This paper presents a battery powered 8-bit tonal resolution colorimetric sensor circuit for paper microfluidic assays using low cost photo-detection circuitry and a low-power LED light source. A colorimetric 3×3-pixel array reader was developed for rural environments where resources and personnel are limited. The device sports an ultralow-power E-ink paper display. The colorimetric device includes integrated GPS functionality and EEPROM memory to log measurements with geo-tags for possible analysis of regional trends.
The device competes with colour intensity measurement techniques using smartphone cameras, but proves to be a cheaper solution, compensating for the typical performance variations between cameras of different brands of smartphones. Inexpensive methods for quantifying bacterial assays have been shown using desktop scanners, which are not portable, and cameras, which suffer severely from changes in ambient light in different environments. Promising colorimetric detection results have been demonstrated using devices such as video cameras5, digital colour analysers6, flatbed scanners7 or custom portable readers8. The major drawback of most of these methods is the need for specialized instrumentation and for image analysis on a computer.
A colorimetric array reader system was developed for environments such as rural villages in the developing world where resources and medical personnel are extremely limited. The goal is to quantify bacteria in a sample on a paper-based array strip by measuring the colour intensity of dots in an array. The focus lies on the investigation and implementation of colorimetric sensing techniques rather than different techniques for quantifying bacteria, which is the broader goal of the project. The GPS functionality along with the database generated in the memory of the device will be used to generate geo-tagged medical data for research purposes.
Colour is in its nature an extremely hard packet of energy to measure objectively, as it depends on multiple factors such as the observer, lighting, reference colours and the sensitivity of the sensor. For this device it was decided that low power usage is the most important feature and thus the minimum amount of source light needs to be supplied. Various standard colour spaces such as RGB, HSI and HSV exist that serves to quantify specific colour and it is easy to convert between the colour spaces. It was however decided that, seeing that for a specific colour test in a medical application it would only be the intensity of the colour that changes depending on its concentration, the intensity value in the HSI colour space would be sufficient to determine the concentration of the analyte with at least an 8-bit resolution. This makes it possible for us to only use one colour of light and thus dissipating less power (a third of the RGB source solution).
The device does however feature an RGB light source in order to expand on functions and tests that the device could later on be developed for. With an RGB value for a colour it is easy to represent it on a PC screen or to convert it to the HSI colour space which is more intuitive for the way we perceive colour with our eyes.
In order to measure colour on a sample it is necessary to have a light source of which the radiated light wavelengths are known and then a photo sensitive device needs to be positioned above the illuminated surface to capture the reflected or transmitted light, depending on the technique used. It was found that measurement by reflection is less sensitive to irregularities on the paper surface and small dirt dots on the sample than for transmission and thus this device operates on a reflection principle as shown in Figure 1.
The most common sensor devices that are used in light sensitive systems are photodiodes and phototransistors. The main differences between these two are the gain and speed at which the device operates. Photodiodes are typically faster than phototransistors but have the disadvantage of having a typical gain of 100 times less than a phototransistor.
A portable colorimetric reader9 that is very similar to the proposed solution was developed in by a research team in Daejeon, Korea but uses an expensive internal reflective coating to reflect light from one source to all the assay dots. This device also does not have GPS connectivity or database logging functions like the proposed solution.
METHODOLOGY AND DESIGN CONSIDERATIONS
Spectrometer and test paper arrays
The first step was to take light intensity measurements of the environment in which the developed device would be operating. An industry standard spectrometer (Avantes USB-2 Fibre Optic Spectrometer) was used to take measurements of both direct/indirect and indoor/outdoor light in units of lux.
Colorimetric assays were created on which the developed device can be evaluated, see Figure 2. Standard food colouring was used and diluted consecutively by a ratio of two (to achieve an 8-bit binary resolution) and then a small amount of each concentration was put on a strip in array form by means of surgical needles.
