Why Millimeter Waves Don't Need a 'Clear View' Like a Telescope: The Keyhole Analogy

You've heard it a hundred times: millimeter waves need a clear path. Like a telescope aimed at the sky. But that's not how they work. Think of a keyhole instead. You peek through, see a sliver of the room, but sounds and smells reach you from all around. MmWave is more like that. It bounces, bends, sneaks through gaps. It's not a straight-line-only affair.

The myth of pure line-of-sight (LOS) comes from early textbooks and simplistic simulations. Real-world 5G networks show something else entirely. Phones connect behind trees, inside cars, even in elevators. How? The physics hasn't changed, but our understanding has. This article walks through the field context, common confusions, what works, what doesn't, and when to ditch the telescope analogy altogether.

Field Context: Where the Keyhole Analogy Actually Matters

5G Urban Deployment Scenarios

Walk any dense city block—Hong Kong, midtown Manhattan, central London—and the geometry of millimeter wave becomes visible in a way propagation charts never capture. Street canyons bounce signals off glass facades. Cross-street reflections fill the shadow behind a delivery truck. The keyhole analogy matters here because a direct line-of-sight between base station and handset is rare. What works instead is a scattered path: a beam that hits a building across the street, caroms off a metal awning, then drops into a phone held at waist height. The odd part is—these reflected paths often outlast a clean LOS shot when the user turns a corner. I have watched field tests where a tight beam aimed at a pedestrian crossing held a gigabit link through three successive bounces. No direct view. Just geometry and metal cladding.

Not every reflection survives. Glass is lossy. Brick absorbs. The trick is mapping which surfaces act as mirrors and which act as sand. That changes block by block.

Industrial IoT and Factory Floors

A factory floor looks like an antenna killer. Racks of metal shelving. Forklifts drifting through aisles. Welding arcs that generate broadband noise. Yet millimeter wave thrives here—not because of clear sightlines, but because industrial environments are cluttered with repeatable reflective surfaces. Conveyor belts, steel beams, polished concrete floors—all become passive reflectors. We fixed this by placing small-cell nodes not for direct illumination, but to spray energy at known metal obstructions. The catch is that a single forklift repositioning a pallet of steel brackets can shift a reflection path by ten degrees. One morning the link works; by lunch it drops to a quarter of the throughput. Most teams who deploy mmWave on a factory floor assume they need a clear shot across the bay. Wrong order. They need three or four alternative bounce paths, and a scheduler that picks the best one every few milliseconds. That's not a telescope problem. It's a mirror gallery problem.

Fixed Wireless Access in Suburban Homes

Suburban deployment looks easier—trees, houses, some foliage—but the keyhole analogy bites hardest here. A customer mounts a receiver on their roof, paying for a service sold as 'fiber speed without the trench.' The installer points the node at the nearest macro tower. Clear view, right? Then spring arrives. Leaves emerge. The signal degrades. What usually breaks first is not the direct path—foliage attenuation at 28 GHz is brutal—but the secondary paths. A signal that had been skipping off a neighbor's chimney, reflecting off a backyard shed, then arriving at the receiver at a glancing angle. That path vanished when the tree between the shed and the house leafed out.

'We sold them a clear view and delivered a seasonal reflection that lasted eight months.'

— comment from a fixed wireless operator after their first suburban rollout

The solution is not taller poles. It's mapping the dozens of secondary paths that exist in winter, then engineering fallbacks for the other three seasons. Some operators now pre-install small passive reflectors—flat metal panels on chimneys or fence lines—to deliberately create non-line-of-sight routes. That sounds like extra cost. It's. But losing a hundred subscribers every April costs more. The keyhole analogy forces this question: are you building for the one moment when the path is clear, or for the thousand moments when it's not?

Foundations Readers Confuse: LOS vs. Effective Propagation

Fresnel zones and blockage

The common belief is simple: if you can see the tower from your phone, millimeter waves will work. That's wrong. Millimeter wave propagation doesn't care much about your line of sight in the classical sense — it cares about the Fresnel zone. This is the invisible ellipsoid between transmitter and receiver, shaped like a bloated rugby ball. Even if the direct path is clear, anything intruding into that volume — a wet branch, a passing truck, a person standing up — can kill the signal. I have watched teams spend hours aligning rooftop antennas, only to fail because a ventilation duct sat two meters off the boresight, inside the zone. The signal dropped 18 dB. Not because the duct blocked the line of sight. Because it polluted the Fresnel zone. That distinction matters.

