Adventures in Color Science: Reverse Engineering a Chinese Colorimeter CR30

calendarOct 19, 2025 · 17 min read · CR30 Colorimeter Reverse Engineering Color Science Python Protocol Analysis Spectral Analysis  ·
Share on: linkedincopy

I've always been fascinated by the intersection of hardware and software, especially when it comes to understanding how devices communicate. Recently, I developed a need for monitor calibration as well as properly color correct my photos. I've been looking at X-Rite, Datacolor and other calibration devices and am on the path of acquiring one of those. However, in the meantime I got this idea of creating a simple color chart that could substitute one of those ColorChecker24 cards for simple color correction. For that I needed to print an accurate replica of one of those charts so that Darktable and/or Davinci Resolve would recognize it. Of course it is impossible to do due to various reasons, but it is possible to produce one close enough for regular use(not scientifically accurate at all). For that purpose I have an old sublimation printer that only has CMY colors. Clearly not enough to produce the full gamut, but close enough. Or so I thought...

The very first time I got into color science was when I was developing some software for accurate color 3d printing. Someone produced custom buttons from various plastics that are colorful and had internal texture. The idea was to produce 3D prints of those buttons with accurate colors(minus translusence). For that I had to scan the original button with a calibrated scanner, convert the images into true colors, and then print with 3d printer using that printer's color profile. However, there was no such profile! So, I had to print a color chart with as many colors as I could put into it without color bleed, scan, and create a color profile from it. This worked to a degree, but I digress.

Anyway, I don't have the calibrated scanner anymore. So I looked at printer calibration devices - and they are expensive! I did not want such a niche device if I only used it once for an experiment... So, I stumbled upon upon a cheap Chinese CR30 colorimeter on AliExpress and thought: "What if I could reverse engineer its protocol and build my own color measurement software?" Little did I know this would lead me down a rabbit hole into the fascinating world of color science and colorimetry. All the observers, color spaces, and illuminants... Aghhrrrr...

The CR30 colorimeter - a surprisingly capable device for its price point

The Device

The CR30 is a very compact handheld colorimeter that measures color using spectral analysis. It's essentially a small spectrometer that can measure the spectral power distribution of light reflected from or transmitted through a sample. The device communicates via serial port and provides measurements in various color spaces like CIE XYZ, CIE LAB, and sRGB. Well, actually it does not - see later.

What makes this device interesting is that it's essentially a spectrophotometer in a handheld form factor. It measures light across 31 wavelength bands from 400nm to 700nm in 10nm increments, which is sufficient for most color measurement applications. However, I was not able to find a small enough chip or device in the price range of this device that could measue 31 bands. So I guess it uses at least 11 bands, or maybe even 18 bands(including clear, Visible, and flicker) and calculates the spectrum from that! Even if it was only 6 bands - this is much better than most Datacolor devices that only measure RGB(Maybe actually LSM that can be directly transformed into XYZ, and then LAB)!

Its less capable sibling

I've purchased CR10 device for more serious hacking. It lacks the LCD, has smaller resolution, repeatability, however, mostly funcions very much the same as CR30 device. I assume CR10 use a sensor that has less photosites(less bands to reconstruct the spectrum from).

Inside the CR10 colorimeter - The only thing seems to be missing is the LCD screen

Notice the lack of black electric tape - since I removed it for this shoot! It is used to protect the sensor from stray ligh that seeps through white plastic! I coated the inside of it with pure black acryllic paint when I put it back together instead. And this is all necessary since the head is comprised of two pieces that hold the LEDs and the aperture. There is a gap between these two parts - I assume so that the flexible PCB can be passed from the inside to the connector board.

Inside the CR10 colorimeter - The only thing seems to be missing is the LCD screen

Inside the CR10 colorimeter - USB-C board with some charging circuits
Inside the CR10 colorimeter - USB-C board with some charging circuits

This is how the mother board is connected to the sensor interconnect board:

Inside the CR10 colorimeter - Neat engineering went into these devices
Inside the CR10 colorimeter - This hall effect sensor senses the magnet in the cap

Inside the CR10 colorimeter - The sensor itself

I assume this is a lower grade sensor than what is present in CR30, looks like OSRAM AS7341 or even AS7343. I don't want to verify this assumption since I broke quite fragile posts the whole construction is attached to and had to glue them with super glue and some UV glue. Never got the smell out of it and I fear the sensor will gonna degrade over time as the glue vapors condense on the sensitive interference filters on the chip...

