Designing a WISP Backhaul That Survives Peak Hour

Wireless backhaul is cheap to install and expensive to get wrong. A link that tests at 400 Mbps on a clear afternoon can deliver a quarter of that at 9 pm in July. Here is how to design one that does not embarrass you at peak hour.

Start with the capacity, not the radio#

Work backwards from subscribers:

Subscribers on tower       : 120
Peak concurrency           : ~35%   → 42 active
Average active throughput  : 6 Mbps
Required peak capacity     : ~250 Mbps
Design target (1.6× margin): ~400 Mbps

Never size for the sum of sold plans — nobody has ever built a network that way and stayed solvent. Size for measured peak, then keep 40–60% headroom for growth and for the modulation drops that weather causes.

Link budget, honestly#

TX power            +25 dBm
Antenna gain (A)    +25 dBi
Antenna gain (B)    +25 dBi
Free-space loss     −118 dB   (10 km @ 5.8 GHz)
Cable/connector     −1 dB
--------------------------------
Expected RSSI       −44 dBm

Then subtract reality: rain fade, foliage growth over a season, and the mast that goes up next year in your Fresnel zone. If your calculated figure has less than 15 dB of fade margin above the sensitivity for your target modulation, the link is not designed — it is hoped for.

Fresnel clearance is not optional#

The first Fresnel zone radius at midpoint:

r ≈ 17.3 × √( d / (4 × f) )      d in km, f in GHz
10 km @ 5.8 GHz  →  r ≈ 11.4 m

You need at least 60% of that clear. A treeline that is "just below line of sight" today is inside the zone, and it grows.

Channel planning on a tower#

• Survey before you commit: /interface wireless scan and a spectral scan on every candidate channel, at the hour you care about (evening, not lunchtime).
• Keep backhaul and access on different bands where you can — 5 GHz access with 6 GHz or licensed 11/18 GHz backhaul removes an entire class of self-interference.
• Wider is not always faster. On a noisy tower, 40 MHz at high modulation beats 80 MHz that keeps dropping to MCS 3.
• Vertical separation and cross-polarisation between links on the same mast buy real isolation for free.

RouterOS configuration notes#

For MikroTik point-to-point links, nv2 beats plain 802.11 for long distances: TDMA scheduling removes hidden-node collisions and stabilises latency.

/interface wireless
set [find default-name=wlan1] mode=bridge band=5ghz-onlyac \
    channel-width=40mhz-XX frequency=5780 wireless-protocol=nv2 \
    nv2-security=enabled nv2-cell-radius=12 distance=indoors ssid=BH-SITE-A

Set nv2-cell-radius to slightly more than the real link distance in km. Leaving it at default on a long link quietly costs throughput.

Monitoring that catches problems early#

Poll and graph, per link: RSSI both ends, SNR, CCQ or modulation rate, TX/RX capacity, and retransmits. The signature to watch for is a slow SNR decline over weeks — that is foliage or a drifting mount, and it is fixable before it becomes an outage. See the ISP monitoring stack for the tooling.

When to stop and lay fibre#

Wireless backhaul stops making sense when any of these is true:

• You need more than about 1 Gbps sustained on the hop.
• The path crosses a corridor you cannot control (new construction, growing trees).
• Downtime on the link would take out more than a couple of hundred subscribers.
• You are already spending on licensed spectrum and taller masts to keep it alive.

At that point the fibre trench that looked expensive is the cheaper answer over three years.