Electronic Hardware Design
The electronic hardware was broken down into two parts: a power board and a sensor board. There are multiple reasons for this design choice stemming from the objectives set out in the introduction: small size and cost effectiveness. By distributing the electronics across the strap, an increase in width has been successfully avoided making the bracelet more comfortable for the user. An alternative solution to this would have been using High Density Interconnect technology and multi-layer boards, but this option would have increased costs by 40%[1]. Furthermore, the separation between power and sensor modules was chosen to minimise the number of wires going from one board to the other.
The power board was broken down into 3 parts: charging unit, energy harvesting unit and battery gauge. The charging unit takes care of charging the battery when the device is plugged into a USB power source. The energy harvesting unit provides a front-end to the Peltier module used for self-charging as well as takes care of power regulation and charging the battery from the energy harvester. The battery gauge is used to keep track of the charge in the battery.
The energy harvesting unit (EHU) consists of two parts: a Peltier module with a heatsink and a power management unit(PMU). The Peltier module and the heatsink were chosen for their size (15x15mm) which makes them practical for wearable use. The energy harvesting unit was based on the ADP5092[2] IC from Analog Devices. This IC was chosen because of its superior startup characteristics and the fact that it is optimised for extremely low power use. The main criteria was to be able to self-start the energy harvesting circuitry from input powers as low as 20µW as well as to maintain high conversion efficiency. The ADP5092 can go through cold start from input power levels as low as 6µW and is able to convert power very efficiently by using its built-in maximum power point tracking circuitry. It is also capable of circumventing the cold start voltage requirements with the aid of external energy storage - the battery in this case. Most external components were chosen based on the Applications Information section of the datasheet while the programming resistors were chosen for the specific battery type and output voltage.
The main challenge was the steering of the power path to make sure that the two charging and power supply circuits (EHU and CU) do not conflict. To avoid conflict, the charging paths and the power paths needed to be separated. For this, the ADG884BRMZ[3] analogue switch was chosen for having the lowest ON state resistance out of devices available from suppliers. The advantage of using an analogue switch is the high speed at which they can switch the two power paths while still having a short disconnect time, making sure that power only flows one way. The schematic of this circuit is shown on the figure below:
The charging unit consists of two parts: a charge controller and a linear regulator. The charge controller design was adapted from the datasheet of the LTC4079[4] IC, while the linear regulator(LD39015M33R [5]) only required decoupling capacitors. Both of these devices were chosen because of their low price, high availability and small footprint. The charging unit turns on when a USB power source is connected to the device, taking over the driving of the power path and battery charge path from the EHU. Regarding the thermal considerations related to the loss on the linear regulator there is no need for any heatsink or a large airflow path as the maximum power that is dissipated is around 5mW while the device is plugged in.
The circuit of the battery gauge was adapted from the datasheet of the battery gauge IC, the MAX17048G[6] from Maxim Integrated. This component was chosen as it only requires one external component(apart from pull-up resistors) reducing the total occupied area as well as it consumes very little power while measuring the battery status with very high precision. During layout decisions, the considerations outlined in the datasheet were taken into account.
The sensor board was divided into 4 modules: SoC, green light PPG sensor, IR PPG sensor, and IMU (including temperature).
In order to reduce the number of components on the board, and simplify the design process an integrated BLE SoC module was chosen. The Fanstel BT832 with the nRF52832 chip was selected for its low power consumption (7.5 mA TX current consumption), fast processor (64 MHz), large RAM (64 KB) and flash (512 KB) size and NFC capability. While there are options available that outperform in certain categories, the nRF52832 strikes a good balance in the aspects that are important for this project. Additionally the large amount of documentation and support available from the Nordic forums is an invaluable advantage.
Comparison of low power BLE modules (from Predictable Designs)
Photoplethysmography(PPG) is a measurement technique where the skin is illuminated from close proximity and the reflected light is measured and analysed. Some of the health signals that can be extracted using this method are the Pulse waveReisner et al., blood oxygen levels(SpO and SpO2)Allen J, blood pressureKhalid et al., and respirationAllen J. We have decided to use two different wavelengths one in the green regime(550nm) and one that is IR(920nm). The reason for this was the fact that many of the aforementioned health signals can only be extracted using either or both of these wavelengths.