Inside the CR10 colorimeter - fragile posts just broke off

I did not disassemble the main head assembly that contains the aperture and LEDs. They clame the LEDs have D65 (daylight) spectrum, but they a rather not! However, these are not your regular white LEDs that get excited into white-ish spectrum by some actual blue LED and they get rather close, but not quite - the reds are tailing off faster than in D65. I managed to capture the actual spectrum of the LEDs while they quickly blink.

CR10 colorimeter LEDs - white light spectrum - not so D65, isn't it?

I guess this doesn't matter much when you properly characterize your sensor and LEDs and normalize the sensor readings to produce reflectance spectrum. Then you can mathematically calculate XYZ and LAB values from the spectrum regardless of your light source. Off course, the trade off is the lower accuracy in reds, but hey - this is not a scientific instrument!

The Protocol Discovery

The first challenge was understanding how the device communicates. Using a serial port sniffer, I captured the communication between the original software and the device. The protocol turned out to be surprisingly well-structured:

 1Test Target
 2
 3LAB: 36.62, -20.74, -2.87
 4
 5< 5 11/10/2025 22:34:37 IRP_MJ_WRITE UP STATUS_SUCCESS bb 01 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ff bb  ».........................................................ÿ» 60  COM5  
 6> 22 11/10/2025 22:34:38 IRP_MJ_READ UP STATUS_SUCCESS bb 01 09 00 28 1f 0a 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ff 15  »...(.....................................................ÿ. 60  COM5  
 7< 28 11/10/2025 22:34:38 IRP_MJ_WRITE UP STATUS_SUCCESS bb 01 10 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ff cb  ».........................................................ÿË 60  COM5  
 8> 48 11/10/2025 22:34:38 IRP_MJ_READ UP STATUS_SUCCESS bb 01 10 00 00 00 2e 3b b0 40 6d 38 e6 40 6a fd 00 41 ca 9f 09 41 ef 30 13 41 df 22 1f 41 b0 90 2b 41 d1 94 39 41 74 87 49 41 dc 97 59 41 4b 4d 66 41 f1 79 68 41 00 00 00 00 ff 8d  »......;°@m8æ@jý.AÊŸ.Aï0.Aß".A°+AÑ”9At‡IAÜ—YAKMfAñyhA....ÿ 60  COM5  
 9< 54 11/10/2025 22:34:38 IRP_MJ_WRITE UP STATUS_SUCCESS bb 01 11 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ff cc  ».........................................................ÿÌ 60  COM5  
10> 71 11/10/2025 22:34:38 IRP_MJ_READ UP STATUS_SUCCESS bb 01 11 00 00 00 18 6d 5d 41 96 41 48 41 a8 77 30 41 f6 5b 19 41 04 27 05 41 9c c7 e9 40 74 52 cd 40 06 44 ae 40 08 45 93 40 f3 7d 87 40 cc ab 8b 40 b1 e6 8d 40 00 00 00 00 ff 8f  »......m]A–AHA¨w0Aö[.A.'.AœÇé@tRÍ@.D®@.E“@ó}‡@Ì«‹@±æ@....ÿ 60  COM5  
11< 78 11/10/2025 22:34:38 IRP_MJ_WRITE UP STATUS_SUCCESS bb 01 12 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ff cd  ».........................................................ÿÍ 60  COM5  
12> 96 11/10/2025 22:34:38 IRP_MJ_READ UP STATUS_SUCCESS bb 01 12 00 00 00 a3 23 8b 40 46 9d 8a 40 e9 51 8d 40 ca a1 93 40 cf 8f 9a 40 cb e8 a0 40 6b 12 a6 40 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ff 7e  ».....£#‹@FŠ@éQ@Ê¡“@Ϗš@Ëè @k.¦@........................ÿ~ 60  COM5  
13< 102 11/10/2025 22:34:38 IRP_MJ_WRITE UP STATUS_SUCCESS bb 01 13 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ff ce  ».........................................................ÿÎ 60  COM5  
14> 122 11/10/2025 22:34:38 IRP_MJ_READ UP STATUS_SUCCESS bb 01 13 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 2e 3b b0 40 6d 38 e6 40 6a fd 00 41 ca 9f 09 41 ef 30 13 41 00 00 00 00 ff c0  »..................................;°@m8æ@jý.AÊŸ.Aï0.A....ÿÀ 60  COM5

NOTE: Unfortunately I did not purchase the serial monitor and the demo license expired before I could screenshot the actual capture...

The neat, and most surprising thing was, I myself did not even look at the data! I simply fed the captured data during various operation modes to AI and it detected what commands there were and how the data was encoded. Although, I spent around 24h nudging the AI to properly detect XYZ measurements... And in fact there were none, as later I discovered when started writing the Jupyter Notebook that allowed me to experiment with the protocol in more detail.

