Forty-five lessons of this course have been about quantities. Decibels that had to add up, bandwidths that had to be labelled one-sided or two-sided, a marketing claim of “4× range” that turned out to need 12.04 dB and be handed 9.03. A lesson about careers is where a course like this usually goes soft — where the arithmetic stops and the advice starts, and you are told to develop strong communication skills.
So here is the ground rule for this lesson. Every role below is described by its artefacts — the specific documents, measurements, files and numbers that a person in that role produces and is judged on — and by the mathematics they use on an ordinary Tuesday. No personality types, no salary tables, no encouragement. Where this course taught the thing they use, the lesson names the module; where it did not, the lesson says so plainly, because a map that shows only the roads it likes is worse than no map.
Five roles, one radio. They are not five opinions about the same job: they own different, adjacent pieces of the same signal, and most of what goes wrong in real projects goes wrong at the seams between them. That is the last section, and it is the one that matters most.
The RF engineer owns the physical layer as hardware. Everything upstream of the antenna port and downstream of the digital baseband — power amplifiers, low-noise amplifiers, filters, mixers, matching networks, the board itself and the enclosure it sits in — is theirs. This is the role where the electromagnetics of Module 1 stops being a picture and starts being a soldered object that either meets its specification or does not.
The day. Mostly measurement, and mostly of things that disagree with the simulation. A vector network analyser to see how much power a port reflects rather than accepts; a spectrum analyser to find the spurious emission that appeared when the amplifier was pushed 2 dB harder; a noise-figure measurement to find out why the receiver is 1.5 dB worse than the cascade calculation promised. Then a layout change, a component substitution, a new board, and the same measurements again. Design cycles are set by how long a board takes to fabricate, which is why RF engineers care so much about getting the first one nearly right.
The artefacts. Schematics and layouts. S-parameter files (Touchstone .s2p and friends) for every component and every measured port. A test report with plots: gain against frequency, return loss, noise figure, output power against input power showing where compression begins, adjacent-channel leakage, and error vector magnitude. A calibration record, because an uncalibrated measurement is not a measurement.
The mathematics. Decibels, constantly, and the discipline of knowing whether a ratio is power or amplitude (M2-L4). Complex impedance and reflection coefficients. The noise-figure cascade, which is the formula this role lives inside:
When it goes wrong. It goes wrong late and expensively. A radio that passes on the bench fails regulatory emissions testing because a harmonic three octaves up leaks through a seam in the case. A part goes end-of-life and the replacement is 0.4 dB worse, which nobody notices until the field returns start. A power amplifier that met its specification at 25 °C droops at 70 °C inside a sealed enclosure on a roof. The characteristic failure of this role is that the fix requires new hardware, so the schedule, not the physics, is what hurts.
Adjacent roles. Analogue and mixed-signal IC design (the same problems one abstraction level down, in silicon); antenna design, which is a specialism in its own right and much more electromagnetic-solver work than this course suggests; EMC and regulatory compliance; hardware test and validation. Systems engineering is the natural step sideways for an RF engineer who wants to decide what to build rather than make it work.
What this course gave you, and what it did not. Modules 7 and 9 are the anchors. M7-L2 is the closest thing here to an RF specification sheet — gain, radiation pattern, beamwidth, polarisation, VSWR and return loss, and what a mismatch actually costs. M7-L1 and M7-L3 cover radiation and the common antenna families; M7-L4 covers arrays. M9-L1 is the noise-figure lesson, worked as a cascade with real component numbers. Not covered here at all: S-parameters as a formalism, Smith charts, and error vector magnitude — while amplifier compression and intermodulation appear only as interference mechanisms in M9-L3, not as design quantities you budget. Those four are the honest gap between this course and a first RF job, and they are all learnable from Pozar’s Microwave Engineering plus a cheap vector network analyser.
The systems engineer owns the numbers that constrain everyone else. Nothing on their desk is a physical object. Their job is to decide what the link must be — which band, how much bandwidth, which waveform and numerology, how much transmit power, which antenna configuration, what error rate is acceptable at what data rate — and then to defend those decisions with arithmetic when the hardware team says the amplifier cannot do it and the product team says the coverage is not good enough.