The colours that were used are blue, red, green and yellow. The Avantes spectrometer was used to take measurements on the colorimetric assay strips both in reflective and transmission mode for 400-900nm wavelengths, see Error! Reference source not found.. The reflective mode achieved a much higher accuracy and better response than measurements for transmission through the assay paper. The paper’s texture influences the readings a lot and it can be suggested to use smoother paper than cellulose paper such as Whatman (I) chromatography paper. Dirt dots or scratches on the paper significantly influence measurements and should be avoided. The colour strips in Figure 2 were used to obtain illuminance values as Figure 3 with the spectrometer.
Functional block diagram
The system’s main blocks are depicted in Figure 4, it is important to realise that these need not all be physical elements as the microcontroller already contains internal hardware like an ADC and internal memory.
The microcontroller is the main component that determines the voltage levels for most of the components in the system. The maximum voltage for the microcontroller is 3.6 V. The GPS, external EEPROM and the E-ink paper display can operate on this supply level and thus it was decided to have one common voltage level of 3.6 V for the whole system. A Li-Po battery is used to power the system, when fully charged it has a voltage of around 4.2 V and cuts of to 0 V when the voltage drops below 3.5 V. A switch-mode voltage regulator circuit from Pololu (0J7031) is implemented which can buck and boost the voltage level to keep it at a constant 3.6 V for the system power supply. It is important that this voltage is constant because the reference voltage, Vref, of the ADC needs to be constant for all the measurements and the brightness of the LED light source also needs to be kept constant.
The 44-pin microcontroller that was used is a PIC24F32KA304 from Microchip, which is a 16-bit MCU with 13 (12-bit) ADC channels (needed for all the optical detectors) that can sample at 100 samples per second. A PICKIT3 was used to program the microcontroller. The MCU also has 512 bytes of Data EEPROM which are used for the logging of measurement results. The oscillator of the microcontroller was set up as “FRC” which means it oscillates at a frequency of 8 MHz and then an instruction cycle is defined as .
The device needs to measure 9 colour dots simultaneously and thus 9 ADC channels are required. The ADC for the optical receiver circuit is set up in continuous sampling mode in order for all 9 channels to be sampled and converted consecutively as elegantly and quickly as possible. The ports are configured as inputs via the TRISx registers and then the digital buffers are disabled on the used ports with the ANSx registers. The Tad is equal to the clock period and the ADC was set to sample for 60×Tad and once the conversion is finished the next channel is sampled until all 9 channels have been sampled/converted and an interrupt is generated where the ADC1BUF with the results is accessed.
An external EEPROM was added to the system via the I2C protocol in order to reach the specification of storing 50 data measurements in the EEPROM memory. The internal EEPROM memory of the PIC is only 512 bytes and this is not sufficient, thus the 24LC256 from Microchip was added, it sports 256k bit (32k bytes) of memory and can operate on a 3.6 V supply. The external EEPROM uses I2C to communicate with the MPU at a frequency of 400 kHz but the closest rate that the MPU can achieve is about 385 kHz by loading the clock speed register with the value of 9. The SDA and SCL lines then require pull-up resistors of about 2.2 kΩ. It is extremely important to allow the EEPROM time to complete a write when a page-write was done because any commands while it is still writing will stop the write cycle and start the new command, thus a delay of 10 milliseconds was implemented after each write.
The GPS that was used is an EM-408 from USGlobalSat based on the SiRF StarIII chipset. It communicates via a serial UART connection with the microcontroller, continuously transmitting to the microcontroller. The WAAS function for the GPS is disabled as it is not available in our country (RSA). The GPS module will draw up to 75 mA while operating and this will drop to 0.4 mA as the enabled pin is pulled low and the sleep mode is activated. The frequency at which the GPS outputs a NMEA (RMC, GGA, GSA and GSV consecutively) sentence is 1 Hz and the baudrate used is 4800. The RMC (Recommended Minimum Specific GNSS data) sentence is used for this application seeing that only the date and coordinates would be required to be extracted. An RMC sentence has a total length of 69 bytes.
An RGB light source was developed with focus on small size and low power dissipation. The LEDs were chosen to emit only narrow bands around 485 nm (Blue), 518 nm (Green) and 632 nm (Red). Transistors (NPN) are used to switch the LEDs with the microcontroller, see Figure 5.
