Do I Need Mesh Wi-Fi: What to Test Before You Buy

The nodes arrive, they go in the hallway and the back bedroom, the app turns every room green, and the 10 a.m. call still falls apart at the same point in the same sentence. That outcome is common enough to be predictable, and the reason is narrow: a mesh system solves one specific problem very well, and most of the complaints that send people shopping for one are a different problem wearing the same clothes.

The problem mesh solves is coverage — the signal from a single access point not reaching far enough through the particular walls of your particular house. The Wi-Fi Alliance says as much in its own EasyMesh technology overview, noting that "structure surroundings, such as the type of walls, floors, and wiring, can affect Wi-Fi coverage" and that single-AP setups are "frequently insufficient to blanket the space." That is a real condition with a real fix.

It is not the same condition as a line that delivers 40 Mbps on a 300 Mbps plan, or one that loses packets every weekday at seven in the evening, or one whose latency triples the moment anything uploads. Nodes do nothing for those. Worse, they make the diagnosis harder afterwards, because you have added two more radios and two more places for a fault to hide, and because the app now reports a green dot that is measuring the wrong leg of the path.

So the useful question is not which system to buy. It is which of those two conditions you actually have, and that takes about forty minutes to settle.

The one comparison that proves nothing, and why everyone makes it first

The instinct is to walk around with a phone, run a speed test in the bad room, run another in the good room, and compare. Almost everyone starts here. It is close to useless, for a reason that has nothing to do with the phone.

Those two readings were taken at different times. Between them, the shared segment your line sits on changed load, the test picked a possibly different server, and the device may have switched bands. A 90 Mbps result in the kitchen and a 240 Mbps result in the office tell you the two measurements differ. They do not tell you the kitchen is the cause, and they cannot, because nothing in the procedure held the line constant.

Any comparison that is going to support a purchase has to isolate the wireless leg. That means one reading taken with the wireless leg removed entirely — a laptop on an Ethernet cable in a LAN port on the gateway, with its own Wi-Fi radio switched off. The twenty-minute wired diagnosis is the full version of that procedure, and it is the prerequisite for every paragraph below. If the wired control is bad, stop. Nothing in a box will help, and the evidence you need is a logging record rather than hardware.

Assume from here that the wired control is clean.

A node without a cable spends its airtime twice

Here is the arithmetic that decides whether mesh will feel like an upgrade or a lateral move.

Wi-Fi is a shared, half-duplex medium: on a given channel, one transmission at a time. A node that has no Ethernet cable behind it has to receive your laptop's packet and then retransmit it to the router, on radios it is also using to serve clients. Every byte crosses the air at least twice. Whatever throughput that node can give the laptop sitting next to it is therefore some fraction of what the node itself can reach the router with — and the node is usually further from the router than you are.

The Wi-Fi Alliance's own vocabulary makes the structure visible. Your device-to-node link is the fronthaul; node-to-node is the backhaul; and in the EasyMesh overview the Alliance notes that "fronthaul and backhaul links may use separate radios for better performance, depending on the capabilities offered by each AP." Separate radios are what a tri-band system is selling. They do not remove the double transmission, they just stop it from colliding with your traffic on the same channel.

The version with no compromise is a cable. NETGEAR's support article on Ethernet backhaul for Orbi documents three topologies — star, daisy chain, and everything into a switch — and that is the whole trick: a node with a cable behind it is not a repeater at all. It is a second access point, and it gives clients its full radio.

Two things in that article are worth reading closely before you buy anything. One is the order of operations: NETGEAR puts "make sure that you update the firmware on all your Orbi devices" ahead of any of the wiring steps. The other is a sentence that has cost people real money: "The Ethernet ports on your Orbi satellites and Orbi router are rated at 1 Gbps." If you are paying for a multi-gigabit line, a 1 Gbps backhaul port is your new ceiling on every node behind it, and the box will not mention that on the front. Check the port rating per model, not per brand.

If the walls are still open, the cable is cheap and the decision is easy — that is the entire argument in what to run before the drywall closes. If the walls are closed, one cable to one good central location, even surface-run along a baseboard, converts the single most important node from a repeater into an access point.

Steering is a suggestion, and your laptop is allowed to ignore it

The second common disappointment is the device that clings. You walk from the office to the kitchen, the kitchen node is four metres away, and the laptop stays attached to the office node at one bar, dragging the call down with it.

