X-Ray Baggage Scanner PCB Design: Low-Noise Readout, HDI Stackup and Supplier Selection for Security Screening Detectors
The signal acquisition board of a security X-ray machine became the hardest
Metal Detector PCB Design: Coil Layout, Noise Control, and How to Choose a Multilayer or HDI Manufacturer
- sprintpcbgroup
Why the Board Matters More Than the Schematic
Some time ago I helped a friend tinker with a handheld metal detector project, and found that what really holds people back is not the schematic but the board itself, the Metal Detector PCB. The ready-made modules you can buy on the market drift beyond recognition as soon as you change the coil, the phase cannot lock, and after adjusting ground balance for ages it still screams at wet sand. Later I simply stopped fiddling and went directly to a multilayer board manufacturer to re-prototype. The multilayer pcb manufacturer I contacted even reminded me that if the coil on the bottom layer is drawn the conventional way, parasitic capacitance can completely skew the frequency selectivity. In fact, a metal detection board most fears treating the coil and the signal chain as two independent matters. If you let an ordinary PCB factory with insufficient coil copper thickness make it, an inductance difference of a few microhenries makes the receive-end RC filter lag, and signals from small targets are simply submerged. Later I turned to a scheme using blind and buried vias from an hdi pcb manufacturer’s process, shrinking the loop between the preamplifier and the coil to extremely short, and only then was the noise floor pressed down to a presentable level. Now when I look at those designs whose ground balance runs sluggishly, mostly they ignored the cooperation between Metal and PCB. Once the turn-to-turn distributed capacitance of the coil and the trace inductance are mismatched, no matter how fancy an algorithm the DSP runs later, it cannot save it. Sometimes when chatting with peers, they say high-density interconnect costs too much and that it is better to make the dynamic range large and tough it out, but I have measured boards from several of those hdi pcb suppliers myself, and the signal-to-noise ratio in weak magnetic fields is not at all on the same level. If you also plan to tune a sensor front end from zero yourself, do not save on the multilayer board. Directly find a reliable multilayer pcb supplier and have them also do the current density simulation of the coil copper foil, and later you can avoid half a year of detours.
Manufacturing Precision Is Part of the Design
In these years of working on metal detectors, I found a particularly counterintuitive thing: the link most likely to flip over is not the algorithm or the coil shape design but choosing the wrong factory to make the board. Many newcomers think that once the schematic is drawn and the simulation passes everything is fine, so they casually find a prototyping factory, throw the files over, and when the board comes back, is soldered and powered on, the waveform shakes like Parkinson’s. Taking it apart to look, the traces are all connected and the components are not soldered wrong, but it just does not work. In this situation, nine times out of ten the PCB manufacturing precision has not kept up with your design requirements.
A metal detector’s PCB is not equal to an ordinary digital circuit board. The signals running on it are not switching signals but induced signals at the microvolt level. Especially in the coil part, the width, spacing and copper thickness of every turn of trace you draw directly affect inductance and distributed capacitance. The biggest loss I took was in one upgrade where I used a four-layer board to isolate the transmit coil and the receive coil, and found a multilayer pcb supplier that had started out on ordinary double-sided boards. When the board came back, layer alignment was off by about 3 mil. This offset is not an issue at all in an ordinary circuit, but on a differential receive coil it directly destroyed the initial balance, and zero-point drift cut the sensitivity of the whole machine by nearly half. Later I sent the coil layers separately to a factory that made boards as an hdi pcb manufacturer to redo them, using their laser direct imaging to handle those few critical turns of traces. After the board came back and was installed in the body, it hardly needed tuning, and consistency was two grades better than the enameled wire bobbins I wound by hand.