Equation 1 was used to calculate the current limiting resistor:
In this equation Vcol is the specific voltage drop over each colour (ex. Vred = 1.95, Vgreen = 2.85, Vblue = 2.75) and ILED is the desired current through a single LED.
The option for using a PWM signal to drive the LEDs through the transistors was investigated in order to minimize power dissipation. The fact however is that the ADC capacitor only takes about 10 clock cycles to charge up and the lights can be switched off already, making pulse width modulation redundant. This means that the procedure would work in the way that a whole array of 9 lights of one colour is switched on simultaneously through one transistor, the current per LED is chosen as 2 mA for (30, 40, 20 mcd) as from the datasheet, adding to a total current of 18 mA per transistor if all the LEDs are switched on simultaneously. The RGB LED that was chosen is a SML-LX0404 surface mount LED.
There exist several optical sensors and each comes with a different advantage for either sensitivity or speed. The main options are phototransistors, photodiodes and light dependent resistors. A few designs and alternatives for optical detector circuits were considered. There is a trade-off between cost, performance, simplicity and the requirement to design a circuit from first principles. Simple integrated circuits like the TLS235 or TCS3200 light-to-frequency converters from TAOS were investigated but the size of the individual chips might be a problem. Further options would comprise of differential amplifier designs which would entail several operational amplifiers. A phototransistor circuit (amplification inherent so no external amplification needed) would be preferred over a photodiode circuit (faster than phototransistors but speed is not a factor, varies less with temperature changes) where operational amplifiers would be needed and the cost would be a lot higher.
The circuits in Figure 6 are application circuit examples for the sensor devices.
RL is chosen such that the Vout will be the maximum value of the ADC input of the microcontroller when the light source is at the maximum radiation and the maximum current flows from the collector to the emitter of the phototransistor. This configuration allows for the Vout to increase linearly as the light intensity increases.
A phototransistor from OSRAM called the SFH3310 was investigated as it has a viewing angle of 150 ° and a sensitivity function which ranges from 350-970 nm and peaks at 570 nm which is quite similar to the sensitivity of the human eye and the RGB spectrum falls well within these boundaries. The transistor can accommodate a maximum of 20 mA through the collector. A similar transistor with a smaller viewing angle of 40 ° (smaller is better because any external light sources need to be corrected for in the measurement) from Vishay called the TEPT5600 (or the bigger package TEPT5700) was chosen. The dynamic range for the TEPT5600 is from 440-800 nm and it peaks at 580 nm. For a light exposure of 100 lux the device will generate a photo current in the collector of about 350 μA. As the board was ported to a surface mount implementation the phototransistor that was chosen is the SFH3710 transistor from OSRAM. It has a total viewing angle of 120 degrees, a maximum wavelength sensitivity of 570 nm and a spectral range of 350-950 nm. The dark current is claimed to be 3 nA.
As explained above, the phototransistor that was chosen is the SFH3710. The SPICE model for the transistor was obtained from Osram and imported into LtSPICE for analysis, see Figure 7. It is firstly important to look at the collector current versus the voltage between the collector and emitter. This is commonly referred to as ICVCE curves, see Figure 8.
If the maximum collector current is 600 μA then Eq. 2 where IPCE is the collector current holds
The photocurrent is directly related to the amount of light shining on the surface of the phototransistor and this is referred to as luminous flux, see Figure 9.
According to the datasheet a maximum of 600 μA will flow from the collector to the emitter if 2000 lux luminous intensity is present on the base sensor. The maximum voltage between the collector and emitter would always be less than 3.6 V seeing that the resistor inserted at the emitter would have a voltage drop over it when a current flows from the collector to the emitter. The VCE(sat) is specified to be 100 mV so in order to never reach that point it is assumed that, when calculating the resistor, there will always be at least a 200 mV drop over the transistor and thus the maximum voltage over the resistor is 3.4 V and the maximum current is then 600 μA. This leads to Eq. 3:
When designing the casing (see Figure 10) it was important to keep in mind that low cost is a requirement and that the casing should minimise the amount of external light reaching the phototransistors and also preventing cross-talk between neighbouring phototransistor-light source pairs. The casing was designed in eMachineShop which is a free CAD software program and it can export files as .STL for 3D-printing. The casing sports side flanges to properly align the device sensors to the paper dots. The casing was printed using PLA plastic and the total cost amounted to below R100.
