| Episodes | 131 |
| Mentions | 160 |
| Cited here | 38 |
| First — last | #7 — #723 |
| Top guests | Elecia White, Ming Zhang, Greg Charvat |
| Related | arduino · firmware · microcontroller · analog to digital converter · bluetooth |
A sensor is a device that converts a physical quantity — strain, temperature, pressure, particle flux or similar — into an electrical signal that a measurement or control system can use.[476] At its core sits a transducer, the element of instantiated physics that performs the conversion, surrounded in a complete sensor by linearisation, computation and interface circuitry.[476] Transducing a real-world activity into an electrical signal is the first stage of any measurement system; thereafter the signal itself is the thing of interest, persisting through digitisation, compression and transport.[476] Because a sensor reading only means something when normalised against a standard, calibration and traceable reference measurement are inseparable from sensing as a discipline.[72]
Sensor and transducer
A transducer is the piece of instantiated physics that turns a quantity such as strain or temperature into an electrical signal; a sensor contains a transducer but ordinarily adds linearisation, computation and interface circuitry around it.[476] The distinction matters in practice: once the physical quantity has been transduced, everything downstream — conversion to digital, compression, transport — handles the same signal, and the engineering questions shift from physics to signal integrity and data handling.[476]
Much of the signal processing that was once designed discretely has moved inside integrated parts. Filtering in particular is now embedded in converters to a degree that the buyer never specified and does not need to know about; a digital filter block is therefore not marketed as a replacement for an analogue filter, because it completes a function the customer wanted rather than substituting for a component.[476]
Signal conditioning and conversion
Small differential signals frequently ride on large common-mode voltages, and extracting them is the classical role of the instrumentation amplifier: three operational amplifiers in its standard form — two taking the inputs, a resistive network between them, and a third combining the result — which makes it suited to cases such as electrodes placed on the body.[34]
The analogue-to-digital conversion stage is shared by so many sensors that it constitutes a single point of leverage across an entire class of devices: defeating a converter defeats every sensor that depends on it.[352] Not every sensor presents a digital output, however, so any analysis or interface strategy that assumes a digital interface does not cover the field.[352]
Calibration and standards
Any sensor or transducer must be normalised against a standard for its output to mean anything; this is the function calibration laboratories perform across pressure, temperature and the other engineering parameters.[72] The range of quantities that can be measured and calibrated against reference standards extends well beyond electrical ones, encompassing properties such as flatness and material hardness.[72]
Some integrated circuits carry the calibration problem on-chip. Devices intended for radiation environments embed internal sensors monitoring bias, voltage and current conditions, so that a dedicated on-chip processor can detect when a parameter has drifted outside its expected range and run a calibration routine to restore specification.[483] On Adrian Tang’s radiation-tolerant system-on-chip work, this approach depends on knowing in advance where every parameter should sit: the internal measurements are compared against known-good values rather than interpreted, and extreme temperature swings are handled by the same mechanism.[483]
Interfaces
Buses, drivers and host software
Pin count is a recurring constraint when many sensors attach to one controller. A shared two-wire bus allows a dozen sensors to be connected using only two pins, though the component cost accumulates quickly; an addressable expander extends the same technique to general-purpose input and output.[202]
On the software side, the Linux Industrial I/O subsystem now carries almost every category of sensor driver. It originated with a converter manufacturer wanting its own parts well supported, and has become the place a driver is contributed when a new device appears in a phone or other product.[378] Whether to write a sensor driver or adopt an existing one is decided by the reliability required of the production system and by auditing the candidate code, rather than by preference.[556] A real-time operating system with a built-in shell can expose sensor and converter readings as interactive commands over a serial connection, allowing a device to be interrogated for arbitrary readings without writing application code.[713]
Programmable logic at the sensor
Where sensor counts or timing requirements exceed what a microcontroller handles comfortably, programmable logic offers an alternative. Implementing serial ports in software consumes processor time that scales with the number of ports, whereas a programmable logic device can give each sensor its own dedicated hardware port that polls it and deposits the reading in a register, removing the periodic work from the main program entirely.[395] The broader argument for a soft microcontroller on programmable logic is that the peripheral set can be built to match the design rather than selecting a processor whose fixed peripherals nearly fit, and modern tooling permits this without writing hardware description code.[395]
Optical sensors in particular demand timing tight enough that programmable logic suits them better than a microcontroller running an operating system, which would have to be substantially more capable to keep up.[588] More broadly, the emerging requirement in vision and industrial sensing is low-latency processing placed next to the sensor rather than a bridge device that merely moves data, which drives demand for logic capacity in very small, low-power packages.[535]
Power
On a deployed remote sensor, the power budget is frequently the overriding design constraint. Sealed battery-powered units may be expected to last five years without service; on one such programme of Greg Charvat’s, the transducer itself was the harder technical problem while power consumption was well understood.[179]
Certain sensing technologies are intrinsically hostile to battery operation. Chemical gas sensors of the common resistive type contain an internal heating element and can require warm-up times measured in tens of minutes, which rules them out of battery-powered applications and motivates work to convert older sensing methods to lower-power equivalents.[376] At the opposite extreme, energy harvesting can supply enough power for a low-power microcontroller and a sensor on a continuous basis, including from unconventional sources such as a potential difference developed between two spikes driven into a tree.[7]
Power consumption is not always where intuition places it. In an automotive electrical system the current is consumed by the sensors, actuators and relays rather than by the processors, and a mechanical relay retains the advantage of drawing no current at all in one of its states.[93] Power profiling of a sensing device proceeds by removing one contributor at a time — substituting a fabricated message for a real sensor reading, for example — to establish how much each part of the system consumes, since optimising a small contributor while a large one dominates yields nothing.[527]
