← all posts

What AFC Requires From a Standard Power Access Point

September 20, 2026 · by Brent Leekley · ~8 min read

A request and a response. The access point sends a horizontal position with one horizontal uncertainty, drawn as an ellipse or a polygon, plus a height above ground with its own vertical uncertainty, to the AFC service. The service replies with authorized channels, permitted power levels, exclusions around incumbents, and a grant that requires contact at least once per day to keep. Caption: a fuzzy position is evaluated as if the AP could be anywhere inside it.

Second post in the 6 GHz power mode series. The first post covered the LPI versus Standard Power decision and the three things every design pays for: power, spectrum, and operations. This post is about the operations bill. Automated Frequency Coordination is the system a Standard Power AP must satisfy before it may transmit at all, and it is where an indoor deployment gets complicated.

Who is AFC protecting, and where did it come from?

The 6 GHz band has incumbents: licensed users who were there first, chiefly fixed point-to-point microwave links. The regulatory deal that opened the band to Wi-Fi requires that unlicensed devices at higher power provably avoid interfering with them. (Satellite receivers use the band too, but the FCC decided AFC was not needed to protect them and instead capped how much power an outdoor Standard Power AP may radiate skyward.)

AFC is the enforcement mechanism, and it follows a lineage you have already worked with. DFS made 5 GHz APs listen for radar and vacate the channel when they heard it, which put the detection job on the AP itself. CBRS changed the model by adding a central database that devices had to ask before transmitting. AFC is the 6 GHz version of that database model: a cloud service that knows where every registered incumbent is, evaluates each AP's request, and hands back a list of channels and power levels that AP is allowed to use at that location.

The distinction between those two models is worth holding onto, because it explains most of what follows. DFS asked the AP what it could hear. AFC asks the AP where it is standing.

Three spectrum-sharing models left to right. DFS asked the AP to listen for radar and vacate the channel, so detection lived on the AP. CBRS added a central database the device had to ask before transmitting. AFC is the 6 GHz version of that database model, and it asks the AP where it is standing rather than what it can hear.
Each model moved more of the decision off the AP and into a database.

The request and the daily reauthorization

Before transmitting at Standard Power, the AP must send the AFC service its position in three dimensions with two uncertainties, one per plane: a horizontal position carrying a single uncertainty region, modeled as an ellipse or a polygon rather than a separate figure for latitude and longitude, and a height above ground carrying its own vertical uncertainty. The service checks that position against the incumbent database and returns a grant.

The grant is not permanent. The rule requires contact at least once per day, and if a check-in is missed, the AP may keep operating only until 11:59 p.m. the following day before it must stand down. It is a daily calendar obligation with a grace deadline, not a rolling 24-hour clock, and the answer can change between one check-in and the next either way. That is a different operational posture than anything else in your wireless stack, because the thing that decides your channels and your power is not in your building and does not answer to your change process.

What happens when the grant does not come back?

That depends on the hardware, and you need to know which behavior yours has before you build a coverage plan on top of it.

If the AP cannot reach the service, cannot produce a usable location, or fails its check-in, it does not keep transmitting at Standard Power past the grace period the rules allow. Cisco's dual-mode indoor APs drop to LPI, with LPI's lower power and the smaller cells that come with it. An AP built for Standard Power only, like some outdoor models, has no LPI to drop to, and its 6 GHz radio goes quiet instead.

Both of those are the right engineering decision and both are an operational problem, for the same reason. The AP stays up, stays joined, and keeps answering your monitoring. What changed is the power mode, and a coverage plan that only closes at Standard Power has just developed holes that nothing in your alerting is looking for. Confirm which behavior your AP has, and design so your worst-case coverage survives it either way.

This is the part I would weigh most heavily as an operator. Channel and power decisions that used to belong entirely to your RRM now partially belong to an external database you do not control. That is workable, but only if you monitor it, which is what post four is about.

Why is indoor location the hard part?

Because the coordinate AFC cares about most is the one a building is worst at giving you.

Reporting latitude, longitude, and height sounds simple until you try it on the fifth floor of a concrete building. GNSS does not reliably penetrate indoors, and height above ground is exactly the coordinate a microwave link running past your window at rooftop height cares about. A street address gets you two of the three, and the third one is the one that decides whether you are pointed at somebody's link.

On Cisco, latitude and longitude come from GNSS or a supported derivation method, and height above ground and its uncertainty are configured separately from that. An indoor rollout needs a reliable method for both halves, and they do not necessarily come from the same source.

One documented derivation approach is anchoring: a few APs near windows or exterior walls hold a GNSS lock, and the rest of the fleet derives its position from those anchors over a measurement protocol. HPE has described accumulating uncertainty in its own version of this in a Tech Field Day presentation, separate from the roundtable that prompted this series. I have not verified that accumulation, or a specific hop-count-to-uncertainty number, as Cisco or C9800 behavior, so treat the figure below as illustrative of the general anchoring idea rather than a measured result from this controller.

The reason uncertainty matters at all: AFC treats it as risk, so a fuzzy position gets evaluated as if the AP might be anywhere inside the reported radius. It is the difference between giving a delivery driver a street address and giving them a zip code. With the address they go to one door. With the zip code they plan for the whole area, and you get the answer that works everywhere in it rather than the answer that would have worked at your door. A restricted grant on one AP can then push a channel change that ripples through the plan around it.

