Proceedings ofthe 20!}: Annual International Conference aflhe IEEE Engineering in Medicine and Biology Society, Vol. 20, No 4, I998
`
`The Ring Sensor: a New Ambulatory Wearable Sensor for Twenty-
`Four Hour Patient Monitoring
`
`Sokwoo Rhee, _Boo—Ho Yang, Kuowei Chang and Haruhiko H. Asada
`d’Arbe1off Laboratory for Information Systems and Technology
`Department of Mechanical Engineering
`Massachusetts Institute of Technology
`Cambridge, MA 02139, USA.
`Email: sokwoo@mit.edu
`
`This paper describes the development of a ring
`Abstract
`sensor for twenty-four hour patient monitoring. The ring is
`packed with LBDs and photo detectors where the technology
`of pulse oximetry is implemented for blood oxygen saturation
`monitoring. The measured data are transmitted to a computer
`through a digital wireless communication link. The ring
`sensor is worn by the patient at all times, hence the health
`status is monitored 24 hours a day. Detailed descriptions of
`the hardware and the software of the ring sensor will be
`presented. Also, the effects of motion artifact and ambient
`light will be investigated.
`
`1.
`
`Introduction
`
`As the population of aged people increases, close and
`continuous monitoring becomes more important. Real-time.
`continuous monitoring would allow not only for emergency
`detection but also for long-term assessment to establish the
`tight dose
`and timing of medication. Especially.
`an
`ambulatory system that would allow long-term monitoring of
`otherwise difficult
`and noncompliant patients
`such as
`demented elderly people is highly in demand. A couple of
`compact. continuous monitoring devices have been developed
`[1-2] for elderly care. However, these devices have not been
`widely accepted due to the lack of functionality and comfort
`for wearers.
`
`To answer these needs. we have developed a compact.
`non-obtrusive telemeterd wearable patient-monitoring device
`in a ring configuration. Figure 1 shows a photograph of the
`miniaturized ring sensor. This
`sensor
`is equipped with
`optoelectric components that allow for long-term monitoring
`of the paticnt’s arterial blood volume waveforms and blood
`oxygen saturation non-invasively and continuously [3-4].
`These signals are transmitted to a home computer
`for
`diagnosis of the patient’s cardiovascular conditions. This
`continuous monitoring system can provide unique and useful
`information for preventive diagnosis in which long-term
`trends and signal patterns are more important. The ring
`sensor is completely wireless and miniaturized so that the
`patient can wear the device comfortably twenty—four hours a
`
`CPU LEDS and
`
`Photodiode
`
`Figure 1 : The photograph of the finger ring sensor. The size of
`the circuit board on top of the ring is 0.3 by 0.8 inch.
`
`to provide detailed
`is
`day. The objective of this paper
`descriptions of the hardware and software of the ring sensor.
`Also, the effects of motion artifact and ambient light will be
`investigated.
`
`2. Method
`
`A finger ring is a unique form of wearable sensors and.
`probably,
`the only thing that
`the majority of people will
`accept to wear at all times. To monitor a patient twenty-four
`hours a day continually, a miniaturized sensor in a ring is a
`rational design choice. Other personal ornaments and portable
`instruments, such as car rings and wrist watches. are not
`continually worn in daily living. When taking a shower, for
`example, people remove wrist watches. Bathrooms, however,
`are one of the most dangerous places in the home. More than
`10,000 people. mostly hypertensives and the elderly, die in
`bathrooms every year. Miniature ring sensors provide a
`promising approach to guarantee the monitoring of a patient
`at all times.
`LEDs with two different wavelengths, red and near infra-
`red. as well as a photodiode are imbedded in the ring facing
`inwardly. The red and infra-red LEDS are alternately turned
`on and the output from the photodiode is amplified and
`
`0-T303-5164-9!93J'$I0.00 © 1998 IEEE
`
`1 9'06
`
`Apple Inc.
`Apple Inc.
`APL1021
`APLl02l
`U.S. Patent No. 8,989,830
`U.S. Patent No. 8,989,830
`
`

`
`switched to a sample-and-hold filtering circuit to generate a
`piecewise constant wave for each wavelength of light. This
`alternative and sample-and-hold sequence is repeated at the
`frequency of 1000 Hz to eliminate light interference even in a
`quickly changing background of room lights. The resultant
`waves are filtered and conditioned as photoplethysmograrns.
