In this article, we will explore the specifics behind the University of California, Berkeley's wearable sweat sensor they recently published about in Nature Letters.
View of the wearable sweat sensor worn on the wrist. Image courtesy of UC Berkeley.
One of the more interesting classes I have taken thus far in grad school has to be Biosensors.
In this class, we detail the development of biosensors over the last few decades in research and consumer technologies. A biosensor can be described as a device that measures a specific biological quantity and transduces the biological quantity into a form that can be read and interpreted by a person.
One common type of biosensor is the glucometer, which allows diabetics to monitor their blood sugar levels and their insulin regimen. Of course, being the EE at heart that I am, I tend to pay pretty close attention to the electronics design that accompanies some of these sensors.
Gao et al. detail their design for “fully integrated wearable sensors arrays for multiplexed in situ perspiration analysis.” In their paper, they analyze human sweat for a few important biomarkers, namely glucose, lactate, sodium (Na+), and potassium (K+). These biomarkers provide important information regarding hydration status and overall fitness. Surprisingly, very simple circuits can perform these measurements.
Let's take a look at the circuit configurations.
Schematic diagram of the glucose and lactate sensor. Image content recreated from Nature Letters.
Glucose and Lactate Sensing with a Transimpedance Amplifier (TIA)
At the heart of their glucose and lactate sensor is a transimpedance amplifier (TIA), or “current-to-voltage converter.” A transimpedance amplifier, as the name suggests, converts an input current to a voltage. Let’s analyze this further.
General schematic of a transimpedance amplifier (TIA) or "current-to-voltage converter".
Satisfying Kirchhoff's Current Law (KCL)
Remember your basic Kirchhoff’s Current Law (KCL) relationship. If a current, let’s say I1, flows into a node, another current, equal to I1, flows out of the node. Looking at our schematic of a basic transimpedance amplifier, if a current, I1, flows into Node A, we must also have an equal current, I2, flowing out of Node A to satisfy KCL.
Now we get into Ohm’s Law which states that V = I×R. Our op amp produces a voltage at its output to drive a current across Rf in order to satisfy KCL (or sink a current depending on the direction of current flow and your preferred convention of positive current flow). In this way, we obtain current-to-voltage conversion with a pretty simple relationship.
Next, they add an inverter to correct the phase inversion from the transimpedance amplifier. Notice the resistor from the non-inverting pin to ground. This is a slight modification to the typical inverting amplifier configurations that many are used to. The purpose of this resistor is to help correct for input bias currents that can add noise at the output of the circuit distorting the glucose or lactate measurement.
From our basic op amp rules, we know that no current flows into the inputs of the op amp and that the voltages at the op amp inputs are the same. Unfortunately, it's not quite that simple.
Bias currents can cause the non-inverting input and inverting input to have different voltages. For a low-gain amplifier, this is probably not that big a deal. It’s also not problematic if the ratio of the input voltage (Vin) to the input resistance (Rin) is relatively large. Decent amplifiers have input bias currents in the nanoamperes (nA), so we figure if Vin/Rin is on the order of milliamps, we are a few orders of magnitude higher than our noise.
Gao et al. use an input resistance of 1MΩ. And we can figure that the input voltage (from the output of the transimpedance amplifier) is probably in the millivolt range. This means that Vin/Rin will probably be around a few nanoamps, which is right in our noise margin. Gao et al. place a resistor at the non-inverting pin to help correct for the bias currents. As you examine the full schematic a bit further, you will see that the researchers use this trick in pretty much all their op amp stages. They are being quite careful in managing their input bias currents.
Second-Order Low-Pass Filters
Next, we have two 2nd-order low-pass filters to get rid of unwanted noise using the popular Sallen-Key topology. Connecting these two filters in series leads to a four-pole response that provides steep roll-off for frequencies above the cutoff. These biosensors are operating at about 1Hz, so bandwidth is certainly not a limiting factor in the amplifier or filter design.
