Why Does Wi-Fi Show Full Bars in a Villa but Stall When You Go Upstairs? How to Choose Between Mesh, AP + AC, and Fiber
Why does a call freeze upstairs despite a strong Wi-Fi signal? Compare wired mesh, managed APs and fiber through roaming, directional bandwidth and acceptance tests.
Imagine this: you walk from the first-floor living room to a second-floor bedroom with your phone, and your video call suddenly freezes. The Wi-Fi icon is still there, and the signal even looks good. Once you stop and run a speed test, the result looks excellent again.
Someone suggests buying a more expensive Mesh system. Someone else says you must use enterprise-grade APs and an AC. Another person thinks a five-story villa should go straight to 10-gigabit fiber.
Each suggestion might solve part of the problem. You might also spend the money and still have your call drop at the stairs.
That is because at least three different things are involved: whether your phone can connect well to a nearby AP, whether it can switch to the next AP in time, and whether the network behind the AP can carry its data onward. One speed-test screenshot can hardly prove that all three are working.
This article uses a hypothetical unfinished villa with two basement levels, three above-ground floors, and about 200 m² per floor to separate those questions. It is an architectural exercise, not a completed installation: without floor plans, a wireless survey, or roaming records from actual devices, it cannot promise that “ten APs will definitely cover the whole house.”
1. Before Choosing Mesh or an AC, Ask How the Data Gets Back
An AP is a wireless access point that lets phones and other clients join the network. An AC is a wireless controller responsible for tasks such as centralized configuration and management; depending on the product, it may be dedicated hardware, software, or cloud-based management. Mesh products may bundle node connectivity, management, and roaming assistance.
The point is that “Mesh” and “AP + AC” are not entirely equivalent, mutually exclusive technical choices.
Mesh can use wireless or wired backhaul. TP-Link’s official Deco documentation explicitly supports connecting nodes with Ethernet cables and distinguishes the wiring rules for router mode and AP mode.[1] Likewise, some centrally managed APs support wireless Mesh uplinks.
The combinations worth comparing are these three.
| Combination | Where it helps | What it costs |
|---|---|---|
| Wireless-backhaul Mesh | Flexible deployment when running cables is difficult | Backhaul is also affected by walls, distance, and wireless contention |
| Wired-backhaul Mesh | Avoids wireless backhaul between floors; home-oriented management is usually simpler | VLAN, logging, authentication, and tuning capabilities depend on the model |
| Wired APs + centralized management | Makes centralized planning, segmentation, and troubleshooting easier | More configuration, compatibility validation, and maintenance work |
For this five-story home, where cables can still be installed, I would decide on wired backhaul first. Whether the equipment is called Mesh or AP + AC would then depend on management, power, logging, and maintenance requirements.
If the goal is simply one household Wi-Fi network with straightforward setup, and a specific product’s wired backhaul, roaming, and coverage have been validated, wired Mesh is a viable candidate. If the system must also manage APs across several floors, cameras, a guest network, and smart-home devices, and you want to investigate why a particular handoff failed, centrally managed APs generally offer more scope for engineering configuration. You still need to check the actual models; “enterprise-grade” is not an acceptance report.
2. Full Bars Describe Only One Link; There Is Another Link Beyond the Floor Slab
Your phone is connected to a nearby Mesh node, so it shows full bars. But that node has to send the data wirelessly to another node through a reinforced-concrete floor slab. Your phone’s signal icon does not show the quality of that backhaul link.
Even if both links have good signal, how they share wireless resources matters.
Consider a deliberately simplified example: suppose the same data must traverse two hops in sequence using the same wireless resource. The first hop serves data at 400 Mbps and the second at 200 Mbps, and they cannot transmit simultaneously. Ignoring additional overhead, sending 1 Mb takes at least:
1 / 400 + 1 / 200 = 0.0075 seconds
Ideal throughput upper bound = 1 / 0.0075 ≈ 133.3 Mbps
That is neither 400 Mbps nor simply the slower hop’s 200 Mbps. The example illustrates how forwarding the same data again consumes airtime; it is not a measured speed for any Mesh product.
A dedicated backhaul band, multiple radios, or spatial reuse could change those assumptions. Retransmissions, contention, and management overhead could reduce real-world performance further. So “every additional hop always halves the speed” is also too absolute. Check the actual backhaul method and path, then measure them.
In an unfinished house, the main value of Ethernet is removing the between-floor link from the set of wireless variables. It will not eliminate interference on the phone’s wireless link, but it removes one layer of uncertainty.