The day. Spreadsheets and simulation, punctuated by arguments. A link budget in the morning that says the design closes with 6 dB of margin; a message at lunchtime that the antenna gain has dropped 2 dB because the industrial design changed; an afternoon spent deciding whether to spend those 2 dB on a lower-order modulation, a stronger code, or a smaller cell radius. Then a link-level simulation to check that the choice actually holds the target block error rate over a fading channel, because M8-L3 established that an average signal-to-noise ratio does not tell you the error rate in fading.
The artefacts. The link budget, which is the central document of the role — a line-by-line accounting from transmit power to receiver sensitivity with every loss and every margin named and sourced. A simulation model, usually MATLAB or Python, that produces error-rate curves against signal-to-noise ratio for the candidate configurations. A parameter table that the implementation teams work from. A trade-study memo: three options, the quantities that separate them, and a recommendation someone can disagree with on the evidence.
The mathematics. Everything in Modules 6, 8, 9 and 10 at once. Shannon and Nyquist as bounds you quote to say what is impossible (M6-L1, M6-L2). Spectral efficiency and the bandwidth–power–error-rate trade (M6-L3, M6-L4). Path-loss models and margins for fading and shadowing (M8-L1, M8-L4, M9-L4). Probability, because every real requirement is a percentile: coverage at the 95th percentile of locations, not at the average.
When it goes wrong. Quietly, and in one direction. Margins are the currency of the role and they are spent by other people — a 2 dB antenna concession here, a 1 dB implementation loss there, a fade margin chosen because it was the number in the last project’s spreadsheet rather than because anyone measured that environment. The result is a design that closes on paper and fails in the field, and it fails everywhere at once because it was one number that was wrong. M9-L4’s remark that a drive test replaces a guess two months later, and lets you shrink the margin honestly, is the whole professional discipline of the role compressed into a sentence.
Adjacent roles. Modem and physical-layer architecture; standards work, which is systems engineering conducted in public with competitors in the room; network planning, which is systems engineering applied to a geography rather than a product; performance analysis. Systems engineers also end up as the translators between the RF, digital and software teams, which is not a soft skill so much as a consequence of being the only role that holds the whole budget.
Course anchors. Modules 6, 8, 9 and 10 — more of this course than any other role uses. M9-L4 for the link budget, M6 for the bounds and the trades, M8 for the channel, M10-L2 for what a numerology choice actually is (subcarrier spacing against symbol duration, cyclic prefix against delay spread) and M10-L4 for duplexing. If you enjoyed M9-L4 — and specifically if you enjoyed arguing with its margins — this is the role you enjoyed.
Everything the previous two roles build has to interoperate with equipment they have never seen, made by a company that may not exist yet, and keep working with equipment sold ten years ago. The protocol engineer owns that: the state machines, message formats, procedures and timers that let two radios agree on what is happening. Their medium is text — specifications — and their loyalty is to the specification rather than to the product.
The day. Reading. A great deal of reading, of documents written to be unambiguous rather than pleasant. Then implementing or reviewing a state machine against that text, then testing it against another implementation and discovering that the other side read a sentence differently. Conformance and interoperability testing is a large part of the working life: a test case says the device shall respond within so many milliseconds with such a message, and the log says otherwise, and somebody has to decide whether the device is wrong, the tester is wrong, or the specification is ambiguous. The third answer is more common than newcomers expect.
The artefacts. Specification text and change requests. State-machine diagrams and message-sequence charts. Protocol logs and packet captures, read at the bit level. Conformance test cases and their results. If the role is standardisation rather than implementation, the artefact is a contribution document argued through a working group, and the skill is drafting language that says exactly one thing.
The mathematics. Less continuous mathematics than any other role here, and more discrete: state machines, timers, sequence numbers, combinatorics of message orderings, and the arithmetic of overhead — how many bits of header per payload, how much airtime a procedure costs. What replaces the calculus is a very high tolerance for detail and an instinct for the case nobody wrote down.