The PC board (see Figure 11) was designed using Altium Circuit Designer and it was manufactured at Central Circuits, Centurion SA. Most of the components like the GPS, E-ink screen, power regulator and battery are external components and headers were simply put in place for each of these.
What is an EPD?
As described in the application note “Driving electronic displays” by EFM32, an Electronic Paper Display (EPD) is a display that has the property of being reflective and bi-stable. The screen is reflective in the sense that it uses no backlight and relies only on ambient light. Bi-stability means that the display can retain its image for an extended period of time (1 month) with no power applied. EPDs are commonly used in e-readers, industrial signage and electronic shelf labels. The EPD does not have a very fast frame rate updating speed (typically 1 second) and is thus suitable for static display applications. The screen draws no current while displaying, only when it needs to update the image.
How does the EPD work?
The pixels in an EPD are made up of small transparent capsules which contain a dark oil. The oil contains white titanium dioxide particles which are negatively charged. Electrodes are placed on the top and bottom of the capsules. In one polarity where the positive potential is on the front electrode the pixel appears white because light is reflected off the white particles. In the opposite case the white particles are drawn to the back electrode and the pixel appears black because light is absorbed by the oil.
The extension board from Pervasive Displays come with the EPD on purchase, note that this specific version is at the End-Of-Life stage on Digikey. The screen chosen was the 1’44 screen which works with the COG2 V231 driver but this only works for Arduino and Texas Launchpad so it had to be converted to be PIC Microchip compatible. Various libraries, C++ files and font tables were manipulated to make the screen work for the PIC24F32KA304. The solution for the PIC was only done for a screen of size 1’44 (smallest available) because of the large array needed to store pixel array to be displayed.
It is important to realise that not all the LEDs will have the exact same brightness, neither will all the sensors give the same ADC value for the same intensity of reflected light. This should be corrected for in code by firstly and most importantly averaging the ADC values, and then secondly there will be values added or subtracted for each sensor to balance all the outputs to a uniform level. A typical colour assay would look similar to the array in Figure 12, where the intensity of the colour decreases linearly as the concentration of the colourant is halved for each consecutive dot.
The following test colour assays in Figure 13 were made using food colouring and surgical needles and each consecutive dot has half the concentration of the previous dot. The top papers are used to calibrate the sensors for full concentration. The colours that were chosen to test the device on are red, green, blue and yellow.
The device currently has a size of 15 cm (Length) 8.6 cm (Width) 4.0 cm (Height) but could easily be designed to be a quarter of this size, it was chosen to be as big in order to be able to still debug all the separate components after the proof-of-concept has been assembled. The device, shown in Figure 15, has three pushbuttons for measurement, GPS coordinate updates and flushing the memory via a UART connection to a PC. The switch on the front enables the user to put the device in an off-charging mode (connector for charger included) or to switch the device on. The GPS is deactivated during normal operation and the E-ink screen only draws current when it is being updated. Figure 14 and Figure 15 each shows the final device as an integrated system.
It was noticed that the blue light source measurements rendered significantly larger ADC values than the green or red light even though the lights have been current regulated in a way that the luminous intensity value for each light is similar. It might be that the white of the paper, which becomes more prominent as the colour concentration decreases, reflects the blue light the best. This phenomenon might also be due to the sensitivity curve of the phototransistor that is not exactly as indicated in the datasheet and shown in Figure 16.
It was thus decided to use only blue light as a source seeing that for 8-bit tonal resolution it rendered sufficiently large differences between subsequent colour dots for each tested colour. The voltage drop over the blue LED is also higher than over the red or green LED and thus less power will be dissipated over the series resistor.
Figure 17 shows where the typical colours fit in the spectrum with their corresponding wavelengths. The LEDs that were used have wavelengths of 485 nm (Blue), 518 nm (Green) and 632 nm (Red).
The reflected light intensity depends on three main factors: The emitted source light wavelength, colour sample and phototransistor peak sensitivity. For this device it was found that the peak sensitivity was obtained when using a blue light source. One would rather expect that the green light source would be closer to the peak sensitivity of the phototransistor.
