The advantages of custom designing each sub-system and component of a .... implement data-acquisition and wireless media-access control protocols for the ...
Proceedings of the 26th Annual International Conference of the IEEE EMBS San Francisco, CA, USA • September 1-5, 2004
A TinyOS-Based Wireless Neural Interface Shahin Farshchi1, Istvan Mody2, and Jack W. Judy1 1
Electrical Engineering Department, and 2Department of Neurology University of California, Los Angeles, USA
Abstract- The overlay of a neural interface upon a TinyOSbased sensing and communication platform is described. The system amplifies, digitally encodes, and transmits two EEG channels of neural signals from an un-tethered subject to a remote gateway, which routes the signals to a client PC. This work demonstrates the viability of the TinyOS-based sensor technology as a foundation for chronic remote biological monitoring applications, and thus provides an opportunity to create a system that can leverage from the frequent networking and communications advancements being made by the global TinyOS-development community. Keywords—Brain-Machine Interface, Telemetry, TinyOS, Epilepsy, Smart Dust
EEG,
Wireless,
I. INTRODUCTION The rhythmically varying electrical impulses (i.e., field potentials) of large neural populations can vary at slow (1 Hz) or fast (100 Hz) rates. Such brain-wave activity has been correlated to specific physiological outcomes, such as sleep, excitation, and epilepsy. In order to quantify fieldpotential brain activity, an electroencephalogram (EEG) is recorded by measuring the potential difference between a pair of electrodes placed in or on the brain region of interest. Although EEG recordings are frequently performed as acute experiments (e.g., < 6 hrs), some studies require chronic or longer-term measurements. For example, the study of epilepsy requires continuous recordings to be made over a period of several days. Conventional EEG techniques use a direct-wired connection between the subject and the measurement tool. Typically, this connection consists of a bundle of fine wires that can frequently limit animal behavior. In addition, the wired connection prevents the environment from containing natural elements such as tubes and tunnels. The constraints of such direct-wired connections have the potential for skewing the obtained results. A wireless recording system could be used to remove the aforementioned constraints. Such a wireless neural recording system must be capable of sensing, amplifying, and transmitting neural signals with a sampling frequency of at least 100 Hz per channel, while being small, low cost, lightweight, and low power. The system also requires a receiver to receive, demodulate and display the transmitted neural signals. Existing approaches to develop a wireless EEG measurement tool have ranged from designing a custom microfabricated recording and telemetry system, to the use of commercial-off-the-shelf (COTS) PC technology. Each approach has its own set of advantages and drawbacks.
0-7803-8439-3/04/$20.00©2004 IEEE
The advantages of custom designing each sub-system and component of a neural recording system include (1) a greater degree of design flexibility and (2) the ability to optimize each sub-system in order to minimize power consumption and noise. Such custom-designed systems have combined bulk micromaching with on-chip CMOS circuitry to integrate neural probes with dataacquisition/transmission systems, which decrease the overall size and increase the signal-to-noise ratio [1]. However, advances can only be incorporated with a very significant and costly re-integration effort. In contrast, designing a neural interface that uses mainstream PC (COTS) technology, such as a scaled-down PC motherboard with an 802.11b PCMCIA card, requires a smaller development effort while providing current cuttingedge data-processing, networking, and communication technology [2]. However, these systems are very bulky and power intensive, since they were not originally designed for truly low-power applications. A novel approach would be a compromise between custom designing each sub-system and using PC-COTS components. This approach would achieve a balance between low-noise and low-power signal transmission, data communication, and networking performance.
Fig. 1. MICA2, MICA2DOT, and MIB510 PC-Interface Board [7].
II. TECHNOLOGY Global efforts, led by the Computer Science Department at the University of California at Berkeley, have succeeded in developing an operating system with a component-based runtime environment designed to provide support for embedded systems with a minimal amount of physical hardware to keep the system size very small (Fig. 1). The primary goal of this effort has been to facilitate the creation of ultra-low-power miniature devices that can be widely distributed in mesh networks (otherwise known as “multi-hop”) to remotely monitor low-frequency phenomena [3].
4334
The wireless sensor nodes, which are commonly referred to as “motes”, have been designed to operate using TinyOS and are currently being used in wildfireinstrumentation, habitat-monitoring, and global-positioning applications to mention just a few [4,5,6]. The datathroughput performance of the motes is severely constrained by their ultra-low-power operation and large-scale mesh networking capabilities. Conventional data-acquisition and communication protocols could increase data throughput at the expense of increased power-consumption and mote-tomote networking capabilities.
wireless neural recordings from several freely moving subjects simultaneously. III. DEVICE DESIGN The overall system design can be divided into two major components: hardware and software. A neural preamplifier circuit is required to properly amplify and level-shift the differential neural signals. TinyOS software components are required to implement data-acquisition, signal-transmission, signal-reception, and wireless mediaaccess protocols that achieve high data throughput. Finally, a client-side user interface is required to properly interpret, store, and display the received waveforms. A top-level diagram of the neural interface system is displayed in Figure 3.
