Chapter 33: Small Satellites, CubeSats, and Constellations

"Faster, better, cheaper." — Daniel S. Goldin, NASA Administrator (1992–2001), on a new philosophy for space missions

Overview

For the first fifty years of the space age, a satellite was a monument. It was a bus-sized, one-of-a-kind machine, hand-built over the better part of a decade by a government or a giant aerospace prime, costing hundreds of millions to billions of dollars, and launched — if all went well — exactly once, on a rocket that cost as much again. Space was the province of superpowers and the largest corporations, and the reason was not secrecy or politics. It was arithmetic: the launch was so expensive, and the environment so unforgiving (theme 2), that the only rational thing to build was a single, exquisite, gold-plated spacecraft engineered so that everything must work, because you would never get a second one to orbit.

That world is ending, and this chapter is about what is replacing it. Two independent revolutions collided at the turn of the millennium. First, the same microelectronics revolution that put a supercomputer in your pocket made spacecraft-grade computing, sensing, and radio cheap, tiny, and available off a commercial catalog — you no longer had to custom-build every component from scratch. Second, the reusability revolution and the rise of the rideshare launch (theme 5) drove the cost of reaching orbit down by something like an order of magnitude. Put those together and the economics invert. A satellite can now be the size of a loaf of bread, cost tens of thousands of dollars, be built by a university lab or a startup in a year, and reach orbit as one of a hundred payloads sharing a single rocket. And if a satellite is cheap enough, you can afford to fly thousands of them at once — a constellation that does, as a coordinated system, what no single satellite ever could.

This is the most disruptive shift in spaceflight since staging, and it is the clearest expression of the book's fifth theme: reusability and cheap access are changing everything. We are at the beginning of this transformation, not the end.

In this chapter, you will learn to:

• Explain the two revolutions — cheap electronics and cheap launch — that democratized space, and classify a spacecraft by mass.
• Describe the CubeSat standard and why a standard, more than small size, is what unlocked the movement.
• See every Part IV subsystem shrunk onto a circuit board, and understand the COTS bargain: cheaper and faster, in exchange for accepting risk.
• Compare rideshare against dedicated launch on cost per kilogram, and follow a satellite out of its deployer.
• Compute, from simple geometry, how many satellites a constellation needs for continuous coverage — and see why Starlink flies thousands.
• Weigh the promise of cheap, ubiquitous satellites against the debris and sustainability bill that comes due in Chapter 35.

Learning Paths

🚀 Space Enthusiast: Read 33.1 for the story of how space got cheap, then 33.5 for the constellation idea that is reshaping the sky over your head. The coverage-geometry math in 33.5 is worth slowing down for — it is the reason there are thousands of Starlinks and not a dozen.

📐 Engineering Student: Read all of it, and do the coverage-geometry worked example in 33.5 and the cost-per-kilogram comparison in 33.4 by hand. Section 33.3 is a rapid tour of Part IV from a new angle; the ⭐⭐/⭐⭐⭐ exercises on link budgets, power, and constellation sizing are the payoff.

🎮 KSP Player: You have built rideshare payloads without knowing the name. Focus on 33.2 (the standardized form factor), 33.4 (how deployers fling satellites free), and the coverage geometry of 33.5 — the same "how many relays do I need for continuous contact" problem the game's comm-network mods solve.

🛰️ Industry Prep: This is the fastest-growing corner of the industry and the one most likely to employ you. Sections 33.4 (rideshare economics), 33.5 (constellation design), and 33.6 (the debris question) are the daily vocabulary of the NewSpace sector. The Checkpoint asks the question every new mission now asks first: should this be a small satellite, or a constellation of them?

33.1 The democratization of space

Start with the word that names the whole shift: democratization. For most of the space age, only a handful of actors on Earth could fly a spacecraft. Today, more than eighty countries have operated a satellite, thousands of companies have flown hardware, and a capable satellite has been built by high schools. The barrier that fell was not scientific — the physics of Chapters 1 through 29 has not changed — it was economic. Two costs collapsed at once.

The first was the cost of capability. A modern smartphone contains a processor billions of times faster than the Apollo Guidance Computer that flew astronauts to the Moon, a suite of MEMS gyroscopes and accelerometers that would have been a national-asset inertial navigation system in 1970, a multi-megapixel camera, and multiple radios — all for a few hundred dollars, all weighing a few grams. Moore's law, marching for half a century, turned the exotic electronics of a spacecraft into commodity parts. The engineer no longer has to invent an onboard computer; they can buy one. This is the world of COTS.

Definition (COTS — commercial off-the-shelf). A COTS part, component, or subsystem is one bought from a commercial catalog — a processor, radio, camera, battery cell, or reaction wheel produced in volume for a broad market — rather than custom-designed and formally space-qualified from scratch. COTS parts are cheap, immediately available, and often outperform their space-rated equivalents, because the commercial market that funds them is enormous. Their cost is risk: they are not radiation-hardened, not vacuum-screened, not qualified to survive years in the space environment. Using them means accepting that some will fail, and buying reliability back another way — through redundancy, screening, short mission lifetimes, and sheer numbers.

The second cost that collapsed was access — the price of a kilogram to orbit. This is the story of Chapters 22 and 38: reusable first stages and, just as importantly, the rideshare launch (33.4) cut the cost of reaching low Earth orbit from the tens of thousands of dollars per kilogram that prevailed for decades toward a few thousand, and falling. When launch was the dominant cost, it made no sense to fly a cheap satellite — you would put your expensive rocket's capacity to work carrying an expensive, capable spacecraft. Once launch got cheap, a cheap satellite finally made economic sense. The two revolutions are multiplicative: cheap parts make a cheap satellite possible, and cheap launch makes it worth building.

What these forces produced is a whole zoo of spacecraft far smaller than the traditional multi-tonne bus. The field sorts them, loosely, by mass:

Definition (small satellite). A small satellite (or smallsat) is, by common convention, a satellite with a launch mass below roughly $500\ \text{kg}$ — an order of magnitude or more lighter than a traditional large spacecraft. The category is subdivided by mass, though the boundaries are conventions, not physics, and vary between sources:

Most CubeSats (33.2) are nanosatellites; the broadband-constellation satellites of 33.5 are typically mini- or microsatellites of a few hundred kilograms. The umbrella term "small satellite" spans the whole range below the traditional big bus.