The catch? Most network planners ignore it.

Fresnel clearance at 28 GHz demands a radius of roughly 0.3 meters at 100 meters distance. That's not large. But it's large enough that a single leafy branch becomes a problem. Tree canopies are worse — not because the trunk blocks the beam, but because leaves create hundreds of tiny scatterers that diffract energy away. The practical fix is not to cut down every tree. It's to model the Fresnel zone as a dynamic volume, not a straight line. We fixed one deployment by raising the antenna height by 1.2 meters. The budget hurt. The link budget didn't.

Rain fade and atmospheric absorption

Rain fade is the most cited bogeyman in mmWave. And yes, at 60 GHz, oxygen absorption is real — that band is practically a wall at 500 meters. But at 28 GHz, typical rain fade adds only 1–2 dB per kilometer. That's manageable. The real issue teams conflate: rain creates a temporal problem, not a geometric one. You don't need a clear view through the storm. You need enough link margin to absorb the loss. Most commercial links budget 10–15 dB for rain. That works — until the beam is already marginal because Fresnel blockage cost you 6 dB. Then a light drizzle finishes the link. That's not a rain problem. That's a stacking problem.

It's rarely the rain that kills your link. It's the rain plus the tree you didn't clear.

— field engineer, after an all-night debug in August monsoon season

Atmospheric absorption at higher frequencies creates a hard range limit, but it's predictable. You can calculate it. The harder variable is dynamic blockage — a person walking past a window, a car parking in a driveway. These events happen in seconds. Rain builds over minutes. The irony: teams over-engineer for rain fade yet ignore the Fresnel zone that costs them more every day.

Foliage loss and tree canopies

Foliage is the silent killer of mmWave links. A single oak tree in full leaf can add 20–30 dB of loss at 28 GHz. That's not attenuation — that's destruction. The conventional wisdom says "cut the branches." Practical reality says you can't cut a neighbor's tree. Or a protected one. Or the one that grows back in two seasons. The better approach: treat foliage loss as a time-varying parameter, not a static obstacle. Deciduous trees lose leaves in winter; your link may work for six months and then fail. We saw this at a fixed-wireless installation in Portland. The autumn outage was repeatable. The fix was not trimming. It was moving the customer antenna 15 feet to the west, avoiding the canopy entirely. That's effective propagation — not clear view.

The trade-off is subtle: more margin means more power, which means more heat, which means bigger radios. You can't brute-force foliage loss with gain alone. The math doesn't close. What usually breaks first is the assumption that "line of sight" means "free space." It doesn't. Not at these frequencies. Not ever.

Odd bit about technology: the dull step fails first.

Odd bit about technology: the dull step fails first.

Patterns That Usually Work: Beamforming, MIMO, and Small Cells

Beamforming and steering

The common image of millimeter wave—a straight, brittle ray that shatters on the first leaf—is wrong. Beamforming is not a flashlight beam. It's a phased array of dozens of tiny antenna elements, each emitting the same signal at slightly different times. By adjusting those tiny timing offsets, the combined wavefront can be bent, focused, or even split. I have watched a 28 GHz link hold steady while a delivery truck rolled between the transmitter and receiver. The array simply re-targeted around the truck, using a reflection off a glass storefront. That's not magic. That's physics.

The catch is that beamforming requires computation. The radio must sample the channel, compute the optimal phase shifts, and apply them—all within milliseconds. Old gear can't do it. Cheap gear half-asses it. But when it works, you don't need a clear view. You need enough scatter paths to compute a valid solution. Puddles on a parking lot, corrugated metal siding, even a chain-link fence—each provides a usable reflection. A single dominant reflector beats a dozen weak ones, but any reflector beats a dead channel.

Wrong order: assuming that beam steering means the link can chase a moving client through a brick wall. It can't. Phase-coherent steering works over angles, not through dense obstacles. The trick is knowing which reflections are stable and which flicker out.

Massive MIMO spatial multiplexing

Massive MIMO multiplies the same principle—more antennas, more paths, more tolerance for blockage. A 64-element array doesn't produce 64 separate beams; it produces overlapping patterns that can serve multiple users simultaneously, even when no user has a direct line to the tower. Each user sees a composite signal built from many reflected copies. The mathematics of this is brutal—matrix inversions at microsecond scale—but the result is ordinary: your phone gets data, even under a tree, even around a corner.