When it goes wrong. Backward compatibility is the hard constraint, and it is unforgiving in a way product engineers rarely believe until they meet it. You cannot fix a deployed protocol; you can only add to it. Every legacy behaviour that must keep working is a permanent restriction on every future design, which is why standards accumulate optional features and capability negotiation rather than clean redesigns. M11-L4 has two clean examples of the constraint shaping the technology: NB-IoT is 180 kHz because that is exactly one LTE resource block (M10-L2), so it fits an existing carrier, an existing filter and an existing band plan — and Bluetooth 5 added two new physical layers on the same 40 channels rather than replanning the band. Elegance was not the objective; not breaking anything was.
Adjacent roles. Embedded firmware and modem software; test engineering and test-equipment development; regulatory and spectrum policy; systems engineering. Standards delegates are frequently senior systems engineers, because the arguments in a working group are usually about whose link budget the specification should favour.
Course anchors, stated honestly. Modules 10 and 11 are where this course meets the standards world. M11-L1 covers the 802.11 generations and contention for the medium; M11-L2 covers cellular from 2G to 4G, including the LTE architecture where each base station holds its own state machine and talks to its neighbours over the X2 interface; M11-L3 covers 5G NR release by release, including the Non-Standalone mode that is entirely a backward-compatibility artefact; M11-L4 covers the short-range and IoT specifications. M10-L4 shows band plans as compatibility constraints. But this course does not teach protocol engineering: there is no conformance testing here, no message-sequence chart, no reading of an actual specification. If this role attracts you, the honest next step is to download a real specification and read it — and the fact that you can, for free, is the subject of the resources section below.
This is the role that decides what the receiver does with the samples. Given a sequence of complex numbers corrupted by noise, a channel that smears them together and an oscillator that is not quite the right frequency, recover the bits. The industrial version of the job is algorithm design for a modem; the research version is proving things about the algorithm and publishing them. Both are the same mathematics, and it is the heaviest mathematics in this lesson.
The day. A simulation of a link you can break on purpose. You add a channel model, a frequency offset, a phase noise process, then measure the error rate against signal-to-noise ratio and compare it with the theoretical curve. When the gap is 1.5 dB you find out where the 1.5 dB went, and it is nearly always one of the estimators. Then you rewrite the algorithm in fixed point, discover it needs 14 bits where you budgeted 10, and negotiate. In research the loop is the same but ends in a paper rather than a hardware handoff.
The artefacts. Simulation code and its error-rate curves — always plotted against the theoretical bound, because a curve without a bound to compare it against says nothing. Algorithm specifications precise enough for someone else to implement. Fixed-point analyses. Papers, if that is the path, and the reviews of other people’s papers that come with it.
The mathematics. Probability and random processes first, then estimation and detection theory — which is the formal name for the two questions that fill the day: what was the channel (estimation) and which symbol was sent (detection). Linear algebra, unavoidably, once there is more than one antenna. Fourier analysis. Coding theory if you go that way. Optimisation. Increasingly, machine learning, honestly applied and honestly compared against the classical estimator, which is more often than not still winning.
When it goes wrong. Two characteristic failures. The first is an algorithm that works beautifully in simulation and dies on hardware, because the simulation contained no phase noise, no automatic-gain-control transient, no quantisation and no clock drift — the model was kinder than the world. The second is a real improvement of 0.2 dB that costs three times the silicon area, which is a correct result and a rejected proposal. In research the failure mode is different again: an idea that is genuinely novel, provably better under assumptions no deployed system satisfies, and therefore never implemented.
Adjacent roles. Modem and digital design (the same algorithms as gates); systems engineering; radar and sensing, which is the same estimation theory pointed at a different question; positioning; academia. The research path normally requires a PhD, and the honest reason is not credentialism — it is that the role is judged on original results, and a PhD is the standard apprenticeship in producing them.
