| Episodes drawn on | 265 |
| Cited here | 51 |
| Claims extracted | 115 |
| Gathering | concept union |
| Other themes | |
Radar and radio-frequency engineering are treated within electronics as a specialism beyond analog circuit design rather than a subdivision of it, forming a third tier above general electrical engineering and analog work.[66] The field is commonly described as black magic, a characterisation that reflects its working method rather than any absence of physics: the number of parameters entering the equations that govern a radio section is large enough that the practical approach is iterative tweaking to bring an impedance to a match rather than calculation to a result, and the instrumentation required to observe what is happening is expensive.[226][479] Radar itself is one application of that method, distinguished as a sensor by all-weather operation, range in three dimensions, and the ability to characterise targets by their Doppler signature.[179] The same techniques and instruments recur across through-wall imaging, millimetre-wave communication, software-defined radio and amateur microwave practice.[115][228][161][107]
Defining the bands
Microwave was conventionally taken to begin around 1000 MHz, but defining it by an arbitrary frequency is less useful than defining it by design consequence: the meaningful boundary is whether transmission-line behaviour must be accounted for, and that boundary falls at different frequencies for discrete circuits than for integrated ones.[107][459] Millimetre wave conventionally denotes frequencies above 30 GHz, where the free-space wavelength falls below about ten millimetres; inside silicon the dimensions are far smaller, with wavelengths of hundreds of microns near 100 GHz and about 1.6 millimetres at 90 GHz in a dielectric of relative permittivity 4.2.[228] Terahertz operation sits between millimetre wave and visible light, with wavelengths in the hundreds of microns — below visible light but well beyond millimetre wave — and the same region was historically described as far or long-wave infrared.[729]
Method
Matching by iteration
On a small radio module the radio-frequency section is the hard part of the design, both because it requires expensive instrumentation and because so many parameters enter the performance equations that the working method is to tweak components until the impedance matches.[226] Radio chipsets differ from one another in idiosyncratic ways and are individually particular about layout, so radio boards need frequent respins; a two-week fabrication turnaround per iteration is a real obstacle, which is what makes in-house milling attractive for radio and analog work despite its compromises.[280] Entering the field involves capital outlay before competence — a three-thousand-dollar vector network analyser bought together with a shelf of books — and radio frequency is deep in the same way firmware is deep, so the equipment does not shorten the learning.[479]
Engineers who combine printed circuit board layout and embedded firmware with a working radio-frequency and analog background are rare, and the combination pays off directly in review: someone holding it reads a layout and identifies a trace running alongside another, or a structure that will behave as an antenna, before the board is built.[11] The failure such a review catches is easy to create and easy to miss — a coaxial connector whose shield is tied to the board ground makes the whole cable part of the radiating structure, so that moving the cable makes the measured response move with it.[464]
Diagnosis by comparison and observation
Because bench instruments cannot exercise a complete radio path, amateur microwave groups diagnose home-built equipment by comparative field measurement: receiving six decibels worse than another station using the same antenna indicates a fault, and a deficit that appears equally on transmit and on receive points at the antenna rather than at the transceiver.[107] An equivalent method for understanding what a circuit board emits is to write code that toggles a pin on and off at a chosen rate using a shift register while watching the result on a software-defined radio, since playing with the code and observing the spectrum builds intuition for how generated signals, intentional and unintentional, fit together.[667] Comparing antennas empirically requires hardware that will accept them: a test box built to explore the tradeoffs between omnidirectional and directional antennas housed five different antennas, and the binding constraint on the radio chosen was finding an inexpensive, small USB wireless adapter with an SMA connector, since most such devices integrate their antenna or use a chip antenna.[420]
Radio experiments that transmit on licensed bands, such as spoofing a cellular signal, are conducted inside a Faraday cage, which contains both the intended emissions and the harmonics a crude transmit chain produces.[391] Production testing can be arranged along similar lines by generating the signal the product needs: on one FM-receiver line, the same single-board computer that flashed the customer’s chosen frequency into the unit’s EEPROM also drove a software FM transmitter on one of its general-purpose pins to emit a test tone on that frequency, confirming the unit heard it at the moment of programming.[350] High-power equipment brings a separate class of constraint, since transmitters carry human-exposure separation requirements in their installation instructions: a unit in the region of 140 watts specified that its antenna be installed no closer than about 1.4 metres to anything else, people included.[520]
Instrumentation