What usually breaks first is the channel estimate. If the array can't distinguish between the direct path and a reflected ghost, the spatial multiplexing collapses into interference. I have debugged sites where the only fix was to rotate a small reflector by ten degrees. Ten degrees. The link went from 200 Mbps to 850 Mbps. That sounds fragile, and it's. But it's also fixable once you stop treating the link like a telescope that needs a clear aperture.

Most teams skip this: they deploy MIMO with default antenna patterns, hoping the environment sorts itself out. It doesn't. The pattern must be tilted, weighted, or even partially disabled to match the actual scatter geometry. A site with a single brick wall and a metal awning behaves differently than a site with double-pane glass and a row of parked vans. Massive MIMO amplifies both signal and noise—you can't run it blind.

Small cell densification and handover

Small cells are the ugly practical answer to millimeter wave myths. You don't need one tower that sees everything. You need forty small boxes, each covering a block, each handing off to the next as the user walks. The handover is where most deployments fall apart. A moving phone crossing a millimeter wave cell boundary at 3 mph loses synchronization in under 20 milliseconds if the link drops. That is tighter than any LTE handover. But modern 5G NR protocols have a trick: they maintain two simultaneous connections—one on a lower-frequency anchor, one on millimeter wave—so the mmWave link can fail and recover without the user noticing.

I have seen a small cell array on a traffic light pole beam-form through a bus stop shelter, then hand off seamlessly to a lamppost unit 40 meters away. The key was that both cells shared the same millimeter wave channel, tuned to the same reflection plane—the flat side of a loading dock. Without that common reflector, the handover would have dropped. The trade-off is that small cells demand site-specific tuning. Generic deployment recipes kill performance.

'We installed forty nodes in one weekend. Three weeks later, we had to reposition sixteen of them. The trees leafed out.'

— field engineer, mid-Atlantic deployment, recounting why spring maintenance schedules matter

Small cell densification works. But it works only when you budget for environmental drift—trees, construction, seasonal parking changes. The pattern that usually works is a dense grid, active beamforming, and MIMO that adapts to the clutter. The pattern that fails is hoping one tall pole will see across the neighborhood. It won't. Millimeter wave is not a telescope. It's a conversation with the environment. You need many ears, placed where the echoes live.

Anti-Patterns and Why Teams Revert: Over-engineering LOS Assumptions

Treating mmWave Like Microwave Backhaul

I have watched teams burn three months of deployment budget because they treated a 28 GHz link exactly like a 6 GHz microwave shot. Same tower height specs. Same Fresnel zone clearance rules. Same assumption that if you can see the other antenna with binoculars, the link works. That logic holds for 11 GHz backhaul—it shatters for millimeter wave. The difference? A microwave beam at 6 GHz bends around a tree crown maybe 2 dB. At 28 GHz that same tree is a 20 dB wall. The mistake isn't the LOS check—it's stopping there. Teams over-engineer tower placement because they demand a clean Fresnel zone that mmWave never actually needs. They spend weeks negotiating rooftop access for a sightline that beamforming would have bent around a billboard.

The catch is hidden in procurement: project managers who cut their teeth on traditional microwave assume the old planning rules transfer. They don't. And the budget hemorrhage starts early—steel reinforcement for a tower that could be 12 feet lower if they'd trusted the reflection map instead of the binoculars.

Ignoring Reflection Maps in Planning Tools

Most planning software ships with a reflectivity layer turned off by default. Engineers skip it because the setting feels optional. Wrong choice. In one urban deployment I audited, the team placed a node on a water tower because the static LOS model showed a perfect line to the target block. What the model missed: a glass-curtain building three blocks south that acted as a passive repeater. The water tower node saturated one street; the reflected path would have covered four. The missed coverage gap required two additional small cells to fix—each mounted on leased streetlights at $1,200 per month per pole.

Static LOS models lie. They treat every surface as a radio graveyard. Real mmWave propagation treats glass, metal, and wet concrete as mirrors. The fix is cheap: turn on the reflection layer, run the simulation at three polarizations, and look for ripples in the coverage heat map. Most teams skip this because it adds two hours to the planning phase. That two-hour save costs six months of retrofits.

'We mounted the node where the LOS tool said to. Then we drove the coverage truck and found signal behind the building we couldn't see.'

— field engineer, after the third truck roll

Odd bit about technology: the dull step fails first.

Odd bit about technology: the dull step fails first.