Course anchors. Modules 5, 6 and 9. M5-L2 through M5-L4 build keying, constellations and QAM; M5-L3’s decision regions are detection theory in its simplest possible dress. M6-L1 gives Shannon; M6-L2 gives Nyquist and symbol rate; M6-L4 sets the bandwidth-against-power-against-error-rate trade you will spend a career inside. M9-L1 gives noise and signal-to-noise ratio, and M9-L2 is the error-rate lesson, worked in terms of energy per bit over noise density — the axis every paper in this field plots against. Two further anchors sit outside those modules: M8-L3, where multipath produces the intersymbol interference that motivates equalisation, and M10-L3, where channel estimation and pilot overhead appear as MIMO’s real cost. Beyond that, this course is deliberately light on the mathematics of estimation, and Tse and Viswanath or Goldsmith is where it continues.
Everything so far concerned one link. This role concerns thousands of them, sharing spectrum, in a real geography, for paying customers who complain. The planning half decides where cells go, what they radiate, at what tilt and azimuth, on what frequencies. The optimisation half looks at what the live network is actually doing and changes it. It is the most empirical role in this lesson and the only one where the answer is measured rather than derived.
The day. Data. Counters from the network by the million, aggregated into key performance indicators — accessibility, retainability, throughput, handover success, and the ratio of signal to interference plus noise across the coverage area. A planning tool with terrain and building data, predicting coverage over a map. Drive tests and, increasingly, measurements reported by the handsets themselves. Then a change: two degrees of downtilt on one sector, a neighbour list corrected, a frequency reassigned, a handover threshold moved — and a week of watching the counters to see whether it helped or moved the problem next door.
The mathematics. Statistics before anything else: distributions of measurements, percentiles, and enough discipline to tell a real change from noise in a week of counters. Propagation models and their calibration against measurement (M8). Interference and reuse arithmetic (M9-L3, M10-L1, M11-L2). Traffic and queueing theory for capacity forecasting — how many users at what offered load before the cell saturates. Geometry, unglamorously and constantly.
When it goes wrong. It goes wrong locally and visibly, and someone is already complaining. The characteristic trap is that every fix is a trade against a neighbour: tilt down to cure overshoot and you open a coverage hole at the cell edge; add power to fill it and you raise interference in the cell beyond. Optimisation in an interference-limited network is a zero-sum game played on a map, and the second characteristic trap is optimising the average while the 5th percentile of users — the ones who complain — gets worse.
Adjacent roles. Network operations and performance engineering; spectrum and site acquisition; automation and data engineering, which is where much of this role is heading as the counters outgrow spreadsheets; systems engineering. It is also the most common entry point into wireless for people without a signal-processing background, because it is learnable from the data.
Course anchors. Modules 8, 9 and 11. M8-L1 gives free-space path loss and where the log-distance model comes from; M8-L2 gives reflection, diffraction and scattering, which is why the map matters at all; M8-L4 gives path-loss exponents for real environments and the indoor and outdoor cases separately. M9-L3 is the interference lesson and the one that defines the ratio above; M9-L4 supplies the margins. M11-L2 covers cellular reuse and handover. Not in this course: key performance indicators as named quantities, planning tools, and traffic theory. The concepts are all here; the operational vocabulary is not.
Laid out together, so the differences are visible rather than asserted. Read the last column as an instruction: it is where this course sends you next.