So location accuracy is the input that decides whether Standard Power was worth turning on in the first place, and it is worth pricing that work, and confirming your vendor's actual mechanism, before you commit to the mode.

Illustrative, vendor-general diagram, not measured on the C9800 bench. A floor plate seen from above. APs near the exterior walls hold a GNSS lock and report a tight position. Interior APs derive their position from those anchors over a measurement protocol, and the uncertainty circle grows with every hop inward. HPE has described this kind of accumulation for its own indoor Standard Power approach; the hop counts and radius shown here are for illustration, not a Cisco-measured figure. AFC sizes its protection to the circle you report.
Illustrative: error accumulates per hop in HPE's documented approach. AFC prices its answer off the number you report, whatever the source.

The notch

On a channel chart, LPI can occupy spectrum right up against restricted frequencies. Standard Power mostly cannot, at least not at full power. AFC evaluates frequency and power together, so near a protected incumbent it may still grant a channel, just at a lower EIRP (equivalent isotropically radiated power, the transmit power a receiver would see) than you designed for. Plotted as the maximum EIRP AFC will grant across the band at a given site, that produces a notch: full power away from the incumbent, falling off as you get closer, and genuinely excluded only inside whatever threshold your design's target power runs into. The diagram below picks one such threshold to illustrate the shape; your actual grant will draw its own curve.

Contiguity is what makes that shape expensive for a wide channel. A traditional 802.11 channel has to be one unbroken block of spectrum, so if the notch drops any part of an 80 or 160 MHz block below the power your design needs, the whole block is unusable at that target, not just the slice AFC actually restricted. Wi-Fi 7 preamble puncturing complicates that all-or-nothing picture in general, since a punctured channel can drop the narrow piece under the notch and keep the rest running. It does not rescue this specific case: Cisco's IOS-XE 17.17 guide excludes static puncturing when Standard Power and AFC are both in use, so on that platform today a notch inside a wide Standard Power channel still costs the whole channel. A 20 or 40 MHz plan needs a much smaller contiguous block to clear the same notch and keeps its reuse intact, which is why narrow plans survive grants that would gut a wide one. This is the spectrum you spend when you buy Standard Power, and post three works through what it does to a real channel plan.

Conceptual diagram at one illustrative EIRP threshold, not a real grant. A channel chart under one AFC exclusion drawn as a notch: below the chosen threshold the frequencies are unsuitable for the design's target power, not necessarily prohibited outright at any power. Sixteen 40 MHz channels lose two at that threshold and keep fourteen. Across exactly the same spectrum, four 160 MHz channels lose two and keep two, because on Cisco's Standard-Power-with-AFC implementation a notch that clips any part of a wide channel removes the whole block at that power target.
Conceptual, at one illustrative EIRP threshold: the same notch, and contiguity is what makes the wide plan expensive.

What does AFC not do?

It does not sense what is actually happening in your room.

AFC protects registered incumbents, chiefly fixed microwave links but also radio astronomy sites, and its model can take terrain and building data as an input to that calculation. What it does not do is observe your building's live RF environment. It does not see the wireless microphone kit and camera rigs that show up in your auditorium on event day, because those are not incumbents and are not in the database. Local interference remains your problem, detected and handled by your own infrastructure, in both modes.

Think of it like a building permit office rather than a site inspection. The permit office checks your plans against the map of what is already recorded, and it is very good at that. Nobody drives out to see what somebody parked on your street this morning. AFC provides modeled incumbent protection. It does not replace local spectrum analysis or your infrastructure's interference management, and the name invites people to assume otherwise.

Bench notes from a Catalyst 9800

Two operational details from my own lab work on a 9800-CL running 17.17.1. First, the virtual controller cannot talk to the AFC service until it is onboarded to Cisco's cloud with a one-time-password token import, and the config guide's prerequisites include DNS, outbound HTTPS, and OCSP reachability from the controller. If you are standing one of these up in a lab and wondering why nothing is happening, work that list before you go looking for RF problems.

Second, the controller checks AFC service health on a timer and reports its state in a readable field. On my bench, in its current not-onboarded state, that check ran about every 30 seconds. Cisco's documentation describes longer intervals for a healthy, onboarded controller, and I have not measured what mine does once it clears onboarding, so take the 30-second number as what a failing check looks like on this box, not as a general polling cadence. My bench unit reports the exact blocker in that field: not-otp-upgraded. That leaf is one case of a YANG choice, and device-not-onboarded is a separate, mutually exclusive case, so you get one blocker at a time rather than a list. That is genuinely useful, because it means the controller will tell you why Standard Power is unavailable instead of leaving you to infer it. Post four covers how to read that over NETCONF instead of a console.

The rest of the series

  1. Choosing a 6 GHz Power Mode: Low Power Indoor or Standard Power.
  2. What AFC Requires From a Standard Power Access Point (this post).
  3. Why I Start at 40 MHz for Enterprise 6 GHz.
  4. Configuring and Monitoring 6 GHz Standard Power on a Catalyst 9800, Programmatically.

← all posts