Fig. 2. System-level schematic of MICA2DOT mote
The two types of motes used in this work are the MICA2 and MICA2DOT motes produced by Crossbow Technology Inc. [7]. A basic system schematic of the MICA2DOT mote is displayed in Figure 2. The MICA2DOT has six input channels, each with its own 10bit analog-to-digital converter (ADC). Data is processed by an Atmel Atmega128 microprocessor with 512 kilobytes of flash memory. Data transmission to and from both the MICA2 and MICA2DOT is handled by a Chipcon CC1000 radio chip. When the two 1.5-V dry-cell batteries are installed, the MICA2 is approximately the size of a matchbox (58 × 32 × 15 mm). The MICA2DOT uses essentially the same computational and communications hardware as the MICA2, but in a much smaller form-factor. The diameter of the MICA2DOT is roughly that of a United States quarter Dollar (25 mm), and its thickness is approximately 6 mm. An MIB510 serial PC interface will also be used to allow the motes to communicate with a PC. In addition, an MIB600 ethernet interface could also be used to broadcast sensor readings directly over the Internet. The MICA2, MICA2DOT, and MIB510 are shown in Figure 1 [7]. The work in this paper has been directed toward investigating new software and hardware designs that can be used to enable one or more motes to perform multi-channel
Fig. 3. Top-level diagram of the neural interface system.
A. Hardware Each EEG channel is sensed differentially by a pair of electrodes. A neural preamplifier circuit is used to take the differential signals and amplify, level-shift, and convert them into to a single-ended waveform ranging from 0 V to the MICA2DOT-battery voltage (nominally 3 V) in order to be properly digitized by the MICA2DOT ADCs. This preamplifier circuit has been designed to interface directly with the MICA2DOT mote. The heart of the neural preamplifier is an Analog Devices AD627 Instrumentation Amplifier. The gain of the AD627 is set by an external resistor to 200, and the output has been referenced to half the supply voltage by a simple resistive divider followed by a voltage-follower circuit. To avoid high-frequency noise from being aliased into the sampled signal, the AD627 output is followed by an RC-filter.
4335
B. Software TinyOS-software components have been written to implement data-acquisition and wireless media-access control protocols for the MICA2DOT. The MICA2 will operate on a standard TinyOS component to receive packets and broadcast them via a serial port or an ethernet connection. The data-acquisition component implements a twochannel signal-acquisition and packetization algorithm to maximize data throughput while maintaining acceptable data resolution. The media-access-control protocols have been adjusted to allow for greater data throughput. A Java-based program has been designed for use on the client PC. This program acquires data from a TCP/IP port, and displays them either as raw data points, or a reconstructed waveform. Signal reconstruction is performed by padding the original signal and passing it through an 8thorder Chebyshev filter.
data-acquisition and transmission system, a 5-mV 200-Hz differential sinusoidal signal was applied to the signal input of the system (Fig. 5). To assess the overall performance characteristics of the system, a 5-mV simulated cardiac signal was applied to the preamplifier with respect to ground (Fig. 6). This signal was generated by an HP 33120A Waveform Generator.
IV. EXPERIMENTAL TESTING Experimental testing was performed in two categories: bench testing and in-situ testing. Bench testing was performed to assess the specific performance metrics of the system, such as data rate, range, power consumption, and signal resolution. In-situ testing was used to evaluate the overall performance of the system in its respective application environment.
Fig. 6. Input, amplified, and received/reconstructed 3-Hz ECG.
B. In-Situ Testing The system was tested with a living mouse in a typical laboratory environment (Fig 7). Normal brain activity has been captured, followed by seizures induced by injecting kainic acid at 15 mg/kg to model temporal-lobe epilepsy [8].
Fig. 5. Input and received/reconstructed 200-Hz sinewave.
A. Bench Testing Fig. 7. Recorded normal and induced seizure activity.
To assess the performance of the client-side signalreconstruction program, as well as the total bandwidth of the
4336
TABLE I WIRELESS MICROSYSTEM PERFORMANCE COMPARISON (ADAPTED FROM [1]) Institution
UCLA
UCLA
Michigan
Duke
Aachen
Tokyo U.
Clev. Med.