📜 From History: from Vanguard to the garage. The very first American satellites were small by necessity — Vanguard 1, launched in 1958, was a $1.5\ \text{kg}$ grapefruit-sized sphere — but as rockets grew, so did their payloads, and for forty years "more capable" meant "bigger and costlier." The pendulum swung back in the 1990s when NASA Administrator Dan Goldin pressed the mantra in this chapter's epigraph — faster, better, cheaper — arguing that flying more, smaller missions would return more science per dollar and per year than a few flagship behemoths. The philosophy was controversial and had painful failures, but its logic was sound and its timing prophetic: within a decade, cheap electronics and the CubeSat standard delivered on "faster, better, cheaper" more completely than Goldin's own programs ever could. The satellite stopped being a monument and became a product.

💡 Intuition: a monument versus a product. A traditional flagship satellite is engineered like a cathedral — built once, at enormous cost, to stand for decades, with every stone perfect because there will never be another. A smallsat is engineered like a consumer product — designed to be built in quantity, cheaply enough that losing one is an acceptable cost of doing business, and replaced by a better version next year. Almost every strange-looking choice in the rest of this chapter (COTS parts, no propulsion, planned short lifetimes, thousands of identical units) follows from that single change of mindset: from this one must last to the fleet must work.

🔄 Check Your Understanding 1. Name the two independent cost collapses that together democratized access to space, and explain why they are multiplicative rather than merely additive. 2. What is the central trade a designer accepts by using COTS parts instead of space-qualified ones?

Answers

1. The cost of capability (cheap, powerful electronics from Moore's law) and the cost of access (cheap launch from reusability and rideshare). They multiply because cheap parts make a cheap satellite possible while cheap launch makes it worth flying — neither alone is enough; a cheap satellite on an expensive dedicated rocket is still pointless, and an expensive satellite gains little from cheap launch.
2. Lower cost and faster availability (and often higher raw performance) in exchange for lower per-unit reliability in the space environment — COTS parts are not radiation-hardened or space-qualified, so the designer must buy reliability back through redundancy, screening, short lifetimes, or numbers.

33.2 The CubeSat standard

If cheap electronics and cheap launch were the fuel, the CubeSat standard was the spark. And here is the crucial insight, the one most people miss: the revolution was not small size. Engineers had built small satellites for decades. The revolution was standardization — agreeing on a single, fixed form factor and a single mechanical interface to the rocket, so that a satellite became a modular, interchangeable, mass-producible thing that any launch provider could carry without a custom engineering project each time.

📜 From History: a satellite you could finish before you graduated. In 1999, Jordi Puig-Suari of California Polytechnic State University and Bob Twiggs of Stanford confronted a teaching problem: a real satellite took longer to build than a graduate degree lasted, so students never saw one through. Their fix was to define a satellite small and standardized enough that a small team could design, build, and fly it within a couple of years. They settled on a $10\ \text{centimeter}$ cube and, just as importantly, published an open specification — the CubeSat Design Specification — that anyone could build to. That openness is why it spread. A standard that one university owns is a curiosity; a standard everyone can build to becomes an industry.

Definition (unit, "U"). A unit, written U, is the basic building block of the CubeSat standard: a cube $10\ \text{cm} \times 10\ \text{cm} \times 10\ \text{cm}$ ($1\ \text{U} = 10\ \text{cm}$ on a side, a volume of one liter), with a mass originally capped at about $1.33\ \text{kg}$ (recent revisions of the specification allow more — up to roughly $2\ \text{kg}$ per U). Satellites are assembled by stacking units, and the standard's magic is that any conforming satellite, whatever its internals, presents the same standard outer dimensions and rails to the launch vehicle.

Definition (CubeSat). A CubeSat is a small satellite built to the CubeSat Design Specification, in integer multiples of the $10\ \text{cm}$ unit, so that it fits a standardized deployer (33.4) and can ride to orbit as a standardized payload. A CubeSat is defined not by what it does — they carry every kind of payload — but by the standard it conforms to.

The unit stacks into a family of sizes, and a few have become workhorses:

(Masses are per-specification approximate figures — Tier 2 — and rise as the standard evolves; treat them as the order of magnitude, not exact limits.)

The 10-centimeter choice is a Goldilocks compromise. Smaller, and there is no room for a useful payload, a real radio, or enough solar cell area to power anything. Larger, and you lose the cheap, rideshare-able modularity that is the whole point. Ten centimeters is small enough to be cheap and to slot into a standard dispenser, and just big enough to hold commercial electronics and do real work.

🔧 Engineering Reality: the standard is mostly a set of promises to the neighbors. Much of the CubeSat specification is not about the satellite's function at all — it is about not endangering the primary payload and the other rideshares sharing the rocket. Nothing may protrude beyond the $10\ \text{cm}$ envelope while stowed; deployables (antennas, solar panels) must stay latched until after ejection; there are rules on materials that outgas, on stored energy, on batteries, on how the satellite stays inert inside the dispenser until it is safely clear. A launch provider integrates a standard deployer once and then trusts that any conforming CubeSat is a safe, inert box until it is flung free. That trust — encoded as a standard — is what lets a rocket carry a hundred satellites from a hundred strangers on one flight. The standard is a social technology as much as a mechanical one.

🔗 Connection: small does not mean local. It is tempting to file CubeSats under "cheap toys for low Earth orbit," but the physics of the earlier chapters does not stop at $10\ \text{cm}$. In 2018, two 6U CubeSats named MarCO-A and MarCO-B flew all the way to Mars alongside the InSight lander, becoming the first interplanetary CubeSats, and relayed InSight's entry-descent-landing telemetry back to Earth in near-real time. They ran the same interplanetary trajectory mathematics as any Mars mission, in a package the size of a cereal box. We will meet their descendants when we go to Mars for real in Chapter 34. The standard shrinks the spacecraft, not the solar system it can reach.

🔄 Check Your Understanding 1. The chapter claims the CubeSat revolution was standardization, not miniaturization. Restate the argument: what does a standard buy that a merely small satellite does not? 2. A team wants a $10\ \text{kg}$, moderately capable science satellite. Which CubeSat size is the natural fit, and roughly what are its dimensions?

