2.2. Hardware System Description

The developed prototype system consists of sensor module pairs where each has one integrated 3D accelerometer, 3D gyroscope, 3D magnetometer IMU with digital readout (LSM9DS1, STMicroelectronics, Geneva, Switzerland), and either an ultrasonic loudspeaker (custom-made, SINTEF, Oslo, Norway) or an ultrasonic microphone (Knowles, Itasca, IL, USA). Figure 1 shows the module concept and building blocks. The module pairs can provide inertial sensor data (3D acceleration and angular velocity) of each module, as well as the relative ToF distance between the modules. The modules are integrated on a development board (Nordic Semiconductor, Trondheim, Norway) with a microcontroller and a radio frequency wireless communication unit, a custom-made printed circuit board for ultrasound and IMU electronics as well as a battery. The modules are controlled from a central node, which also receives the measured data from the ToF sensor and the IMUs.

System concept (a) with one pair of modules attached to the lower leg seen from the side (b) and from the front (c). The modules consist of an Inertial Measurement Unit (IMU) and either an ultrasonic loudspeaker (in this case, left leg) or a microphone (in this case, right leg). The ultrasonic signal is marked with a dashed arrow, and the coordinate system of the IMUs with solid arrows.

The ultrasonic loudspeakers in the system are piezoelectric micromachined ultrasonic transducer (PMUT)-based ultrasonic transducers developed at SINTEF. These ultrasonic loudspeakers are based on SINTEF’s thin-film PZT (lead zirconate titanate, 2 µm film thickness) bulk micromachining microelectromechanical system (MEMS) technology. The thin-film PZTs have been deposited by high quality chemical solution deposition. The piezoelectric coefficient e_31,f of these films is −18 C/m². The transducers are based on a membrane structure made up of 8 µm silicon-on-insulator, 1 µm stress compensation oxide, a platinum bottom electrode, 2 µm PZT, and a gold top electrode. The MEMS loudspeaker is 1.7 × 1.7 × 0.4 mm in size. The loudspeaker devices were characterized with respect to frequency response of the sound pressure level for different actuation voltages:

95–100 dB @ 10 cm;

sound pressure at resonance saturated already at 5 V AC;

bandwidth continues to widen using higher voltages (−6 dB BW ~14% using 10 V AC).

The ultrasonic microphone in the system is a commercial Knowles analogue silicon MEMS microphone, SPH0642HT5H-1, with an acoustic sensor, a low noise input buffer and an output amplifier. The microphone is stated to have a flat frequency response from 100 Hz to 10 kHz. We tested that the microphone operates out of range in the ultrasound frequency range. The SINTEF PMUT loudspeaker was used to send a frequency sweep from 70 to 110 kHz, and around the resonance frequency of the loudspeaker (90–100 kHz) the receiver microphone signal was found to have sufficient quality to be used in the present experiments. Figure 2 shows the frequency response of the microphone resulting from the frequency sweep transmitted by the SINTEF PMUT loudspeaker.

Frequency response of the received raw signal of the Knowles microphone transmitted by the SINTEF piezoelectric micromachined ultrasonic transducer (PMUT) loudspeaker at 10 cm distance.

The purpose of the electronic circuit board is to generate suitable excitation signals for the loudspeaker and to amplify the signals received from the microphone, where the noise level should be as low as possible. The transmitter and receiver boards are identical, except that the transmitter board was equipped with a loudspeaker while the receiver board was fitted with a microphone. The boards also contain an inertial measurement module. The electronic circuit board has the following main components (Figure 3): a 3V Low Dropout (LDO) regulator, a 12V charge pump DC/DC converter, an LSM9DS1 inertial measurement unit (IMU), a SINTEF PMUT loudspeaker (TX), a Knowles SPH0642HT5H-1 microphone (RX), an astable multivibrator pulse (AST) generator and an amplifier circuit, as well as several resistances and capacitors. The power supply is from a Nordic Semiconductor developer board (see Section 2.2.3) where a 5V DC supply is available. A custom-made dedicated circuit was used to produce a 12V loudspeaker excitation signal. Another custom circuit was used to amplify the microphone signal in front of the microcontroller ADC.

Transmitter and receiver electronics boards functional building blocks supported by text description: NORDIC: Nordic Semiconductor development board. DC/DC: charge pump. LDO: low dropout regulator. IMU: inertial measurement unit. AST: astable multivibrator pulse generator. TX: ultrasonic loudspeaker (transmitter board only). RX: ultrasonic receiver (microphone board only). AMP: amplifier circuit. MCU: microcontroller.

An nRF52 development kit from Nordic Semiconductor was used as a basis module as it contains a radio with an embedded microcontroller. The software may be configured to accommodate different radio protocols such as Bluetooth Low Energy, Thread, ANT, and Enhanced Shock Burst (ESB). The kit also allows easy access to some basic peripherals such as GPIOs and a 14-bit analog-digital converter (ADC). Powering of the board and the other electronic components was carried out using a power bank (MO8735, Mobility on board (MOB), Paris, France). Figure 4 shows a block-schematic representation of the hardware setup.

Block schematic representation of the hardware setup of the sensor module pair.

The distance between the sensors was estimated using a ToF measurement of the ultrasonic pulse. Accurate range estimates require a common timing reference for both the loudspeaker and microphone modules. A distance accuracy of 1 cm dictates a timing accuracy of approximately 30 µs. This is achieved by the loudspeaker modules sending a short radio packet prior to emitting the ultrasonic pulse. Timing uncertainties can arise from delays in the protocol stack, both for the loudspeaker and microphone. To minimize the uncertainty, a simple real-time protocol (ESB) was chosen where both listen-before-talk (LBT) and retransmissions of lost packets may be switched off, as both these mechanisms introduce uncontrolled delays. Similar synchronization solutions using low latency wireless systems have also been reported in [19,20]. The remaining stack delays were all constant values and could easily be adjusted in the software.

The loudspeaker sent bursts of 20 cycles of 100 kHz ultrasonic pulses. On the receiving side, the electronics filter and amplify the signal and present it to the ADC with a 1.65 V DC bias. This signal was sampled at 150 kHz and digitally processed to provide an estimate of the elapsed time. The maximum sampling frequency provided by the nRF52 was 200 kHz, which is at the Nyquist rate. It was therefore decided to subsample to avoid sampling too close to the signal DC crossing.

The presence of a signal will cause an oscillation around the abovementioned 1.65 V DC bias, where the sign of the deviation is irrelevant. Hence, only the absolute value of the signal needs to be considered. The algorithm employed for the energy detection can thus be expressed by the following relation:

where x(n) is the signal presented to the ADC, 1.65 is the DC bias, and F{} is a moving average filter. The ToF was estimated to where the signal e(n) crossed a certain threshold that was experimentally determined.