WO2007072487A1

WO2007072487A1 - System and method for synchronization of a communication network

Info

Publication number: WO2007072487A1
Application number: PCT/IL2006/001467
Authority: WO
Inventors: Dov Wulich; Motti Goren
Original assignee: Ipclock Ltd.
Priority date: 2005-12-22
Filing date: 2006-12-21
Publication date: 2007-06-28
Also published as: IL172756A0

Abstract

There is provided a synchronization system for a communication network, including an optimal frequency estimator based on a vector of time stamps constituted by an observation vector that estimates the period of incoming packets. A method for synchronization in a communication network is also provided.

Description

SYSTEM AND METHOD FOR SYNCHRONIZATION OF A

COMMUNICATION NETWORK Field of the Invention

The present invention relates to communication networks and, more particularly, to synchronization transfer over packet networks of communication networks. Specifically, the invention is concerned with genuine algorithm for clock recovery over packet networks. Background of the Invention

Deployment of next generation networks results in a transition from circuits to packets at the metro and access network segments. In particular, Ethernet at the edge is replacing traditional El/Tl TDM circuits, providing higher capacity at lower costs. The negative implication is that the traditional TDM circuits, widely used for clock distribution, will no longer be available.

There is therefore a need for an alternative clock distribution scheme that provides reliable and standards compliant clock distribution over packet networks that support sync sensitive applications such as cellular base-stations, WiMAX base station and circuit emulation.

The lack of reliable and standards' compliance sync over packet has negative technical and economical implications. For example, ensuring reliable handoff and minimizing dropped connections, 3G wireless technologies like universal mobile telecommunications system (UMTS) demand strict synchronization from their backhaul circuits. Today's 3G cellular backhaul is based on asynchronic transfer mode (ATM) over plesiochronous digital hierarchy (PDH). An alternative internet protocol (IP) or Ethernet solution must carry clock with similar performance characteristics in order to meet the operator's quality of service (QoS) requirements and to comply with the relevant telecom standards such as ITU-T G.823, ITU-T G.824, TS 145.010, TS 125.104 and others. The forthcoming 4G wireless technology will be an all IP network and become even more demanding driver for sync over packet. As seen in the prior art Figs. 1 and 2, the sync over packet engine (SPE) has to "duplicate" the phase and frequency of the master clock generator of the transmitter (TX) while keeping the time error between SPE and transmitter (TX) within requested accuracy in terms of phase and frequency. The frequency of the transmitter (master) clock equals fa .

A periodic signal, τ(t) is transmitted to the receiver (synchronizer) via asynchronous IP network, as a synchronization packets with a rate/₅. The packet rate

Λ =f, (D where K » 1 is some natural number.

T(t) denotes the packet position on the time axis and is called time signal. Its units are sec. Synchronization packets are delayed by the network in a random way yielding packet delay variation (PDV). The distribution of the PDV is usually unknown and depends on the specific network and traffic load. However it may be well approximated by a distribution decaying as e^~a .

The time function T_R (t), at the output of the packet switch network (PSN) is equal to:

T_R{t) = τ(ή+x{ή, (2) where x(t) represents the PDV.

Fig. 2 graphically illustrates the signal τ(t) as well as the time function T_R(t) .

The instantaneous frequency of the SPE' s clock, local oscillator (LO), equals/^ . The nominal frequency of the LO equals fa . For various reasons, however, f_R ≠ fa . The aim of SPE is to produce a time function T₀ (t) with a frequency f₀ that is much closer to fa than f_R , i.e.,

\fo -fτ\ «\f_R

(³) U.S. Patent 6,175,604 proposes a solution based on minimum mean square error (MMSE) that minimizes the mean square error. This algorithm has bias (is only asymptotically unbiased), and therefore, is inferior, especially for small observation intervals (N_obs). Disclosure of the Invention

It is therefore a broad object of the present invention to provide a system and a method for synchronization of communication packet networks or for synchronization of applications over packet networks, enabling reduction of oscillators' stability requirements, reduction of the needed packet rate and improving the synchronization performance.

It is a further object of the present invention to provide a system and a method for synchronization of communication networks based on an optical frequency estimator such as a maximum likelihood (ML) period estimator.

In accordance with the present invention there is therefore provided a synchronization system for a communication network, comprising an optimal frequency estimator based on a vector of time stamps constituted by an observation vector that estimates the period of incoming packets.

There is further provided a synchronization system for a communication network, comprising an optimal frequency estimator that reaches the Cramer-Rao bound such as an maximum likelihood estimator, receiving input signals from a packet switch network and clock signals from a local oscillator, and a controlled oscillator fed by said estimators' output signals and by said local oscillator to produce an output signal representing the output time function and used to estimate probability density function (PDF) of a packet delay variation PDV.

There is also provided a method for synchronization in a communication network, comprising forming an observation vector by a time stamp, assuming an initial probability density function (PDF), computing a PDF of the observation vector, determining the estimated values of the output packets period and the time shift, and generating an output time function which is used to estimate the PDF of the packet delay variation (PDV).

