Description of Wireless System Design

0892-EX-CN-2017 Text Documents

MIT Lincoln Laboratory

2017-12-11ELS_202167

Lightweight, On-Body, Wireless System for Ambulatory Voice and Ambient Noise Monitoring Patrick C. Chwalek, Daryush D. Mehta, Member, IEEE, Brendon Welsh, Catherine Wooten, Kate Byrd, Edward Froehlich, David Maurer, Joseph Lacirignola, Thomas F. Quatieri, Fellow, IEEE, Laura J. Brattain version that was used to evaluate the Lombard effect in a  Abstract— In this paper, we present a lightweight, on-body, wireless system designed for monitoring real-world voice laboratory setting with increasing background noise level [3]. characteristics and behavior. The system has the potential to provide important assessments of voice and speech disorders, II. EVOLUTION OF AMBULATORY VOICE MONITORING psychological and emotional state, and the impact of In the laboratory, acoustic and non-acoustic vocal sensors environmental sound levels. The system’s transmitter is have been applied previously to robustly characterize speech positioned on the neck and synchronously streams dual-channel in the presence of various types of background noise by fusing sensor data from an on-board MEMS microphone and a high- features from multiple sensors, including a bone conduction bandwidth accelerometer, which acts as a noise-robust and confidential contact microphone. These data are recorded to a MIC, radar sensor, and contact MICs [4]. In the field, receiver that can store the data locally and stream a real-time ambulatory voice monitoring devices have typically feed to a computer. We also report on the design considerations incorporated either acoustic MICs, contact MICs, or high- of this novel system and discuss progress leading up to the latest bandwidth ACC sensors to capture vocal characteristics and iteration, especially of the transmitter components on a flexible behavior [5]. The primary application of such devices was the circuit. clinical assessment of voice use in the daily lives of individuals working in occupations associated with higher incidences of I. INTRODUCTION voice-related complaints. These technologies have allowed for In this paper, we discuss a wireless voice monitor system the computation of parameters that have been associated with that we developed for recording synchronized acoustic heavy voice use (increased talk time, inappropriate pitch and (MEMS microphone; MIC) and non-acoustic (accelerometer; loudness, etc.) [6]. ACC) data. Our system offers a tether-less method of Recent work has developed a smartphone-based platform capturing voice-related features that are important for speaker for ambulatory voice monitoring that records the raw signal identification, noise reduction, and, most notably, for from a wired high-bandwidth ACC, allowing for the exploiting non-acoustic vocal signatures in real-world exploration of novel parameters related to the acoustics and environments to provide long-duration monitoring and real- aerodynamics of vocal function [7, 8]. In addition, real-time time biofeedback. Since naturalistic environments make it computation of voice features on the smartphone enables the challenging to estimate many important voice characteristics study of more sophisticated biofeedback scheduling known to in noisy conditions, recordings of neck-surface vibration have be critical for behavior modification [9]. been the subject of ongoing investigation due to their robustness to acoustic environmental noise, low profile, and lack of speech intelligibility (alleviating confidentiality concerns) [1, 2]. However, MIC recordings continue to be desirable to capture the airborne acoustic signal that can be analyzed to quantify speech features (e.g., formants) and environmental characteristics. For clinical testing, it is also desirable to have a small form-factor device so that it minimizes discomfort and leads to increased compliance using the device. Furthermore, we discuss designing a system of this size with a wireless communication requirement on a flexible substrate. There are several improvements discussed that were made to this current system from the previously reported P. C. Chwalek, D. D. Mehta, B. Welsh, C. Wooten, K. M. Byrd, E. DISTRIBUTION STATEMENT A. Approved for public release: Froehlich, D. Maurer, J. Lacirignola, T. F. Quatieri, and L. J. Brattain are distribution unlimited. This material is based upon work supported by the with MIT Lincoln Laboratory, Lexington, MA 02421 USA (e-mail: Assistant Secretary of Defense for Research and Engineering under Air Force [patrick.chwalek, daryush.mehta, brendon.welsh, catherine.wooten, Contract No. FA8721-05-C-0002 and/or FA8702-15-D-0001. Any opinions, kate.byrd, edward.froehlich, david.maurer, lacirignola, quatieri, findings, conclusions or recommendations expressed in this material are brattainl]@ll.mit.edu). Corresponding author email: brattainl@ll.mit.edu. those of the author(s) and do not necessarily reflect the views of the Assistant D. D. Mehta is also with Massachusetts General Hospital, Harvard Secretary of Defense for Research and Engineering. Medical School, and the MGH Institute of Health Professions, Boston, MA USA (e-mail: mehta.daryush@mgh.harvard.edu).

