What is it?

This is a small (17.9 mm x 10.3 mm) breakout for PixArt Imaging's PAF9701C1 low-power IR thermal array sensor.

From the data sheet: "The PAF9701C1 is an 8 x 8 pixel low-power infrared (IR) thermal array sensor with integrated temperature computation function. It measures target object's temperature without direct contact (with) the object."

The sensor uses simple I2C communications at up to 400 kHz bus speed allowing the host to read directly both 8 x 8 pixels of object (target) temperature as well as ambient temperature. The sensor has an interrupt pin that can be configured for data ready or over-/under-temperature threshold (either absolute or differential) alerts. Furthermore, each pixel that exceeds the threshold limit can easily be identified. This makes it straightforward to track motion in the temperature field for directional transit monitoring, estimating people count in a room, or simple object detection.

The breakout board exposes 3V3 (3.0 V to 3.6 V only) and GND, SDA/SCL for I2C communications as well as an I2C address pin which selects either 0x34 (LOW) or 0x57 (HIGH) as the PAF9701 I2C address allowing multiple sensors to be used on the same I2C bus. The breakout exposes the interrupt pin, a reset pin (active LOW), and a shutdown pin (active HIGH) that disables I2C communication and puts the sensor in the lowest-power state. The sensor can also be placed in a suspend state by writing 0x00 to the OUTPUT_ENABLE register. Pullup (SDA/SCL/SHUTDOWN/Alert) and pulldown (ADO) resistors are already on the board.

The PAF9701C1 is similar to Panasonic's AMG8833 8 x 8 pixel thermal IR sensor as can be seen in the following table:

But there are notable differences. The PAF9701 sensor (8 x 8 x 3.85 mm) is 62% of the volume of the AMG8833 (8 x 11.6 x 4.3 mm) so there is less thermal mass and lower latency in measuring changes to ambient temperature. The normal run mode power of the PAF9701C1 is less than half that of the AMG8833 while the accuracy is more than twice as high. The object temperature range of the PAF9701C1 is significantly wider than that of the AMG8833.

While both sensors offer 8 x 8 pixels at 10 Hz, the PAF9701 sample rate is continuously variable from 10 Hz to as low as 0.0007 Hz (i.e., 1342 second inter-frame time). Furthermore, the PAF9701C1 has an auto power save mode which allows the sensor sample rate to automatically drop until an alert condition is met at which time the sensor returns to normal run mode. In the lowest auto power save mode the sensor uses just 52 uA. This feature can save significant power making battery-operated applications with the PAF9701 feasible.

The PAF9701 has a variety of data filters available to smooth the object temperature field and improve the SNR (signal-to-noise ratio). The user can select an IIR (Infinite Impulse Response) filter, a Moving Average filter, or a Normal Average filter. For IIR, the user can select the IIR level for the filter. For the latter two, number of frames to be averaged from one to eight can be selected.

Lastly, the user can select the emissivity (default to 0.98) of the object and the decay rate of the object temperature (default to 0) due to the metal sensor cover. These make it possible for the user to calibrate the sensor further for specific measurement conditions.

How does one use it?

There is a github repository here containing well-commented Arduino sketches that should run on most 3V3 microcontrollers with an I2C port for the sensor and an SPI port for the display.

The normalmode sketch shows how to initialize the PAF9701 image sensor in normal run mode, configure the data filters and image orientation, set up the data ready interrupt, read the data and plot the properly scaled data on the serial monitor and on a 160 x 128 pixel Adafruit TFT color display.

The autoPowerSaveMode sketch demonstrates how to take advantage of the built-in power management mode. In autoPowerSaveMode, the sensor starts out in normal run mode. If one sets a data ready alert, the sensor simply behaves as it does in the normal run mode sketch and updates the display at the requested sample rate (I typically use 4 Hz). If the user instead specifies a temperature threshold limit (either absolute as in this sketch or differential) the normal run mode checks for an alert condition at the sample rate but only updates the display when the alert condition obtains, i.e., when there is an object in the field of view with enough pixels above/below the programmed thresholds.

Normal run mode operation (2 mA) transitions to detectMode1 (310 uA) after a user specified delay time (60 seconds in the sketch) without an alert. In detectMode1, the sensor checks for an alert condition every 20 seconds. If there is no alert condition observed within the delay time in DetectMode 1 the sensor drops down into detectMode2 (52 uA), where the sensor checks for an alert every 120 seconds. Once an alert condition is detected the sensor returns to normal run mode and the process starts over again.

So the power usage drops by factors of ~6-7 at each stage and the latency increases by about the same amount. This provides a way for the user to manage power usage that is very convenient and effective, and offers enough flexibility that the power usage and latency can be tailored to the specific application without elaborate host programming.

Why did you make it?

I love sensors (who doesn't?), especially imaging sensors.

I designed my own version of the cheap ESPEye as a security device, but this really isn't practical for low-power, battery-operated applications except maybe to catch an intruder in the act, once. This is because the ESP32 uses a lot of power and there isn't an API for using the cheap cameras with lower power MCUs like the STM32L4.

Fortunately, there are several new imaging sensor that make practical low-power imaging applications possible. Sensors like ST's VL53L5 16 x 16 pixel ranging camera, PixArt Imaging's PAA3905 35 x 35 pixel optical flow camera, and the PAF9701C1 8 x 8 thermal IR imaging array.

These are not traditional imaging cameras, and each offers a new way to map events in 2D either via range (distance), optical flow (velocity) or temperature. These sensors are easy to use with simple serial (I2C or SPI) peripherals, don't require a lot of GPIOs to make work, and are compatible with ultra-low power applications.

Panasonic (AMG8833) and Melexis (MLX90621) have offered thermal IR imaging sensors for years now and I have used them both. But I am impressed with the accuracy and performance of PixArt Imaging's PAF9701C1 offering. Especially its small size, very low power usage, ease of use, and many features missing from either of the other two sensors. The data filtering and auto power save modes, in particular, make the PAF9701C1 a winner!

What makes it special?

This is a small breakout board that is easy to integrate into your application.

There are well-commented Arduino sketches that make it easy to use the full features of the sensor.

This is an ultra-low-power sensor that is well-suited to remote, battery-operated device applications.

Order some breakout pcbs from OSH Park and assemble some of your own or order the fully assembled and tested PAF9701C1 breakout board here and see how easy and useful low-power thermal imaging can be!