| Role | Owns | Main artefact | Maths | Tools | Fails when | Course anchors |
|---|---|---|---|---|---|---|
| RF engineer | Everything between baseband and the antenna port | Board plus a measured test report | Decibels, impedance, noise cascades | Network and spectrum analysers, EM and circuit simulators | Late — the fix needs new hardware | M7 (all four), M9-L1, M2-L4 |
| Systems engineer | The numbers everyone else is constrained by | Link budget and a trade study | Probability, bounds, budgets, percentiles | MATLAB or Python, link-level simulation, spreadsheets | Quietly — a margin spent by someone else | M9-L4, M6, M8, M10-L2, M10-L4 |
| Protocol engineer | Agreement between radios that never met | Specification text, state machines, test cases | Discrete: states, timers, overhead arithmetic | Protocol analysers, log tools, conformance testers | Permanently — you cannot un-deploy a protocol | M11 (all four), M10-L4 |
| Signal processing | What the receiver does with the samples | Algorithm plus an error-rate curve against the bound | Estimation, detection, linear algebra, coding | Simulation, fixed-point analysis, sometimes an SDR | On contact with hardware the model omitted | M5, M6, M9-L2, M8-L3, M10-L3 |
| Planning and optimisation | Thousands of links sharing spectrum on a map | Cell plan, parameter change, KPI report | Statistics, propagation, traffic and queueing | Planning tools, drive-test kit, counter databases | Locally — every fix is a trade with a neighbour | M8, M9-L3, M9-L4, M11-L2 |
A real design flows in one direction: standards define what is permitted → systems engineering decides what to build within it → RF and digital implement it → planning and optimisation deploy it and measure what actually happened. Each arrow is an interface, and in every project the expensive problems live on the arrows rather than inside the boxes. Four of them, concretely:
The useful conclusion is not that engineers should communicate better. It is more specific and more actionable: every quantity handed across an interface needs a stated owner, a stated tolerance and a stated source. A link budget line that reads “implementation loss 2 dB” with nobody’s name against it is a defect in the document, not a detail.
The most reliable signal available to you is which lessons of this course you did not want to stop reading. Not which you found easy — which you found interesting when they got hard. Five honest mappings:
TS 38.211 and read one clause end to end, then capture and decode real packets from something you own.Two of those five need no permission, no employer and no money: the link budget and the simulation. Both are also, not coincidentally, the artefacts that get people hired.
Take the conclusion first, because it saves money: in most of wireless engineering, certifications are not the credential that matters. Hiring in RF, systems, protocol and signal-processing roles is done on demonstrated work — a link budget you built, a board you measured, a specification you implemented, a paper you wrote. The one genuine exception is enterprise WiFi and network operations, where vendor-neutral and vendor-specific certifications do carry real weight with employers.
Primary sources beat everything else, and the important ones are free.
3gpp.org, along with the working-group meeting documents and contributions. Start with the physical-layer series and read one document rather than surveying many: TS 36.211 for LTE physical channels and modulation, TS 38.211 for the 5G NR equivalent. They are exactly the specifications M11-L2 and M11-L3 describe, and reading twenty pages of one teaches more about protocol work than any summary.P.1411 and P.1238 are the formal versions of what M8-L4 taught by hand — and the Radio Regulations are where band allocations actually come from.One caution about all of the above: the specifications and Recommendations are revised continuously, and a textbook edition on cellular standards ages faster than one on electromagnetics. Check the version, and prefer the primary document when the two disagree.
This is the forty-sixth and last lesson of Wireless 101. Before the map of roles closes, here is the map of the ground you covered, because the arc is easier to see from the end than from the middle.
The through-line, stated once: nothing in wireless is free and everything is an exchange. Bandwidth for power, power for error rate, range for data rate, spectral efficiency for hardware cost, elegance for backward compatibility. Every lesson in this course was one instance of that, and every role in this lesson is a job description for someone who negotiates it on a particular axis.
3gpp.org (TS 36.211, TS 38.211), many IEEE 802 standards are free through the GET Program, and the ITU-R P-series is what planning tools implement. Read one specification rather than ten summaries.You have finished Wireless 101 — all 46 lessons, Modules 1 through 12. You started with what a wave is and a spectrum to divide up; you are leaving able to read a link budget, defend a modulation choice, explain why a range claim needs a path-loss exponent attached, and say what an OFDM numerology actually settles. That is a real working foundation, and it is the same foundation the five roles above are built on.
What is genuinely still to come: the syllabus specifies eleven hands-on labs — one for each of Modules 1 to 11 — plus the capstone project, “Design a Wireless Link”, and none of the twelve is built yet. The capstone is the one to want: a scenario with constraints, and you choose the frequency, modulation, coding and antenna, compute the link budget, estimate capacity and coverage, and defend the design. Until it exists, the best possible next step is to do it anyway, on a link you care about, with M9-L4 open beside you.