The Increasing Role of AI in Telecom
The term "AI" is everywhere, and can mean different things depending upon the application. Those who work with computerized networks know that self-programming systems have been around for years, though most still require significant human interaction. Newer, truly intelligent features are coming which will reduce service impacts and ease system complexity, while automatically performing all provisioning for new or changed circuits, helping to improve overall customer experience.
A Ciena 6500 T-Series high-capacity transport system. The 6500 series is capable of significant circuit-level auto-provisioning, reconfigurable optical add-drop multiplexing, plus it can automatically respond to outages or impairments based on many different alarm conditions.
Telecommunications and IP networks evolved separately over the years, but each industry helped push state-of-the art computerized systems. Simple actions such as making a cross-connection with a physical cable required a person to move a cable between two ports. Needless to say, this required a person to do the work, but it also added possible problems including the interpretation of labeling (if there was any at all), making mistakes, plus added points-of-failure in both electrical contacts and the cable itself. Future changes or repairs would require a person to revisit the connection points, and start the process over
Digital-access cross connects (DACS) were a huge improvement, making the connection process electronic which significantly improved reliability. However, using an electronic system still required a person to log in and make changes to programming. This required training, but also lead to possible challenges with understanding an overall network, or even a "network of networks", leading again to time delays and possible mistakes.
IP network designs, largely lead by Cisco in the 1990s, would benefit greatly with the advent of Routing Protocols, an automatic program that can recognize adjacent connected equipment via the network ports on a local router. This allowed a network to essentially program interfaces all by itself, with minimal human interaction. That was great, but what made the routing protocol so powerful was the ability for all equipment to then send the learned connection information to all other adjacent devices, allowing every device on the network to know the entire network topology. Protocol's like Cisco's RIP and EIRGP, or the Internet Engineering Task Force's (IETF) OSPF and BGP, are all considered essential elements of modern IP networks today. While the automatic learning used by such protocols might today be considered pre-AI and even rudimentary to some, they fulfill the promise of AI by automatically learning extremely complex networks quickly and accurately, and with the ability to dynamically update if any equipment is added or removed, or even if a fault occurs on single network, or a network of networks. (See below: An example of an OSPF network)
Routing protocols are terrific, but what if the link speed between two routers needs to change? Or what if a new link is being turned-up, but the installation team doesn't yet know the correct parameters to enter. Could this all be provisioned by the system automatically? The answer, most of the time, is yes.
Image provided from the Cisco Community. An example of a network programmed automatically by the Open Shortest Path First (OSPF) routing protocol. This relatively simple network would have taken hours to program manually. OSPF can map out much more complicated networks than this, saving very significant amounts of time, while also preventing errors.
Early "Telco AI"
Autoprovisioning is another feature that has existed for years, but has become much more capable in determining changing-circuit parameters by itself. For example, many SFP+ transceivers are capable of supporting 10G Ethernet, OC-192 SONET, an OTU2 rate, and 10G Fiber Channel. After installing such a transceiver, it will automatically detect the header information being received from the remote end, even in a circuit idle state (with no customer traffic). This means the system can automatically enable the transceiver port, and set it to the correct speed for the correct protocol. In most modern transceivers, an inadvertent reversal of transmit (Tx) and receive (Rx) can also be detected, causing this specific alarm condition so that technicians know the exact type of problem. Sometimes, the system can even automatically reverse Tx and Rx (usually just for copper circuits), providing service while still noting the alarmed condition, so that the cabling can eventually be corrected.
These kind of autoprovisioning features not only ease the process of programming complex networks, but help prevent and notify users of any mistakes. The idea can be taken further when working with protection circuits.
Many will already recognize the concept of a working circuit, which is being backed-up by a protection circuit. In its original principle, a protection circuit is used only when the working circuit fails or degrades in quality. While a simple design to implement, traditional protection schemes essentially double the cost of a network, offering no ongoing return on the investment for the protection circuit that sits idle most of the time, other than the guarantee of service that is needed for service level agreements. But what if the protection circuit could also be used to send data, and so generating additional revenue?
This is where Virtual Concatenation (VCAT) and Link Capacity Adjustment Scheme (LCAS) come in. Often used together, VCAT and LCAS allow traffic in one direction to use several different paths at the same time, acting as one big path. This can greatly speed communication, while generating revenue on the protection paths that might otherwise sit idle. These multi-path links can also be adjusted dynamically, so that if a fault does occur, the overall path might drop in speed, but not entirely lose service. While this might result in a network slow down, it may be undetectable for many or most users, particularly if the terms of their SLA agreement did not require very high, always-on speeds. The diagram below shows a simple VCAT/LCAS example, but network designs can become much more complex.
Virtual Concatenation (VCAT) and Link Capacity Adjustment Scheme (LCAS) can automatically learn and allocate throughput on more than one path, both in normal circumstances, and after a network failure.
What's Coming
Some interesting new features are already arriving. For example, during conditions of stress on a fiber optic cable, such as cable sag during summer months or ice loading during the winter, the light sent through the fiber will suffer from reflections, a phenomena due to the nature of how light moves through a fiber, where some of the forward-moving light is reflected back towards the transmitter. What makes this interesting is that light reflections can be characterized for a given degree of sag on the fiber span, so that a given reflection level indicates a given amount of cable sag. If specifications from the fiber vendor are provided, the fiber break point can be predicted in advance, allowing traffic to be re-routed via another route, such as via a leased circuit that would normally add unnecessary cost, but may be valuable now when a service interruption seems imminent. Used along with features like VCAT and LCAS, leasing just-in-time network capacity when needed can theoretically eliminate network outages, at the lowest possible cost, and while also maintaining the highest level of service. What's not to like?