This is not a defect in the mesh system, and no amount of restarting fixes it. It is how the protocol allocates authority. The Wi-Fi Alliance is unusually direct about it: a controller "may choose to send control messages to 'steer,' or suggest, a client move its connection from one AP to another," and those messages work through client-side features — the Alliance names IEEE 802.11v BSS Transition Management, part of its Agile Multiband programme — "so clients will likely experience faster transitions when being steered."

Suggest. Likely. The roaming decision belongs to the client. A laptop, phone or smart plug that does not implement BSS Transition Management gets a suggestion it cannot act on and does what its own driver decides instead, which is often to hold the existing association until it fails outright. Older devices and cheap IoT radios are the usual offenders, and no node you buy will change their firmware.

Two practical consequences. First, if your complaint is specifically "my laptop won't let go," test that laptop before spending, because the variable may be in your hand. Second, mixing vendors is more realistic than it used to be — EasyMesh exists precisely so that APs "from a variety of vendors" can join one network, with onboarding by push button, Wi-Fi Easy Connect, or simply plugging the AP into an Ethernet port — but the certification governs the APs, not your clients.

The 6 GHz band on the box is legally the quiet one

Every current mesh system advertises 6 GHz, and it is genuinely useful: 5.925 to 7.125 GHz is 1,200 megahertz of spectrum with none of the microwave-oven and Bluetooth company that 2.4 GHz keeps. But the power rules are not the ones the marketing implies, and they are written in a place you can read.

I pulled 47 CFR 15.407 from the eCFR versioner API on 2 October 2026 (current through the 30 September 2026 issue) expecting the indoor 6 GHz limit to be a single total-power figure. It is not. Paragraph (a)(5) sets it per megahertz: for an indoor access point in the 5.925–7.125 GHz band, "the maximum power spectral density must not exceed 5 dBm e.i.r.p. in any 1-megahertz band," with total e.i.r.p. across the band capped at 30 dBm. Paragraph (a)(6) applies the same two numbers to subordinate devices under that AP's control.

Compare the standard-power class in paragraph (a)(4) — the devices that have to clear their frequencies with an Automated Frequency Coordination system before transmitting, per paragraph (k)(1) — at 23 dBm per megahertz and 36 dBm total. Then compare the bottom of 5 GHz, where paragraph (a)(1)(ii) gives the indoor limit for 5.15–5.25 GHz as conducted power: 17 dBm per megahertz, provided antenna gain stays within 6 dBi.

Those last two figures are in different units, so convert before drawing a conclusion. Add the 6 dBi the rule permits and the 5 GHz number becomes roughly 23 dBm e.i.r.p. per megahertz, against 5 dBm e.i.r.p. for 6 GHz indoors — about 18 dB of difference, measured the same way over the same 1 megahertz. Note that the limit is per band and not per "5 GHz": the DFS bands at 5.25–5.35 and 5.47–5.725 GHz are tighter at 11 dBm per megahertz, and 5.725–5.850 GHz is more generous still.

Three things follow, all of them useful at the shop.

• 6 GHz is a short-range band by regulation, not by accident. It is the band for the node and the laptop in the same room, not for reaching the far bedroom.
• Wider channels do not buy reach. At 5 dBm per megahertz, a 160-megahertz channel adds up to roughly 27 dBm and a 320-megahertz channel to roughly 30 — the cap. Past that width you are buying bandwidth, and spending it on a shorter radius.
• Indoor means indoor as a matter of law. Paragraph (d)(3) limits devices operating under (a)(5) and (a)(6) to indoor locations, and (d)(4) requires the device and manual to carry the statement "FCC regulations restrict operation of this device to indoor use only," along with the prohibition on oil platforms, cars, trains, boats and aircraft. The detached garage and the deck are not 6 GHz problems you can solve by moving a node outside.

Dedicated 6 GHz backhaul in a tri-band system is still a good design. Just place it on that basis: node-to-node over 6 GHz wants line of sight and one wall, not three.

Thirty minutes off the air: the rule behind drops nobody can reproduce

This is the paragraph that explains the strangest category of complaint — connections that fail on something resembling a schedule, for roughly half an hour, with no pattern in the household's own usage.