Copper Thickness Uniformity and Q Factor
Another point underestimated by most people is the uniformity of copper foil thickness. The coil is essentially a planar inductor, and the thickness of the Metal layer directly affects the impedance of each turn. A conventional PCB factory’s copper thickness tolerance is considered passing if it reaches plus or minus 10%, but for the transmit coil on a Metal Detector PCB, if one turn is 18 microns thick and the next becomes 22 microns, the Q value of the resonant circuit scatters. Once, to test this matter, I took five boards from the same batch and cross-sectioned them to look at copper thickness, and the most outrageous one was nearly 15% thinner than nominal. If it had not been that the sensitivity of that batch of boards varied so much between individuals that they could not ship, I would not have gone to the trouble. Later I settled on making them at a multilayer pcb manufacturer, who could do separate plating leveling for copper thickness in the coil area, and the inductance deviation of boards in the same batch could be pressed within 1%, which is more stable than machine-wound coils.
Why a Chip Alone Cannot Fix a Poor Coil Layout
Many people ask me why I do not use ready-made TI inductance-to-digital converter schemes to bypass front-end tuning and go straight to the LDC series. It indeed saves trouble, but you should know that however strong the chip, it only compensates in the latter half of the signal chain. If your coil traces and layer structure themselves introduce uncontrollable parasitic parameters, the range the chip can calibrate is also limited. Especially when making high-sensitivity models, for the ground balance algorithm to take effect, the premise is that the electric field distribution of the coil is within your designed expectation. I once helped a friend modify a large-depth pulse machine. To save cost, his board squeezed the four-layer board coil and the digital part into the same stackup, without strict partitioning, and digital noise ran from the ground plane into the receive coil and could not be filtered out however he filtered. Later I honestly split it into two boards: one multilayer PCB specially for running coil signals, and one ordinary PCB for digital and power, and the problem disappeared directly.
Choosing Suppliers for Coil Boards
In choosing suppliers, my principle now is that for coil-related boards I never use the cheapest prototyping. It is not that expensive necessarily means good, but that you must find an hdi pcb supplier who can understand your needs. For example, laser drilling alignment accuracy, dielectric layer thickness tolerance and copper tooth length control are things you cannot see default parameters for on the order page, and you must confirm them with their engineering staff. I even mark the Gerbers of those coil layers separately and tell them this is the core loop used for Metal induction, ask them to tighten the copper thickness tolerance, and set etching line width compensation according to the actual value of the finest trace. Sometimes spending a few hundred yuan more in engineering fees can save you from scrapping several batches of boards.
Ultimately, the PCB for metal detection is essentially a precision sensor, not a simple connection board. If you understand this matter in place, then from drawing the coil to choosing the factory you will be calm, instead of facing a spectrum analyzer every day looking for causes.
Differential Receive Coils on Multilayer Boards
After working on several metal detectors, I increasingly feel the board is what really drags things down. The circuit principle was thoroughly digested long ago, and what is hard to deal with is how the receive coil part lands on the PCB. If the coil you draw cannot achieve precise differential balance, the residual voltage can easily rise to dozens of microvolts, and then hanging a 24-bit ADC behind it is in vain, with effective bits all swallowed by noise. This stuff cannot be solved by pulling a few turns of spiral in simulation software, and prototype measurement deviation is often beyond imagination.
My approach now is to cut the coil structure into multiple narrow copper strips distributed over a multilayer board, stringing an equivalent multi-turn receive loop through interlayer vias. It sounds simple, but if line width and spacing deviate by a few tenths of a mil, the balance of the whole board collapses. So when choosing a multilayer board supplier, I do not trust their so-called high precision advertised on their website at all. I must see whether they can actually control lamination registration and etching tolerance tightly. Especially for a Metal Detector PCB used in metal detectors, once copper thickness goes up, any carelessness in line width compensation causes the coil’s inductance and distributed capacitance to disperse, and finally the characteristics of the entire receive chain are all disordered.