3. If There Is an AC, Why Does the Phone Keep Holding On to the Downstairs AP?
The client usually decides when to initiate roaming. Apple’s deployment documentation explicitly says that a device makes roaming decisions based on factors including received signal and candidate APs. The network can supply assistance, but it cannot promise identical switching behavior from every phone.[2]
Three common mechanisms can be understood this way:
- 802.11k: Tells the phone which nearby APs are candidates, reducing the work needed to find them.
- 802.11v: Provides network-side suggestions to help select a candidate target.
- 802.11r: Optimizes authentication during a fast transition, reducing the associated overhead.
Apple and Cisco Meraki describe these mechanisms from the client and access-network perspectives, respectively.[2][3] They can improve the relevant processes, but they cannot replace coverage design or guarantee that every client supports the same features.
If there is no suitable candidate AP by the bedroom door, faster authentication has nothing to work with. If the phone still considers the downstairs AP usable, providing a neighbor list does not mean it will switch immediately. Older clients and different authentication configurations may also have compatibility differences; turning on every “fast roaming” option is not the end of troubleshooting.
Another approach is to forcibly disconnect clients with weak signals. That may prompt a client to choose a new AP, or it may simply take the client offline when no suitable alternative exists. Forcing a disconnect is a control mechanism, not proof of a seamless transition.
I would therefore check for reasonable coverage overlap along the walking route first, then channels and power, and finally the roaming configuration. More neighboring APs and higher power are not necessarily better: excessive overlap can increase contention, and an AP’s ability to transmit farther does not mean a phone can transmit back equally well.
4. If the Wi-Fi Name Is the Same, Why Can a Session Still Drop After a Switch?
Using the same SSID, password, and authentication method is a common baseline, but the same Wi-Fi name does not mean the entire network is the same.
Suppose each floor has an independent router doing its own NAT and DHCP, and all their wireless networks are given the same name. When the phone moves to another access point, its underlying address and route out may change. It still appears to be connected to Wi-Fi with the same name, but the original session may not survive.
For this villa’s everyday Wi-Fi, I would connect the APs on all floors to a consistent client network with a common gateway and DHCP service, avoiding separate routing on each floor. If Layer 3 roaming across IP subnets is required, use an architecture that explicitly supports it rather than merely copying the SSID. Layer 3 means the network layer, not three floors of the house.
It is also worth asking where the AC sits: does it only manage devices, or must user traffic pass through it? Local forwarding, centralized tunnel forwarding, and an all-in-one gateway have different failure implications. Before buying, ask what happens when the controller goes offline: can connected clients still communicate, can new clients join, and which authentication and roaming capabilities change?
“Internet access still works when the cloud is down” and “Internet access still works when the controller fails” are not the same test. The product’s behavior needs to be established from its documentation and a pilot setup. A box labeled AC should not lead you to assume either that all traffic must pass through it or that its failure cannot affect service.
5. Should the Floors Be Connected by Fiber? Draw the Traffic Directions First
“Five floors and 1,000 m²” describes space, not bandwidth demand.
Suppose each floor has four cameras recording at an assumed 8 Mbps each, with the recorder in the main equipment cabinet. Four clients on the same floor each receive a 25 Mbps video stream. All these figures are hypothetical inputs, not default bitrates for any product.
For the full-duplex uplink between that floor and the main cabinet:
| Traffic | Floor to main cabinet | Main cabinet to floor |
|---|---|---|
| Four camera feeds sent for recording | 32 Mbps | Small amount of control traffic excluded |
| Four downstream video streams | Small requests and acknowledgments excluded | 100 Mbps |
| One computer backing up to a NAS in the main cabinet, requesting an additional 960 Mbps | 960 Mbps | Small amount of acknowledgment traffic excluded |
Without the backup, demand is about 32 Mbps in one direction and 100 Mbps in the other. The two directions of a full-duplex link are counted separately; you cannot simply add them into 132 Mbps and treat that as utilization in one direction.
With the NAS backup, demand from the floor rises to 992 Mbps. If we assume a 940 Mbps budget for application traffic on this gigabit link in the current environment, it cannot satisfy all those requests simultaneously. Subtracting the recording traffic leaves a theoretical application-traffic budget of 908 Mbps for the backup, or 113.5 MB/s; protocol behavior, contention, and disk performance could make the backup slower still. The 940 Mbps figure is a budget chosen for this example, not a fixed measured result for every gigabit link.
An upgrade to 2.5G or 10G might then make sense, but the NAS port, disks, computer port, and upstream aggregation link also need checking. Replacing only the floor uplink’s optical modules with 10-gigabit ones will not give a single computer a 10-gigabit backup path to a NAS still connected through a gigabit port.