`An 8-bit AID converter samples each photoplethysmogram at
`the frequency of 30 Hz and the digital signals are transmitted
`by a RF wave through the standard RS-232 protocol. The
`whole process is
`scheduled and controlled by a single
`microprocessor on the ring.
`received and
`are
`Transmitted photoplethysmograms
`analyzed by a home computer. The technology of pulse
`oximetry [4] is implemented on the computer for continuously
`monitoring the patient’s pulses and blood oxygen saturation.
`Although the signal
`is already filtered and refined by the
`analog signal conditioner in the ring sensor, it still contains
`the high frequency noise due to an ambient light source and
`motion artifact. For example. Figure 2 (a) shows a steady
`photoplethysmograrn without having any artifact, whereas
`Figures 2 (b) and (c) show the signal contaminated with the
`influence of ambient light and motion artifact respectively. It
`is clearly seen that
`the contaminated signal carries high
`frequency noise even though it already passed through a
`hardware lowpass filter. When the host computer detects
`these high-frequency noise, the computer does not display the
`wave forms on the screen not uses them for pulse oximetry.
`
`‘IUD (E) 50 0
`
`D
`
`100
`OD
`U3) O0
`400
`159
`100
`60
`II
`
`(C)
`
`0
`
`2
`
`4
`
`G
`
`8
`
`50
`
`I2
`
`E‘
`r
`
`2
`
`I
`r-
`
`ti
`
`B
`-r
`
`6
`
`10
`I
`r- —-—-r-
`
`B
`
`IO
`
`time(sec)
`
`I2
`
`12
`
`3. Hardware Description
`
`A block diagram of the ring sensor is shown in Figure 3.
`All of these components are contained in a small circuit board
`of 0.8 by 0.8 inch, as shown in Figure 1. Each of the blocks in
`the diagram is explained in detail as follows.
`
`3.}. LED: and Phatodiodc
`
`One red LED and two infra-red LEDs are used as the
`
`light sources. The peak wavelength of the red LED is 660 nm,
`and that of the infra—red LEDs is 940 nm. The photodiode has
`the peak wavelength of 940 rim and spectral sensitivity ranges
`from 500 nm to 1000 nm, which meets our needs. The voltage
`drop of the red LED is 1.6 V and that of the infra-red LEDs is
`1.2 V, and two infra—red LEDs are connected in serial. We
`used LEDs in a die form and the diameter is less than 0.1mm.
`
`3.2. First-Stage Amplifier.
`
`The first stage amplifier must be fast enough to keep in
`pace with the flickering speed of the LEDs, which means t.l'1at
`it must have a high slew rate. On the other hand, it is not
`desirable if this amplifier consumes a lot of power. We chose
`OPAB36 surface mount style amplifier from Burr-Brown.
`This amplifier has 0.03 Wits of slew rate which is quite high
`for a 20 ].l.A low power amplifier. Furthermore. this amplifier
`is designed to be used as a pre-amp for photodiode, which
`also satisfies our need.
`
`3.3. Sample-and-Hold Circuit
`
`Since one photodiode is shared by two channels of signal
`conditioner, the sample and hold circuit is necessary to hold
`the right signal for a brief moment. The two LEDs are
`alternatively lit and the two bilateral switches are also in
`synchronization with the LEDs. When the red LED is on. the
`bilateral switch (MCl4066B from Motorola) of the firs!
`channel is turned on to make the signal flow into the first
`channel. When the infra—red LED is on. the switch on the
`second channel
`is turned on and the signal
`is held by the
`sample-and-hold circuit. With
`these
`sampling—and-hold
`channels. the single photodiode can generate two wave fonns
`from the different LEDs at the same time. The sample-and-
`hold circuit comes with a 1000 pF capacitor which is enough
`
`Figure 2 : Various signals detected by the ring sensor
`
`Red LED Signal
`
`
`
`Figure 3: Block diagram of the ring sensor
`
`1 907
`
`

`
`to hold the signal for a while. To reduce the circuit size to that
`of a real ring. die-form chips are used and the connections
`were done by wire bonding machine that uses extremely thin
`gold wire.