HDI as a Necessity for Coil Boards
Interestingly, the HDI process has instead become a necessity on this kind of board. You might think only phones use the hdi pcb manufacturer‘s whole set, but in fact the via density of strip coils is outrageously high, with segment after segment of coil sections routed tightly side by side, and laser microvias and via-in-pad are nearly mandatory. At that time you have to find an hdi pcb supplier who can really chew hard bones. I am never stingy on process cost, because a receive coil is essentially a device extremely sensitive to spatial distribution, and the differential induced voltage waveform from an ordinary through-hole board is often crooked, smearing the secondary field signal of the target metal into a lump. After switching to stacked microblind vias and finer lines, the electrical symmetry of the coil immediately improved. Drop a coin in and the transition of the receive waveform is crisp and clean, and the system can catch a much weaker signal.
By the way, a gripe: quite a few factories in China that claim to be a multilayer pcb manufacturer, once they hear that the board has a large area of exposed coil, first prepare to coat the whole thing in solder mask. They do not understand the metal detection block, and the thin layer of solder mask on the coil surface directly changes near-field coupling, with even a few more microns of thickness letting sensitivity collapse. You must make it clear, even specify no ink covering or a very thin protective layer. In the end, a truly good Metal Detector PCB, from stackup design and copper thickness control to surface treatment, has to be worked out around that microvolt-level receive signal of the coil. It is not just a matter of winding copper foil. If process details are not in place, the hardware foundation is like a leaking probe rod, and the best algorithm cannot make up for it.
Where the Series Resistance Really Comes From
In all the time since I entered the industry, I have modified no fewer than twenty metal detector schemes, and my deepest takeaway is that for coils, calculation on paper and the thing finally held in your hand are two different matters. Many people stare at fluctuations in the quality factor, as if optimizing omega L over Rs by a few tenths could make the machine transform, but they are going in the wrong direction. What you should really care about is where that Rs actually comes from and whether you can find a reliable person to press it down, rather than facing a formula and tuning parameters.
I first used an ordinary double-sided board for the coil with 35 micron copper, and after winding and measuring resistance, my heart sank by half. The series resistance was outrageously large, the resulting thermal noise could not be covered by the weak signal at all, and no matter how beautiful the transmit-end power amplifier was made, the receive end was full of noise floor. Later I simply stopped tinkering myself and outsourced the coil part to a factory that specialized in multilayer boards, finding a multilayer pcb manufacturer that had long made prototypes for RF and sensors, and had them use a four-layer board with the coil in four-way parallel and copper thickness increased to two ounces, and the resistance instantly dropped. Note that it is not just any factory that can make double-layer boards that can do this job, as their requirements on line width uniformity and solder mask dielectric constant consistency are not the same thing as for ordinary digital boards.
Talking about multilayer, some may feel that a coil is just those few turns of wire, so why bring up HDI, purely a waste of cost. I thought so at first too, until on a certain pulse model I could not press down the transmit waveform overshoot at all, and only then went back to re-examine the PCB’s parasitic parameters. At the moment the transmit coil current switches, dozens of amps pull up within a few microseconds, and the parasitic inductance on the trace can make your MOSFET bear extra voltage spikes. Later I switched to a factory acting as an hdi pcb manufacturer, and stacked the drive loop with buried holes directly on adjacent layers, cutting loop area by nearly two thirds, and that ringing immediately settled down. You see, many problems are not at all about how clever the circuit topology is but about whether you are willing to put a bit more thought into board material and structure.
Now I look at only two points when choosing a supplier: whether he has long made boards for industrial control or medical equipment, and whether he has his own impedance test reports. I do not need him to say he can do it. I need to see the data. The water is deep among multilayer pcb suppliers. Some factories pile up layers for you without controlling dielectric thickness tolerance, and in the end you cannot even hold the coil’s nominal inductance steady, let alone batch consistency. An hdi pcb supplier is relatively better, because those that can use laser drilling mostly have a decent process foundation, but you also have to look at the loss tangent of his interlayer material, as at high frequency a small difference lowers transmit efficiency a notch.