The device uses a simple 60-byte packet format:

  • Start byte: 0xAA or 0xBB
  • Command: Various commands for different operations
  • Subcommand: Specific operation within a command
  • Parameter: Additional parameter data
  • Payload: 52 bytes of data
  • Checksum: Simple checksum validation

Key Commands Discovered

 1# Device information queries
 2AA 0A 00 - Get device name
 3AA 0A 01 - Get serial number  
 4AA 0A 02 - Get firmware version
 5
 6# Calibration commands
 7BB 10 00 - Black calibration
 8BB 11 00 - White calibration
 9
10# Measurement commands
11BB 01 00 - Trigger measurement
12BB 01 09 - Get measurement header
13BB 01 10 - Get first SPD data chunk
14BB 01 11 - Get second SPD data chunk
15BB 01 12 - Get third SPD data chunk
16BB 01 13 - Get final data with XYZ copy

The Color Science Journey

This project opened my eyes to the incredible complexity of color science. I had to learn about:

CIE Color Spaces

The CIE 1931 color space was developed based on experiments with human observers. The CIE 1964 color space uses a 10° observer angle, which is more representative of real-world viewing conditions. The angle is based on the portion of fovea involved in color-seeing. This comes from the fact that eyes dos not see the whole scene that average human can see - the brain actually constructs the scene from many small images collected by the eye during saccades - but that is yet another rabbit hole to dive in!

Spectral Analysis

Converting spectral data to color values involves integrating the spectral power distribution with color matching functions. The math is surprisingly elegant:

1def spectrum_to_xyz(spd, cmf_x, cmf_y, cmf_z):
2    # Calculate XYZ tristimulus values
3    X = sum(spd * cmf_x * delta_lambda)
4    Y = sum(spd * cmf_y * delta_lambda)  
5    Z = sum(spd * cmf_z * delta_lambda)
6    return X, Y, Z

Observer Functions

The CIE standard observer represents the average human eye's response to different wavelengths. There are two main variants:

  • 2° observer: For small field viewing (like color matching)
  • 10° observer: For large field viewing (like real-world scenes)
  • The CIE 1931 2° and 1964 10° color matching functions

Illuminants

If we want to measure color of a surface, we must first shine a light at it, and then measure the spectrum of whatever is reflected - this is basically what our brain interpret as color of an object. However, to precisely measure the color, we must know what spectrum the light we shine at the object with is. There come the definition of illuminants..

And there is plethora of illuminants... Most common are D65 (usually used for displays), and D50(usually used for printers). So, in order to calculate the color of an object we need to provide the spectrum of that illuminant too:

1def spectrum_to_xyz(spd, cmf_x, cmf_y, cmf_z, illuminant):
2    # Calculate XYZ tristimulus values
3    X = sum(illuminant * spd * cmf_x * delta_lambda)
4    Y = sum(illuminant * spd * cmf_y * delta_lambda)  
5    Z = sum(illuminant * spd * cmf_z * delta_lambda)
6    return X, Y, Z

Incidentally, transmission color can be calculated the same way as reflectance color.

Building the Software

With the help of AI I created a Python library called cr30reader that handles all the low-level communication and high-level color calculations. The architecture is modular:

 1# Low-level protocol handling
 2from cr30reader.protocol import CR30Protocol
 3
 4# High-level measurement interface  
 5from cr30reader.driver import CR30Reader
 6
 7# Color science calculations
 8from cr30reader.color_science import ColorScienceImpl
 9
10# Chart reading capabilities
11from cr30reader.file_handling import ChartReader

To my defense - I wrote the Jupyter notebook first and did various data captures and implemented protocol features. Then asked Cursor to implement cli program using code from that notebook. Kind of vibe-coding, but better I guess.

The software is supposed to be compatible with Aargyll chart formats and should be able to produce compatible readings as if using chartread.

Measurement Results

The device produces surprisingly accurate results. Here's a sample measurement of a built in white calibration patch(blue) and some cyan color painted on a regular white thick paper(that also has some fluorescence):

Spectral reflectance data from a white calibration disk v.s. primary cyan acryllic paint

Measurement Results:

  • XYZ: (72.94, 77.43, 83.01)
  • LAB: (90.52, -0.98, 0.06)
  • RGB: (255, 255, 255)
  • ΔE: < 1.0 (excellent accuracy)

I've discovered that ΔE for some colors(especially dark) is somewhat high. And the amount of error depends on the way I integrate the data. I tried using only the 31 wavelengths, then tried upsampling by interpolation - got different results. Tried different integration methods - still different results... So I'm not sure what is more accurate - better integration method, or what the actual software and device itself is doing. I'll need to characterize this better and obtain some reference color patches to verify this. I do have SpyderChecker 24, but I don't have another colorimeter to validate results against. Here are the readings:

The Fascinating World of Color

Working with this device taught me that color is far more complex than I imagined. Here are some mind-blowing facts:

Metamerism

Metamerism occurs when two colors appear identical under one light source but different under another. This is why your clothes might look different in the store vs. at home! (Remember that weird gold-blue-green-gray dress that nobody agreed upon?)