The data for Figure 18 was obtained by first measuring on a blank white piece of paper, and then on a darker paper in order to characterise the differences between the readout values from each sensor in order to compensate for it in code. It can be seen that the sensor differences at least stay constant for different colour assays.
The graphs as seen in Figure 19 were generated to show how the code compensates for sensor differences by doing non-uniformity correction. The fact that the device should work the same for all colours makes this a difficult task.
Phototransistors have an internal gain like any normal transistor and this gain is highly dependent on the batch and process used to manufacture the transistor. This means that there could be a great gain variation between the 9 phototransistors in the array of the colorimetric reader. It is therefore proposed that for subsequent versions of the reader device that photodiodes with transimpedance amplifiers be used instead of phototransistors as photodiodes have a more constant and predictable output, they react faster to changes in light and they are less temperature dependent.
The difference in measurements is also due to the light that is emitted that can differ in intensity due to small differences between the LEDs and consequently different currents flowing through the LEDs. It is therefore suggested that a LED driver be used for subsequent versions of the device.
The following results were obtained for the four different colours. The ADC values were recorded and then scaled to compensate for the characteristic differences in sensor readout values. From this point on the readout values can be binned and assigned to specific concentrations. It is important to remember that the device can work for ranges of lighter or darker concentrations as well and this means that the resolution may in fact be more than just 8 bits, this will be confirmed during ongoing research.
It can be seen that for red, green and yellow paper assays the curves are predictable and easy to linearize but for blue it is slightly harder. These curves can greatly vary due to the differences in paper used and the accuracy with which the colour concentrations were prepared. A dynamic binning algorithm was developed to determine each bin according to the scaled ADC values obtained for the calibration assay. This binning algorithm would only be successful if no values of bins overlap for measurement. The colorimetric device was used to take 40 measurements of a green food colouring assay (see Figure 13) in order to obtain standard deviation data (red bars) for the measurements, see Figure 21.
It can be seen that the standard deviation of the sensor values, for each different concentration of an analyte, is small enough to confidently and correctly bin/cluster measurement values. It should be noted that the standard deviation is larger for analytes of a lower concentration, meaning that the sample would be lighter and reflect more light. The binning algorithm takes the scaled ADC values and compares these values of the calibration curve to create larger bins for higher measured intensities. This is more accurate than simply using a linear binning algorithm.
A typical output of the colorimetric device is shown in Figure 22 for a measurement on a green food colouring assay. The values shown are scaled ADC values and the values indicated in brackets are the assigned bins for each analyte.
The power dissipation of the device was investigated. The current drawn by the device was measured for each of the operations during a normal measurement cycle, see Table 1.
Current drawn by colorimetric device for different operations at 3.6 V
|Measure (LEDs on)||9.24 mA|
|EPD update||9.73 mA|
|GPS coordinates update||41.3 mA|
The capacity of the battery is 3000 mAh. If the average current drawn is about 10 mA, the device can achieve a maximum of 300 operating hours on a single charge.
It can be concluded that the colorimetric array reader accurately discerns between paper-based colour dots with a tonal resolution of 8-bits. The device complies with the ASSURED standards as set out by the World Health Organisation. The device stores up to 255 sets of measurements and GPS geo-tagged data that can be sent via a UART to a PC. The device minimises change in measurements to less than half a significant bit for changes of 100-100 000 lux in ambient lighting for both indoor and outdoor environments. The device operates normally for temperatures between 15 and 40 degrees Celsius. The device uses minimal power as the GPS draws 100 mA only on start-up, 45 mA while running and 0.4 mA in disabled state (where it is most of the time when it is not updating its coordinates), the E-ink screen only draws current when updating its image and switching currents are in the order of 0.5 uA, the light source uses less than 1 mA per light and this is only on for the time it takes the ADC to fully load its internal capacitor, which is about 400 microseconds. The microcontroller operates in a Microchip Sleep mode and only gets triggered by interrupts from the pushbuttons. On average the device draws 10 mA of current when the GPS is disabled. This means that the device can operate for 300 hours on a single charge. The curve in figure 62 shows that the standard deviations of 40 values for each bin are small enough that no values could be incorrectly clustered to the wrong bin. This is true for measurements done on a food colouring assay that has significant colour variation within the analyte. It can thus be confirmed that the device will accurately determine concentration levels of analytes with a resolution of 8 bits. The noise present in the sensor measurement values, combined with the small differences between scaled sensor measurement values, are small enough to fit in the indicated standard deviation regions.