The vector network analyser is the critical instrument for measuring radio-frequency and antenna behaviour, and its cost is high enough that practitioners have built their own rather than buy one.[446] Single-box analysers reach about 70 GHz, extensible to 72 GHz; above that, measurement moves to millimetre heads mounted at the device covering the 70 to 120 GHz bands, and the heads are a necessity rather than a convenience, because at 100 GHz a one-metre cable leaves no signal at its far end.[533] A vector network analyser is itself a radar device, measuring the reflection from a stimulus it emits, which is why the same add-on hardware could plausibly serve both network-analysis and radar functions on one platform.[214]
The microwave component industry is populated by very small specialist companies supplying waveguides and machined parts whose function is to take a signal and make it turn a corner; as operating frequencies rise these parts shrink, so the limiting skill migrates from machining toward micro- and nano-machining.[714] Component availability is a first-order constraint on millimetre-wave design: few companies make millimetre-wave parts, and one manufacturer of such equipment solved its sourcing problem by mixing the signal down inside the box to a frequency where parts can be bought and doing the work there.[714]
Radar
Principles
Every radar is fundamentally clocking the speed of light — the round-trip time between emitting a wave and receiving its reflection — so the transmitter’s firing instant must be known precisely; in an ultra-wideband linear-FM system the trigger is taken at the start of the up-ramp, and that timing is what makes range-Doppler processing, synthetic aperture and high signal-to-noise ranging possible.[214] Low-power radars reach long ranges through pulse compression: they transmit long linear-FM pulses and recover range resolution by Fourier processing of the return, so transmitted power can be very low while the processing supplies the capability.[214] Radar resolution is proportional to occupied bandwidth, which is why the move to higher carrier frequencies benefits radar for the same reason it benefits communications.[553]
Radar returns range and angle unambiguously, whereas an optical or camera-based system requires an algorithm to infer range from the image; radar also resolves rate of change effectively instantly, which is why closing speeds of a hundred kilometres per hour between two vehicles present no detection difficulty.[115] Its distinguishing capability as a sensor is the combination of all-weather operation, range in three dimensions, and target characterisation through Doppler signature.[179] When selecting a sensor for a problem, radar therefore belongs on the shortlist alongside infrared, LIDAR and acoustic methods, the practical position being not that it must be used but that it is achievable and should be remembered as an option.[179]
Timing and synchronisation
The hard part of building a radar from software-defined radio hardware is timing and synchronisation between transmitter and receiver, because such platforms are designed for streaming applications rather than for the deterministic, triggered acquisition a radar requires.[179] Exporting a trigger is as useful as importing one: a TTL line that goes high when the receiver starts acquiring is sufficient, because radar systems are designed either to supply or to accept the trigger.[214] A receiver can also be synchronised to a transmitter it is not wired to by using a second radio as a reference — one antenna pointed at the transmitter to capture the direct path, a second collecting the scattered return, and the two recordings cross-correlated — which supplies the timing reference without a physical trigger line.[179] That arrangement scales to installations with no receiver at all: a bistatic space-surveillance radar can run as a continuously transmitting fan beam of nearly a megawatt, with anything crossing the beam scattering energy back to earth and separate receive stations geolocating the scatterers.[179]
Imaging through obstacles
Through-wall radar sets its frequency by trading penetration against aperture size. Any microwave signal passes through concrete to some degree, so the question is how much; typical through-wall work sits at UHF around 440 or 1200 MHz, and an operating frequency of 3 GHz is about the highest ever attempted.[115] That frequency was chosen for resolution: dropping to 400 MHz would require an aperture roughly ten times the eight-and-a-half-foot aperture used, making the antenna larger than the room being imaged.[115] At the extreme of resolution, a 680 GHz imaging radar with 30 GHz of bandwidth achieves centimetre resolution and can image concealed objects from a standoff of several metres, but the limitation there is interpretation rather than hardware, since a trained operator is needed to identify what is in the image and a part-time operator will not classify the threat reliably.[483]
Passive and backscatter techniques
Radar-illuminated backscatter exists to move the power budget off the target device: a self-powered transmitter must run from a battery sized for however long the installation must last, whereas a passive modulator can be flooded with high-power radar from outside and simply reflect the modulated return, which is why permanently embedded listening devices are built passive.[182] Such an implant can be attached to the red line of an analog monitor cable without transmitting anything of its own — it is illuminated with a radar signal, the modulated return is captured, and the dot clock is recovered from it to reconstruct the screen contents.[182] The same physical principle supports ambient backscatter communication, in which an antenna and a board either absorb or reflect existing radio energy, such as a television station’s transmission at the antenna’s resonant frequency, to convey information between two devices that carry no batteries and draw their operating power parasitically from the same field.[160]