One more thing about the transmit part. Many designs treat the transmit driver as an independent module, with the coil as the load and the power amplifier as the source, caring only about matching output impedance and neglecting one thing: the transmit trace on the Metal Detector PCB is itself a radiator. If you do not handle its return path well, it can couple interference directly into the receive coil without ever passing underground targets. Once this kind of crosstalk forms, however strong the filter that follows, it cannot save it. My habit is that the transmit loop must use an entire copper pour as a reference, and it must never run parallel to the receive coil leads. It is all details when said, but stacking these details up is the fundamental basis for a machine truly distinguishing an iron nail from a hard coin.
Layout of Transmit Power and Receive Chain
When I was making metal detectors, the biggest headache was actually not circuit principle but how to land that pile of ideas on the board steadily. Many people feel metal detection is just a matter of a coil and a few op-amps, and when you really start, you find there are endless tricks hidden on that small Metal Detector PCB.
The transmit part alone can easily see current jump to several hundred milliamps, and if the power loop on the board is not held down, ground bounce and spikes can stir the receive end into a pot of porridge. I learned later: when doing layout, I first circle the large-current path of the transmit stage and let them run on the shortest copper, sprinkle via fences next to them, and the ground plane is firmly soldered on the inner layer of the multilayer board. At that time you will understand how critical it is to find a reliable multilayer pcb manufacturer. A poor board factory gives you slightly misregistered lamination or substandard copper thickness, and with just slightly more current, the voltage drop all shows up, and the transmit waveform directly changes flavor.
The receive chain is another matter. The signal induced back by the coil is pitifully weak, microvolt-level stuff, and if a little transmit frequency harmonic or digital ground noise creeps in, no matter how high the amplification behind it, it is in vain. I have tried soldering the receive coil leads directly to the board end, then leaving the first-stage preamplifier alone in the analog area, with an isolation slot dug around it and a whole clean ground plane underneath. At this point the difference in board layer structure shows: a four-layer board can barely work, and only six layers or more can truly bury sensitive traces in inner layers so that electric field interference cannot stick to the signal. So later I tended to find factories that could do hdi pcb manufacturer work for prototypes, using laser drilling to press blind and buried vias next to critical signals, with shorter paths and smaller parasitics, and the signal-to-noise ratio at the receive end rose immediately.
The layout of the coil has a huge influence on detection depth and directivity, and this is bound to the design of the board itself. The transmit coil and receive coil are not just placed casually. How tightly they couple directly determines the zero drift of the whole system. On one board I integrated transmit drive and receive conditioning at the same time, and it felt like adding weights to both ends of a seesaw: if board-level coupling is slightly more by even a little, the receive end shows a baseline signal that cannot be shaken off. Later I simply twisted the transmit coil’s drive wires into a twisted pair, and specially left a column of header jumpers on the board so the grounding point could be fine-tuned on site. This seemingly simple small change cost me three or four wasted board revisions before I found a comfortable balance point.
Finding suppliers also stepped in quite a few pits. Cheap through-hole boards simply cannot cope with that kind of routing so dense it dazzles the eyes, especially when you cram the transmit power amplifier, receive preamplifier and phase-sensitive detector all onto one card and also have to save space for the coil interface and battery management. I changed several multilayer pcb suppliers one after another: some factories’ Tg150 board material began to delaminate after several solderings, and some had poor copper adhesion so that when the coil terminal post was tightened with torque, the pad curled up directly. Only later did I lock onto one hdi pcb supplier, using high-Tg material with inner layer copper thickness that dared to go to 2oz and solder mask controlled finely, and board-level reliability stabilized.
Ultimately, with metal detection, transmit and receive are like the left and right hands, and the board is the nerve between them. However carefully the coil is wound, if the PCB loses its strength at any step, the earlier work is all in vain. Later, in chats with some veteran detector makers, we all agreed on one thing: do not skimp on the board. Every link in the signal chain is competing with noise for that bit of margin, and spending a little more on PCB manufacturing is often more direct than piling up compensation algorithms later.