Sodium and Potassium Sensing with a Differential Amplifier
Schematic diagram of the Na+ and K+ sensor. Image content recreated from Nature Letters.
For the Na+, K+ measurement systems, Gao et al. utilize a simple differential amplifier stage with two buffered inputs. An interesting aspect of the design is the use of the buffer amplifiers to interface the circuit to the bioelectrical system.
Metals are often used as electrodes for performing bioelectric measurements. Metals are great conductors of electricity. In bioelectrical systems, however, there are a number of complications that affect the impedance of a metal. I won’t go into too much detail but, when a metal is placed into a biological system, an electrode-fluid interface develops. This electrode-fluid interface has a highly variable impedance. In order to nullify the effect of the variable electrode-fluid impedance, a buffer amplifier is used as the first stage.
A buffer amplifier has a very large input impedance, much larger than the variable impedance of the electrode-fluid interface. As a result, we obtain a faithful measurement of the voltage of our bioelectrical system without loading the system, itself. The resulting voltage generated from the system is subtracted from a reference voltage—a standard technique in bioelectrical measurements—using a differential amplifier. The circuit is further conditioned with two 2nd-order low-pass filters before being processed by a microcontroller.
The entire circuit design is on a flexible printed circuit board, allowing the device to conform to the skin.
Finally, and this is a personal favorite of mine, the analog signal is digitized by the ADC on an ATmega328 microcontroller, a device that is used extensively in Arduino and Arduino-compatible boards.
The board's flexible sensor array. Screenshot courtesy of the UC Berkeley.
In Summary: A Research Paper with Implications for Health and Wellness
Overall, I was really impressed by the detail the researchers went into in developing their sensing circuit. I would also like to propose a few suggestions for thought.
For starters, they built four separate amplifier stages for each of their four biosensors (glucose, lactate, Na+, K+). Admittedly, glucose and lactate required a different sensing configuration compared to Na+ and K+. Nevertheless, the use of a simple two-channel multiplexer like the TS5A9411 from Texas Instruments could very quickly simplify the circuit by essentially cutting their design in half. I have personally found the TS5A9411 to be a pretty good option for interfacing between electrochemical cells due to its low on-resistance (10Ω max at VCC = 5V) and on-capacitance (8.5pF). Furthermore, with a low operating voltage (down to 2.5V) and small footprint (SC70-6, 2.2mm × 2mm), the TS5A9411 is a decent option for sensitive switch applications for bioelectrical systems.
Several ICs have been developed for bioelectrical sensing. The LMP91000, also from Texas Instruments, is an integrated analog front-end for electrochemical sensing (I’ve been using "bioelectrical" instead of "electrochemical"—but, in this context, they can be considered interchangeable). It is capable of performing both the current-to-voltage conversion for glucose and lactate and the differential measurements for Na+ and K+, which makes the IC pretty versatile. Use of the LMP91000 essentially cuts their design down to a single chip, which is outstanding. Maybe they found that it was necessary to explicitly design each amp stage in order to maintain sensitivity and configurability for each biomarker.
Furthermore, research papers are often presenting proof-of-concepts. From the title, we understand that the focus was more on detecting these biomarkers in sweat, in real-time, with a wearable device, than on the circuit design, itself. We can surmise that further optimization of the circuit design will soon follow.
All that being said, Gao et al. built an impressive device for simultaneous measurement of four important biomarkers of hydration and fitness. Their project also demonstrates that there is a need for experienced electrical and electronics engineering in a variety of fields including chemistry, biological science, and biomedical engineering.
The sensor and its accompanying app. Image courtesy of UC Berkeley.
Biosensors have extraordinary impact in the consumer and research space ranging from applications in glucose monitoring to cancer detection. In recent years, the quantified-self movement has increased the demand for wearable biosensors, prompting researchers to investigate new ways to bring laboratory technologies to the hands and homes of consumers.
Featured image used courtesy of the University of California, Berkeley.