Another easily missed detail: count only traffic that actually crosses the link. If the NVR and cameras are on the same floor switch, and their forwarding path really remains local, their recording streams will not cross the backbone merely because those devices exist in the house. Conversely, if recordings from all five floors converge on the main cabinet, the NVR’s interface faces a combined 160 Mbps of that input, plus other relevant traffic.
Draw the locations of the cameras, NVR, NAS, and frequently used clients first. Then calculate concurrent traffic in each direction on each link. Buying the most expensive backbone before identifying the bottleneck reverses that order.
6. Fiber Is Worth Provisioning For, but It Is Not “Faster Wi-Fi”
If client Ethernet runs from every floor can return to the main cabinet along compliant routes, centralized switching may be easier to manage: power, UPS, patching, and maintenance remain in one place. Conventional structured copper cabling is typically designed for a permanent link no longer than 90 meters and a channel including patch cords no longer than 100 meters. The cable category, target speed, and installation conditions must also match.[4]
The vertical height of five floors is not the length of a cable run. Routing around rooms, using risers, entering patch panels, and leaving maintenance slack all count toward the actual path. You cannot just multiply floor height by five, nor assume copper will be too long simply because there are five floors.
When there are many outlets, crowded conduits, or unsuitable route lengths, a floor switch with a star-topology fiber uplink becomes more attractive: the main cabinet connects separately to each floor, so an access switch on one floor is not on the required path to the next.
During construction, it may be worth evaluating a provision for multi-core OS2 single-mode fiber, such as one six-core cable to each floor’s equipment enclosure. A conventional duplex link would use two cores, leaving the rest for maintenance and expansion. That is a candidate provision, not a quantity every villa must copy. The final choice must account for termination, modules, link budget, installation requirements, and maintainability.
Pay particular attention to three things:
Spare cores are not a spare route. If all six cores are in the same cable and conduit, one cut may take them all out.
Fiber does not power equipment on the floors. Floor switches need power, cooling, and service access. If they supply PoE to cameras, any requirement to keep recording during a power outage must cover those switches too. A UPS only in the main cabinet cannot keep a floor switch running after it loses power.
Fiber and 10-gigabit speed are not synonyms. Link speed depends on the ports at both ends, the modules, and their compatibility. A fiber backbone will not automatically increase a phone’s wireless-link speed or fix a bad roam.
The better priorities during construction are replaceable conduits, suitable routes, reliable terminations, labeling, and test records. Fiber loss, polarity, and end-face condition, along with copper-link performance for its specified category, should be part of acceptance. “The link light is on and the Internet works” proves only a small part of it.
7. How I Would Draw the First Network Design for This Villa
Without floor plans, I would draw an architecture with explicit conditions, not a shopping list that only looks precise.
First, I would provision two primary candidate AP locations and one spare location per floor so placement options can be compared. The primary positions should be near actual areas of use; stairs and corridors should be checked as continuous paths between floors and rooms. A spare location does not mean a third AP must be installed. An open-plan 200 m² floor and a 200 m² floor divided into many rooms call for different coverage designs.
Second, I would prioritize wired backhaul. If cable length and capacity permit, I would run connections back to the main cabinet. Where a floor access switch is needed, I would use a star backbone and include floor-level power in the failure analysis. The main router and core switch may still be shared points of failure; a star topology does not automatically create a high-availability system.
Third, I would select products according to maintenance needs. If simple home access is all that is required, wired Mesh remains a candidate. If more network segmentation, logging, and parameter control are needed, I would prioritize evaluating centrally managed APs. I would also check local management, operation during outages, configuration export, and replacement and recovery procedures.
Fourth, I would provide appropriate isolation for everyday clients, guests, and cameras without treating “more VLANs” as a mark of professionalism. Matter, casting, and device discovery may require communication across network segments. The allowed paths between controllers, border routers, and clients must be explicit; otherwise, wireless handoffs may work while the phone cannot find devices in the house.
Only then would I decide uplink speeds. Whether gigabit is enough depends on concurrent traffic in the same direction and expected growth. Whether 2.5G or 10G is worthwhile depends on whether the complete path can use it. Wi-Fi 6 and Wi-Fi 7 are parts of equipment selection, not substitutes for this groundwork.
8. Do Not Just Run Speed Tests at Acceptance; Walk Around With the Phone
“A speed test works in every room” checks a set of stationary positions. “Does the call continue as you walk from the first to the third floor?” checks something else.
I would divide acceptance into three rounds, each addressing a distinct set of questions.