Analog Grounding, Isolation Slots and Parasitics from Buried Vias
Recently I helped a friend who makes metal detectors tune a board, and only then found how deep everyone’s misunderstanding of the Metal Detector PCB runs. Many people think it is just a circuit board: solder the amplifier on, get the signal through, and done. But in fact this thing’s sensitivity to noise is even harder to deal with than many RF boards. I took a look at his previous board: the ground wires were split in a mess, the amplifier supply was taken directly from an LDO in the digital section without any filtering, so it was no wonder that every time a hand approached the coil, the noise floor drifted like an electrocardiogram.
This reminded me of a time dealing with an hdipcbsupplier, where that factory’s technical support absolutely could not understand why a strange isolation slot had to be dug under a preamplifier. They thought that by conventional design, the ground layer of a multilayer board is complete enough. But a metal detector is different: the signal induced by the coil is itself weak to an outrageous degree, and any parasitic coupling in the PCB will be faithfully amplified by the amplifier. Later I switched to another multilayerpcbsupplier, who directly asked me how large the coil impedance was and what the operating frequency was, and then suggested using a 4-layer board with a dedicated ground plane for the analog front stage, and pulling this ground and the digital ground separately to the star ground point at the power entrance. That is more like it.
Now many multilayerpcbmanufacturers boast at every turn of how tall and fast a digital board they can make, but in our analog small-signal field, what is tested is not line width and spacing but understanding of return paths and noise currents. HDI technology can indeed help metal detectors shrink in size, such as putting the DSP chip and amplifier into a tighter layout and reducing trace loop area, but the premise is that the hdipcbmanufacturer truly knows how to handle the parasitic effects brought by buried vias. I have seen a negative example: a first-order HDI board that, to save trouble, used a blind via to run an analog signal from the top layer straight through to the third layer, and the result was that the hole happened to pass through the digital power layer, noise coupled in directly, and the sensitivity of the whole Metal Detector PCB dropped by two orders of magnitude.
The choice of the amplifier itself is also interesting. Some in the circle favor ultra-low-noise op-amps, such as models with equivalent input noise density below 1nV per root Hz, but this is actually a trap. In a low-frequency scenario like a detector, 1/f noise is often the real master, and the noise voltage drop produced by the amplifier’s current noise flowing through the coil impedance is sometimes much larger than voltage noise. So I have always felt that rather than blindly piling money into amplifier chips, it is better to spend energy on PCB routing, power decoupling and coil shielding. These invisible details are the decisive factor of the final noise floor of the whole board. After all, looking at a clean baseline on an oscilloscope is more reassuring than any fancy digital algorithm.
Ground Balance, I/Q Demodulation and Why Phase Errors Matter
Anyone who has worked on metal detectors knows how much the process of the board itself affects the final result. I used to always feel the algorithm was the core, until one time on a heavily mineralized beach, a prototype could not hold ground balance steady no matter how I tuned, and later found the problem was in that four-layer board’s handling of mixed signals: analog ground and digital ground were not cleanly separated, the I/Q components from quadrature demodulation were a disaster, and the ground balance algorithm simply could not pull the useful signal out of the noise.
This forced us to run off and find a multilayer PCB manufacturer again, and not an ordinary kind. The weak signal processing chain on a metal detector PCB is too sensitive: the transmit coil drive section carries large current, the receive end has to process microvolt-level induced signals, and if there is even slight interlayer crosstalk or impedance discontinuity on the board, after phase-sensitive detection there is a pile of false targets. Later I switched to an HDI PCB supplier and redid the stackup design, burying critical analog traces on inner layers and leading them out with laser blind vias, and only then was signal quality pulled up. They had made RF boards for other customers and knew how to control the phase error between quadrature components, which is critical for baseband extraction in the digital domain: after the received signal is multiplied with the reference signal and low-pass filtered, if the dielectric constant of the board material drifts severely, the imaginary part eats the real part, and when distinguishing ferrous and non-ferrous metals the judgment turns chaotic.