Figure 1. System framework of wireless voice monitor. All previous devices, however, required a tethered connection which can be cumbersome during data collection. Users may snag cabling while walking or find wiring uncomfortable during long-term use, leading to user non- compliance and the potential for noisy recordings. Thus, our system’s wireless design was a high priority, as well as the ability to program user-specific algorithms that can perform computations on real-time data streams captured by the receiver (see Fig. 1). Like the VoxLog device, we maintain the desirable dual-channel recording of both acoustic MIC and high-bandwidth ACC sensors for the noise-robust monitoring of voice use in naturalistic environments. An early prototype of our wireless system was initially reported as part of a comprehensive multimodal system for animal behavior monitoring [10]; the module continues to undergo significant Figure 2. Wireless voice monitor, showing (A) transmitter circuitry on the anterior neck surface, (B) components on the front layer of the improvements and optimization for human voice analysis [3], flexible circuit, and (C) components on the back layer. with the latest version being the subject of the current paper. single-axis, high-bandwidth ACC (BU-27135, Knowles III. SYSTEM COMPONENTS Electronics). Our improved circuit includes multiple The system consists of two components: (1) a wearable, amplification stages before the analog sensor signals are on-body transmitter that is worn around the neck below the digitized by our transceiver to increase the signal-to-noise ratio thyroid prominence and (2) a wireless receiver that can either (SNR). be carried on body, in a pocket, or plugged into a computer to The BC128 Bluetooth module (BlueCreation, Cambridge, visualize raw signal streams. A summary of the system UK) is the chosen transceiver due to its low power specifications is shown in Table 1. consumption, software configurability, 16-bit analog-to- A. Wireless Transmitter digital converters, and ability to use an external antenna to As in the previous system iteration [3], the wireless minimize radio frequency (RF) noise inherent in a transmitter (Fig. 2) consists of several components soldered on miniaturized printed circuit board (PCB). RF noise issues were a flexible circuit substrate. For sound level analysis, we use an prevalent in the previous system iteration [3] that included a omnidirectional MEMS MIC (SPA2410LR5H-B, Knowles BC127 module, which consisted of a built-in antenna that we Electronics, Itasca, IL) because of its small form factor and found subjected the small circuit to a substantial amount of RF wide bandwidth that includes frequencies that range from noise. The BC128 also offers several software features that can 100 Hz to 15 kHz. For non-acoustic voice sensing, we use a be toggled depending on the application (e.g., sending channel data to a Bluetooth-enabled phone, streaming to multiple TABLE I. WIRELESS VOICE MONITOR SPECIFICATIONS FOR THE receivers, etc.). DUAL-CHANNEL STREAMING OF MICROPHONE (MIC) AND ACCELEROMETER (ACC) SENSORS The ACC, MIC, and active circuit components are Feature Specification powered by a single-cell, rechargeable, lithium-ion polymer Sample rate 44.1 kHz (per channel) battery that can be charged through a micro-USB port on the Resolution 16 bits circuit. The micro-USB input also allows for communication Bandwidth ACC: 0–5 kHz, MIC: 100 Hz–15 kHz to the BC128 to modify firmware settings (e.g., channel gain) MIC Sensitivity −38dB ± 3dB @ 94dB SPL and troubleshoot. Additional features include electrostatic ACC Sensitivity −45 dB @ 1kHz discharge protection, an on/off switch for the battery, status Power Consumption 56 mW (transmitter) LEDs, and a power multiplexer integrated circuit that enables (active streaming) 330 mW (receiver) the BC128 to be fully functional when powered via USB Weight (circuit) 5 g (transmitter), 14 g w/ 0.4 Ah battery and/or battery. The battery size is dependent on the application 13.5 g (receiver), 47.5 g w/ 1 Ah battery and the usual tradeoff of size and capacity. We chose a Size (circuit) Transmitter: 68 × 14.5 × 5 mm 400 mAh battery for the transmitter because of its small form Receiver: 68 × 14.5 × 5 mm factor and ability to run the system uninterrupted for at least Wireless Protocol Bluetooth 4.0 24 hours.