Learning and teaching radar
Radar expertise is frequently self-taught rather than programmatic: at a university with no radar reputation, an ambitious graduate student may simply be given room to teach the subject to themselves, which is how some practitioners enter the field.[115] Entry also comes through hands-on apprenticeship rather than coursework — a high-school intern at an imaging-radar company did wiring, cables, connectors and boards, was taught to use a microwave network analyser to sweep amplifiers, and travelled to repair the azimuth actuators of radar ranges.[115] The subject is teachable as a project-based one to students with no prior background: a high-school outreach programme had participants build the coffee-can radar, which works well enough to measure passing cars and people running down the street.[115]
That coffee-can radar course — a project-based class that builds a working radar system from coffee cans and wood, created by Greg Charvat at a defence laboratory alongside the through-wall imaging radar — ran continuously for at least seven years and was taken up and in some cases substantially improved by other universities and institutions.[407] Radar instruction aimed at practitioners is organised as application first, then basic theory, then a worked example against that theory, then case studies that each carry a bill of materials and step-by-step build instructions, so a reader can reproduce a working system rather than only follow the mathematics.[179]
Origins
The cathode-ray tube was the enabling technology for radar rather than the transmitting valves, because radar requires displaying amplitude against time and there was no straightforward method of doing so beforehand; operational systems such as the British defensive chain of 1936 followed shortly after the tube became available.[179] Wartime radar development in the United States was concentrated at Signal Corps sites such as Camp Evans in New Jersey, and it seeded the postwar instrument industry: Howard Vollum worked on radar at that site before founding Tektronix.[117] The same organisation ran Project Diana in the 1940s, which bounced a radio wave off the moon for the first time and demonstrated that radio signals could pass out through the atmosphere and return, the result underpinning satellite communication and manned spaceflight.[117] The wartime radar laboratory at MIT produced a twenty-eight-volume series covering everything needed to build radar with 1940s technology, from cathode-ray tube construction through vacuum-tube amplifiers to microwave transmitters and detectors; the volumes are long out of print but available in full as PDFs.[421]
Software-defined radio
Architecture
The defining architecture of software-defined radio is an analog front end followed by digital processing for everything else, including filtering; because the digital part is reconfigurable, one piece of hardware becomes many different radios through recompilation rather than rewiring.[48] A software-defined receiver can dispense with the front-end amplifier and often with the front-end bandpass filter, which removes the insertion loss those stages contribute, and direct conversion then takes the signal from radio frequency straight down to a very low intermediate frequency.[52] Because that lands the signal in the kilohertz range, the amplifiers can be ordinary off-the-shelf operational and instrumentation amplifiers and the filtering can be done with resistors and capacitors, which simplifies the analog chain and lowers its power consumption.[52] A capable platform generally needs a processor, programmable logic and a digital-signal engine together, because implementing the algorithms purely in FPGA logic is hard and slow to iterate; writing them in C rather than in a hardware description language buys iteration speed at the cost of throughput.[254]
A general-purpose software-defined receiver is independent of modulation format — any signal can be demodulated provided a decoder exists on the host computer — which is why the same hardware is used to recover weather-satellite imagery and for arbitrary experimentation.[73]
Hardware
Before purpose-built hardware existed, experimental receivers were assembled from repurposed instruments: an early open high-definition television receiver was built from a fifteen-hundred-dollar data acquisition card and a television tuner evaluation board from a cancelled, unsupported programme.[101] The capabilities that motivated the first general-purpose software radio hardware — multiple-input multiple-output operation, wide bandwidth, and signal processing in an FPGA — had no practical implementation below tens of thousands of dollars, which is the gap the USRP was designed to close.[101]
The HackRF was designed by Michael Ossmann as a deliberately general-purpose low-cost platform rather than an optimal instrument for any single task, on the reasoning that a tool usable across a wide variety of jobs lowers the barrier to entry more than a specialised one would.[161] It was funded by a United States defence research programme on an explicitly open proposal: prototype a sophisticated software-defined radio, build five hundred units, give them away, and publish everything the project produced.[177] The FunCube Dongle Pro, a roughly two-hundred-dollar software-defined receiver tuning continuously from about 150 MHz to 1.3 GHz, arrived by a different route, having been built as the ground downlink for a student satellite project.[73]