`
`3.4. Signal Conditioner
`
`The signal conditioning part is composed of filters and
`amplifiers. Since the signal from the first stage amplifier is
`weak in a milli-volt range, it must be amplified by the order
`of 1000 times. We used a MAX407 operational amplifier
`from Maxim for the signal conditioner stage. One of the
`major reason for choosing this amplifier is that it consumes
`extremely low power which is around 1.2 ].tA per amplifier. In
`this stage. the slew rate is not an important factor since the
`frequency of interest at this stage is less than 10 Hz. This
`amplifier is also used in a lowpass filter circuit. The lowpass
`filter cuts off most of the frequency components higher than
`20 Hz. Also there is a simple highpass filter circuit composed
`of a resistor and a capacitor that removes DC components.
`We used die—forrn integrated chips and wire-bonding-style
`resistors of which size is on the order of 20 by 40 mil.
`
`3.4. CPU
`
`The CPU on the board controls all the operations of the
`ring, from scheduling LEDS to digitally convening acquired
`analog signals to fonnatting the signals in a RS-232 form for
`transmission. Since the CPU is one of the major components
`of power consumption, it has to be chosen carefully. For the
`purpose, we chose a PICl6C71l from Microchip. This CPU
`has two channels of embedded AID converter, 8 channels of
`
`digital U0 line. It has I KB of EPROM which is enough for
`the code that satisfies our task. An advantage of this chip is
`that it consumes very low power (usually less than 40 pm with
`32 kHz clock speed.) in the normal operation mode and
`almost no power in the sleep mode. This chip even comes
`with built—in RS-232 signal generation function. However, we
`didn’t use that function, since a much higher clock speed is
`necessary to obtain a satisfactory baud rate if we are to use
`this built-in RS-232 generator which will
`result
`in more
`power consumption.
`
`3.5. RF Transmitter
`
`the LED
`The piecewise constant waves generated at
`circuit are converted to digital signals by an 8-bit AID
`converter and transmitted through a RF wave by the
`microprocessor. The transmitter
`is
`simply a. ON.-‘OFF
`transmitter. In other words, it transmits signal when the input
`is high, and does not transmit anything when the input is low,
`hence, the power is consumed only when the input is high.
`We can save the power by reducing the width of the ‘I’ bit.
`which will happen when we use a higher baud rate. Currently
`we are using 600 and 1200 bps.
`
`4. Software Description
`
`4. l’. Software for the Microprocessor on Ring Side
`
`The assembly program was loaded in the microprocessor
`on the ring sensor. The first process of the code is the
`initialization of the CPU. Then it triggers the AH) conversion
`of a channel. During the AID conversion delay, it retrieves 3.
`number from another AID channel. Then it
`transmits the
`number in RS-232 format. To match the timing of a certain
`baud rate. we counted the number of the instructions executed
`to send one bit, and calculated the time for one bit
`transmission. Each 8 bit number is sent with a start bit and a
`
`stop bit which complete the RS—232 protocol. This process is
`done in both channels in turns.
`
`4.2. Sofrwarefor the Host Compmer
`
`A RF receiver receives transmitted signals in the RS-232
`form and transmits the data to the home PC thought a serial
`port. The software on the host computer is run under the
`Windows 95 or Windows NT environment. It is programmed
`using Microsoft Visual C++ 5.0, and uses the standard serial
`cornmunication programming technique. It is natural that the
`program has to be in the standard Win32 program format, and
`this format has to start with initialization of the windows and
`the variables.
`
`Apart from the standard windows prograrnniing routines,
`most of the program is dedicated to detecting the faulty signal
`and removing those noise signal from the clean heart beat
`signal. The multi-threading routines check the serial port
`continuously to find out if any data have arrived. As soon as
`some data arrive at the serial port, the program counts the
`number of red LED signal and the number of infra-red signal.
`If
`the signal
`transmission was done correctly,
`the two
`numbers must be almost exactly the same. If this is not the
`case,
`this means that
`there was some problem with signal
`transmission. In this case, the program considers the received
`signal as noise and ignore them. If the number of red LED
`signal and that of infra—red are more or less the same, then the
`program checks the frequency component of the received
`data.
`If the data includes any strange high frequency
`components such as that of Figure 4 and Figure 5,
`the
`program thinks that these data were contaminated by motion
`artifact or ambient light influence, and discard them. If the
`received signal passes through these filtering processes, the
`program accepts the data as valid ones and display them on
`the screen. The program also measures the peak-to-peak
`distance at this stage and calculates the pulse rate. The same
`process is done continuously whenever any data is detected
`available at the serial port.