Section 15.407(h) requires any U-NII device with any part of its 26 dB emission bandwidth in 5.25–5.35 GHz or 5.47–5.725 GHz to run Dynamic Frequency Selection: it must check a channel for 60 seconds before using it, and once radar is detected, "all transmissions shall cease on the operating channel within 10 seconds," with normal traffic allowed for at most 200 milliseconds after detection. Then comes the part that produces the symptom: a channel flagged as containing a radar system "is subject to a non-occupancy period of at least 30 minutes," and the clock "starts at the time when the radar system is detected." The same paragraph also calls for transmit power control in those bands — the capability to operate at least 6 dB below the mean e.i.r.p. value of 30 dBm — while exempting systems that run below 500 mW.

Why this lands on mesh owners specifically: the radar bands sit in the middle of 5 GHz, and what is left on either side is narrow. Below them is 5.15–5.25 GHz, 100 megahertz wide. Above them is 5.725–5.850 GHz, another 125. Neither reaches 160 megahertz on its own, and they are not adjacent — the DFS spectrum is exactly what separates them. So on the channel lists mesh systems actually offer, a 160-megahertz channel has to borrow from 5.25–5.35 or 5.47–5.725 GHz, and it inherits the radar rules when it does. A system set to wide channels for a fat wireless backhaul is, in practice, a system sitting on DFS spectrum — and within range of airport, naval or weather radar it will vacate, quickly and silently, taking the backhaul with it.

One caveat, so you do not over-read that as a law of physics: paragraph (a)(3)(ii) also admits indoor access points into 5.850–5.895 GHz, and it expressly contemplates "indoor access points operating on a channel that spans the 5.725-5.850 GHz and 5.850-5.895 GHz bands." That stretches the upper non-radar run to 170 megahertz, which means a 160-megahertz channel clear of DFS is not forbidden — it is just not what consumer mesh hardware is shipping. Read your own node's channel list rather than assuming it either way.

What it looks like from the sofa: everything stops for a few seconds, recovers somewhere worse, and behaves oddly for the next half hour. Reboots appear to help because half an hour passes while you are rebooting.

How to tell: open the node's own radio settings and write down the backhaul channel and width, then check it again after the next incident. A changed channel number, or a width that silently dropped from 160 to 80, is the fingerprint. If that is your fault, the fix is configuration and costs nothing — pin the backhaul to a non-radar channel and accept the narrower width, or move it to a cable.

Forty minutes, three numbers, one decision

Do this before ordering. Every step is free.

1. The wired control. Laptop into a gateway LAN port, laptop Wi-Fi off. Record download, upload, idle latency, and latency while the link is saturated. This is your line's ceiling; nothing wireless improves on it. The repeatable method is in the piece on speed tests that mean something.
2. A hundred pings to your own gateway from the bad room, over Wi-Fi. ping -n 100 192.168.1.1 on Windows, -c 100 elsewhere. Nothing in that path belongs to your provider, so any loss, or a maximum twenty times the average, is a house finding — and a house finding is what mesh is for.
3. The same hundred pings from the bad room, over the cable. If the wired run is clean and the Wi-Fi run is ugly, you have confirmed a coverage problem and nodes are the correct purchase. If both are ugly, the fault is at or past the gateway, you need the loss, jitter and latency numbers that actually break calls written down over several days, and the next conversation is with your provider rather than a retailer.

Then one more reading, in the room where you would put the first node: signal strength and the connected band, from the laptop's own Wi-Fi details. A room that already shows a strong 5 GHz association does not need a node. It may need the existing access point moved out of the media cabinet, which costs nothing and is the single most underrated fix in this whole category.

If step 3 says coverage, the shopping rule is short. Count how many nodes you can reach with an Ethernet cable, including an ugly surface-run one, and buy a system whose non-primary units have Ethernet ports rated at least as fast as your plan. Prefer tri-band if no cable is possible. Treat 6 GHz as the in-room band and the node-to-node band, never the through-three-walls band. And check whether the box supports Wi-Fi EasyMesh if you ever intend to add a unit from a different vendor.

One last caution about what mesh does to your paperwork. Once nodes are in place, a provider's support line will have a new and permanent explanation for everything: it's your mesh. That is not always wrong, which is what makes it effective. Keep the wired control reading from step 1, dated, in the same file as the rest of your line history — the one you would attach to a bill dispute or a complaint — and add a fresh wired reading the week after installation. Two timestamped wired numbers, taken either side of the nodes arriving, are the only thing that keeps a line fault from becoming your hardware's fault for the rest of the contract. And keep the receipt until you have run the three measurements again in the rooms that prompted the purchase, because the return window is usually shorter than the time it takes a new symptom to appear.