056 - Advanced algorithms and techniques for resilient time provision
Market trends show a clear increase in number of timing-dependent services (including highly critical ones), with increasing demand for improved performance and capability. RF signals broadcasting is becoming increasingly dependent on GNSS for accurate timing and frequency references. 5G networks will be increasingly dependent on precise and robust timing and synchronisation. Integer timing signals are essential to keep communication systems and information systems synchronised, while providing a frequency reference for radio systems. Other market segments include the power industry, operating widely distributed infrastructure with stringent timing and availability requirements to improve the efficiency of power generation and distribution, and finance corporations that are also operating critical infrastructure deployed worldwide, with rapidly evolving practices (e.g. high-frequency trading) and associated regulatory constraints.
In these market segments today, GNSS remains the prime source of time information for system synchronisation. In order to improve resilience of time provision and cope with GNSS vulnerabilities, different combinations of alternative time sources can be used (e.g. local clock to guarantee hold-over, packet-based time protocols, or any other signal of opportunity).
Combining different time sources call for advanced techniques and algorithms to provide the required level of time performance to users. In addition, the availability of such robust algorithms shall allow the various providers of timing solutions to monitor their network or technology for fault detection and system optimisation.
One of the possible solutions could be based on combined time and frequency Kalman filters applied to multiple sources of time. Although the idea of a frequency Kalman was introduced long ago, there is no trace in the literature of any real application to date, most probably due to the problem of steering to an external time reference, which is better solved in the time domain. The idea of optimizing Kalman parameters around some natural resonance frequency has both intuitive and strong physical meaning while, in contrast, setting optimum Kalman parameters to minimize the time error seems quite artificial and has no analogous physical meaning. Other solutions also include various types of Machine Learning techniques, which could also be combined with the Kalman filter approach.
The applicability of these techniques could be very broad, from time service providers for critical infrastructure and network operators to trade organisations, oil and gas companies, scientific research, power grids’ corporates, etc., at different scales and based on the available time information sources for each of them.
The objective of this activity is to analyse and develop advanced techniques and algorithms for the generation of a resilient time reference based on a multiplicity of time sources (e.g. local clocks, GNSS, signals of opportunity). In particular, algorithms based on hybrid (frequency-time) Kalman filters, Machine Learning techniques shall be considered, as a minimum.
The tasks to be performed include:
- review and trade-off analysis of algorithms and techniques for generation of a time reference based on a multiplicity of time sources and information;
- development of algorithms and techniques for optimum combination to reach the targeted accuracy, stability, continuity and availability requirements;
- preparation of an experimentation plan for the assessment of the benefits of such combinations as compared to state-of-the-art solutions;
- tuning algorithms, including an exploration of ways to use AI to adapt the thresholds of the corresponding frequency and time Kalman detection barriers;
- carrying out extensive experimentation over a sufficiently representative set of data in order to properly assess performance of algorithms;
The main results of the activities will provide:
- advanced techniques and algorithms for the generation of a resilient time reference, including detailed processing models;
- outcome of extensive experimentations to assess performance and capabilities;
- Hardware to operate and collect time information data.