Now when I choose a multilayer PCB supplier again, I do not look only at the quote. You have to talk with them about specifics such as residual copper rate, dielectric thickness tolerance, and even the effect of glass cloth warp and weft direction on signal delay during lamination. Especially for HDI PCB manufacturers doing high-density interconnect, if their additive process is controlled well, trace side etching is small and phase consistency of differential pairs can be preserved, which is worth more in a detector needing fine demodulation of quadrature components than just stacking layers. Since the analog era when four op-amps were stitched together to form a phase detector, we have known that if there are even a few extra degrees of phase shift on the signal path, the discrimination logic is inaccurate, and it is the same with DSP: if the board contaminates the signal components, the best algorithm is in vain.
Baseline Drift, Multilayer Registration and the Physical Foundation
I have been in metal detection for several years and found a rather contorted phenomenon: many people would rather spend several months debugging baseline tracking algorithms than put half a bit more energy into the Metal Detector PCB itself. They always think the coil is just a few turns of copper foil printed on the board and that finding a cheap factory to make it is enough. The result is that the signal chain is tilted from the physical layer onward, and no amount of pulling with algorithms later can pull it back.
Anyone who has really tinkered with physical objects knows how critical finding a suitable multilayer pcb manufacturer is. The receive coil of a detector mostly relies on differential symmetry to cut out common-mode interference, but once a four-layer board is laminated, if layers are slightly offset by even a few hundredths of a millimeter, the inductance on the two sides is no longer consistent. I once threw the same Gerber at two multilayer pcb suppliers: one produced boards whose zero position was steady like a rock, while the other’s baseline after demodulation crept slowly upward as soon as the temperature shifted. Can you blame the algorithm? It was clearly copper going to places it should not go.
Later I simply flipped my supplier selection criteria entirely. I no longer look at how fine a line they can print but at how rigorously they control dielectric thickness. Once I met an hdi pcb supplier who worked on automotive radar, and he told me something very frank.
Another very deeply hidden pit on the coil is the solder mask. Many people, to prevent oxidation, habitually use immersion gold over the whole area or coat a thick layer of ink. But as long as a layer of high-dielectric-constant medium lies over the coil, parasitic capacitance climbs immediately and the resonance point runs toward low frequency. I once had a board that was extremely sensitive and responsive when first powered on, but withered after half a day outdoors, and after checking for a long time I found the solder mask ink had absorbed moisture and its dielectric constant had changed. After that, whenever I draw a coil area, I open a window for it, and even if a thin plating layer for rust prevention is unavoidable afterwards, it is better than letting solder mask drag the resonant frequency drifting all over the world.
Piling all these things together points to a view I increasingly agree with: seventy percent of the baseline depends on the coil and the PCB’s physical design, and only the remaining thirty percent turns to software to slowly grind with the environment. The board’s own temperature coefficient, moisture absorption characteristic and interlayer registration deviation are the roots that decide whether the baseline wobbles around as if drunk. Counting on real-time compensation by algorithms can compensate first-order drift but cannot compensate phase shift caused by changes in parasitic capacitance.
So now when I deal with any multilayer pcb manufacturer, I do not talk about how many layers they can do or how many mils of line width at all, but directly throw over a test board with a coil and have them run thermal cycling before pulling out the impedance curve and inductance shift to look at. Only factories that are willing to be serious about lamination uniformity and dielectric material selection produce boards that can withstand the temperature torment of a whole day in the field, with a baseline that sits quietly and does not throw tantrums at you.
Ultimately, with the Metal Detector PCB, eighty percent is physical labor and twenty percent is brain work. If you lay the physical foundation solidly, the digital filtering and threshold judgment behind it become meaningful. The board itself is that 1, and the algorithm is the 0 added afterward. Without that solid vertical stroke at the front, drawing any number of zeros is wasted effort.
Three Months of Tinkering: Ground Splitting, Temperature Drift and Material Choice
I spent three months recently tinkering with a metal detector, stepped in far too many pits, and feel it necessary to write these experiences down.