Answers

1. A standard makes satellites interchangeable and mass-rideshare-able: a launch provider integrates one standard deployer and can then carry any conforming CubeSat without a bespoke engineering effort, and component vendors can sell off-the-shelf CubeSat parts to a broad market. Small size alone gives none of that — a small but non-standard satellite still needs a custom launch interface and custom parts. The standard turns "small satellite" into a modular product with an ecosystem. 2. A 6U CubeSat ($10\times20\times30\ \text{cm}$, up to roughly $8$–$12\ \text{kg}$) — the workhorse size for capable science payloads, and the size the interplanetary MarCO spacecraft used.

33.3 Small-sat subsystems: Part IV on a circuit board

A small satellite is not a different kind of physics; it is Part IV of this book, shrunk. Every subsystem you studied for a large spacecraft still exists on a CubeSat — power, thermal, communications, attitude control, command and data handling, structure, sometimes propulsion — but each is miniaturized, and each leans hard on COTS parts. Walking through them is a rapid review of Part IV from a new vantage.

Power (Chapter 25), the binding constraint. A CubeSat's power comes from small solar panels — either body-mounted on its faces or deployed on hinged wings after ejection — feeding COTS lithium-ion cells (often literally the same $18650$ cells found in laptops and power tools). The power available is tiny, and it is usually the constraint that limits everything else the satellite can do.

Worked Example: how much power does a 3U CubeSat make? Take a 3U CubeSat and, at best, point one of its large $10\ \text{cm} \times 30\ \text{cm}$ faces at the Sun. That face has area $A = 0.10 \times 0.30 = 0.030\ \text{m}^2$. The solar flux at Earth is about $1{,}361\ \text{W/m}^2$ (Chapter 25). High-efficiency space cells convert roughly $\eta \approx 0.28$ of that, and after packing gaps and imperfect pointing, keep a factor $\approx 0.8$. So the instantaneous power from that one illuminated face is $$P = 1{,}361 \times 0.030 \times 0.28 \times 0.8 \approx 9\ \text{W}.$$ Nine watts — about the draw of a dim household bulb — is the peak, with the best face square to the Sun. Average it over an orbit (the satellite tumbles or points elsewhere, and spends part of every orbit in Earth's shadow, Chapter 25's eclipse problem) and the orbit-average is often only a couple of watts. Deployable panels can multiply the area several-fold, but the lesson stands: a CubeSat runs on a power budget measured in watts, and every other subsystem must fit inside it. Theme 4 — mass is the enemy — has a sibling here: power is the enemy too, and on a smallsat it bites first.

Communications (Chapter 26), squeezed by the link budget. The earliest CubeSats used amateur-radio UHF/VHF bands and simple whip antennas; capable ones now use S-band and X-band with modest deployable or patch antennas. But the link budget is brutal on a small platform: low transmit power (you only have watts), small antennas (low gain), and therefore low data rates. Many CubeSats close their link only when passing over a ground station, dumping a burst of data in the few minutes of a pass — the same "searchlight" limitation we saw for any single LEO satellite in Chapter 9. Networks of low-cost ground stations, and inter-satellite relay (33.5), are the workarounds.

Attitude determination and control (Chapter 14), miniaturized. A CubeSat still needs to know and control which way it points, and the hardware of Chapter 14 shrinks onto a board: MEMS gyroscopes, sun sensors, small star trackers, tiny reaction wheels, and — the signature small-sat actuator — magnetorquers, coils that push against Earth's magnetic field to generate torque. Magnetorquers are beloved on CubeSats because they have no moving parts, sip milliwatts, and never run out of anything — though they only work where there is a magnetic field to push against (fine in LEO, useless at Mars). The quaternions and reaction-wheel dynamics of Chapter 14 are exactly the same; only the scale changed.

Propulsion (Chapter 20) — often none at all. Here is the sharpest break from a large spacecraft: many CubeSats have no propulsion whatsoever. They cannot raise, lower, or circularize their orbit; they cannot dodge a collision; they cannot deorbit themselves — they simply go where the deployer puts them and wait for atmospheric drag to bring them down. This is cheap and simple, and it is the single biggest sustainability problem of the smallsat era (33.6). When CubeSats do carry propulsion, it is miniaturized: cold-gas thrusters, tiny electrospray or ion units, even water-fed steam thrusters. The broadband constellations of 33.5 are the exception that proves the rule — their satellites carry real electric propulsion precisely because they must maneuver and deorbit responsibly.

Command, data handling, and structure (Chapters 23–24). The onboard computer is usually a COTS processor — an ARM chip not far removed from a phone's — protected not by expensive radiation-hardening but by software: watchdog timers that reboot a hung processor, error-correcting memory, and redundant copies that vote. The structure (Chapter 23) is the machined aluminum frame and rails that give the unit its shape and its standard interface, and thermal control (Chapter 24) is mostly passive — coatings and the thermal mass of the structure itself, since there is no power budget for heaters.

⚠️ Common Misconception: "COTS electronics will be fried by radiation instantly." This worry is real but overstated, and the nuance is the whole art of small-sat engineering. Low Earth orbit sits beneath the worst of the Van Allen belts (Chapter 9), so the radiation dose over a short LEO mission is survivable for many commercial parts. Missions manage what dose there is with three cheap tools: screening (test a batch, fly the parts that pass), redundancy (Chapter 32 — carry two or three and vote), and short lifetimes (a two-year mission accumulates far less dose than a fifteen-year one). It is only when a smallsat ventures into the belts, to GEO, or to deep space that radiation-hardening becomes unavoidable — which is exactly why MarCO's trip to Mars was a genuine engineering achievement and not a routine one.

🔗 Connection: reliability moves from the part to the system. Everything above is a worked application of Chapter 32's reliability thinking, stood on its head. A flagship satellite achieves reliability by making every part nearly perfect. A CubeSat — and even more, a constellation — achieves it by accepting that individual parts (and whole satellites) will fail, and arranging for the mission to succeed anyway through redundancy and numbers. Same goal, opposite strategy. We will see this reach its logical conclusion in 33.5, where the "system" that must be reliable is a fleet of thousands.