Brief Description of the Drawings

The invention will now be described in connection with certain preferred embodiments with reference to the following illustrative figures, so that it may be more fully understood.

With specific reference now to the figures in detail, it is stressed that the particulars shown are by way of example and for purposes of illustrative discussion of the preferred embodiments of the present invention only, and are presented in the cause of providing what is believed to be the most useful and readily understood description of the principles and conceptual aspects of the invention. In this regard, no attempt is made to show structural details of the invention in more detail than is necessary for a fundamental understanding of the invention, the description taken with the drawings making apparent to those skilled in the art how the several forms of the invention may be embodied in practice.

In the drawings:

Fig. 1 is a block diagram of a prior art synchronization system; Fig. 2 graphically shows PDV of a system according to Fig. 1; Fig. 3 is a block diagram of a SPE₅ according to the present invention; Fig. 4 is a flow diagram of the ML period (frequency) estimation, and Fig. 5 is a block diagram of a sync engine for carrying out the process of Fig. 4.

The sync over packet engine (SPE) 2 of a communication network, according to a preferred embodiment of the present invention, is shown in Fig. 3. It consists of an optical frequency estimator 4 that reaches the Cramer-Rao-bound, such as an unbiased ML period estimator, receiving input signals T_R(t) during 0 ≤ t ≤ T_obs and clock signals elk from a local oscillator (LO) 6. The output signal f_τ from the estimator 4 is fed to an oscillator 8, e.g., a digitally controlled oscillator (DCO), a voltage controlled oscillator (VCO) or a numerically controlled oscillator (NCO), or the like, which also receives signals f_R from the local oscillator 6. The output from the SPE 2 is T₀ if). The description of the estimator 4 is as follows:

Let r_k denotes the k-th. time stamps, i.e., the time when the k-th. packet arrives according to the time axis created by the receiver of LO 6. There is obtained: r_k = ak + b + x_k ,\ ≤ k ≤ N_obs (4)

where a = fτ _> b denotes the time shift; x_k is the k-th packet delay;

N_Obs is the total number of synchronization packets taken for estimated values a and b, and

^■N obs ⁼ J s ^' * obs -

Let f_r(r/a,b) denotes the conditional probability density function (PDF) of the observation vector r = [r\₎r^,...,r_Ngbs ] given a and b . Let f_pct_v(x) denote the joint PDF ofthe vector x = [X₁, X₂V! λ"_/^ , ] • There is obtained:

/_r (r/a,b) = f_pdv (Jr₁ -a-b,r₂ -2a - b,...,r_k - ak - b,...,r_N(jbs - N_obsa - b]) . (5) According to the maximum likelihood (ML) principle, the best estimate of a and b for given r = [r₁,r₂,...,rj_\f ] are such a and b for which f_r (r/a,b) seen as a function of a and b reaches maximum value:

(6)

where a and b denote the estimated values, and arg is the value of a and b for which %la,b) reaches its maximal value. The flow diagram of the described above ML algorithm is shown in Fig. 4 while the detailed block diagram of the sync engine is shown in Fig. 5. As mentioned, the PDF of PDV is not known, however to start the process some initial PDF should be assumed. Usually a truncated Gaussian distribution is assumed. Using such assumption temporary T₀(t) is generated. Such T₀(t) may be used to estimate the actual PDF of the PDV. The freshly estimated PDF is used to estimate new T₀ (t), that is used to estimate the actual PDF of PDV, but more accurate, etc.

In fact the PDF of PDV estimation algorithm as described above and shown by the flow diagram of Fig.4 as well as by the block diagram of Fig. 5 results with feedback mode algorithm.

The algorithm described above is repeated using sliding window routine which means that the observation interval is defined as fy ≤ t ≤ t_[ + T₀^ , 0 < fy < T_L or in the discrete time l ≤ k ≤ N_0Jj3 + 1 , I = 0,1,..., L . T_L denotes the entire time when the window slides and L = T_L -f_s . where / is the number of the sliding window, and

L is the total number of sliding windows. Example 1 :

It is assumed that the random variables {x_k)_k °_\ ^{s are} independent and identically distributed (HD). In such a case

^Nobs fpdv(^x) = Tlgpdv(^xk), C⁷) k=\ where g_pcj_v (•) denotes the PDF of each one of x^, k = 1,2,..., N₀^₈.

The conditional f_r (r/a,b) may be substantially simplified and is equal to:

^Nobs f_r (r/a, b)= γl g_pdv {r_k -ak-b). (8) k=\

Example 2:

In addition to the HD assumption it is assumed that

may be well described by the family of PDF's S/wfrW- ∞pj — ₂ ^"p ^x > 0 (9.1)

or "full" Gaussian distribution

^{for a11 x} • (9.2)

The PDF's of expressions 9.1 and 9.2 cover wide range of random variables such as Gaussian, single side normal, absolute value of normal, truncated Gaussian, etc.