Twenty-dollar digital-television USB dongles repurposed as receivers are more limited than purpose-built hardware but are sufficient to start experimenting, and they run the same host software, so the learning transfers later.[161] The practical division between the two is transmit capability and frequency coverage: the dongle receives only and covers a narrower, though still wide, range, which is enough to explore spectrum but not to originate signals.[161] The dongles also displaced purpose-built alternatives, and an open-source receive-only project built around an ARM processor, an external converter and a tuner was abandoned when they appeared — the purpose-built receiver performed better, but its price-performance could not compete, a recurring outcome for open hardware displaced by a commodity part used off-label.[467]
Not every radio-hacking tool is a software-defined radio: a wireless microcontroller on a USB dongle is the appropriate instrument for low-speed digital systems such as industrial control links, smart meters, home automation, remote keyless entry and garage door openers, where full flexibility is unnecessary.[265] Software-defined radio is likewise ruled out by power budget in battery-operated designs, the tell being that a host computer’s cooling fans spin up while running one, which indicates a continuous processing load incompatible with a low-power device.[442]
Deployed systems
Modern cellular base stations split the radio from the protocol hardware: a radio head, which is a software-defined radio, is mounted next to the antenna and carries already-encoded samples over a fibre link to a baseband unit that handles the protocol stack.[467] In a satellite messaging system the ground terminals are themselves software-defined radios, so the waveform, the congestion control and the transmission scheduling can be reconfigured on the fly, including on units already deployed in the field.[427] Such a terminal can hold an orbit propagator on board: from a broadcast packet giving the time and satellite position it computes when the next satellite will be overhead, sets a timer, and returns to deep sleep, which is how it spends roughly ninety-nine percent of its time asleep.[427] The terminal is dominated by its antenna rather than its electronics — the radio module measures around two and a half centimetres and needs only power and a serial interface, while the antenna is a dome about eight centimetres in diameter and five centimetres high, with a patch version of roughly eight by eight by one and a half centimetres under development.[427] At the hobbyist end, a radar-like traffic display can be assembled from commodity parts: a single-board computer with a software-defined receiver decoding aircraft transponder broadcasts produced an alarm that maps approaching aircraft and sounds when one comes within five kilometres of an airfield below three thousand feet.[538]
What the technique does not solve
A software-defined radio solves only the modulation layer — how bits are transmitted and received — and leaves untouched the routing, store-and-forward behaviour and application that determine whether a link is useful; teams routinely spend a long time producing a radio that puts a bit in and gets a bit out with nothing built on top.[401] Commercial cellular radios are software-defined, but the companies that ship them staff whole teams above the radio layer to turn recovered bits into applications and platforms, which is the work an individual project usually does not account for.[401] Amateur radio’s practical gap is likewise not modulation flexibility but medium-speed data: there is no readily available way to hook up a radio and obtain on the order of thirty kilobits per second within a developed infrastructure, and choosing a fixed-modulation transceiver rather than a software-defined one keeps the power budget low while the effort goes into the data layer.[401]
Teaching
There is no introductory book on software-defined radio that is both at the level newcomers want and practical rather than purely mathematical, which is why the field’s teaching has been carried by intensive in-person courses instead.[214] The people who take it up from outside are typically security professionals — around eighty percent with substantial programming experience — who know little about radio or digital signal processing, which sets what an introductory course must and must not assume.[214] The working prerequisite is therefore not programming or radio knowledge but general competence at making a computer do what one wants.[214] Much of a curriculum can be completed through simulation without any radio hardware, learning the theory and the computer side without dealing with real over-the-air signals, which keeps the material accessible to students with varying equipment.[214] One stated motivation for designing the HackRF was to have a platform suited to teaching such a class, so the hardware was shaped by the pedagogy rather than the other way round.[318] Self-teaching the same material — demodulating and sniffing a protocol such as LoRa, or building one’s own signal-processing graphs — has a real learning curve, and practitioners describe progress as depending on having contacts already working in radio frequency to consult.[698]
Regulation
