 )T f . ak1Tm (1 a )T f 4 K Tm (1 )T f 23 T f2 (1 a ) ak 2 Tm2 12 .(1 a ).T f ak1Tm 2 T L( 2) K 2 2 Tm (1 )T f T L(3) K 3 6. Tm (1 )T f . k1Tm2 (1 )T f2 2 k 2 Tm3 (1 )T f3 2.K ak 2 Tm2 (1 a)T f2 4. ak 2 Tm2 (1 a)T f2 . Tm (1 )T f 6 K 2 Tm (1 )T f 3.K 2 ak1Tm (1 a)T f 6.Tm (1 )T f .ak1Tm (1 a)T f 3 3 12 K Tm (1 )T f . ak1Tm (1 a)T f 12 .(1 a)T f a.k 3 .Tm 32 .K .(1 a)T f ak1Tm 2 3.(1 )T f Tm . ak1Tm (1 a)T f 2 ak1Tm (1 a)T f .ak 2 Tm2 (1 a)T f2 6 K Tm (1 )T f 2 a where a (1 a ). k1Tm2 (1 )T f2 2 k1Tm2 (1 )T f2 is the relative frequency of MSUs. Tm Tf The kth moment of the distribution of Tvir is given by: (k ) k Tvir E[Tvir ] (1) k * d k Tvir ( s) ds k * , where Tvir ( s) is the Laplace-Stieltjes transform of the distribution of s 0 Tvir. The transform is obtained as follows: * Tvir ( s | N n) E e sTvir (t| N n) E e s ( n 1)TMSU .E e sTL (t1 ) ... TL (t n ) [n 1]s . TL* ( s) * Tvir ( s) n Tvir* (s | N n).Prob{N n} Prob{N n}.[n 1]s .TL* (s) n 0 n 0 n The first 3 moments of the distribution of Tvir are thus: (1) Tvir Tm 1 N (1) N (1) TL(1) 1 3T 2 ( 2) Tvir TM(2) N (2) 2N (1) 1 2TmTL(1) N (2) N (1) TL(2) N (1) TL(1) . N (2) N (1) (3) Tvir TM(3) N (3) 3N ( 2) 3N (1) 2 ( 2) (1) M TL N (3) 2 N ( 2) N (1) 3TL( 2) Tm N ( 2) N (1) N 3Tm TL(1) . N (3) N (1) TL(3) N (1) 3TL( 2) TL(1) N ( 2) N (1) TL(1) SG11\REPORT\R205E.WW9 (102326) 3 (3) 3N ( 2) 2 N (1) 09.02.00 - 38 COM 11-R 205-E In addition, aeff .Tvir(1) .Tm (1) Tvir T (1) a vir Tm Tm aeff and these moments of Tvir replace a and the corresponding moments of TM in the formulas for Qa and a2 to give: Qt 2t B.1.3.2 Tf2 12 ( 2) a eff Tvir 1 Tf (1) 2 1 a eff Tvir 2 ( 3) 1 aeff T ( 2) 1 aeff Tvir vir (1) (1) 4 1 aeff Tvir 3 1 aeff Tvir Relationship between SU error rate and MSU error rate (references [3] and [6]) From the previous section, a eff eff PSU eff f f .Pu eff f Pu T L(1) 1 Tm a 1 Pu . The SU error rate is given by: a eff Pu 6 Pu / m .PFISU (1 a eff ). , where Tm T a eff (1 a eff ) a eff (1 a eff ) m Tf Tf eff Tm f T f 1, eff Tm aeff and PFISU is the FISU error rate. Here, the mean MSU length is m octets. If the bit error rate is Pb, then, since a FISU is 6 octets long, Pb m 1 a eff 1 1 a eff a eff PSU and Pu 8m Pb a eff 8 6 m 6 P SU and at changeover PSU = ~0.004. These equations furnish a quadratic equation for Pu which can be solved for the particular MSU length and link loading. B.2 PCR Loop delay As in the BEC case, the loop delay is influenced by the link loading and MSU size. The FISU loading on the link is calculated as a function of the loop delay, but the loop delay is a function of the FISU loading. The equation for the FISU loading can be solved by iteration. The first three moments of the loop delay replace its corresponding powers in the formulae for H and F, this is shown below. The Laplace-Stieltjes transforms for H*(s) and F*(s) in the original PCR paper contained TL as a constant. H*(s) is the Laplace-Stieltjes transform of the distribution of the time for retransmitting all the MSUs in the retransmission buffer (RTB), F*(s) is the LaplaceStieltjes transform of the distribution of the time for transmitting MSUs in the RTB ahead of an MSU previously transmitted with errors. If we write H*(s | TL) and F*(s | TL) for the original paper's H*(s) and F*(s) respectively, we can cater for TL being a variable by taking the expectations of H*(s | TL) and F*(s | TL) with respect to TL. Thus we have: TL (t ) K (1 ).T f .TMSU (t ) RSU (t ) , where is the relative frequency of MSUs, TMSU (t) is the SG11\REPORT\R205E.WW9 (102326) 09.02.00 - 39 COM 11-R 205-E emission/absorption time of an MSU and RSU(t) is the remaining emission time of the SU before the SU containing an acknowledgement is sent. Now, denoting the Laplace-Stieltjes transform of TL(t) by T*L(s), we have: TL* ( s ) E e s.TL (t ) e sK . .