The dilemma of C upgradeable to L Band OLS

Errors do not occur due to a lack of choices.

They occur when the choices are too many.

Dear All,

I am writing this article after analyzing several networks that are currently considering the implementation of C Band, upgradeable to L Band Optical Line Systems (OLS). While being “future-ready” is an essential KPI for every network planner, it is equally important to keep a close watch on current network utilization, realistic traffic projections, and optimization of the existing spectrum.

Let us carefully evaluate each aspect of this OLS dilemma so that we can determine when and how it makes sense to implement C-to-L Band OLS.

View of the DWDM Spectrum with C and L Band
Source: Researchgate

1. C Band already offers significant capacity

Gone are the days when C Band meant 96–120 channels of 50GHz spacing, each carrying 100G optics. Today, C Band can support flexible spectrum configurations—such as 64 channels of 400G or even 32 channels of 800G wavelengths.

This represents enormous capacity with more efficient grooming options. For instance, with 32 channels of 800G per segment (approx. 500 km), we achieve 25.6 Tbps of bandwidth—all on a cost-effective C Band OLS.

2. The importance of accurate traffic projections

An OLS is fundamentally designed based on traffic projections. If a five-year forecast indicates that C Band spectrum will not be exhausted (even after applying grooming optimizations), then deploying an L Band-ready OLS is an unnecessary expense.

Such an approach increases the upfront cost of the solution, while L Band usage may not be required until after year five. By then, the additional cost of a C+L OLS becomes an overhead, impacting provisioning economics and overall profitability.

There have been instances where bandwidth providers chose C+L OLS upfront but later struggled to sell services due to higher entry costs—leading to reduced margins and lower market competitiveness.

3. Cost implications of C-to-L upgradeable OLS

While it may sound appealing to demand C+L upgradeable OLS, the CAPEX and OPEX implications must be understood:

Component manufacturing for C+L systems is not yet cost-effective.

Switch-gain amplifiers with dual directions are expensive compared to EDFA-based C Band OLS.

Low-degree ROADMs (for smaller sites) are often unavailable; most C+L ROADMs are 20–32 degrees, adding cost.

Some vendors require separate ROADMs for C and L Bands, effectively doubling costs. Others mitigate this using couplers or integrated designs, but pricing remains high.

Amplification remains a challenge. No single amplifier can handle both bands today. Instead, L Band amplifiers must be cascaded with C Band amplifiers—further inflating costs.

For operators whose traffic will remain stable for the next five years, deploying a C+L OLS is commercially unjustifiable, as competitors running pure C Band deployments will enjoy lower costs and better margins.

4. The evolving industry landscape

Just two years ago, C+L ROADMs and WSS were not commercially available. Today, they are. Similarly, research is ongoing for day-one C+L amplifiers.

If an operator invests prematurely in today’s C+L upgradeable OLS, they may miss out on more cost-effective and technically advanced solutions that will arrive in the near future.

Thus, the decision to deploy C+L OLS must be purpose-driven and strategically timed.

When should C+L OLS be considered?

Despite the above constraints, there are specific scenarios where C+L OLS may be justified today:

a) Confirmed near-term growth: Bandwidth is projected to exceed 12.8 Tbps within two years.

b) Captive demand: Customers with steady, predictable YoY growth (e.g., ~25%).

c) Hyperscalers as primary customers: These clients often demand full-spectrum allocations (multiples of 500GHz), which validates building a C+L network.

d) Spectrum slicing strategy: C Band is reserved for captive bandwidth, while L Band is leased for alien wavelengths—enabling day-one dual-band utility.

Concept of C Upgradeable to L band 
Source: Ribbon Communications

Conclusion

An OLS is a long-term investment—this much is agreed. However, the cost must be balanced against realistic capacity growth and market demand.

Engineering-driven choices that overlook commercial realities can result in revenue shortfalls and reduced margins. Therefore, the decision to adopt C+L OLS must be based on accurate traffic projections, well-defined timelines, and market conditions, ensuring that the network delivers maximum yield for the investment.

So till then my friends, 

Happy Network Planning. 

Cheers, 

Kalyan

Understanding AI like a layman engineer

 The best way to understand a complex thing is to break it down into very simple terms.

--- My Uncle (Shankar Mukherjee)

I would not be a successful person today and definitely not an efficient engineer if I would not have followed this principle laid down by my late Uncle.  However complex the problem may have been, he used to break it into several simple blocks and then the understanding would be so simple that it would enter not only the brain and intellect but also be engrained in the conscience. 

With this very principle, I would like to break down the concept of AI into several blocks that are important to understand how this new animal works. This blog post is especially for those of my engineer friends who are a bit intimidated by this new technology and I would like to tell them that possibly, as the learned say, this is the first time that the human kind has created something that may be more intelligent than the Human mind itself. Well let us first understand the building blocks of this AI from a basic engineering point of view. 