`
`5. Validation of the Device
`
`To establish the validity of this instrumentation system,
`two kinds of experiments were
`conducted. The
`first
`
`1 908
`
`

`
`and the monitoring system have the following distinctive
`features:
`
`oxygen
`and
`0 Measurement of photoplethysrnograms
`saturation for diagnosis of the patient‘s cardiovascular
`conditions.
`
`0
`
`Continuous monitoring to provide unique and richer
`physiological data.
`0 Discrimination of valid heart beat signal from the noise
`generated by other sources.
`
`The hardware and the software of the system were described
`in detail. and the methods of avoiding a faulty heart beat
`detection due to the interference of the motion artifact and the
`
`ambient light were suggested and verified. The experimental
`results show that the invalid noise can be discriminated from -
`
`the valid signal effectively by detecting the high frequency
`component and the saturation of the signal.
`150
`
`I00
`
`‘I50
`
`0
`
`0
`
`n
`5
`
`|
`10
`
`'5
`
`I
`15
`
`.
`mttme (sec?
`
`5
`
`mtifilé (seals
`
`(:1)
`influence.
`light
`Figure 5 7 Signal contaminated with ambient
`Actual signal received by the host computer. The person took off the
`ring sensor at t=ll (sec). Alter diis point the signal is purely the
`noise from ambient light. (b) The host computer detected the high
`frequency noise from ambient light, and displayed just a flat line
`after the detection.
`
`experiment was designed to test how well this system detects
`motion artifact. After the signal from the ring sensor becomes
`stabilized,
`the wearer shakes his hand so that
`the signal
`becomes distorted due to the motion artifact. Figure 4-(a)
`shows the actual signal received at the host computer and the
`Figure 4-(b) shows the data displayed on the screen. As the
`signal begins to be contaminated by the motion artifact from
`t=l2 (sec), the software on the host computer detected a high
`frequency noise as well as a saturation of the signal
`to
`conclude that this is not
`the valid signal. and ignored the
`received signal and displayed just a flat line. Figure 5 shows
`the next experiment to test the detection of the noise from the
`ambient light source. At around tall (sec). the person took
`off the ring and put it on the table. Naturally the signal after
`this point is not a valid signal and only noise from ambient
`light would be acquired. Figure 5-(a) shows the actual data
`received by the host computer. The software detected the high
`frequency noise but no saturation. Therefore it concluded that
`this signal
`is not
`the valid signal. Figure 5-0:) shows the
`actual display on the screen as a flat line after the person took
`off the ring. which indicates that the system was not confused
`by the ambient
`light source and clearly discriminated the
`noise from the right signal.
`151]
`
` 0
`
`5
`
`in
`
`(I3)
`
`is
`
`“time (see?
`
`Figure 4 : Signal contaminated with motion artifact (a) Actual signal
`received by the host computer from the ring sensor. The person
`began to shake his hand from t=l 2 (sec). (in) Signal displayed on the
`screen. The host computer detected the motion artifact and ignored
`the signal. putting only a flat line after t=l2 (sec).
`
`Acknowledgement
`
`The authors would like to thank Ms. Yi Zhang for her
`intensive help in fabricating the ring sensor.
`
`6. Conclusions
`
`References
`
`In this paper. a twenty—four hour patient monitoring system
`using the ring sensor has been presented. The ring sensor is
`equipped with optoelectrical components for monitoring a
`patient‘s arterial blood flow in a finger base. A wireless
`transmitter on the ring sensor sends measured signals to a
`home computer through multiple receivers for diagnosis and
`abnormality detection. The host computer analyses
`the
`received data and discriminate the valid signal front the noise
`from motion artifact or ambient light source. The ring sensor
`
`1909
`
`[1]
`
`K. lkeda. A. Watanabe. and M. State. "A Vital Sign Sensor
`for Elderly People at Home." Biotelemetry. Vol. ll. l99l
`[2] M. Yamashita.
`K.
`Shimizu.
`and G. Matsumoto,
`“Development of a Ring-Type Vital Sign Telemeter."
`Biotelernetry XIII, I995
`J. C. Veraart. A. M. Van Der Kley. and H. A. M. Neumann.
`“Digital Photoplethysmography
`and Light Reflection
`Rheog1'aphy." J. of Dermatol Surg Oncol. Vol. 20. I994
`J. P. Welch, R. DeCesare, and D. H. Hess. “Pulse Oximctry:
`Instrumentation and Clinical Application." Respiratory Care.
`Vol.35. No. 6. June. 1990
`
`[3]
`
`[4]