Mid-Split vs High-Split

What 'the split' refers to

An HFC network carries both directions on the same coaxial cable, separated by frequency. The lower portion of the band carries upstream traffic; above a transition region, the rest carries downstream. The frequency where that changes over is the split.

Legacy plant used a low split, giving upstream a narrow band at the bottom. That was the right allocation for the traffic of the time. It stopped being right as upstream demand grew, and on a lot of plant the return path is now the binding constraint on what the network can offer.

Raising the split gives the upstream more room. Within a fixed upper-spectrum limit it is not free: increasing the upstream allocation reduces the spectrum available to downstream, plus the transition region between them, which is not usable by either direction. Extending the plant's upper frequency limit — the 1.2 GHz and 1.8 GHz modernization path — can recover and add downstream capacity. The available frequency plans are defined in the CableLabs DOCSIS 3.1 PHY specification.

Mid-split

Mid-split raises the upstream ceiling to roughly 85 MHz. It is the more modest of the two moves and has been deployed widely.

What it buys. A substantial multiple of the upstream spectrum a legacy low-split provides — enough to lift upstream service tiers meaningfully for most residential use.

What it costs. A relatively small slice of downstream, and plant work on the devices that constrain the return path. Diplex filters in amplifiers determine where the split sits, so raising it means those devices have to support the new plan — which in a lot of plant means a module change or a change-out.

Where it fits. Mid-split is the pragmatic choice when the upstream constraint is real but not yet acute, when the plant has a large population of devices that would be expensive to replace wholesale, or when it is a staging step toward something larger. Its appeal is that it is contained.

High-split

High-split raises the upstream ceiling to roughly 204 MHz, with the forward path resuming above the transition region. It is a considerably larger intervention.

What it buys. Substantially more upstream than mid-split — enough to support the kind of symmetrical-leaning service tiers that competitive fiber builds have made table stakes in some markets.

What it costs. Three things, and the third is the one that surprises people.

• More downstream surrendered. Enough that most operators pair a high-split with a plan to recover that capacity — higher-order modulation, extending the upper band, or both. That connects the decision directly to DOCSIS 4.0 planning.
• Broader plant impact. More devices are affected, and the transition region imposes tighter requirements on the diplex arrangement.
• CPE and legacy service interaction. The spectrum a high-split reclaims was carrying something. Anything still using those frequencies — legacy services, older CPE — has to be accounted for before the change, and that coordination frequently sets the program schedule rather than the plant work.

Where it fits. High-split is the choice where the upstream constraint is acute, where competitive pressure demands a symmetrical-leaning offer, or where the operator wants to make the move once rather than twice.

Comparing the two

The honest summary is that these are not competing products so much as two points on a cost curve.

• Upstream gained: high-split provides considerably more than mid-split.
• Downstream surrendered: proportionally more for high-split, usually enough to require a recovery plan.
• Devices affected: both require attention to the diplex arrangement across the cascade; high-split reaches further.
• CPE coordination: mid-split is more contained; high-split reclaims spectrum that is more likely to be carrying something today.
• Longevity: high-split buys more headroom, which matters if the alternative is doing this again in a few years.
• Stageability: mid-split is easier to deploy incrementally.

The comparison that actually decides it is rarely mid-split versus high-split in isolation. It is mid-split now and high-split later versus high-split once. Doing it twice means touching a large share of the same devices twice, and the second visit costs nearly as much as the first.

How this interacts with segmentation

Splitting and segmenting are the two levers, and they are not substitutes.

Segmentation reduces how many homes share a service group. It does not change the spectrum available to that group; it changes how many people are dividing it. Depending on the architecture it may require new fiber, optics or RPD capacity — and where new node locations are involved, the fiber brings make-ready and permitting with it.

A split changes the usable upstream/downstream spectrum across the affected plant, without requiring new fiber — but it touches the actives across the cascade and interacts with the CPE base. Operators may deploy that change in phases by node, hub, service area, or market.

Most operators need both, and the sequencing has real money in it. Segmenting a market first reduces the number of subscribers affected by a subsequent cutover and can reduce cascade lengths in ways that make the split easier. Splitting first delivers upstream relief across the whole footprint sooner. Doing them in uncoordinated sequence means visiting the same nodes twice, which is the outcome worth engineering around. Our modernization playbook covers this sequencing question at program scale.

The part that actually sets the schedule: CPE and legacy services

Plant work on a split program is predictable. You can count the actives, price the modules, and schedule the crews. What is far less predictable — and what repeatedly determines when a high-split market can actually cut over — is everything that is currently using the spectrum you are about to reclaim.

Modem capability

A subscriber's modem has to support the upstream plan the plant is moving to. Legacy CPE may continue operating within the spectrum and DOCSIS capabilities it supports, but it may not take advantage of the additional upstream capacity or the higher service tiers the new split enables — so operators need to inventory CPE capability and plan migrations where required. That leads operators to an important question: how much of the installed CPE base can use a high-split, and what does replacing the rest cost?

The answer is specific to each operator's deployment history and is one of the few genuinely large uncertainties in the program. It is worth establishing early, because it can move the sequencing decision entirely — a market with a modern CPE base is a very different proposition from one carrying a long tail of older devices.

Other services using the reclaimed spectrum

The reclaimed spectrum is rarely empty. Depending on the network it may be carrying legacy services, status monitoring, or other operational traffic that predates the current plan and that nobody has needed to think about for years. Each of those has to be identified, and either migrated or retired, before the band can be handed to the upstream.

This is the work that surprises programs. It is not engineering-hard; it is discovery-hard. Finding out what is using a frequency range across a large footprint means asking questions of teams who were not part of the capital project, and the answers arrive on their schedule rather than yours.

What to do about it

Start the CPE and legacy-service inventory at the same time as the plant verification, not after the design is finished. On a mid-split the exposure is smaller and can often be absorbed. On a high-split it is frequently the critical path — and a program that discovers it during construction has already lost the quarter.

It is also the strongest practical argument for the “high-split once” option. This coordination cost is largely paid per market rather than per megahertz, so doing it twice means paying most of it twice.

What a split program involves

Whichever position is chosen, the program shape is consistent.

1. Device inventory. What is deployed, in what configuration, and what supports the target plan. Sizing this from records rather than verification is the most reliable way to be wrong about the budget.
2. Design. Per-node identification of what changes, plus the level and cascade implications of the new plan. Covered on our HFC design page.
3. Headend and CMTS readiness. The platform has to support the plan and the channel arrangement it implies.
4. CPE and legacy service plan. What is using the reclaimed spectrum today, and how it is migrated. On high-split programs this frequently sets the schedule.
5. Construction sequencing. Touching actives across markets against finite crew capacity — see construction coordination.
6. Sweep, QA/QC and closeout. The plant now carries a different frequency arrangement and has to be verified against it. See activation and closeout.

Field practice throughout is shaped by the operator's own standards alongside those published through the SCTE standards program.