Number of Data Channels
2
1
3
16
2
1
8
Telemetry Link Frequency
916 MHz
3 GHz
88-108 MHz
2.4 GHz
88-108 MHz
80-90 MHz
902- 928 MHz
Communication Scheme
FSK
Analog FM
TDMA
802.11b
FM Stereo
Analog FM
FSK
Power Supply
3V
Inductive
±1.5 V
3.3 V and 5 V
±1.4 V
3V
9V
Max. Power Dissipation
150 mW
13.8 mW
< 2.5 mW
4.0 W
-
10 mW
-
Transmission Range
152 m
< 1 meter
a few meters
9m
a few meters
~16 m
< 46 m
System Clock Frequency
4 MHz
-
70 / 138 kHz
66 MHz
-
-
-
Detectable Signal
0.06-15 mV
0.015-15 mV
0.1-5 mV
-
-
-
-
Detectable Frequency
240 Hz
10 kHz
10 kHz
-
-
-
-
Dimensions (cm)
2.6 × 2.6 × 1.8
1.4 × 1.2 × 0.4
1.8 × 1.3 × 0.16
5.1 × 8.1 × 12.4
2.5 x 1 x 0.5
1.5 x 0.8
6.4 x 5.1 x 1
Total Weight (w/ battery)
12.8 g
3.1 g
> 0.1 g
> 68 g
Connection with Probe
External
External
Integrated
External
External
External
External
Reference
This Work
[9]
[1]
[2]
[10]
[11]
[12]
[3]
V. CONCLUSIONS In this paper, we have demonstrated the design of a TinyOS-based wireless neural interface. The system is capable of amplifying, sampling, transmitting, and reconstructing input signals at a rate of 480 8-bit samples per second. This data rate allows for the reliable transmission of two 100-Hz-bandwidth EEG channels simultaneously from live, mobile test subjects. Table 1 illustrates the performance metrics of this technology, in comparison with other reported technologies. VI. ACKNOWLEDGEMENT The authors would like to thank Aleksey Pesterev for his development of the TinyOS data-acquisition component, and Paul Herag Nuyujukian for his work in signalreconstruction and preamplifier circuit-board design. The authors would also like to thank Dr. Jamie L. Maguire for preparing the test animal and setting up the live-animal experiment. Lastly, we would like to acknowledge input from the members of the Judylab, the CENS center, the Speech Processing and Auditory Perception Laboratory at UCLA, and the TinyOS community. REFERENCES [1]
[2]
P. Mohseni and K. Najafi. “A wireless FM multi-channel microsystem for biomedical neural recording applications,” IEEE Southwest Symposium on Mixed Signal Design, pp. 217-222, February 23-25 2003. I. Obeid, M. A. L. Nicolelis, and P. D. Wolf, “A multichannel telemetry system for single unit neural recordings” Journal of Neuroscience Methods, 133, pp. 123–135, 2004.
J. L. Hill, “System architecture for wireless sensor netowrks,” Ph.D. Dissertation, Computer Science Program, University of California, Berkeley, California, United States. 2003. [4] M. M. Chen, C. Majidi, D. M. Doolin, S. Glaser, and N. Sitar, “Design and construction of a wildfire instrumentation system using networked sensors (poster)” Network Embedded Systems Technology (NEST) Retreat, June 17-18, 2003, Oakland, CA. [5] J. J. Polastre, “Design and implementation of wireless sensor networks for habitat monitoring,” M.S. dissertation. Computer Science Program, University of California Berkeley, Berkeley, CA, USA, 2003. [6] D. Niculescu and B. Nath, “DV based positioning in ad hoc networks” Kluwer Journal of Telecommunications Systems, 2003. [7] Crossbow Technology, Inc. http://www.xbow.com [8] G. De Sarro, E. Russo, G. Ferreri, B. Giuseppe, M. A. Flocco, E. D. Di Paola, and A. De Sarro, “Seizure susceptibility to various convulsant stimuli of knockout interlukin-6 mice,” Pharmacology Biochemistry and Behavior, In Press, Corrected Proof [9] P. I. Pastor., I. Mody, and J. W. Judy, “Transcutaneous RF-powered neural recording device” Proc. of the 24th Annual Conference and the Annual Fall Meeting of the Biomedical Engineering SocietyEMBS/BMES Conference Vol. 3, pp.2105-2106. October 2326, 2002. [10] A. Nieder, “Miniature stereo radio transmitter for simultaneous recording of multiple single-neuron signals from behaving owls,” Journal of Neuroscience Methods, 101, pp. 157-164, 2000. [11] S. Takeuchi and I. Shimoyama, “An RF-telemetry system with shape memory alloy microelectrodes for neural recording of freely moving insects,” Proc. of the 1st Annual International IEEE-EMBS Special Topic Conference on Micro-Technologies in Medicine and Biology, Lyon, France, pp. 491-496, October 12-14 2000. [12] M. Modarreszadeh and R. N. Schmidt, “Wireless, 32-channel, EEG and epilepsy monitoring system,” Proc. of the 19th International Conference of IEEE/EMBS, Chicago, IL, pp. 1157-1160, Oct 30-Nov 2, 1997
4337
