PIR Motion Sensor PCB Design: How Layout, HDI and Stackup Choices Eliminate False Triggers

Building a PIR Board from Scratch: Temperature-Adaptive Thresholds and Split Structures

I came into contact with PIR motion sensors relatively early, about six years ago, because of a smart lighting control project that required building a circuit by hand. At that time it was convenient to buy ready-made modules online, but I always felt that thing was like a black box, and if something went wrong the only option was to replace the whole thing. So I decided to start from chip selection and draw my own PIR Motion Sensor PCB.

The first trouble when drawing the board was signal processing. The raw signal from a PIR sensor is too weak and drifting, like listening to a person speaking softly in a noisy vegetable market. I had to put full effort into the analog front end, amplifying and filtering, and not one link could be omitted. Many people like to use the window comparator routine, setting a fixed reference voltage, and trigger when the signal exceeds this threshold. But I found that if you pin down a fixed threshold, the temperament of this thing becomes particularly strange. When it is cold, it is as sensitive as a neurotic, alarming at the slightest breeze; when it is hot, it simply lies flat and plays dead, with no reaction even when people walk right up to it.

Only later did I figure out that the core problem is not the threshold itself but the swallowing of PIR signal amplitude by ambient temperature. When it is thirty-seven or thirty-eight degrees in summer, the weak voltage change of the sensor may not even reach thirty percent of that at normal temperature. At that time, even if you quibble about whether the comparator hysteresis is 10% or 20%, it does not matter much, since the signal simply cannot reach that threshold.

I thought at that time: since temperature can directly affect signal strength like this, why not simply let the threshold follow temperature? On the next version of the PIR Motion Sensor PCB I added an NTC thermistor, stuffed it next to the sensor, and had the microcontroller sample temperature in real time. In software I set a simple piecewise function, with the trigger threshold dropping a bit for every degree of temperature rise. The benefit of doing this is that in cold weather it is not too sensitive, and in hot weather it can basically maintain detection distance. In actual debugging I found that full linear compensation cannot be used: above 35 degrees the compensation strength must be increased, but the threshold also cannot be below a floor limit, otherwise the noise floor begins to trigger falsely and frequently.

Another easily overlooked point is the physical structure of the PCB itself. PIR sensors actually have rather poor tolerance to mechanical stress and thermal gradients. If you use a single-board design with the sensor soldered on the board, slight thermal expansion and contraction of the board can introduce low-frequency noise. I previously used an ordinary double-layer board, and after sunning for a while in summer there were occasional false alarms. After checking for a long time, I finally suspected the piezoelectric effect caused by board deformation. Later I gritted my teeth and made the sensor and the main control section a split structure, connected in the middle with a flexible cable, with a thick thermal silicone pad under the sensor, and only then was the problem cured at the root. From this experience I realized that a PIR board cannot be treated merely as a circuit board. It is itself a stress-sensitive element.

Speaking of the design of the PCB itself, multilayer boards have an absolute advantage here. For my first prototype I used an ordinary double-sided board, with routing that could only roughly separate one ground plane, and analog and digital signals inevitably crossed each other, so noise coupling was severe. Later I changed to a four-layer board, dedicating one whole layer to a complete ground plane, with the signal layer and power layer completely separated, and signal integrity improved by more than one grade at once. At that point I began stably cooperating with a reliable multilayer pcb manufacturer, who could guarantee interlayer dielectric thickness and copper foil uniformity, which is critical for analog circuits. For more complex models later I simply went to the HDI process, because the sensor circuit plus the wireless module made component density too high, and ordinary vias simply could not route it. At that time you need an hdi pcb manufacturer who is strict in controlling microvia and buried-blind-via processes, otherwise one misalignment can scrap the whole board.

Of course, choosing suppliers is also a lesson. At first I felt that a multilayer pcb supplier with a low quote was good, and I got burned. One batch of boards was said to be four layers, but in fact the inner copper foil was thin, and the impedance of the ground plane was quite a lot higher than my simulation, directly affecting the amplifier’s noise floor. Later I learned my lesson and switched to an hdi pcb supplier that emphasized quality control, who could do impedance control at plus or minus 5% and attached a test report with every batch. The price was about thirty percent more expensive, but that feeling of not having to worry about board-level hidden dangers was worth the price.

