Clock recovery method over packet switched networks based on network quiet period detection

1. A method of recovering timing information in a packet network, comprising:

receiving timing packets from a network;

detecting quiet periods in the packet network when network packet delay variation (PDV) for the timing packets is low;

training a frequency prediction unit during said quiet periods to learn output clock variations of a timing recovery circuit to permit the frequency prediction unit to predict frequency update values for a local oscillator during non quiet periods taking into account the historical output clock variations during quiet periods; and

using the output of the timing recovery circuit as the active frequency update values during quiet periods and said frequency prediction circuit as the active frequency update values during non quiet periods.

2. A method as claimed in claim 1 , wherein the output of the timing recovery unit is compared with the output of the frequency prediction unit, and when the frequency update values provided by the frequency prediction unit differ from the frequency update values provided by the timing recovery unit by a predetermined threshold, the frequency update values provided by the frequency prediction unit are selected as the active frequency update values.

3. A method as claimed in claim 2 , wherein the threshold is adaptively varied based on the degree of confidence in the output of the frequency prediction unit.

4. A method as claimed in claim 3 , wherein the degree of confidence is based on the time required for the frequency prediction unit to predict the frequency update values.

5. A method as claimed in claim 4 , wherein the frequency update circuit comprises a Kalman filter.

6. A method as claimed in claim 5 , wherein the Kalman filter produces the frequency update values based on the equation:

{circumflex over (x)} n − =F n−1 {circumflex over (x)} n−1 +

where F n represents the state transition matrix and {circumflex over (x)} n − and {circumflex over (x)} n + are respectively a priori and a posteriori state estimates of the state vector x n .

7. A method as claimed in claim 1 , wherein the local oscillator is a digital controlled oscillator (DCO).

8. A timing recovery unit for use in a packet network, comprising:

a local oscillator for producing a local output clock;

a timing recovery circuit responsive to timing packets received over a network to produceg frequency update values for the local oscillator;

an activity detector for detecting quiet periods in the packet network when network packet delay variation (PDV) of the timing packets is low;

a frequency prediction circuit for predicting update values for the local oscillator during non quiet periods, and wherein the frequency prediction circuit is configured to learn output clock variations of a timing recovery unit to permit the frequency prediction unit to use the historical clock variations to predict frequency update values during non quiet periods; and

a selection circuit for selecting the output of said frequency prediction unit as the active frequency update values for the local oscillator during non quiet periods and the output of the timing recovery circuit as the active frequency update values for the local oscillator during quiet periods.

9. A timing recovery unit as claimed in claim 8 , further comprising a selection unit that compares the output of the timing recovery unit with the output of the frequency prediction unit, and when the frequency update values provided by the frequency prediction unit differ from the frequency update values provided by the timing recovery unit by a predetermined threshold selects the frequency update values provided by the frequency prediction unit as the active frequency update values.

10. A timing recovery unit as claimed in claim 9 , wherein the selection unit is configured to adaptively vary the threshold based on the degree of confidence in the output of the frequency prediction unit.

11. A timing recovery unit as claimed in claim 10 , further comprising a timing circuit for measuring the time required for the frequency prediction unit to predict the frequency update values, and wherein the degree of confidence is based on the measured time.

12. A timing recovery unit as claimed in claim 11 , wherein the selection unit selects the update values based on the relationship:

(1−β( T p ))× f e ≦f c ≦(1+β( T p ))× f e →f d =f c

where f c is the output of the timing recovery circuit, f e is the output of the frequency prediction circuit, and 0<β(T p )<<1 defines the acceptance threshold for f c .

13. A timing recovery unit as claimed in claim 8 , wherein the frequency prediction unit comprises a Kalman filter.

14. A timing recovery unit as claimed in claim 13 , wherein the Kalman filter is configured to produce the frequency update values based on the equation:

{circumflex over (x)} n − =F n−1 {circumflex over (x)} n−1 +

where F n represents the state transition matrix and {circumflex over (x)} n − and {circumflex over (x)} n + are respectively a priori and a posteriori state estimates of the state vector x n .

15. A timing recovery unit as claimed in claim 8 , wherein the activity detector comprises a feature extraction circuit, a minimum maximum estimator, and a comparator for comparing the difference between the minimum and maximum values of the extracted features.

16. A timing recovery unit as claimed in claim 15 , wherein the extracted feature is the variance of the raw delays.

17. A timing recovery unit as claimed in claim 15 , wherein the extracted feature is the variance of the minimum delays over a series of time periods.

18. A timing recovery unit as claimed in claim 8 , wherein the local oscillator is a digital controlled oscillator (DCO), and the update values are the frequency update values for the DCO.