Colorimetric assays were prepared with both food colouring and paint chart samples. An industry standard spectrometer was used to illuminate red, blue, green and yellow samples and then determine the amount of light reflected by each analyte sample. The curves generated by the spectrometer tests were analysed to develop a binning/clustering algorithm based on the Jenks natural break classification to divide the data into 9 bins. A functional colorimetric assay reader device was designed and implemented by using discrete OTS surface-mount components on PCB. A 3D-printed casing with a specialised light confining/blocking array (for the LED-sensor pairs at the bottom) was designed and manufactured. A complete rewrite of the driver for the 1’44 E-ink display for PIC microcontrollers was done and implemented. A GPS, external EEPROM memory and E-ink display were integrated with the colorimetric deviceMultiple verification tests in different lighting and temperature conditions were done to ensure that the device truly has an 8-bit tonal resolution for all practical applications.
There are still numerous upgrades possible to improve on the performance of the device. Firstly, the device needs to be tested on a wider range of concentrations of samples that have been tested by an industry standard spectrometer. The device can be modified to be a lot smaller and its stability and temperature dependency could be improved by replacing the phototransistors with photodiodes and transimpedance amplifiers. The device currently only detects colour intensity while it already has all the components necessary to do specific colour detection into a colour space like RGB or HSI for biosensing applications where the sample undergoes a change in colour.
Yager, P., Domingo, G. J.., Gerdes, J., “Point-of-Care Diagnostics for Global Health,” Annu. Rev. Biomed. Eng. 10(1), 107–144 (2008).Google Scholar
Martinez, A. W., Phillips, S. T., Carrilho, E., Thomas, S. W., Sindi, H.., Whitesides, G. M., “Simple telemedicine for developing regions: Camera phones and paper-based microfluidic devices for real-time, off-site diagnosis,” Anal. Chem. 80(10), 3699–3707 (2008).Google Scholar
Tseng, D., Mudanyali, O., Oztoprak, C., Isikman, S. O., Sencan, I., Yaglidere, O.., Ozcan, A., “Lensfree microscopy on a cellphone.,” Lab Chip 10(14), 1787–1792 (2010).Google Scholar
Wu, G.., Zaman, M. H., “Low-cost tools for diagnosing and monitoring HIV infection in low-resource settings.,” Bull. World Health Organ. 90(October), 914–920 (2012).Google Scholar
Tohda, K.., Gratzl, M., “Micro-miniature autonomous optical sensor array for monitoring ions and metabolites 2: color responses to pH, K+ and glucose.,” Anal. Sci. 22(7), 937–941 (2006).Google Scholar
Suzuki, K., Hirayama, E., Sugiyama, T., Yasuda, K., Okabe, H., Citterio, D., Analytic Chemistry 74, 5766–5773, (2002).Google Scholar
Soldat, D. J., Barak, P.., Lepore, B. J., “Microscale colorimetric analysis using a desktop scanner and automated digital image analysis,” J. Chem. Educ. 86(5), 617–620 (2009).Google Scholar
Lee, D.-S., Jeon, B. G., Ihm, C., Park, J.-K.., Jung, M. Y., “A simple and smart telemedicine device for developing regions: a pocket-sized colorimetric reader.,” Lab Chip 11(1), 120–126 (2011).Google Scholar
Drescher, P., “EaEpaper”, mbed, 25 June 2014, <https://developer.mbed.org/users/dreschpe/code/EaEpaper> (10 July 2016).Google Scholar
Orr, G., “Understanding colour”, Willamette University, 13 June 2006, <https://www.willamette.edu/~gorr/classes/GeneralGraphics/Color/physics.htm> (10 August 2016).Google Scholar