🔄 Check Your Understanding 1. Why is power, even more than mass, often the binding constraint on what a CubeSat can do? 2. Why are magnetorquers such a popular attitude actuator on LEO CubeSats — and where would they be useless?

Answers

1. A CubeSat's solar area is tiny (a 3U face is only $0.03\ \text{m}^2$), so its orbit-average power is a few watts; every subsystem — radio transmit power, computing, heaters, actuators — must fit inside that tiny budget, which limits data rate, duty cycle, and payload. Mass is capped by the standard, but power is what you run out of first when you try to do something. 2. Magnetorquers have no moving parts, draw almost no power, and never deplete a consumable — they generate torque by pushing against Earth's magnetic field. That last point is also their limitation: they only work where there is a strong magnetic field, so they are ideal in LEO and useless far from Earth (e.g., in deep space or at Mars, whose global field is negligible).

33.4 Rideshare and deployment

A small satellite creates a mismatch. A CubeSat might mass $4\ \text{kg}$; a Falcon 9 can lift more than $22{,}000\ \text{kg}$ to low Earth orbit. Buying a whole rocket to launch one CubeSat would be like chartering a container ship to mail a postcard — the launch would cost thousands of times more than the satellite. The resolution is to share the ride.

Definition (rideshare). Rideshare is the practice of launching multiple independent payloads — often from many unrelated customers — on a single rocket, so that the fixed cost of the launch is split among them, rather than dedicating a rocket to one payload. It comes in two flavors: a small satellite can fly as a secondary payload, filling the spare mass and volume left over on a large primary mission's rocket; or it can buy a slot on a dedicated rideshare mission, where the launch provider sells the whole rocket as a large number of standardized small slots.

The dedicated-rideshare model has scaled to astonishing degrees. On its Transporter-1 mission in January 2021, SpaceX deployed 143 satellites on a single Falcon 9 — a record — for customers around the world, each paying for a slice of a rocket none of them could have afforded alone. Rideshare turned launch from a bespoke service into something closer to a scheduled cargo flight.

How does a satellite actually leave the rocket? Through a deployer.

Definition (deployer). A deployer (or dispenser) is the mechanism that houses a satellite during launch and ejects it into orbit on command. For CubeSats, the canonical deployer is the P-POD (Poly-Picosatellite Orbital Deployer), a spring-loaded rectangular tube that holds a stack of up to 3U of CubeSats, isolates them from the rocket's vibration, keeps them safely inert, and then — at the commanded moment — releases a door so a spring pushes them out at a gentle relative velocity of about $1$–$2\ \text{m/s}$. Larger dispensers exist for larger small satellites. The deployer is the physical embodiment of the standard: it is the one piece the launch provider integrates, and everything that conforms to it can fly.

That gentle spring push matters. Every payload on a rideshare is released at nearly the same place and time, so they must drift apart slowly and safely — a metre or two per second of relative velocity spreads them out over hours without any risk of recontact. It also means every satellite from one rideshare starts in nearly the same orbit, which becomes important when you are trying to build a constellation whose satellites need to be spread evenly around the planet (33.5).

Now the economic heart of the section: what does rideshare actually save you? Compare three ways to reach low Earth orbit.

Worked Example: cost per kilogram, three ways. All prices here are approximate, widely reported figures (Tier 2) chosen to show the ratios, not to quote a current price list; real prices move and are negotiated. Take a $200\ \text{kg}$ microsatellite bound for a sun-synchronous orbit.

Option	Approx. price	Payload basis	Cost per kg	You control…
Dedicated small launcher (Rocket Lab Electron)	$\sim\$7.5\ \text{M}$ | $\sim 300\ \text{kg}$ to LEO | $\sim\$25{,}000/\text{kg}$	orbit and schedule
Dedicated rideshare (SpaceX Transporter)	$\sim\$1\ \text{M}$ for $200\ \text{kg}$ | the $200\ \text{kg}$ slot | $\sim\$5{,}000/\text{kg}$	almost nothing (fixed orbit/date)
Buy the whole rocket (Falcon 9)	$\sim\$67\ \text{M}$ | $\sim 22{,}800\ \text{kg}$ to LEO | $\sim\$2{,}900/\text{kg}$	everything — if you can fill it

Read the table as a trade, not a ranking. Buying the whole Falcon 9 is the cheapest per kilogram, but only if you have $22{,}800\ \text{kg}$ of payload to fill it — useless for a lone $200\ \text{kg}$ satellite. The dedicated small launcher is roughly five times more expensive per kilogram than rideshare, and you pay that premium for one thing: control. On Electron you choose your orbit and your launch date; on a Transporter rideshare you go to the orbit the provider picked, on the provider's schedule, packed among a hundred strangers. The question a mission designer must answer is whether that control is worth a $5\times$ price — and for most small satellites, whose requirements the standard rideshare orbit already satisfies, it is not.

🔧 Engineering Reality: rideshare gives you a seat, not a destination. The hidden cost of rideshare is that you inherit the primary's orbit — usually a common sun-synchronous or mid-inclination LEO. If your mission needs a specific altitude, inclination, or local time that the rideshare does not offer, you have three choices, each with a price: pay far more for a dedicated launch; carry your own propulsion to maneuver after release (mass and cost you may not have); or buy a ride on an orbital transfer vehicle (a "space tug") that drops you off in a custom orbit for a fee. This is why orbit selection (Chapter 9) and launch-vehicle selection (Chapter 30) are coupled decisions in your mission design: the cheap launch is only cheap if you can live where it takes you.

🔄 Check Your Understanding 1. Why is buying a whole Falcon 9 the lowest cost per kilogram yet still the wrong choice for a single $200\ \text{kg}$ satellite? 2. What is the one thing a dedicated small launcher sells that rideshare cannot, and why might a mission pay $5\times$ for it?