Using Only Static LOS Models

The static LOS model assumes nothing changes. No leaves on the tree that grows 18 inches taller each year. No construction crane that parks for six months. No delivery truck that idles in the same loading dock every afternoon. I've seen a perfectly planned mmWave link die every day at 4:17 PM because a metal-sided box truck parked in exactly the wrong spot. The team had modeled the street as empty—they had a photo from Sunday morning. That static assumption cost them a 45 dB fade every rush hour. They solved it by adding a second node 30 meters away, reusing the reflection off a brick wall the original model ignored as "not LOS."

The better approach? Run a time-series propagation model across a full business day. Or, cheaper: park a van where the truck sits, measure the loss, and adjust the beam-steering weight vector. That takes afternoon, not a PhD. Static models produce static failures. The environment moves—your beam pattern should too. What usually breaks first is the corner case nobody modeled: a seasonal awning, a new billboard, a leafed-out tree in May. Over-engineering the LOS assumption creates a brittle system that works on paper and fails at 4:17 PM.

Maintenance, Drift, or Long-Term Costs: The Changing Environment

Seasonal Foliage Changes

Most teams test in summer. Leaves are full, humidity is high, and the path loss looks acceptable. Then autumn hits. Deciduous trees shed their canopy, and suddenly your link budget looks suspiciously generous. The odd part is—teams celebrate the improvement instead of questioning it. They should not. Come spring, leaves return with a vengeance, soaking up millimeter waves like a sponge. I have seen a perfectly fine backhaul link drop 8 dB in two weeks because oak trees flushed new growth. That is not a glitch. That's physics you ignored during deployment planning.

What usually breaks first is the margin. You designed for 5 dB fade margin; after foliage gain in winter you had 12 dB. Then April arrives and you scramble. The fix? Bury assumptions deeper. Instead of a single seasonal snapshot, model the worst month—August, when leaves are thickest and rain is frequent. One team I coached used satellite imagery dated to peak foliage. They still got burned because the trees grew taller. Wrong order. You need growth-rate data, not a static photo.

New Buildings and Reflective Surfaces

A clear path today is tomorrow's dead zone. That empty lot across the street? Developers are breaking ground next quarter. New concrete and glass facades don't just block—they redirect. A building going up two blocks away can create a multipath nightmare you never modeled. The keyhole analogy breaks here: that narrow window of propagation can slam shut without warning. Most teams skip this: they map the environment once and assume it holds for five years. That hurts.

Reflective surfaces are worse. Fresh glass coatings on an office tower can turn a stable path into a party of constructive and destructive interference that dances through the day. You lose a day every time a technician drives out to realign a node. The long-term cost isn't hardware—it's labor. We fixed this by scanning municipal construction permits and updating propagation models quarterly. Sounds tedious. Cheaper than sending a crew twice a month.

The catch is that permit data lags reality. By the time a building is framed, you have already lost two months of signal quality. Consider deploying temporary repeaters during construction phases. Not elegant. But a concrete wall doesn't care about your elegant model.

“We spent more on truck rolls in year two than we did on the entire initial install. The environment changed. Our assumptions didn't.”

— field engineer, after a 30 GHz fixed-wireless rollout in a developing suburban corridor

Device Movement and User Orientation

Handheld devices are the wildcard. A user rotates their phone 90 degrees—the antenna pattern shifts. A millimeter-wave beam that was hitting a reflector now points at a wall. Signal drops. This is not a failure of beamforming; it's a failure of orientation awareness. The trade-off is clear: you can chase perfect alignment, or you can accept that people move their hands.

I have watched demos where engineers hold a phone in a death grip to maintain lock. That is not a product. That's a lab trick. Real users tuck phones in pockets, hold them sideways, cover them with cases made of millimeter-absorbent rubber. The drift happens in seconds. What helps? Dense small cells that hand off faster than the user can rotate—lower the cell radius to sixty meters, increase overlap zones. The cost rises, yes, but the alternative is a connection that works only when the phone is mounted on a tripod.

Try this experiment. Walk a test route with the phone in a jacket pocket. Then repeat with the phone held to your ear. The failure patterns will differ completely. Design for the worst grip, not the best—and budget for field testing that captures 90-degree rotations, sweaty hands, and small children grabbing the device. That hurts to think about. It hurts less than 30% drop calls.