As a result the conditional PDF of the observation vector r = [f\,r₂,...,r_N ] given a and b is:

where C is a normalizing constant. From (10) there is obtained:

^~2 (H)

A=I ^2σ

According to ML algorithm, it is required to find, for given observation vector r , a and b such that ln[/(r/α, &)] ->■ max. Define

max(\n[f(r/a, έ)j) is achieved for the same a and 6 as min(Λ(α,έ)), therefore the

estimated a andb are given by

(a,*)= arg|min(Λ(α,6))i . (13)

[ α,b J and are solutions of the system of two linear equations:

[c₂ιά + c₂₂b = d₂ ' where

N_obs(N_obs+l)(2N_obs+l)

C₁₁ = (15) 6

_^Nobs(^Nobs+^l)

^C12 = ^C21 = ₂ (16)

^C22 = Nobs, (¹⁷)

^= Z*-*, (18)

A=I where J₁ is a weighted sum of time stamps over N_obs samples;

and

Nobs d₂ = ∑r_k ■ (19)

A=I

where d₂ is a sum of time stamps over N_0J18 samples.

Solution of Expression 14 with respect to a gives: ά= °₂₂d_l -c₁₂d_{2 (20)}

^C11^C22-^C12^C21

The performance of the ML estimator 4 depends on the PDF of PDV. Its performance is measured by two parameters:

where E{-} denotes the expected value and var{ά}, where var{-} denotes the variance. It is required that E{ά} = a (unbiased estimation) and the accuracy of the estimation var{α} -> min .

The ML estimator of Example 2 is unbiased and the most accurate estimator among all unbiased estimators as its variance achieves the Cramer-Rao Bound (CRB), the lowest possible value of the variance. Therefore the ML is an optimal estimator. In other words any estimator that uses the same observation vector r will have at best the same performance as the ML estimator of Example 2. It may be shown that for the Example 2

var{α}= ⁿ(^σ} v, (21)

^Nobsψ_o ² _bs -l) where σ^ is the variance of x_k, k = 1,2,...,N₀J₃₈.

For N_0J38 » 1 , var{α} behaves as

•

There are solutions to sync over packet networks

where p = [_Pl,p₂,...,p_Nobs ] , (22)

Pk = ^rk\\ ~ ^rk (²³)

and where p is a vector of difference of adjacent time stamps

is taken as the observation vector. In such a case part of information contained in r is disregarded and therefore the performance of any estimator based on d will be, at best, equal to that of ML estimator based on r as given by Expression 6.

It will be evident to those skilled in the art that the invention is not limited to the details of the foregoing illustrated embodiments and that the present invention may be embodied in other specific forms without departing from the spirit or essential attributes thereof. The present embodiments are therefore to be considered in all respects as illustrative and not restrictive, the scope of the invention being indicated by the appended claims rather than by the foregoing description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein.

Claims

WHAT IS CLAIMED IS:

1. A synchronization system for a communication network, comprising: an optimal frequency estimator based on a vector of time stamps constituted by an observation vector that estimates the period of incoming packets.

2. The synchronization system as claimed in claim 1, wherein the optimal frequency estimator is a maximum likelihood estimator.

3. The synchronization system as claimed in claim 2, wherein the maximum likelihood estimator reaches the Cramer-Rao-bound.

4. The synchronization system as claimed in claim 2, wherein said estimator is an unbiased estimator.

5. The synchronization system as claimed in claim 2, wherein said maximum likelihood estimator utilizes the time stamps stamped according to the time axis dictated by a local oscillator of the maximum likelihood estimator.

6. The synchronization system as claimed in claim 1, wherein the period estimated is used to generate an output time function that appears at the output of an oscillator with externally controlled frequency.

7. The synchronization system as claimed in claim 5, wherein said oscillator is selected from the group of a digitally controlled oscillator, a voltage controlled oscillator or a numerically controlled oscillator.

8. A synchronization system for a communication network, comprising: an optimal frequency estimator that reaches the Cramer-Rao bound such as an maximum likelihood estimator, receiving input signals from a packet switch network and clock signals from a local oscillator, and a controlled oscillator fed by said estimators' output signals and by said local oscillator to produce an output signal representing the output time function and used to estimate probability density function (PDF) of a packet delay variation PDV.

9. A synchronization system as claimed in claim 8, wherein said oscillator is selected from the group a digitally controlled oscillator, a voltage controlled oscillator or a numerically controlled oscillator.

10. A method for synchronization in a communication network, comprising: forming an observation vector by a time stamp; assuming an initial probability density function (PDF); computing a PDF of the observation vector; determining the estimated values of the output packets period and the time shift, and generating an output time function which is used to estimate the PDF of the packet delay variation (PDV).

11. The method for synchronization as claimed in claim 10, wherein the period estimated is used to estimate the PDF of PDV by generating the output time function that appears at the output of an oscillator with externally controlled frequency and then subtracts it from the time stamps of the input time function to obtain an estimated period.