AI is not magic, it is a masterpiece of co ordination.

Well for sure any technology is not magic. Decades ago when the computer was invented people thought this was a technology that could change the ways of humanity. It really did, but definitely in the way as it was projected maybe four decades ago. There were sci fi movies being made projecting this instrument as some sort of a magical gadget that could do anything that we could not do. Yes it did so but it never went beyond the realms of human thought and imagination. It was not magic. 

Just like computer literacy in 1990s, AI know-how is important in the 2020s

Today at the 2020s we are in the same crossroads where we are thinking that this new technology will be a magical stuff. Possibly we have not learnt enought from our past experiences that machines cannot take our place. However, being AI literate is definitely important just like being computer literate was important in the 1990s and the 2000s. Later computer literacy did not remain an added advantage, it became a necessity and a part of our academic know-how. So will be the concept of AI in the next decades to come. An engineer who can efficiently use AI would be reversed for a decade and for decades to come there after this knowledge will be more commonplace. 

Building Blocks of AI

Now let us see how can we break down the AI structure into different building blocks to understand how this AI is actually realized in any system.

1. Infrastructure. 

For an AI platform to be efficient and robust it is very necessary that it is based on a solid infrastructure. Infrastructure here may refer to bandwidth, power, storage, reliability and a lot of things. Data centers are the new centers of infrastructure that are being panned out now across the world. However, for Data centers to work seamlessly there needs to be ample amount of resilient bandwidth connecting them. Therefore, apart from the data centers what is required is the network that actually connects them together in order to establish that constellation which provides the foundation for the AI Platform. 

2. Data Ingestion, mining and processing 

AI is severely data dependent. It will not at all work if there is not data ingestion mechanism. This data ingestion mechanism has to be constant and has to be very reliable. There has to be also a level of agility and storage resiliency connected to this. Data centers having the ample amount to storage need to be a part of this ecosystem. 

It is also important to search for several data points that can make the platform more realistic and more accurate in nature. The accuracy of an AI platform depends on the granularity of the data and this can be further enhanced by data mining technologies that are there in the market. Statistics and cognitive methods of mining the data can be used to make this work properly. 

After the data is invested and mined it has to be processed and developed into understandable information. Raw data is of no use. Here is where we come to know about the difference between data and information. Information is when you can initiate an action based on the data output. This information is derived when there is an ample amount of data processing. Data processing tools are very much available in the market and being a master of one of them could definitely be a good ticket to being a part of this AI revolution. 

3. Analytics

As we see in the earlier section around converting data into information. Analytics also provides the same part over here. Let us assume that a data processing software or a program is automated in a way that it can process the data and provide meaningful information to us in different manner. This is actually the role of analytics. 

Providing cognitive information, predictive information and inferences is the work of a proper analytics software. The analytics software makes work easy for an engineer to determine future course of planning and to troubleshoot present faults. With a degree of automation in the analytics software there can be elimination of more human interference and this is why knowing analytics is super critical and super important. 

4. Algorithm 

This is the most important part of the AI. Now this is where the so called magic happens. Algorithm is a sort of logic that brings randomness and constancy together with the output of the Analytics that is taken out. 

Let us say that the analytics is giving an output that in the next five years there needs to be more bandwidth allocated to a particular section. The impact of this judgement of the analytics influences the algorithm to proactively probe the section and increase the bandwidth subsequently without any sort of human interference. The algorithm knows when actually to do what. 

The best thing about this algorithm development is that it has to be non predictive in nature so that it cannot be cheated. This only happens when all the other building blocks are rich enough. We have seen in the recent past many such algorithms and intelligent people have been able to even hack or break it. As the time passed there were developments of more and more random algorithms that were difficult to hack. Well, in order to give AI that magical touch such algorithms need to be developed. 

5. User integration 

Have you heard of that term, "computing without the computer".  Fancy, right? We are talking about an entire hardware and software ecosystem being integrated to the most simple peripherals. 

Imagine if you could determine an action by just voice commands? Imagine if you could go through security doors based on your biometric footprints? Well these are today possible because if peripheral integration with the software.  This is the fifth block of AI that makes it accessible to masses and makes it more popular. 

User integration can be seen in professions like digital image processing. Biometric processing. It can be seen in sensors, peripherals etc. Basically any input and output point that can connect a human to the software.

So my friends as you see, this is not as complicated to understand as it seems to be. I know, you have come from various events all across the world. Events that are celebrating and propagating this word "AI" with all the pomp and pageantry. 

But then breaking it into simple forms will make you realize that everyone in the tech domain has a place and has something to offer to this wonderful technology that is germinating. 

Cheers,

Kalyan