Wireless and remote nodes
A wireless sensor node can be built to transmit without receiving at all — waking, sending its reading and returning to sleep with no acknowledgement — which is acceptable for quantities such as temperature and humidity where losing occasional samples does not matter.[272] The protocol for such a node has to be simple enough to run in a couple of kilobytes of code without a large software library, because the economics require the device to cost a few dollars and to be viable at hundreds of units rather than hundreds of thousands.[272]
For assets beyond terrestrial connectivity, very low-power remote sensors relaying through satellites allow instrumentation of fields, water tanks and livestock.[679]
Roles in larger systems
In a control system, the design of the control law is roughly a tenth of the work; the remainder is making the hardware function, interfacing the sensors to the controller and keeping the actuators out of saturation.[119] Kent Lundberg, whose teaching framed this division, held that a capable control engineer is therefore willing to be an integration engineer, accepting that half the effort on a six-component system goes into connecting those components to one another.[119]
A stabilised flying platform illustrates the pattern: the sensors report attitude and motion, but the engineering difficulty lies in the control loop that consumes those inputs and recovers the vehicle from a disturbance.[9] Small autonomous aircraft of this kind became practical through a confluence rather than a single advance — low-cost sensors arriving alongside cheaper processing, motors and batteries.[345] The same trajectory has turned capabilities that once represented years of work by a design team into modules bought with a standard header and attached to a product directly.[633]
In industrial settings, many problems appear as trends over days or weeks rather than as sudden failures, which is what makes continuous monitoring valuable enough to justify instrumenting an expensive machine.[511] What the operator of an instrumented process wants is information rather than sensor readings, so presenting a dashboard of raw channels misses the point of the deployment.[511] Processing is correspondingly kept off the sensor node and moved to where computing is effectively free, reducing the node’s job to acquiring and forwarding the reading, which keeps the firmware simple even when the analysis is complex.[511]
Machine learning applied to sensor output replaces hand characterisation: instead of deciding that a particular slope means a particular event, examples of the event are recorded many times and the classifier is trained on them, removing the need to derive thresholds or signal-processing rules by hand.[546] The alternative that most practitioners actually implement is a threshold comparison, which is why an automatically characterised classifier is attractive to engineers without signal-processing expertise.[546]
Sensing capability has also diffused downward into products. Vehicles now carry a processor and sensing in each subsystem down to the level of a door handle, because the function requires local detection.[530]
Sensing modalities and selection
Sensing methods are best chosen at the level of the problem being solved rather than by familiarity. Radar’s distinguishing properties are that it works through all weather, gives range in three dimensions and characterises a target by its Doppler signature, and Greg Charvat argues it deserves consideration alongside infrared, lidar and acoustic options whenever a task calls for a sensor, rather than being treated as a specialism.[179]
For development work, resistance temperature detectors small enough to sit directly on an integrated circuit allow the thermal path and dissipation of a board to be instrumented on several channels at once.[425]
Security and tamper resistance
A deployed sensor whose readings have consequences for the operator will be tampered with. Designs must anticipate the unit being disabled, a different sensor being substituted, or a reading being spoofed, and answer these with authentication that the attached hardware is genuine.[355]
Tyre pressure monitoring illustrates both the architecture and the exposure: it is a widely deployed radio sensor system in which four independent transmitters periodically send to a central controller, radio being the only practical way to get data off a rotating wheel.[265] Because such a sensor carries its own processor and reports onto the vehicle’s internal bus, a substituted or modified unit is a route onto that bus from outside the vehicle.[265] At the component level, the shared analogue-to-digital conversion step means an attack on the converter defeats a whole class of sensors at once.[352]
Failure modes and maintenance
Deployed sensing systems fail through mundane physical causes rather than exotic ones: a connection knocked out, a power cut with no clean restart, a flat battery, or the unit physically disturbed. Any claim of a maintenance-free installation has to answer these.[674]
Component substitution creates failures of a different kind. A forced sensor substitution has a larger consequence than a comparable power-supply substitution, because different sensing parts branch the firmware and leave separate builds to be maintained against each board revision.[573] In a learning context, substituting a different vendor’s sensor for the one specified in a tutorial is a common cause of failure that is very hard to diagnose, because the difference hides under layers of abstraction; duplicating the exact hardware is the reliable way to reach a working baseline, and the value of a kit is precisely that every copy is identical.[330]
Engineering practice
The recurring need to evaluate unfamiliar parts has produced standard working methods. Bridging a sensor to a host computer through a small microcontroller and a serial terminal is the standard first step in evaluating it, and general-purpose interface tools exist to serve exactly that repeated need.[461] Standardised sensor connector families exist so that a measurement can be attached without designing an interface, and a board can carry several of them at once to cover the common ecosystems.[441] A general-purpose sensing platform is best built by keeping the node generic and swapping the attached sensor, rather than by building a distinct product for each measurement.[511] Where a sensing element is still being selected, placing it on a daughter board rather than the main board allows it to be revised repeatedly while better parts are evaluated, without respinning the whole design each time; Piotr Esden-Tempski used this arrangement for the inertial measurement unit on his flight-controller hardware.[356]
The people specifying sensing systems are frequently not electronic engineers at all: mechanical, civil and materials specialists fill the training courses, because what they want is the data and the process improvement, with the electronics incidental.[451] Mechanical engineers typically reach electronics through measurement — needing stress or temperature data leads to attaching a sensor, which has to be powered, which leads to recording the data.[153] A company with a physical process and high volume but no hardware capability is a recognisable customer type for sensing products, since the problem is worth solving at their scale but standing up a hardware team is not something they want to do.[451]
For practitioners, the sensors themselves are approachable; the difficulty appears only once estimation techniques such as filtering and orientation mathematics are required.[256]