Answers

1. Cost per kilogram assumes you use every kilogram. A lone $200\ \text{kg}$ satellite would leave more than $22{,}000\ \text{kg}$ of the rocket's capacity empty, so its effective cost per kilogram is the whole $\sim\$67\ \text{M}$ divided by $200\ \text{kg}$ — hundreds of thousands per kilogram. The low per-kg figure only applies if you can fill the rocket. 2. Control over orbit and schedule: a dedicated launcher takes you to the exact altitude, inclination, and local time you want, on a date you choose, rather than to the shared rideshare orbit whenever the provider flies. A mission whose requirements the standard rideshare orbit cannot meet — a specific inclination, a precise local time, a time-critical launch — may have no cheaper option and will pay the premium.

33.5 Constellations

Recall the searchlight from Chapter 9: a single satellite in low Earth orbit races over any given point in a few minutes and does not return for hours. It is a superb camera — close, high-resolution — but a hopeless utility, because it cannot provide the one thing a communications or navigation service must have, which is to be there all the time. GEO solves that with altitude (one satellite hangs over a hemisphere) but pays in distance, delay, and the inability to serve the poles. The other solution — the one this section is about — is to solve coverage not with a better satellite but with many satellites, coordinated into a system.

Definition (constellation). A constellation is a set of satellites operated together as a coordinated system, arranged in complementary orbits so that their combined coverage or capability exceeds anything a single satellite could provide — most often, continuous coverage of a region or the whole Earth. The satellites are typically identical and evenly distributed; what makes them a constellation rather than a mere fleet is that coverage is a property of the set, engineered deliberately by choosing how many satellites fly, in how many orbital planes, at what altitude and inclination. GPS (Chapter 9, ~$24$–$31$ satellites in MEO) is a constellation; Iridium ($66$ satellites in LEO) is a constellation; Starlink (thousands, in LEO) is a mega-constellation.

Orbital shells: how a constellation is arranged

A constellation is organized into shells — groups of satellites sharing an altitude and inclination, distributed across several orbital planes, with the satellites in each plane evenly spaced around it. The inclination sets which latitudes get covered (recall Chapter 9: a $53^\circ$ orbit covers the populated mid-latitudes densely but never overflies the poles, while a near-polar $\sim 90^\circ$ shell reaches everywhere). The number of planes and the number of satellites per plane are chosen so that, as the satellites and the Earth turn, at least one satellite is always high enough in the sky over every user. The classic recipe for spreading satellites evenly over the globe is the Walker constellation pattern, specified by an inclination and three numbers — total satellites, number of planes, and a phasing offset between planes — but you do not need the formalism to grasp the essential design question, which is simply: how many satellites does continuous coverage take? That we can compute.

How many satellites? The coverage geometry

Strategy first. A user on the ground can only use a satellite that is high enough above their local horizon — say, at least some minimum elevation angle $\varepsilon$ (below that, buildings, hills, and the thick low atmosphere block the signal). We will find the size of the ground footprint one satellite can serve to that elevation, express it as a fraction of the Earth's whole surface, and then the minimum number of satellites for instantaneous global coverage is simply the reciprocal of that fraction. It is a floor — real constellations need more, for overlap — but it captures the physics and explains the fleet sizes.

Set up the geometry. Put a satellite $S$ at altitude $h$, so its orbital radius is $r = R_E + h$ with $R_E = 6{,}371\ \text{km}$. A user $U$ on the ground sees it at elevation angle $\varepsilon$ above the horizon. Consider the triangle formed by Earth's center $O$, the user $U$, and the satellite $S$. Its three interior angles are the nadir angle $\eta$ at the satellite (between straight-down and the line to the user), the angle at the user, which is $90^\circ + \varepsilon$ (the line to the satellite sits $\varepsilon$ above the local horizontal, itself $90^\circ$ from straight-down), and the Earth-central angle $\lambda$ at $O$. The three angles sum to $180^\circ$, so $$\lambda = 90^\circ - \varepsilon - \eta.$$ The law of sines in the same triangle relates the sides $R_E$ (opposite $\eta$) and $r$ (opposite the user's angle): $$\frac{R_E}{\sin\eta} = \frac{r}{\sin(90^\circ + \varepsilon)} = \frac{r}{\cos\varepsilon} \quad\Longrightarrow\quad \sin\eta = \frac{R_E}{R_E + h}\cos\varepsilon.$$ The footprint a satellite serves is the spherical cap of angular radius $\lambda$ around the point beneath it, and the fraction of the Earth's surface inside a cap of half-angle $\lambda$ is $$f = \frac{1 - \cos\lambda}{2}.$$

Worked Example: sizing a Starlink-altitude shell. Take $h = 550\ \text{km}$ (a typical Starlink altitude) and require a usable minimum elevation of $\varepsilon = 25^\circ$. First the nadir angle: $$\sin\eta = \frac{6{,}371}{6{,}371 + 550}\cos 25^\circ = \frac{6{,}371}{6{,}921}\times 0.9063 > = 0.9205 \times 0.9063 = 0.8343,$$ so $\eta = \arcsin(0.8343) = 56.5^\circ$. Then the Earth-central angle: $$\lambda = 90^\circ - 25^\circ - 56.5^\circ = 8.5^\circ.$$ The footprint fraction is $$f = \frac{1 - \cos 8.5^\circ}{2} = \frac{1 - 0.9891}{2} = \frac{0.0109}{2} = 0.0054,$$ i.e. one satellite serves about $0.54\%$ of the globe to a $25^\circ$ elevation. The minimum number of satellites to blanket the whole Earth at one instant is the reciprocal: $$N_{\min} \gtrsim \frac{1}{f} = \frac{1}{0.0054} \approx 184.$$ Sanity check the direction of every knob: lower the elevation requirement toward the horizon ($\varepsilon \to 0$) and $\lambda$ grows to $\arccos(R_E/r) = 23^\circ$, $f$ jumps to about $4\%$, and $N_{\min}$ falls to roughly $25$ — far fewer satellites, but each barely peeking over the horizon (useless for a real link). Demand a higher satellite (bigger $\varepsilon$) and you need more of them. The numbers are physical.

So a few hundred satellites is the floor for continuous global coverage at Starlink's altitude — and the real system flies thousands. Why the gap? Two reasons. First, our $N_{\min}$ assumed the footprints tile the sphere perfectly with no overlap, which is geometrically impossible; guaranteeing no gaps at every latitude, at every moment, as satellites and Earth rotate, takes two to three times the naive floor. Second, and larger: a communications constellation is not sized for coverage alone but for capacity. Each satellite can only carry so much data to so many users at once, so densely populated regions need many satellites overhead simultaneously to supply enough bandwidth. Starlink's thousands are driven more by capacity than by coverage — which is exactly why the constellation keeps growing even after coverage is complete.