Figure 4. Sensor signals being transferred throughout the system. Laboratory data include raw ACC and MIC signals, whereas ambulatory recordings save data channels to preserve confidentiality (raw ACC and averaged MIC rms signals). with the Bluetooth link; accurate timestamps would only be lost within that five-minute period. This scheme also solves issues of large data file preservation; if the system were to lose power inadvertently—e.g., due to a depleted battery—only up to the last five minutes of data would be lost due to incomplete Figure 3. Wireless voice monitor’s receiver. Components on the closure of the last data file. (A) front layer and (B) back layer of the circuit board. C. Design Considerations B. Wireless Receiver Flexible circuit design has its challenges when compared For data logging, we decided not to store the data on the to a rigid PCB, especially when building a relatively small neck-worn circuit since that would require additional space form factor with sensitive analog circuitry that will be and power, making the neck-worn circuit bulkier, heavier, and subjected to repetitive bending. Initially our ACC picked up less desirable for a user to wear throughout the day. We opted noise artifacts from the MIC signal and from a preexisting chip to design a separate receiver device, shown in Fig. 3, that can antenna [3] so we opted for the BC128 with an external log the data sent over Bluetooth from the transmitter, as well antenna that is mounted on the back of a user’s neck. In as output the data streams in real-time to a personal computer addition, when designing the PCB, we employed various noise (PC) via USB. We chose to build the device around a Teensy mitigation techniques to give us the best SNR. In addition to 3.2 microcontroller unit because of its small footprint, a 32-bit separating the ground planes between sensitive components, ARM processor with general-purpose input/output pins for we added multiple operational amplifier stages placed next to future expansion, and active development community. For the the sensors to increase the SNR of the received signal at the Bluetooth receiver, we chose to use the BC127 module that BC128. The amplifier stages also allowed us to scale the has the same internal components as the BC128 but included dynamic ranges of both sensors separately, a feature lacking in the chip antenna in the same package. For the receiver, we did the software-adjustable preamplifier in the BC128 module. not have any RF-sensitive components so we found the BC127 ideal for a compact receiver design. In addition, the receiver Flexible circuits are not designed to be often subjected to has similar power regulation circuitry to the transmitter circuit, compressive and tensile stresses (i.e., multiple users taking the with additional circuitry that allows users to charge the circuit off and on). To promote bending at certain regions receiver’s larger battery at a faster rate. Red/green/blue LEDs while mitigating it at others where integrated circuits and on the receiver serve as system status indicators (recording passive components are mounted, we chose to vary our ground status, Bluetooth pairing successful, etc.). plane density. As can be seen from Fig. 2, we wanted to prevent bending where the sensors were co-located so we As mentioned above, the receiver module can be connected added a hatched copper pattern, versus a solid one underneath to a PC where the data can be streamed in real time from the the sensors, on one side and omitted the copper plane on the transmitter module. The computer recognizes the USB input as a standard audio device, similar to a USB MIC. This allows the users to monitor data streams in real time. Regardless of whether real-time streaming to a PC is performed, the raw ACC and processed MIC data are saved on a Class 10 or higher micro SD card on the receiver module. We did not save raw MIC data due to the requirements of confidentiality during acoustic recording in a public environment. An illustration of the sensor waveforms streamed to PC and saved to SD card is shown in Fig. 4. The receiver has an internal real-time clock that is powered by an external cell battery. When recording, the system saves the channel streams into five-minute data files, labeled with Figure 5. Packaged wireless voice monitor. (A) transmitter components the starting time of the block. This scheme handles the case of sewn on fabric neck strap and (B) enclosed receiver. a user being monitored in an environment that could interfere