Trying to design such complex circuit options at the human level is certainly possible, but it can take significant time. What's worse is such networks are not easily updated when changes occur (either due to equipment failure, addition, or removal), plus manual updates are prone to mistakes. New AI features built into telecommunication and IT systems are now making it possible to both learn and react to network changes automatically. Even systems made by different vendors, often with their own proprietary technologies, are increasingly being supported by just a single vendor's network management system, allowing the entire network of different hardware to be managed by only one system. Ciena's BluePlanet and Cisco's Telco Cloud have broad capabilities to manage many different types of hardware, while providing features for complex networks to self-heal, effectively routing traffic around a problem area until the physical issue can be fixed.
Calix's Broadband Platform is also an early user of Vertex AI and Google Gemini, which will make setup and provisioning of the system much easier for the service provider. If you have ever tried to set-up an FTTP network, using revertive G.8032 rings, Link Aggregation uplinks, and Service profiles for bronze, silver and gold tiers, you will know how much work it is. Using AI to easily set-up a system, and later administer it via a cloud manager, is all a welcome idea.
These are arguably incremental ideas, which improve existing processes, but do not necessarily add new revenue-generating ideas…yet. Using truly intelligent AI agents, rather than today's limited and often annoying Chat-bots, could be very helpful in reducing customer response times, plus overall customer satisfaction, if done well. As with sagging cables, using AI to recognize a degradation in a specific customer's service, well before it becomes noticeable, would be helpful. Once again, Calix has tried to put this intelligence into their Calix Cloud® manager which proactively looks at customer experience levels for Wi-Fi and overall ONU throughput. The service provider is made aware of logged performance issues, and can almost predict when the customer might call, and what they'll say. This is a direct line to discovering how to avoid problems in the first place. It may take time, but data being gathered today will eventually lead to much lower service impacts, and faster methods to mitigate impacts when they do occur. AI can help with this too, analyzing large swaths of data, organizing cause and effect for a given user, or for an entire network. What might take a small team a week of analysis could be done in seconds by AI.
While most of us acknowledge that technology will be helpful in the future, few of us have the gift to see how it will actually help us. AI is still in it's first inning, and it's difficult to imagine the full scope of what it might one day do for us. Still, we can already see it is having a positive impact on communications and IT, and this will continue to accelerate.
Some interesting new features are already arriving. For example, during conditions of stress on a fiber optic cable, such as cable sag during summer months or ice loading during the winter, the light sent through the fiber will suffer from reflections, a phenomena due to the nature of how light moves through a fiber, where some of the forward-moving light is reflected back towards the transmitter. What makes this interesting is that light reflections can be characterized for a given degree of sag on the fiber span, so that a given reflection level indicates a given amount of cable sag. If specifications from the fiber vendor are provided, the fiber break point can be predicted in advance, allowing traffic to be re-routed via another route, such as via a leased circuit that would normally add unnecessary cost, but may be valuable now when a service interruption seems imminent. Used along with features like VCAT and LCAS, leasing just-in-time network capacity when needed can theoretically eliminate network outages, at the lowest possible cost, and while also maintaining the highest level of service. What's not to like?
Trying to design such complex circuit options at the human level is certainly possible, but it can take significant time. What's worse is such networks are not easily updated when changes occur (either due to equipment failure, addition, or removal), plus manual updates are prone to mistakes. New AI features built into telecommunication and IT systems are now making it possible to both learn and react to network changes automatically. Even systems made by different vendors, often with their own proprietary technologies, are increasingly being supported by just a single vendor's network management system, allowing the entire network of different hardware to be managed by only one system. Ciena's BluePlanet and Cisco's Telco Cloud have broad capabilities to manage many different types of hardware, while providing features for complex networks to self-heal, effectively routing traffic around a problem area until the physical issue can be fixed.
Calix's Broadband Platform is also an early user of Vertex AI and Google Gemini, which will make setup and provisioning of the system much easier for the service provider. If you have ever tried to set-up an FTTP network, using revertive G.8032 rings, Link Aggregation uplinks, and Service profiles for bronze, silver and gold tiers, you will know how much work it is. Using AI to easily set-up a system, and later administer it via a cloud manager, is all a welcome idea.
These are arguably incremental ideas, which improve existing processes, but do not necessarily add new revenue-generating ideas…yet. Using truly intelligent AI agents, rather than today's limited and often annoying Chat-bots, could be very helpful in reducing customer response times, plus overall customer satisfaction, if done well. As with sagging cables, using AI to recognize a degradation in a specific customer's service, well before it becomes noticeable, would be helpful. Once again, Calix has tried to put this intelligence into their Calix Cloud® manager which proactively looks at customer experience levels for Wi-Fi and overall ONU throughput. The service provider is made aware of logged performance issues, and can almost predict when the customer might call, and what they'll say. This is a direct line to discovering how to avoid problems in the first place. It may take time, but data being gathered today will eventually lead to much lower service impacts, and faster methods to mitigate impacts when they do occur. AI can help with this too, analyzing large swaths of data, organizing cause and effect for a given user, or for an entire network. What might take a small team a week of analysis could be done in seconds by AI.
While most of us acknowledge that technology will be helpful in the future, few of us have the gift to see how it will actually help us. AI is still in it's first inning, and it's difficult to imagine the full scope of what it might one day do for us. Still, we can already see it is having a positive impact on communications and IT, and this will continue to accelerate.
Copyright, AG Advice and Support, 2026.