🔗 Connection: Starlink versus OneWeb, a tale of two altitudes. The two best-known broadband constellations made opposite bets, and the coverage geometry above explains both (figures are approximate, Tier 2). Starlink flies low (~$550\ \text{km}$) at $53^\circ$ inclination, so each satellite's footprint is small ($\lambda \approx 8.5^\circ$) and it needs thousands — but low orbits mean short latency and, crucially, they decay in a few years if a satellite dies (Chapter 12's atmospheric drag, here a feature). OneWeb flies high (~$1{,}200\ \text{km}$) in near-polar orbits, so each satellite sees a much larger footprint and the system needs only ~$648$ satellites — but at $1{,}200\ \text{km}$ there is almost no drag, so a dead satellite lingers for centuries, a debris problem we take up in 33.6 and Chapter 35. Fewer-and-higher versus more-and-lower is the central constellation trade, and it is nothing but the altitude knob of Chapter 9 pushed to a system scale.

Inter-satellite links and the latency prize

A constellation whose satellites can only talk to the ground beneath them must have a ground station in view of every satellite at all times — impossible over oceans, poles, and remote land. The fix is to let the satellites talk to each other. Inter-satellite links — increasingly, laser crosslinks — let a constellation route a user's data across the network in space, satellite to satellite, until it reaches one that can see a ground gateway. Iridium pioneered this with radio crosslinks in the 1990s; modern Starlinks carry laser terminals that move data between satellites at the speed of light in vacuum. That last detail hides a genuine prize: light travels faster in vacuum than in the glass of a terrestrial fiber (where it crawls at about two-thirds of $c$), so for very long routes a mesh of satellite laser links can actually deliver data with lower latency than fiber on the ground.

And latency is where LEO constellations win outright. The signal delay to a satellite and back is set by distance and the speed of light. To GEO ($35{,}786\ \text{km}$) and back down, a one-way trip is $2 \times 35{,}786 / 299{,}792 \approx 0.24\ \text{s}$ — the famous quarter-second lag of a geostationary link (Chapter 9). To a LEO satellite at $550\ \text{km}$ overhead and back, it is $2 \times 550 / 299{,}792 \approx 0.0037\ \text{s}$ — under four milliseconds, some sixty times less. That difference is imperceptible in a broadcast but decisive for a video call, a trade, or an online game, and it is the single strongest reason to build a broadband service in LEO despite needing thousands of satellites to do what three could do from GEO.

🚪 Threshold Concept: reliability moves from the satellite to the system. For the whole book so far, theme 2 has said everything must work — a spacecraft is a fragile thing in a lethal place, and every part must be near-perfect because there is no repair. The constellation quietly inverts that. When you fly thousands of cheap, mass-produced satellites, you plan for some to fail; the reliability that matters is no longer the satellite's but the constellation's, and it is achieved through numbers and replacement rather than perfection. A single Starlink failing is a non-event — the mesh routes around it and a replacement launches next month — in a way that a single flagship GEO satellite failing never could be. This is the deepest consequence of cheap access: it does not just make satellites cheaper, it changes what reliability means, from a property you build into each unit to a property that emerges from the system. Once you see it, you see it everywhere the cost of a unit falls far enough — and it is the idea that makes mega-constellations thinkable at all.

🔄 Check Your Understanding 1. Raising the required minimum elevation angle $\varepsilon$ (demanding the satellite be higher in the sky) changes the number of satellites a constellation needs — which way, and why? 2. Starlink and OneWeb both provide broadband, but Starlink flies thousands of satellites while OneWeb flies hundreds. Give the geometric reason and the sustainability consequence.

Answers

1. It increases the number needed. A higher required elevation shrinks the Earth-central angle $\lambda$ (from $\lambda = 90^\circ - \varepsilon - \eta$), which shrinks each satellite's footprint fraction $f$, so more satellites ($\sim 1/f$) are needed to cover the globe. Demanding satellites higher in the sky is demanding more of them. 2. Geometric reason: Starlink flies much lower (~$550$ vs ~$1{,}200\ \text{km}$), so each satellite's footprint is smaller and it takes many more of them for continuous coverage. Sustainability consequence: the low orbit that forces Starlink to fly thousands also makes those satellites decay within a few years through atmospheric drag, whereas OneWeb's high orbit lets each dead satellite linger for centuries — so "more but self-cleaning" trades against "fewer but long-lived."

33.6 The economics and the debris question

Everything so far has a two-sided ledger, and this section is where we total it. On one side is an economic transformation with few precedents; on the other, a sustainability bill that is only now coming due.

The economics: mass production comes to space

The deepest change is not that satellites got small — it is that they became something spaceflight had never produced before: identical units, built in quantity. A traditional satellite was a one-off, each one hand-integrated and individually tested. A mega-constellation satellite rolls off an assembly line. That shift — from craft production to mass production — is the same one that reshaped automobiles a century ago, and it changes the cost structure completely. The first satellite of a design carries all the engineering cost; the ten-thousandth is nearly pure marginal manufacturing cost. Reported per-satellite costs for the largest constellations have fallen below a million dollars and continue dropping (Tier 2 — these figures are company claims, not audited), against the hundreds of millions of a bespoke GEO comsat.

This is the fifth theme in its purest form. Cheap, reusable launch made cheap satellites worth flying; cheap, capable electronics made them possible; and mass production made them cheap enough to lose. Recall the threshold concept of 33.5: when a unit costs a million dollars instead of a billion, you can afford to fly ten thousand and let a few percent fail — and that unlocks capabilities (continuous global broadband, daily imaging of the entire Earth, ubiquitous machine-to-machine connectivity) that no affordable number of flagship satellites could ever provide. The economics and the capability are the same fact seen from two sides.

💡 Intuition: the fleet, not the flagship, is the product. A traditional operator sold you a satellite's capability and guarded that one satellite like a crown jewel. A constellation operator sells you a service — connectivity, imagery, position — that no single satellite provides and no single satellite's failure can take away. The satellite has become a replaceable, depreciating component of a larger machine, the way a single server is to a data center. That is why constellation operators talk like factories and network operators, not like the space agencies of old.