Low-Power Design: MCU Choice, Layer Count and Adaptive Baselines

After doing this kind of low-power sensor project for a long time you find that what really gives a headache is never the PIR probe itself but how that small PIR Motion Sensor PCB sits in the whole system and who it cooperates with. Many people start by agonizing over the noise figure of the signal amplification circuit or insist on pushing the threshold compensation algorithm to the extreme, only to neglect something more fundamental: the physical structure and manufacturing tolerance of the board itself. A project I handled last year stumbled on exactly this.

At that time we wanted to make the probe a round board less than two centimeters in diameter and also stuff it into a very thin hemispherical shell. A conventional double-sided board always felt wrong after being drawn. No matter how the small analog front-end signal traces wound, they passed through the digital area, and the ground return was cut up in a mess. Later we simply changed our thinking and found a factory that could make HDI PCBs, changing to laser blind and stacked holes, and put the whole analog front end on a layer of separate microstrip, with a complete ground layer directly on the back. In this way the MCU and wireless part were stuffed on the other side, with a layer of thick copper core board in between, and crosstalk dropped at once. Interestingly, that hdi pcb manufacturer originally made phone motherboards, and during communication their engineers even felt that our kind of sensor board was too simple, but precisely because of their obsession with impedance control, they saved me an op-amp stage: the signal coming from the probe could satisfy the signal-to-noise ratio going directly into the MCU’s differential ADC.

Speaking of MCUs, there are too many choices now, but for a battery-powered PIR, power strategy matters more than the core itself. I do not much like the general-purpose models that hang all peripherals on the bus, because every time they wake up, the charge consumed just opening each clock tree is not small. I lean toward MCUs with an asynchronous event system that lets the analog comparator trigger the RF transceiver directly while the CPU is fully asleep. How low can the average power consumption of the whole board be made this way? I used a silicon-anode coin cell, and at an upload rate of once every 10 seconds, it ran nearly two years and still had power, and the measured average current really can be pressed to a bit over 1 microamp. This number is lower than the theoretical value on many datasheets, and the secret is actually just letting all power-consuming modules be on only during the few milliseconds they should be on.

The choice of board layer count also affects power consumption, though not directly. I tried using an eight-layer board for a PIR signal chain with impedance matching and found that changes in dielectric thickness have a subtle influence on the distributed capacitance of the microstrip line, causing some boards in the same batch to have particularly clean oscillation waveforms while others had small glitches on the edges. If this glitch is not eaten by the back-end filter, the MCU may be falsely triggered awake, and every extra false wake-up is burning the battery. Later, when I chatted with a multilayer pcb supplier I had been cooperating with, they suggested placing all RF and analog key traces between symmetric-structure core layers and giving up using the outer layers for high-speed signals. After this modification, mass-production consistency improved enormously. So whether a multilayer board supplier is reliable is not entirely about how fine a line they can make, but whether they can explain clearly how the thermal expansion coefficient of the lamination structure affects the microstrip line parameters of the board when temperature changes. PIR probes hang outdoors year-round, and when the morning-evening temperature difference comes, the board deforms slightly, the signal shifts a few millivolts, and the threshold has to be readjusted.

I actually do not like to make adaptive thresholds complicated. Some schemes use an MCU to run a simple Kalman filter to track the background baseline, and the effect is very good, the key being that the time constant of baseline update must be much slower than human movement and faster than environmental warming. As for algorithms, what is feared most when writing them into firmware is overfitting. Actual installation environments are all kinds, with curtains fluttering, air conditioner outlets, and pet fur reflections, and these false alarm sources simply cannot all be simulated in a lab. My approach now is to leave a test point on the PIR Motion Sensor PCB to bring out the raw waveform after front-end amplification, sample a batch before volume production in various interference scenarios to record data, and then determine a conservative but not sluggish decision logic by statistical methods and burn it into the MCU. After this process, the false alarm rate is an order of magnitude lower than that of ready-made modules relying on so-called smart algorithms.

As for finding a board factory, I now strongly reject the kind of multilayer pcb manufacturer that only sends gerber files over and then quotes.