United States regulation requires manufacturers of radios under software control to implement a security mechanism preventing end users from modifying that software in ways that change the radio’s operation.[265] The rule is structurally incompatible with open-source radio work, since a project cannot publish its source and simultaneously prevent users from changing it.[265] Originally applied to software-defined radios seeking equipment authorisation, the lock-out requirement was later extended to any radio system with software control, which brought it to bear on the widespread practice of installing custom firmware on 802.11 wireless routers.[265] Software-defined radio is forgiving of designer inexperience on receive — a poorly understood setup still receives, just less well — but transmitting without understanding what is being emitted is a regulatory problem rather than a performance one.[162] Spectrum allocation shapes the component market as directly as it shapes the law: when Japan allocated an oddball 950 MHz band, chip companies would have had to build radios solely for that standard, the price rose because the market was small, essentially nobody built them, and the allocation was later moved down to around 908 MHz to sit closer to the North American bands.[355]
Effect on radio-frequency practice
When software-defined radio first appeared on the horizon it prompted open debate among professors and industry about whether radio-frequency engineering would remain a career; in practice the technology took far longer to become commonplace than expected.[162] It did not remove the need for radio-frequency engineers so much as change the problems they solve: a software radio board is full of switches, up-converters, down-converters and mixers, and although the work is mostly integration rather than discrete design, it can still be done badly.[162] Most amateur transceivers now sold are software-defined radios, a design choice that buys better integration, lower power and lower cost by digitising early and doing as much processing in silicon as possible.[613] Taking the antenna signal to an analog-to-digital converter with minimal analog filtering does, however, make a receiver far more susceptible to out-of-band interference, since a strong signal well away from the wanted one can overload the converter and jam the receiver entirely; on this specific measure, software-defined amateur transceivers perform substantially worse than equipment built in the 1980s and 1990s.[613]
Millimetre wave and above
Devices and arrays
Transistors reached millimetre-wave speeds through process scaling rather than through any change in device concept: intrinsic parasitic capacitances shrank with dimensions until building a transistor that generates signals in the hundreds of gigahertz became routine on a state-of-the-art silicon process.[228] CMOS wins radio sockets on the same logic that it won imaging, being the cheapest technology that works well enough, and good enough at low cost displaces better at high cost.[459] That outcome was not the expectation: in 1985 wireless was widely held within academia to be a solved problem with nothing new left in it, and CMOS was regarded as too slow and too cheap a technology to be worth applying to radio, the counter-argument being simply that CMOS kept scaling and that continued scaling would eventually make credible radio-frequency circuits possible in it.[459]
Integrated millimetre-wave phased arrays have reached large element counts on a single device: a W-band array operating from roughly 85 to 100 GHz carried 384 elements, 256 for transmit and 128 for receive, each with its phase modified coherently to steer the beam electronically.[430] Pushing to millimetre wave lets an antenna of reasonable physical size focus a beam onto a very small spot and steer it electronically, which permits the same spectrum to be reused several times within one volume and so raises aggregate bandwidth without new allocation.[459] One research direction concerns manufacturability rather than performance: array technologies cheap enough to produce by roll-to-roll processing for around a dollar, hung on a wall as sheets, would serve both as a communications fabric and as a means of directing power to devices in a room.[459]
There is nonetheless a ceiling on how much information can be packed into a given bandwidth, because beyond a certain complexity of modulation per hertz, meeting the signal-to-noise requirement forces transmitter and receiver closer together, so spectral efficiency trades directly against range.[430] Beam-steered spatial multiplexing is bounded by processing and geometry rather than by antenna theory: it requires open space because the beams will not pass through a wall, and forming a couple of hundred independent beams for a crowded railway station, with the corresponding processing carried on the handset, is not realistic.[483] Short-range 60 GHz links for applications such as laptop docking face a plainer obstacle, competing against a one-dollar cable while the packaging solution required at those frequencies costs on the order of a hundred dollars, which is why such links have found only scattered use.[483]
Cellular and fixed wireless
Fifth-generation cellular is an umbrella covering distinct technologies — additional spectrum within the frequency ranges already used by the previous generation, and separately millimetre-wave operation in the multi-gigahertz range — which is what makes generic statements about its performance unreliable.[569] It is not primarily a handset technology: the millimetre-wave component is aimed at replacing fibre with an over-the-air link and at enterprise access, and it offers latency orders of magnitude lower than the preceding generation, which the handset case does not exercise.[430] The economics behind that aim are straightforward, since running fibre to a single building can cost on the order of twenty thousand dollars even where other fibres already reach the site, and the cost of laying it cannot be recovered from a single subscriber, so wherever the same service can be delivered over the air it will be.[430]