At first I thought this thing was simple: draw a coil, connect a circuit board, and I could take it to the seaside to hunt for treasure. The result was that when the first revision of the board came back, I was stunned on power-up. Noise from the digital part drowned the whole analog front end, and let alone detecting a coin, it could not even sense an iron lid. Only then did I realize that the metal detector PCB is not a matter of just finding any multilayer board manufacturer to connect the wires and be done.
Later I switched to a factory that specialized in HDI PCBs. Honestly the difference was quite large. When they saw my stackup scheme they pointed out the problem of analog and digital ground splitting. I had completely cut the two grounds apart before and connected them through a 0 ohm resistor, thinking that was isolation. But in fact high-frequency noise crosses through all sorts of parasitic paths, especially on boards of four layers or more, where interlayer capacitive coupling is more severe than you imagine. This HDI PCB supplier suggested that under the ADC I handle the ground plane with a bridging approach, not simply cutting it apart, but leaving an array of parallel ground vias between the analog area and digital area, so that return current takes the shortest path without forming a large ground loop. This trick really worked, and the noise floor of the second revision dropped by nearly half directly.
I also have quite a few lessons on the coil side. Many people agonize over coil inductance and number of turns, but what is really headache-inducing is temperature. When used outdoors in summer, once the sun bakes it, the temperature coefficient of the coil copper foil makes the resonant frequency drift slowly. I tried it on a construction site in Shenzhen in July, and after twenty minutes in the sun, the resonance point ran nearly two hundred hertz. This matter cannot be completely solved by swapping in a C0G capacitor, because the copper foil’s own change with temperature is right there. Later I simply buried an NTC thermistor near the transmit coil to read temperature in real time and let the MCU do dynamic compensation in software. This method is crude, but the effect when running is acceptable, and at least the drift is controlled within an acceptable range.
Speaking of multilayer board suppliers, I also stepped in a pit related to copper plating. The coil of a metal detector is especially sensitive to copper thickness, especially a planar coil wound on the PCB. Some multilayer PCB manufacturers have a rather large process deviation when making outer layer copper thickness, and nominal 1 ounce may actually be a bit above 0.8, with the result that the coil’s Q value was lowered overall and detection depth shrank noticeably. Later I specifically found an HDI PCB manufacturer that could strictly control copper thickness, whose laser blind hole and copper filling processes were indeed stable, and coil consistency became much better, so in batch production there was no need to tune the resonant capacitor board by board.
Temperature does not only affect the coil. After the power amplifier chip heats up, the thermal gradient of the whole board forms a microvolt-level thermoelectric potential drift at the analog front end, and after being amplified by several hundred times of gain in the later stage, it shows up at the output as the baseline slowly wobbling. So when laying out the board, high-power devices must be far from the preamplifier and are best arranged along the heat dissipation direction, not letting hot air blow directly onto the op-amp side. In the third revision I moved the MOSFET of the transmit drive to the board edge, added a small piece of thermal copper, and made a separate thermal relief for the preamplifier area, and baseline stability improved noticeably.
One more thing that many people neglect is the dielectric loss of the PCB itself. FR4 material has a not small loss tangent in the high-frequency range, and if your detection frequency goes up to several tens of kHz or higher, the medium between the coil and PCB will absorb some energy and lower detection sensitivity. Some high-end industrial metal detectors require high-frequency board material, but cost goes up. I compromised a bit, choosing Rogers material for the coil part and keeping ordinary FR4 for the rest, and had the HDI PCB supplier do mixed lamination, so overall cost was controllable and performance was lifted a notch.
After tinkering through these several revisions, my biggest feeling is that making a metal detector PCB is a completely different way of thinking from making an ordinary digital circuit board. It is more like making an analog instrument, where every detail has to be fussed over, from handling of ground to material selection, from temperature compensation to shielding structure, and no link can be fudged. Now taking the machine I tuned out, although I have not yet dug up that legendary gold ring, at least I can precisely find a fifty-cent coin buried ten centimeters deep in sand, and the sense of achievement is far greater than directly buying a finished product.