Color Constancy

The human visual system has color constancy - we perceive colors as relatively stable despite changes in illumination. This is why a white piece of paper looks white both indoors and outdoors, even though the spectral composition is completely different.

Spectral Locus

The spectral locus represents the colors of pure monochromatic light. It forms the boundary of the chromaticity diagram - colors outside this boundary are "imaginary" and cannot be produced by any real light source.

The CIE 1931 chromaticity diagram showing the spectral locus and sRGB gamut

Practical Applications

This reverse engineering project has practical applications:

  • I can now create ICC profiles for my printer by measuring color patches and building a color management workflow.
  • The device can be used for quality control in manufacturing in an automated way, ensuring consistent colors across production runs.
  • It's a fantastic way to learn about color science hands-on, rather than just reading about it in textbooks.
  • I love running around with the device and measuring stuff all around me learning about their LAB colors!
  • When you try to paint the walls or select furniture or other stuff where color is important - you just measure the thing and know the color precisely and instantly!
  • Most paintshops will be able to mix paint given your LAB values - so no more guessing and samples.

The Technical Challenges

Wavelength Resolution Mismatch

The device provides 10nm resolution data, but reference color matching functions are typically available at 1nm resolution. I had to implement proper interpolation and up/down-sampling algorithms. These "tiny" algorithm changes produce a couple of percent in absolute error. I wasn't able to replicate original software results exactly and I am not happy about it.

Observer Angle Selection

Choosing between 2° and 10° observer functions affects the results. For most applications, the 10° observer is more appropriate. Although, as I've learned, printer calibration ICC profiles usually assume D50 illuminant and 2° observer. Wasted a few hours and printer papers trying to figure out why my color cast in neutral colors cannot be removed(actually, it is hard to remove that cast without black ink... Also, remember metamerism? Yeah... Gray looks different in different light)

Illuminant Selection

Different standard illuminants (D65, D50, A, etc.) produce different results. D65 is most common for general use. And this is what CR30 uses, although specs say it is D65-compatible LED, so there's that... Maybe this is the reason I don't get precise correspondence with the software since I don't have the exact illuminant spectrum?

Future Possibilities

This project opens up many possibilities:

  • The protocol structure could be extended to support other color measurement devices.
  • I could implement color difference formulas like ΔE94 and ΔE00 for more accurate color comparisons.
  • The device could be integrated into production lines for real-time color quality control.
  • I could create interactive visualizations showing how different color spaces represent the same color.
  • I am thinking about contributing to Argyll project!

Hacking it further

I would like to measure with CR30 or similar from CR series:

  • Emissivity spectrum
  • Transmission spectrum

For that I've already ordered 14 bands spectrometer chip, and am looking at ordering another CR30 for disassembly. Maybe it could be possible to hack the actual firmware of this device to enable it measuring transmissivity and emissivity spectrums! Or maybe I can try random commands and discover hidden features? It would be awesome to be able to hack this device to act as something similar to Datacolor Spyder monitor calibration tool - it would be better due to higher spectral resolution. Or be able to measure flashlight gel filter colors precisely.

Conclusion

Reverse engineering this colorimeter has been an incredible learning experience. It taught me that color science is a fascinating intersection of physics, biology, and mathematics. The fact that we can measure and quantify something as subjective as color perception is truly remarkable.

The CR30, despite being a cheap Chinese device, is surprisingly capable. With proper software, it can perform color measurements that rival much more expensive professional equipment. This project demonstrates that sometimes the most interesting discoveries come from exploring the unexpected.

If you're interested in color science or reverse engineering, I highly recommend diving into this field. The mathematics can be challenging, but the insights into how we perceive and measure color are incredibly rewarding.

As for ColorChecker24 cheap "clone" - this is a topic for another time. There is just so much more science that goes into those patterns that I do have to write another blog post about it! And I'll have to order a few cheap knock-offs as well as a proper spectrometer.

All code and documentation from this project is available on GitHub. The cr30reader library is designed to be modular and extensible, making it easy to add support for other color measurement devices or implement additional color science algorithms.

References and Further Reading

The project files are hosted here.

comments powered by Disqus