For the green light PPG the BH1792GLC sensor from ROHM was chosen. The reason for this is threefold: both the LED driver and the photodiode components of the sensor are integrated into the IC as well as an optical filter. The first two points are important as these help saving board space as well as reduce noise from interference. The optical filter which is best in its class in terms of selectivity is essential as light around this wavelength is easily scattered under the skin and light from external sources can interfere with the measurementsAllen J. Regarding the choice of LEDs we have chosen a device with a forward voltage that is compatible with 3.3 V, the logic level used across the board while providing the highest illuminance at low currents out of the available LEDs. Regarding layout the LEDs were placed as close as possible to the photodiode so that most of the reflected light can be gathered by the photodiode hence the current of the LEDs can be kept low. The IC also contains an IR photodiode(without optical filter) for skin detection for which the LED was placed so that it aligns well with the skin. The final layout is shown on the figure below:
For the IR light PPG sensor the MAX86141[9] analogue frontend was chosen. This specific module from Maxim Integrated encorporates specific characteristics that mean the module performs significantly better than its counterparts, especially as a wrist device. Above all, is its high resolution 19-bit ADC which allows for a powerful signal-to-noise ratio. It also exhibits ultra low power consumption - typically 10uA when sampling at 25Hz. There is also a built-in FIFO for local storage that can be pre-processed for SpO2 calculations in the micro-controller. For the LEDs and photodiodes the recommended components were chosen where available, the unavailable components were replaced by devices with similar characteristics. The recommended components were obtained from a reference design(available here). Regarding layout the guidelines in the datasheet were followed making sure that there is a separate plane for the ground of the sensor to minimize interference from other parts of the board as well as keeping the analogue frontend connections as short as possible while also keeping the sensor away from the rest of the components. The layout is shown on the figure below:
The Inertial Measurement Unit (IMU) that was decided to be deployed was the LSM6DSL[7] by STMicroelectronics. There were several reasons this specific IMU was chosen to particular advantage. Foremost, its low-power consumption of 0.4mA is noteworthy - less than half of its closest competitor, the LSM6DS33[8]- also by STMicroelectronics. This meant the power budget at hand was not compromised beyond necessary. Its built-in pedometer algorithm meant a well-established methodology could be used without hassle - the threshold for a genuine step was established through testing. In parallel, the IMU accelerator x-, y-, z- values capability could be fed into the other sensor values' signal processing algorithms to reduce motion artefacts and tremors. Despite a lower FIFO storage space (4KB) compared to the LSM6DS33 (8KB), this was not a drawback for us thanks to its Continuous Mode [5.4.3, [7]], which meant delayed pre-processed values were consistently being fed into the micro-controller. The IMU also contained a built-in temperature sensor for an additional vital sign measurement. However, on further iterations it would be preferred to use a specifically-designed temperature sensor for skin (consider MAX30205 by Maxim Integrated) - the IMU suffered inaccuracies due to the surroundings (e.g. clothing) and the chip temperature itself. The communications with the IMU were achieved using a standard I2C protocol.
Due to the nature of the product there were strict size limitations for the PCBs, which made it necessary to use as small components as feasible. This of course increases the difficulty of assembly, hence the option for ordering assembled PCBs has been investigated. While the cost was low, unfortunately the lead time for PCBWay to source the components was too long and the service could not be utilised in this project. This meant that the boards had to be soldered in the lab.
The stencil for applying the solder paste was laser cut from a 0.075 mm mylar sheet. This took a couple iterations to perfect and spending time on this was crucial since having the right amount of paste in the correct places saves a lot of trouble later on. A main issue was having too much solder paste, which caused shorts on the board. Due to the small component sizes and packages with pins on the bottom, this was difficult and time consuming to fix. Reducing the solder mask size for laser cutting has practical limitations due to the thickness of the cut as the laser burns away the material.
The key points when laser cutting the stencils:
- reduce the solder mask layer size to 80% compared to the pad size
- further reduce the large ground planes under the chips
- shrink small pitch pin pads from rectangles to lines
- use raster mode for larger, well spaced pads (resistors, capacitors,etc.) - burns the material away and does not leave small bumps along the edges
- flip the image before cutting - ensures that the small bumps along the cuts are facing upwards when applying the paste
- laser cutter settings for VLS3.50:
- vector cut: 50% power, 100% speed
- raster cut: 70% power, 70% speed
To ease the assembly process a holding bracket was laser-cut to hold the boards in place while applying the paste. The bottom piece contains a cut-out for the components on the bottom side of the sensor board.
The board is then pasted and populated with the components and finally placed in the reflow oven. Preheat to 160°C for 3 minutes, increase to 247°C for 2 minutes and finally a slow 3 minute cool-down to 120°C. For the two sided sensor board, the side facing the skin is soldered first, with the exception of the large LEDs and photo diodes. These can not take the temperature and due to their size might fall of when the the board is flipped and the other side is being soldered in the oven. These components need to be hand soldered at the end.
Download here the list of components required and the scalability available when buying in large quantities. Under economies of scale conditions, there is a reduction of total cost by 49.4%.