The debris question: the bill comes due

Now the other side of the ledger, and the reason this chapter connects forward to Chapter 35. Filling low Earth orbit with tens of thousands of satellites raises a problem the flagship era never had to face at scale: space is a shared, finite environment, and it can be spoiled.

The concerns compound. There are simply far more objects now — every active satellite is a thing that can collide with another, and every collision creates a cloud of fragments that threaten everything else. Many small satellites, as we saw in 33.3, have no propulsion, so they cannot dodge a predicted collision or steer themselves down at end of life — they are passive targets that come down only when drag brings them. The altitude of a shell decides how forgiving it is: a low shell like Starlink's ($\sim 550\ \text{km}$) re-enters within a few years even if a satellite dies, so the orbit cleans itself; a high shell like OneWeb's ($\sim 1{,}200\ \text{km}$) holds its dead for centuries, accumulating risk. The nightmare scenario, in which collisions cascade into more collisions until a shell becomes unusable — the Kessler syndrome — is the subject of Chapter 35; for now, note only that the mega-constellation era is precisely what turned it from a theoretical worry into an operational one. Regulators have responded by tightening post-mission disposal rules — the old "de-orbit within 25 years" guideline is moving toward 5 years — and responsible operators now design for it: autonomous collision-avoidance maneuvers, "design for demise" so a re-entering satellite burns up completely, and deliberate low orbits that self-clean. There is even a cost to astronomy, as trains of bright satellites streak across telescope images and radio-quiet bands fill with traffic.

🧩 Productive Struggle. Before reading on, sit with the genuine tension, because it has no clean answer. A mega-constellation can bring affordable internet to communities that will never be reached by fiber, provide daily imagery that tracks deforestation and disaster, and connect sensors across the whole planet — real goods, unreachable by any other means. The same constellation crowds a shared orbit, raises collision risk for everyone, and may foreclose the night sky and certain orbits for generations. Who should decide how many satellites may fly, and by what rule? Is a low, self-cleaning shell a responsible design or a rationalization? There is no equation that settles this — it is the point where engineering hands the problem to law, economics, and ethics, which is exactly where Chapter 35 picks it up.

🔧 Engineering Reality: sustainability is now a design requirement, not an afterthought. For most of the space age, what happened to a satellite after its mission was somebody else's problem. That era is over. A modern constellation must show regulators a credible disposal plan before it launches: propulsion to deorbit (or an orbit low enough to decay), passivation to vent stored energy so a dead satellite cannot explode into fragments (Chapter 35), collision-avoidance autonomy, and a design that demises on re-entry. In mission-design terms (Chapter 29), end-of-life disposal has become a requirement that flows down into the vehicle — it sizes propellant, constrains the orbit, and shapes the structure, just like any other requirement. The cheapest satellite is no longer the one that ignores its own death.

🔄 Check Your Understanding 1. Why does mass production change not just the cost of a satellite but the strategy for making a constellation reliable? 2. Two constellations carry the same number of satellites, one at $550\ \text{km}$ and one at $1{,}200\ \text{km}$. Which poses the greater long-term debris risk, and why?

Answers

1. Mass production makes each satellite cheap enough to lose, so reliability shifts from making every unit near-perfect (the flagship strategy) to flying enough redundant, replaceable units that the system keeps working even as individual satellites fail — you plan for failures and replace them rather than preventing every one. Cheapness per unit is what makes redundancy-at-scale affordable. 2. The $1{,}200\ \text{km}$ constellation. At that altitude atmospheric drag is negligible, so a failed or dead satellite stays in orbit for centuries, accumulating collision risk; at $550\ \text{km}$ residual drag re-enters a dead satellite within a few years, so the orbit largely cleans itself. Same number of satellites, very different persistence of the hazard.

Mission Design Checkpoint: does a small-sat or constellation approach fit your mission?

For thirty-two chapters your Mission Design Review has assumed a single, traditional spacecraft. This chapter offers an alternative architecture, and a good mission designer at least considers it. Open your MDR and add a short Platform Trade note: could your mission be done better — cheaper, faster, sooner — as a small satellite, or as a constellation of them?

The decision, by track. Argue it explicitly; the reasoning is the deliverable, not the answer.

• Track A — GEO communications satellite. The classic case for the traditional big bus: one large GEO satellite covers a hemisphere from a fixed slot. But note the live alternative — a LEO broadband constellation (33.5) does the same job with far lower latency at the cost of flying hundreds to thousands of satellites. Record which you would choose and why, in the language of coverage, latency, and capital cost. This is the exact real-world trade between the GEO incumbents and Starlink/OneWeb.
• Track B — Lunar cargo lander. A lander itself is not a smallsat, but small satellites ride along: many lunar missions now carry CubeSat secondary payloads as scouts and relays. Note in your MDR whether a small orbiter or relay CubeSat, deployed en route or in lunar orbit, buys down risk for your lander (a MarCO-style communications relay for the descent, for instance).
• Track C — Mars science orbiter. MarCO (33.2) proved that CubeSats reach Mars. Record whether your science could be done — or augmented — by a small satellite or a small pair, and what the COTS/radiation trade (33.3) costs you at Mars, where the belts, distance, and dead magnetic field change the calculus.
• Track D — Asteroid rendezvous. Small-body missions are a natural fit for smallsats (NASA's NEA Scout was a 6U CubeSat bound for an asteroid). Note whether a CubeSat-class spacecraft, riding as a secondary payload on a larger interplanetary launch, could accomplish your rendezvous at a fraction of the cost.

Whatever you decide, write one sentence recording the platform choice and the driving reason — and, if you chose a constellation, a first estimate of how many satellites your coverage needs, using the geometry of 33.5. This choice also feeds forward to your disposal plan in Chapter 35: a constellation owes the orbital environment a credible end-of-life plan (33.6).