Shielding, Metal Housings and Partitioning in the Final Design
After working on several revisions of metal detector circuits, I found that what truly separates results is often not how ingenious the front-end circuit is but that unremarkable Metal Detector PCB itself. Many people start by fiddling with op-amp selection and ADC bit depth, forgetting that the first stop the signal reaches is the coil and traces on the board, and if there is a slight loss of control there, nothing later can make up for it.
For coils, I once had faith in the consistency of PCB printed coils, thinking machine processing must be more reliable than hand winding. After running volume I realized that although printed coils have consistent inductance, the interlayer distributed capacitance depends too much on the board factory’s process capability, especially when you meet that kind of multilayer pcb manufacturer who markets itself as such but has erratic quality control, where tolerances over several batches of boards can drive you to calibration collapse. Later I simply changed my thinking: for models whose sensitivity requirement is not so extreme, I directly use a 4-layer board to make a double-sided printed coil with shielding, but layout must be watched closely myself for line width and copper thickness, and then I find an hdi pcb supplier with real RF board experience to prototype, and only then is the residual balance voltage that comes out stable within an acceptable range.
Speaking of factories, the biggest pit I stepped in was blindly trusting so-called big factories. Some multilayer pcb suppliers, although their equipment is advanced, do not take small-batch special boards, or even if they do take them, they run them on production lines tucked into some corner and will not tune parameters for you at all. Instead, there are two specialized high-end prototyping hdi pcb manufacturers who are willing to micro-adjust impedance and dielectric constant for coil boards, and even press layer registration accuracy within plus or minus 0.03 millimeters, which is critical for the balance of a differential coil. I would now rather pay a bit more and permanently work with such suppliers who can communicate process details, rather than looking only at the layer count ceiling bragged about on their website.
Another pit seldom mentioned in design is the effect of the metal housing on the coil’s magnetic field. Once, a bare board measured fine, but after being installed into an aluminum housing, the false alarm rate took off directly, and after checking for half a day I found that eddy currents in the housing had broken the balanced state of the receive coil. Later I developed the habit of including the housing’s metal material and distance in the simulation at the PCB design stage, digging out slots where they should be dug in advance, and deliberately retracting the coil edge inside the board outline by a sufficient distance. These details are not written in books and all come from flipping over again and again by oneself.
Using a modern high-speed ADC plus DSP demodulation is great, but if the digital part and analog front end share a multilayer pcb that is not carefully handled, then after digital noise couples over, even software filtering written to death cannot save it. I now insist on hard isolation between analog and digital partitions, and even do not mind the cost of making a separate metal shielding frame, preferring a larger board area to enclose high-impedance nodes tightly. Many multilayer pcb manufacturers can make stepped holes and buried and blind vias to optimize partitioning, but the premise is that you must plan your functional areas clearly yourself, otherwise even the best process is in vain. Ultimately, metal detector PCB design competes not on any single piece of black technology but on having a clear mind about every interference path.
More Posts
Automotive Rain Sensor PCB: How Optical Paths, Potting, and Fabrication Quality Decide Field Performance
Do you think a rain sensor is just a simple circuit board?
Automotive Sunroof Control PCB Design: Why Heavy Copper and Hall Signal Integrity Matter More Than HDI or High-Frequency Hype
Sunroof control PCB design is easy to get wrong when teams are
Ihr zuverlässiger Lieferant für PCB-Herstellung und PCB-Bestückung aus einer Hand
- Experte für die Produktion kleiner bis mittlerer Chargen
- Hochpräzise PCB-Fertigung und automatisierte Montage
- Zuverlässiger Partner für OEM/ODM-Elektronikprojekte
Geschäftszeiten: (Mo-Sa) von 9:00 bis 18:30