( s ) (1 ).e sT f . .1 (s) s .1 e since the remaining emission s.T f f time of the SU being emitted is independent of the absorption/emission time of the next SU to be sent. is given by: .Tm f .T f 1 , , f with f the FISU generation rate. The various moments of TL can be obtained from this transform using TL( k ) (1) k 1 H * (s | TL ) e mTL 1( s ) , 1 F * (s | TL ) H * ( s | .TL ) d e mTL 1( s ) d 0 0 d k TL* ( s ) ds k . Now, s 0 1 e mTL 1( s ) m .TL 1 (s) * Hence the new H (s) is given by: 1 ( s ).K (1 ).T f .TMSU (t ) R SU (t ) 1 ( s ). .TMSU (t ) (1 ).T f H * (s) E e m e m 1 ( s ).K .E e m .E e m 1 ( s ).RSU H * ( s ) e Kz . .( z ) (1 ).e zT f . . 1 ( z) z .1 e , where z z .T f f 1 (s) , m. a m .Tm The first three moments of H*(s) can be derived from this, and they are equal to the values in the original paper with TL replaced by TL(1), TL2 replaced by TL(2), etc. To obtain the moments of the new 1 F*(s), replace TL in the original same order moment for H by TL . . d (i.e. 1 2 TL), TL2 by 0 1 TL2 . 2 .d (i.e. 1 .T 2 3 L ), etc., and then replace TL by TL(1), TL2 by TL(2) etc. 0 B.3 Calculation of the kth moments of the MSU emission time for Tod The kth moment of the MSU emission time is required for the factors, k1, k2 and k3 in 4.2: TM( k ) t .t k M (t )dt tM (t) = distribution density function 0 t M (t ) dTM (t ) , TM (t) is the dt probability distribution function. T constant (= Tm): tM (t) = (t – Tm) T negative exponential distributed: t M (t ) t SG11\REPORT\R205E.WW9 (102326) 1 e Tm Tm 09.02.00 - 40 COM 11-R 205-E With the Laplace-Stieltjes transformation ( s) e st . t M (t )dt : 0 t k t M (t )dt ( 1) k . 0 d k ( s) ds k s 0 T constant: (s) = exp(–s Tm) so: 2nd moment = TM(2) Tm2 3rd moment = TM(3) Tm3 4th moment = TM(4) Tm4 1 1 s Tm T negative exponential distributed: ( s) so: 2nd moment = 2. Tm2 3rd moment = 6. Tm3 4th moment = 24. Tm4 This leads to the following values for the factors k1, k2 and k3 (see 4.2): general: T constant: k1 = TM( 2 ) Tm2 k2 = TM( 3) Tm3 k3 = TM( 4 ) Tm4 k1 = 1 k2 = 1 k3 = 1 T negative exponential distributed: k1 = 2 k2 = 6 k3 = 24 B.4 Approximate calculation of the 95% - values of Tod Tod.95% = Qt.95% + Tm.95% SG11\REPORT\R205E.WW9 (102326) 09.02.00 - 41 COM 11-R 205-E Qt.95% and Tm.95% are calculated with the help of the approximation in 4.2.4: Tmean T P(T95% ) exp 95% P(T T95%) = 1 – P(T95%) with P(T T95%) = 0.95 Tmean + 2 with Tmean mean value of T = standard deviation of T where T95% = This gives: Qt.95% Qt + 2 Tm.95% 3Tm for mean MSU length 15, 23 or 50 bytes (MSU length distribution as negative exponential) Tm.95% 2/3 3Tm + 1/3 Tm (2/3 negative exponential distribution + 1/3 constant) and Tm.95% = 7/3 Tm for mean MSU length 140 bytes = Tm for mean MSU length 279 bytes (MSU length distribution as constant). _______________ SG11\REPORT\R205E.WW9 (102326) 09.02.00