The code. Add a convenience helper, astrotools/constellation.py, that turns the coverage geometry of 33.5 into a sizing tool. (Like Chapter 9's orbit_catalog.py, this is a convenience helper for mission design, not one of the canonical astrotools modules — keep your Chapter 6 orbits.py as the real workhorse for orbital mechanics.)

import math

R_EARTH = 6371.0  # km, mean radius

def earth_central_angle(alt_km, min_elev_deg):
    """Earth-central angle (deg) of the footprint a satellite at altitude alt_km
    can serve down to a minimum elevation of min_elev_deg above the horizon."""
    r   = R_EARTH + alt_km
    eps = math.radians(min_elev_deg)
    eta = math.asin((R_EARTH / r) * math.cos(eps))   # nadir angle
    lam = math.pi / 2 - eps - eta                    # Earth-central angle
    return math.degrees(lam)

def coverage_fraction(alt_km, min_elev_deg):
    """Fraction of Earth's surface inside one satellite's footprint (spherical cap)."""
    lam = math.radians(earth_central_angle(alt_km, min_elev_deg))
    return (1 - math.cos(lam)) / 2

def min_satellites_for_global(alt_km, min_elev_deg):
    """Floor on satellites for instantaneous global coverage (ignores overlap;
    the real number is a few times larger)."""
    return math.ceil(1 / coverage_fraction(alt_km, min_elev_deg))

for h in (550, 1200):
    lam  = earth_central_angle(h, 25)
    frac = coverage_fraction(h, 25)
    n    = min_satellites_for_global(h, 25)
    print(f"h={h} km: lambda={lam:.1f} deg, cover={frac*100:.2f}%, >= {n} sats")
# Expected output:
# h=550 km: lambda=8.5 deg, cover=0.54%, >= 184 sats
# h=1200 km: lambda=15.3 deg, cover=1.77%, >= 57 sats

The output reproduces the hand calculation of 33.5 — about $184$ satellites for continuous coverage at Starlink's altitude, but only $57$ from OneWeb's higher shell, because the higher satellites each see a bigger footprint. Feed your mission's altitude and required elevation into this helper and you have a first, defensible estimate of constellation size for your MDR — the number that, through the launch and cost chains of Chapter 29, decides whether a constellation is affordable at all. In Chapter 40 this platform choice becomes one line of your defended design.

Summary

The small-satellite and constellation era is the fifth theme — cheap access changing everything — made concrete. Carry these forward:

Key relationships (coverage geometry):

Numbers worth remembering: $1\ \text{U} = 10\ \text{cm}$ cube, $\lesssim 1.3\ \text{kg}$; a 3U CubeSat makes only a few watts orbit-averaged; deployer ejection ~$1$–$2\ \text{m/s}$; rideshare ~$\$5{,}000/\text{kg}$ versus a dedicated small launcher ~$\$25{,}000/\text{kg}$ (Tier 2); at $550\ \text{km}$, $\varepsilon = 25^\circ$ gives $\lambda \approx 8.5^\circ$, one satellite covers ~$0.54\%$, so ~$184$ satellites (floor) for continuous coverage; GEO round-trip latency ~$0.24\ \text{s}$ one way versus LEO ~$4\ \text{ms}$. astrotools addition: constellation.py (coverage helper).

Spaced Review

Retrieval strengthens memory. Answer from memory before checking, then look back at the cited chapter. This chapter revisits Chapter 9 (orbit types) and Chapter 29 (mission design).

1. (§33.5, Ch. 9) A single LEO satellite cannot provide continuous service to a fixed city, but a single GEO satellite can. State the Chapter 9 reason, and explain why a constellation is LEO's answer to it.
2. (§33.6, Ch. 9) Chapter 9 called low orbits "self-cleaning." Which perturbation makes them so, and why does that make a $550\ \text{km}$ constellation more sustainable than a $1{,}200\ \text{km}$ one?
3. (§33.4 & Checkpoint, Ch. 29) In mission-design terms, why are orbit selection and launch-vehicle selection coupled when you choose rideshare — i.e., why can a cheap launch quietly cost you elsewhere?
4. (§33.6, Ch. 29) Chapter 29 defined a requirement as a testable "shall" statement that flows down into the design. Explain how "dispose of the satellite responsibly at end of life" has become such a requirement, and name two vehicle choices it drives.

Answers

1. A LEO satellite moves at ~$7.7\ \text{km/s}$ relative to the ground and is over any city for only a few minutes per pass, with hours of gaps; a GEO satellite's one-day period matches Earth's spin, so it hangs motionless over one longitude and a fixed antenna always sees it. A constellation answers this by flying enough LEO satellites that, as each races past, another is always coming over the horizon — the continuous coverage GEO gets from altitude, LEO gets from numbers. 2. Atmospheric drag (Chapter 12's orbital decay): residual atmosphere at low altitude saps orbital energy and brings a dead satellite down within a few years, whereas at $1{,}200\ \text{km}$ drag is negligible and a dead satellite persists for centuries — so the low constellation cleans itself while the high one accumulates debris. 3. Rideshare is cheap only because you take the provider's orbit; if your mission needs a different altitude, inclination, or local time, you must pay for a dedicated launch, carry propulsion to maneuver after release, or buy an orbital-transfer-vehicle ride — so the "cheap" launch decision is really a joint decision with the orbit, and choosing the wrong pairing costs mass or money elsewhere. 4. Regulators now require a credible end-of-life disposal plan before launch, so "shall dispose responsibly" is a testable requirement that flows down into the vehicle: it drives (any two of) propulsion sizing to deorbit, choice of a low self-decaying orbit, passivation to prevent post-mission explosions, and a design-for-demise structure that burns up on re-entry.

What's Next

We have now seen the whole modern spectrum of what you can put in orbit — from a single flagship to a constellation of thousands — and the economics that decide which makes sense. And we have seen, in MarCO, that the smallest satellites can already reach other planets. That is the perfect place to turn from the kinds of missions to the single hardest one this book will attempt in full: a mission to Mars. In Chapter 34 every thread of the book converges — the delta-v budget and rocket equation of Part I, the interplanetary trajectory and launch window of Part II, the propulsion of Part III, the spacecraft systems of Part IV, and the mission-design discipline of this Part V — into one integrated case study: how you actually get to Mars, land on it ("seven minutes of terror"), live there, and come home. The tools are all in your hands now; next we point them at another world.