When Not to Use This Approach: Exceptions to the Keyhole Rule

Long-range point-to-point backhaul

Some links demand a laser-straight line. The keyhole analogy—where signal bends around obstacles like light through a partial opening—collapses entirely when you stretch distance beyond a few hundred meters. I have watched teams waste weeks trying to finesse non-line-of-sight (NLOS) millimeter wave into a 5-kilometer rural backhaul. The result was a link that bled out at the first rain fade. The catch is simple geometry: at 28 GHz and above, Fresnel zone clearance becomes non-negotiable beyond 1–2 km. Trees, hills, even a passing truck can punch a 20 dB hole in the link budget. You can't beamform your way out of a missing first Fresnel zone—physics doesn't negotiate. That sounds harsh until you price the cost of a tower relocation. Sometimes the only fix is a taller mast or a relay hop. The odd part is—most backhaul failures I audit trace back to a spreadsheet assumption that NLOS margins would cover the gap. They rarely do.

Satellite and aerospace links

Skyward links break the keyhole rule entirely. No ground reflections, no building scatter—just pure free-space path loss and atmospheric absorption. Here the telescope analogy fits perfectly: you need a clear, unobstructed cone from antenna to satellite. I have debugged a drone tether link where a single utility pole blocked the azimuth for three seconds—the control loop crashed. Most teams skip this: millimeter wave satellite terminals often require at least 10-degree elevation clearance above any local terrain or structures. The exception has no loophole. A leaf in the path at 60 GHz can kill a satellite link faster than a tree at 28 GHz on the ground. What usually breaks first is the assumption that atmospheric absorption models are static—they shift with humidity, temperature gradients, and even bird flocks. Wrong order. You plan for obstacles you can see, but the invisible ones (oxygen absorption spikes, water vapor lines) are the ones that steal your margin.

Highly attenuated materials and sealed enclosures

Metal buildings. Concrete walls with rebar mesh. Double-pane low-E glass. The keyhole analogy assumes partial transparency—that some fraction of signal passes through or diffracts around an edge. That assumption is dead wrong for environments where attenuation exceeds 40–50 dB per barrier. I once watched a team deploy an indoor mmWave sensor behind a security grille. The grille was steel, period. The signal never got out. The keyhole analogy fails because there is no keyhole—just a solid wall. The trade-off here bites twice: you gain nothing from MIMO spatial multiplexing if all paths converge through a single tiny gap (correlation kills rank), and you can't rely on small cell density to solve the problem if every cell faces a metal-clad obstacle. The fix is not a better algorithm; it's a drill bit. You cut a window, reposition the antenna, or accept that some environments—factory floors with heavy machinery, parking garages with rebar slabs—are NLOS-dead zones regardless of how clever your beamforming code is.

'If the material stops light, it stops millimeter waves. The keyhole is a metaphor, not a magic wand.'

— field engineer debrief, after a failed indoor campus trial

Reality check: name the technology owner or stop.

Reality check: name the technology owner or stop.

The pattern to watch for is cost leverage. When the obstacle is temporary (seasonal foliage, construction cranes) the keyhole analogy buys you resilience. When the obstacle is permanent and metallic—electrical substations, steel-framed high-rises, maritime vessels—you're better off budgeting for a physical relay than betting on diffraction. Next time you scope a site, walk the path with a flashlight. If you can't see the other antenna, and the barrier is metal or reinforced concrete thicker than a palm, the keyhole rule doesn't apply. Plan the cable run instead.

Open Questions / FAQ: What We Still Don't Know

How do rain and fog affect 5G at 28 GHz vs. 39 GHz?

The short answer: yes, attenuation increases—but not as catastrophically as the splashy headlines suggest. At 28 GHz, heavy rain (25 mm/hr) adds roughly 5–7 dB/km of loss; at 39 GHz that climbs to about 9–12 dB/km. That sounds terrible until you realize beamforming systems can compensate with 3–5 dB of adaptive gain. Fog is barely a factor—liquid water density in fog tops out around 0.5 g/m³, adding less than 1 dB/km at either band. The weird part is—folks forget that cellular links are rarely 1 km continuous paths. Small cells live 100–200 meters apart. Rain fade at that distance? Under 2 dB. Not nothing, but far from a link killer.

The real pain point is wet foliage. A wet tree canopy at 28 GHz can eat 15–25 dB. That beats rain by an order of magnitude.

“Dry leaves: 3–5 dB loss. Wet leaves after a storm: 18–30 dB. One tree can break a link faster than a monsoon.”

— Field engineering log, urban deployment test, 2022

Can mmWave work reliably indoors through drywall?

