In part 1 of this two-part series on Optical Intelligence, I explored the case for making the journey towards network autonomy and intelligence. In part 2, I look at making the journey.
Part 2: Optical Network Health Prediction: A Key Class of Optical Intelligence
One type of optical network intelligence stands out as required to support and improve – to a lesser or greater degree – a broad range of optical network operations, including all of the following: service configuration and reconfiguration, particularly in tackling high-value, customized, and dynamic premium line services; and resource planning, resource optimization, and service performance & health management (Optical Network Health Prediction). The reach and performance of a prospective optical service connection depends on many factors, including selected transceiver modulation format, rate, and coding; fiber plant characteristics; optical amplification; and wavelength filtering. A sound decision on whether and how to configure an optical service cannot be reliably made without having in advance a good prediction of the prospective service’s performance. Similarly, optical network planning and optimization require the predictive performance assessment of small or large sets of prospective optical services among considered alternative paths, spectral allocations, and modulation and coding formats and rates. Note that the performance of individual optical services cannot be predicted in isolation, as optical services co-propagating on the same fiber links affect each other through fiber nonlinearities and conjoint impacts on optical amplification, rendering accurate performance prediction doubly difficult. Finally, accurate performance prediction is an important assist to optical service performance management – as it provides a benchmark for expected performance under assumed plant and network element conditions.
Delivering Effective Optical Network Health Prediction Capabilities
Accurate optical performance prediction capabilities on operating networks that are network- and time-specific have been limited by (1) the lack of sufficient, relevant sensor capabilities to feed optical performance models and (2) the complexity of the modeling problem itself. In the absence of such capabilities, operators have tended to plan and operate services with substantial optical budget margins, which collectively leave significant network capacity on the table. These capability gaps must be closed if dynamic margin “squeezing” is to be able to unleash this inherently idle capacity.
New methods for measuring optical characteristics and performance – and of delivering such information to software engines using push-sensory data paradigms – must be developed and deployed on optical network equipment and related software. Examples that Huawei has developed include accurate end-to-end channel filter shape measurements and practical methods of independently measuring linear and nonlinear channel noise contributions. Supervised machine learning has proved useful in crafting accurate device performance models, for example, EDFA gain and noise figure models; the modeling of channel filtering impacts on noise (OSNR) penalties; and received BER on optical channels (OCh), transport sections (OTS) and multiplex sections (OMS); and transmitter-receiver back-to-back performance predictions based on measurements done independently on transmitters and receivers at point of manufacture. It should be noted that measurement and prediction are flip sides of the same coin: the more that can be measured on operating networks, the better the quality of information available to feed models in predictive use cases.
Huawei’s new, emerging optical network health prediction capabilities reside largely in a behavioral simulation engine that is cloud-deployed in conjunction with our Network Cloud Engine (NCE) network control and management platform, and which simulates both transient and steady-state operations and resulting optical service performance characteristics. The same modeling engine is used in NCE-based applications from planning to real-time control, with judicious management of model-feeding information ensuring the best predictive results at all times and for all applications. For example, in green field network planning, nominal fiber plant and device model parameters may be used. On a deployed network, such parameters are updated through measurement as operational experience is accrued, and through retrieval of specific model-relevant information from actually deployed equipment instances. In this way, the same planning software application serves green field and incremental planning scenarios with the best available accuracy and quality of results, and the lowest learning overhead for customers. This goes along with improvements in Huawei’s optical control plane implementation that make use of information from the NCE-based performance predictive engine in improving and accelerating configuration and reconfiguration – including restoration-based – operations.
Big Data and AI in the Picture
Beyond push-sensory data and model-based optical network health prediction capability, NCE includes big data and AI engines and capabilities that are key to generating other useful optical intelligence. For example, machine learning – both offline and online – is used to enable correct classification, from signatures on multiple sensory data feeds, of early-stage, late-stage and critical plant and equipment faults. Early signature detection enables pro-active responses ranging from protection switching to preserve service integrity in the face of incipient hard faults, to equipment replacement and maintenance in the case of early-onset soft impairment issues. The potential for such predictive maintenance capabilities to improve service quality assurance and the economics of maintenance operations is obvious.
In addition, big data and machine learning will increasingly be able to treat the transmission network and fiber plant itself as a sensor network. For example, state-of-polarization dynamics, if sensed and reported, can be correlated to mechanical disturbances of fiber plant that may indicate digging and incipient fiber cuts; but also – depending on circumstances – may yield monetizable information such as time-based volumes of proximate road traffic.
Finally, OI tools can learn time-based patterns around service volumes and the use of resources to support them and extrapolate into the future. This has been shown highly useful in the efficient, predictive planning of optical network growth.
Optical Network Autonomy: A Destination and a Journey
Optical network autonomy is a destination: but how to get there? The path to optical network autonomy is a journey: one that will not happen overnight, but rather in managed, controlled and focused steps as capabilities in both network hardware and software are built, deployed and exploited. In many cases, intelligence improvements applied to non-automated processes will come first, followed by closing of the automation loop once operator confidence in them is accrued. Individual operational processes may proceed along the path to autonomy on their own paces, following the business imperatives and other circumstantial priorities and constraints of individual network operators. The point is to leverage optical vendors who are delivering the essential capabilities – both hardware and software – that make the profitable journey to optical network autonomy possible.
Huawei is working with its customers around the world – customers like the Italian network operator Fastweb to deploy NCE-based OI capabilities in support of use cases determined and prioritized according to particular operator business needs and pain points. These capabilities are being used to improve operational processes and outcomes ranging from planning to fast provisioning – including bandwidth-on-demand with hitless bandwidth adjustment and managed latency (premium leased line services) – to predictive maintenance and resource usage prediction. We will continue to work in partnership with our customers to develop and improve OI capabilities that deliver impactful operational benefits.