Round one: verify the wired foundation. From a wired client at the floor switch, transfer data to a test endpoint on the local network. Check throughput in both directions, error counters, and latency under load. The endpoint’s own port and performance must be sufficient, so a server bottleneck is not mistaken for a floor-network problem.
Round two: verify stationary wireless experience. At commonly used positions in bedrooms, studies, stair turns, and the basement, record the AP’s BSSID, band, channel, received signal, retries, and application behavior. Use local-network tests to distinguish wireless issues from ISP issues; do not judge an AP solely by Internet speed tests.
Round three: verify walking and failures. Using the phone models people actually use, walk fixed routes in both directions at similar speeds. Make calls and run local probes both when the network is otherwise idle and when background traffic is present. Record BSSIDs before and after each switch, authentication or association events, application interruption, and recovery. Stopping an AP, stopping the controller, and disconnecting the Internet should be separate tests, not one combined “network outage test.”
“We lost two pings, so roaming took 200 milliseconds” is not rigorous. With probes 100 milliseconds apart, they provide only a coarse indication of application reachability; congestion, a device’s handling of ICMP, and scheduling also affect the result. Measuring the wireless transition itself requires client and AP logs or packet captures with aligned time references. Measuring user experience also requires retaining the observed audio and video behavior. The two are related, but they are not identical.
The following table can hold raw records instead of leaving only a “good signal” heatmap.
| Record | Why it is needed |
|---|---|
| Phone model, OS, AP firmware, and configuration version | Switching behavior may differ between clients |
| Route, direction, walking pattern, and background load | Makes before-and-after comparisons meaningful |
| BSSID, band, and channel before and after a switch | Confirms whether the expected switch actually happened |
| Local probe interval, packet loss, and time reference | Avoids mistaking measurement resolution for actual duration |
| Audio interruption, video freeze, and session re-establishment | Records the actual application impact |
| Failure count, reproduction conditions, and adjustments | Prevents the report from retaining only one successful run |
Acceptance thresholds should follow the target applications and be agreed with the installer in advance. Do not turn the best result from one phone on one route into a claim of “zero interruption” for every client throughout the house. Failed runs must not be removed from the report, and the same route must be tested again after configuration changes.
9. The Parts Omitted From a Quote Often Determine the Experience
Returning to the original question: for this unfinished five-story villa, I would prioritize wired backhaul and centralized management. Both wired Mesh and AP + AC are worth evaluating; the latter better suits a design that needs detailed segmentation, logging, and ongoing maintenance. I would plan fiber provisions around backbone and expansion needs, then decide whether to activate 10-gigabit links immediately based on actual traffic paths and bandwidth budgets.
But I would be more concerned with whether the quote explains four things: why each AP is placed where it is, which same-direction traffic each floor uplink carries, what fails when a particular device breaks, and how calls will be tested while walking.
Without those answers, ten APs may just be ten boxes, six-core fiber may just be six spare cores, and “seamless roaming” may be no more than a switch in a management interface.
Buying equipment that covers the whole house is not the same as getting a network that stays stable as you walk through it. The latter requires coverage, backhaul, client transitions, and failure recovery to be designed together and checked along real routes of use.
Further reading: Wiring and Capacity Budget for a Five-Story Villa, Failure Acceptance Testing After the Central Controller Shuts Down. This article adds Wi-Fi architecture and roaming considerations; it does not repeat the earlier PoE and recording-storage budgets.
Sources checked: 2026-11-16. The bandwidth and two-hop service-time examples were recalculated with the accompanying Python script. This was an arithmetic check using manually supplied inputs, not a wireless simulation, hardware benchmark, or on-site roaming test.
- TP-Link: Deco Ethernet Backhaul, used to verify wired Mesh support and the effect of operating mode on wiring; this product’s behavior is not generalized to every Mesh system.
- Apple: Wi-Fi roaming support in Apple devices, used to verify the client’s role in roaming decisions and the roles of 802.11k/r/v; thresholds for a class of Apple devices are not treated as a universal standard for all phones.
- Cisco Meraki: 802.11k and 802.11r Overview, used to cross-check roaming optimization mechanisms and configuration compatibility limits.
- Fluke Networks: Channel, Permanent Link, Patch Cords, MPTL, E2E… Oh My!, used to verify the distinctions between a copper channel, a permanent link, and the objects being tested.
Companion downloads: Bandwidth calculation script, Results of these calculations. The Python script uses only the standard library; its results are arithmetic checks, not network measurements.
AI-generated scene illustration, not a photograph of this project or a wiring instruction.
Discussion
Share a question or your experience. Comments appear after moderation. Your email is never displayed publicly. Comment privacy
Loading comments…