Yes—but the margin is razor-thin and depends entirely on construction. Standard ½-inch drywall absorbs roughly 2–3 dB per sheet at 28 GHz. Two interior walls plus a hallway corner? You’re down 7–10 dB before furniture reflections kick in. Worse if the drywall contains foil backing (common in energy-efficient builds) — that jumps to 15–20 dB per wall. I have seen a single metal-stud partition collapse a calibrated 28 GHz link from −72 dBm to −112 dBm. The fix wasn’t more power; it was relocating the CPE 18 inches left of a stud line. That hurts.

Most teams skip this: indoor mmWave works best with a clear or near-line-of-sight path through one wall max. Two walls becomes a coin toss. Concrete or brick? Abort. The practical takeaway is that small cells need to be inside the served room, not down the hall. Retrofit costs go up. Deployment density doubles. But the physics doesn’t negotiate.

Will future regulations allocate more spectrum?

Globally, yes—but the bands shift higher, not lower. The 2023 WRC-23 agenda pushed allocations at 24 GHz, 26 GHz, and 40 GHz for IMT-2020. The US FCC already auctioned 28 GHz and 39 GHz; Europe is harmonizing 26 GHz with smaller guard bands. The catch is that higher spectrum means tighter propagation—6 dB more free-space path loss per octave. Regulators can’t legislate physics. What they can do is relax out-of-band emission limits to allow wider channel bandwidths (400–800 MHz instead of 100 MHz slivers). That helps throughput but doesn’t fix the wall problem.

One open question: will 6G push into 100+ GHz (D-band)? That would make current drywall concerns look optimistic—at 140 GHz, a single sheet of paper causes measurable fade. The regulatory trend is toward more spectrum, yes. But also toward spectrum that demands even denser infrastructure. That is not a bug. That is the millimeter-wave bargain.

Summary + Next Experiments

Three Key Takeaways

The millimeter-wave story is not about seeing a straight line—it's about finding the keyhole. Forget the telescope analogy; your signal doesn't need a perfect view through the air, it needs an opening big enough to squeeze through. I have fixed deployments where a tree canopy blocked the link but a single gap in the leaves—barely wider than a person—carried full throughput. That changes how you design.

First: non-line-of-sight (NLOS) propagation works, but only when the environment provides enough reflective surfaces and the beamformer can exploit them. Second: beamforming and MIMO are not magic—they require enough spatial diversity and signal-to-noise headroom to lock onto those secondary paths. Third: small cells are the practical enabler because they shorten the distance, turning a hard NLOS challenge into a manageable keyhole search. The catch is that all three fail if you assume the keyhole stays open.

Field Test Checklist

Before you push a deployment to production, run this sequence. Pick a candidate site without a clear LOS to the base station—then walk the terminal location with a spectrum analyzer and a handheld directional antenna. Rotate 360 degrees slowly. Where does the signal peak? That spot is your keyhole. Mark it with a stake. Then return at three different times of day—morning fog, midday sun, evening moisture. Does the peak shift? It will.

Most teams skip this: measure the angular spread of the dominant path. If it's narrower than 10 degrees, you have a single keyhole—vulnerable to a parked truck or a new sign. If it's wider than 30 degrees, you have multiple reflections working for you. The trade-off is that wider spreads often mean weaker individual paths. We fixed a stadium deployment by accepting three weak paths instead of one strong path—the link stayed up during crowd movement because no single blockage killed all three. That hurts your initial throughput budget but saves maintenance headaches later.

Wrong order? Do the walk first, then simulate the link budget. Rushing to a link calculator without understanding the actual propagation environment gives you numbers that look good in a spreadsheet but fail in rain.

Further Reading on mmWave Propagation

Start with Rappaport's early work on 28 GHz and 38 GHz channel sounding—the papers from 2013-2015 that demonstrated NLOS path existence before commercial gear existed. Then read the 3GPP TR 38.901 channel model documentation, but skip the equations on first pass and focus on the scenarios: indoor, outdoor-to-indoor, and urban canyon. That's where the keyhole analogy breaks and reforms depending on clutter density.

The odd part is—most operator field reports are proprietary, not published. I recommend joining the wireless IoT forum mailing lists (WIF, not Wi-Fi) and searching archives for "small cell deployment hell." You will find engineers describing exactly the keyhole problem without using the word. One comment stuck with me: "We had LOS on the survey, but three months later the neighbor installed a metal awning and the link dropped 18 dB."

— paraphrased from a 2021 deployment thread, identity withheld

Next step: pull a 30-day RSSI log from your test node. Plot it against local weather records. Watch for the 5-7 dB drop during heavy rain—does it recover? If not, your keyhole was already marginal. That is the experiment that separates theory from production reality. Run it before you scale.