other. In addition to localize bending, this method also [3] D. D. Mehta, P. C. Chwalek, T. F. Quatieri, and L. J. Brattain, "Wireless provides a larger ground plane for electrical noise dissipation. neck-surface accelerometer and microphone on flex circuit with application to noise-robust monitoring of Lombard speech," For packaging the transmitter (Fig. 5A), we chose a Proceedings of Interspeech, pp. 684-688, 2017. polyester material that would not irritate the user during long [4] T. F. Quatieri, K. Brady, D. Messing, J. P. Campbell, W. M. Campbell, M. S. Brandstein, et al., "Exploiting nonacoustic sensors for speech duration use. The polyester strap is commonly used for heart encoding," IEEE Transactions on Audio, Speech, and Language rate monitors strapped to a user’s chest. To secure the device Processing, vol. 14, pp. 533-544, 2006. to the fabric and to further ruggedize it, we potted the circuit [5] R. E. Hillman and D. D. Mehta, "Ambulatory monitoring of daily voice with Reprorubber® Thin Pour, a type of polysiloxane. During use," Perspectives on Voice and Voice Disorders, vol. 21, pp. 56-61, encasement, we masked off all ports and LEDs so that they 2011. [6] I. R. Titze, J. G. Švec, and P. S. Popolo, "Vocal dose measures: remain accessible to the user. The battery can be swappable Quantifying accumulated vibration exposure in vocal fold tissues," since different power capacities may be desired depending on Journal of Speech, Language, and Hearing Research, vol. 46, pp. 919- the application (with the usual tradeoff of increased size). 932, 2003. [7] D. D. Mehta, M. Zañartu, J. H. Van Stan, S. W. Feng, H. A. Cheyne II, For packaging the receiver (Fig. 5B), we required a case to and R. E. Hillman, "Smartphone-based detection of voice disorders by protect sensitive circuitry from damage since it would be long-term monitoring of neck acceleration features," Proceedings of the carried either on a belt clip or in a user’s pocket/purse (and IEEE International Conference on Body Sensor Networks, pp. 1-6, could potentially be dropped accidentally). Thus, we designed 2013. [8] D. D. Mehta, M. Zañartu, S. W. Feng, H. A. Cheyne II, and R. E. a rigid case that was 3D printed using Acrylonitrile Butadiene Hillman, "Mobile voice health monitoring using a wearable Styrene (ABS) plastic. The case fits the receiver board and a accelerometer sensor and a smartphone platform," IEEE Transactions battery with enough capacity to last an entire day of data on Biomedical Engineering, vol. 59, pp. 3090-3096, 2012. recording. Ports for USB inputs, switches, and LEDs were left [9] J. H. Van Stan, D. D. Mehta, D. Sternad, R. Petit, and R. E. Hillman, open and accessible to the user. "Ambulatory voice biofeedback: Relative frequency and summary feedback effects on performance and retention of reduced vocal intensity in the daily lives of participants with normal voices," Journal IV. CONCLUSION of Speech, Language, and Hearing Research, vol. 60, pp. 853-864, 2017. In this paper, we presented our latest development and [10] L. J. Brattain, R. Landman, K. A. Johnson, P. Chwalek, J. Hyman, J. implementation of an ambulatory voice monitor designed for Sharma, et al., "A multimodal sensor system for automated marmoset the monitoring of real-world voice characteristics. The ability behavioral analysis," in IEEE 13th International Conference on to stream synchronized sensor data from a MEMS MIC and Wearable and Implantable Body Sensor Networks (BSN), 2016, pp. 254-259. an ACC on a flexible substrate allows us to correlate the two [11] P. Matychuk, "The role of child-directed speech in language sensors. In addition, the acoustic MIC gives us acquisition: a case study," Language Sciences, vol. 27, pp. 301-379, complementary data when processing the raw non-acoustic 2005. [12] D. Mehta, J. Van Stan, and R. Hillman, "Relationships between vocal signal. The ACC is more immune to acoustic noise and may function measures derived from an acoustic microphone and a reveal features not derivable from the MIC. subglottal neck-surface accelerometer," IEEE/ACM Transactions on Future work could consist of adding a more capable Audio, Speech, and Language Processing, vol. 24, pp. 659-668, 2016. [13] M. Zañartu, J. C. Ho, D. D. Mehta, R. E. Hillman, and G. R. Wodicka, processor to the receiver to enable more complex machine "Subglottal impedance-based inverse filtering of voiced sounds using learning algorithms, Bluetooth communication with a neck surface acceleration," IEEE Transactions on Audio, Speech, and smartphone, and additional sensors to the transmitter circuitry Language Processing, vol. 21, pp. 1929-1939, 2013. to increase the system’s capabilities. Streaming more than two [14] V. M. Espinoza, M. Zañartu, J. H. Van Stan, D. D. Mehta, and R. E. Hillman, "Glottal aerodynamic measures in women with data streams would warrant shifting away from the off-the- phonotraumatic and nonphonotraumatic vocal hyperfunction," Journal shelf BC127/128 wireless modules and developing custom of Speech, Language, and Hearing Research, vol. 60, pp. 2159-2169, wireless transceiver circuitry and firmware. 2017. Other vocal function measures available include ACC signal properties of periodicity, harmonic spectral tilt, low- to high-frequency spectral power ratio, and cepstral peak prominence [2]. These types of acoustic-based measures have known relationships with ACC signal properties [12] and can be explored along with aerodynamic measures of vocal function hypothesized to be particularly salient for the assessment of common voice disorders [2, 13, 14]. REFERENCES [1] M. Zañartu, J. C. Ho, S. S. Kraman, H. Pasterkamp, J. E. Huber, and G. R. Wodicka, "Air-borne and tissue-borne sensitivities of bioacoustic sensors used on the skin surface," IEEE Transactions on Biomedical Engineering, vol. 56, pp. 443-451, 2009. [2] D. D. Mehta, J. H. Van Stan, M. Zañartu, M. Ghassemi, J. V. Guttag, V. M. Espinoza, et al., "Using ambulatory voice monitoring to investigate common voice disorders: Research update," Frontiers in Bioengineering and Biotechnology, vol. 3, pp. 1-14, 2015.


Document Created: 2017-12-11 11:40:08
Document Modified: 2017-12-11 11:40:08
© 2024 FCC.report
This site is not affiliated with or endorsed by the FCC