Millimetre-wave cellular retains a low-frequency band for link establishment: the device broadcasts on the lower frequency to reach agreement with the tower, and only then does the tower switch to millimetre wave and begin tracking it with a steered beam.[430] A dedicated search antenna would solve the problem of a steered beam wasting time pointing at empty space, but it doubles the count of devices and components on the tower, so the cost of the additional hardware is weighed against the throughput lost to searching.[430] Low cellular frequencies are used because they are indifferent to obstacles in their path, whereas millimetre waves are stopped by rain, fog and walls, the scale difference between 600 MHz and 30 GHz producing that behaviour; millimetre wave delivers far more capacity if a dense mesh is built, but once carriers costed that infrastructure many chose to lean on existing networks instead.[678] Outside millimetre wave, download speeds are only slightly faster than the previous generation, and the millimetre-wave deployment that would have delivered the promised gain was largely not built because of the tower density it required.[714] Fixed wireless access services marketed as home internet are accordingly mostly not millimetre wave, operating instead at roughly 3.5 GHz, where the equipment and the propagation both work out.[714]
Sensing
Radio-frequency sensing is being explored as an alternative to cameras for presence and movement detection because it is less invasive, producing no photorealistic image and propagating around an environment rather than requiring line of sight; the obstacle is that a radar circuit and a communications radio are not the same integrated circuit, so building both capabilities into one device is not straightforward and carries a cost penalty.[553] Research circuits for the generation after fifth-generation cellular already operate above 110 GHz.[553] Automotive radar volumes have meanwhile pushed millimetre-wave sensing into general engineering reach, with 24 GHz presence and heartbeat-detection modules derived from vehicle backup-detection technology available from Chinese suppliers for around a dollar.[714]
The enduring strength of radar as a vehicle sensor is that it punches through weather, is indifferent to dirt or film on the surface in front of it, and will pass through a plastic bumper, so the sensor need not be exposed.[729] Laser-based ranging by contrast degrades in rain, snow and fog, and also from ordinary road film dried onto the lens, which is a hard limit for any autonomous system required to keep all its sensors working in adverse conditions.[729] Automotive radar at 77 GHz, including recent large multiple-input multiple-output systems, has the opposite weakness, a resolution problem in which returns are interpreted as blobs and targets must move fast enough to be resolved by other means.[729] Terahertz is presented as the remaining unexploited region of the spectrum below visible light for sensing, on the argument that it combines radar’s ability to see through weather with image quality approaching that of LIDAR, a combination neither established option offers.[729] Sensing at a much longer range uses several independent methods rather than one: tracking objects in orbit relies on ground-based radar systems, ground-based optical telescopes, and space-based optical telescopes that observe geostationary and cislunar space from orbit.[679]
Amateur practice
The lower the operating frequency in the amateur high-frequency bands, the more local the traffic tends to be, so band choice is effectively a choice of the geographic reach of the contacts available.[613] Operating practice on those bands therefore amounts to matching band to time of day for a wanted direction — transmitting on 21 MHz at three in the afternoon local time to reach Asia from the Pacific coast, for example — since which paths open is set by propagation rather than by equipment.[613] For worldwide contacts the 20-metre band is generally the most productive, and PSK31, a digital keyboard-to-keyboard mode, works well under weak-signal conditions where voice would not.[57] Efficient digital modes of that kind let a modest station with a poor antenna make long-distance contacts, which lowers the equipment threshold for operating from constrained locations such as a low-lying site or an urban building.[73]
Among the amateur microwave bands, 2.4 GHz is the practical entry point because it sits at the wireless-networking band and therefore has a large supply of readily usable components.[107] Above that, amateur millimetre-wave operation is practised with home-built equipment spanning 10 GHz to 79 GHz, with working contacts made across a room at 79 GHz; assembling such a set of radios represents on the order of a decade of construction.[107]
Scarcity of practitioners
Millimetre-wave integrated-circuit design is a very small labour market: one industrial research team designing such ASICs numbered six people and found the specialism hard to recruit into.[228] The scarcity extends downward to ordinary product work, where the combination of board layout, embedded firmware and a working radio-frequency background is uncommon enough to be worth remarking on.[11] Doctoral advice given to one student entering the field framed the response to that difficulty as a general method rather than a subject: a thesis is not meant to change the world but to change the person doing it, so the task is to pick something hard and learn to teach oneself, because a career will pass through five or six revolutions of the field before it ends.[459]