A Scalable, 2.9 Mw, 1 Mb/S E-Textiles Body Area Network Transceiver with Remotely-Powered Nodes and Bi-Directional Data Communication

A Scalable, 2.9 mW, 1 Mb/s e-Textiles Body Area Network Transceiver with Remotely-Powered Nodes and Bi-Directional Data Communication

The MIT Faculty has made this article openly available. Please share how this access benefits you. Your story matters.

Citation Desai, Nachiket, Jerald Yoo, and Anantha P. Chandrakasan. “A Scalable, 2.9 mW, 1 Mb/s E-Textiles Body Area Network Transceiver With Remotely-Powered Nodes and Bi-Directional Data Communication.” IEEE Journal of Solid-State Circuits 49, no. 9 (September 2014): 1995–2004.

As Published http://dx.doi.org/10.1109/JSSC.2014.2328343

Publisher Institute of Electrical and Electronics Engineers (IEEE)

Version Author's final manuscript

Citable link http://hdl.handle.net/1721.1/98890

Detailed Terms http://creativecommons.org/licenses/by-nc-sa/4.0/ 1 A Scalable, 2.9 mW, 1 Mb/s e-Textiles Body Area Network Transceiver with Remotely-Powered Nodes and Bi-Directional Data Communication Nachiket Desai, Student Member, IEEE, Jerald Yoo, Member, IEEE, and Anantha P. Chandrakasan, Fellow, IEEE

Abstract—This paper presents transceivers and a wireless sends them on to a healthcare provider using conventional power delivery system for a Body-Area Network (BAN) that data networks. uses an e-textiles-based physical layer (PHY) capable of linking Wireless communication devices forming a BAN typically a diverse set of sensor nodes monitoring vital signs on the user’s body. A central base station in the network controls operate in the unlicensed bands, and have to contend with the power delivery and communication resource allotment for every high losses in those bands in the operating environment [5], node using a general-purpose on-chip Node Network Interface [6]. Another scheme for communication with the local base (NNI). The architecture of the network ensures fault-tolerance, station is the use of body-coupled communication (BCC) [7]– reconfigurability and ease of use through a dual wireless-wireline [9]. BCC transceivers avoid a number of multiple-access prob- topology. The nodes are powered at a peak end-to-end efficiency of 1.2% and can transmit measured data at a peak rate of 1 Mb/s. lems when a large number people operate BANs in close prox- Modulation schemes for communication in both directions have imity. However, the path loss in BCC channels is comparable been chosen and a Medium Access and Control (MAC) protocol to wireless channels. In contrast, textiles with electric routing has been designed and implemented on chip to reduce complexity (e-textiles) printed on top of the cloth or woven in as part of at the power-constrained nodes, and move it to the base station. the yarn offer much lower path loss, and have long been a topic While transferring power to a single node at maximum efficiency, the base station consumes 2.9 mW power and the node recovers of research with the aim of building wearable computers [10]. 34 µW, of which 14 µW is used to power the network interface They allow remote powering of the microwatt sensors and circuits while the rest can be used to power signal acquisition their associated telemetry circuits by a network base station circuitry. Fabricated in 0.18 µm CMOS technology, the base [11]–[13]. Building on these advantages, the proposed BAN 2 2 station and the NNI occupy 2.95 mm and 1.46 mm area improves efficiency by using resonant power transfer between respectively. coupled inductors and combines this with electric routing on Index Terms—Body-area networks, e-textiles, continuous clothing to afford better aesthetics and greater comfort to the health monitoring, wireless power delivery, inductive links, in- user by avoiding wires going all the way to sensors adhered tegrated medium access protocol to the body. This allows for cheap, disposable sensors that do not need an integrated battery or radio and can be easily I.INTRODUCTION attached to and removed from the user’s body depending on his/her specific requirements. The system architecture and the ODY-area networks (BANs) allow continuous monitor- circuits used for resonant power transmission and recovery ing of vital signs by linking small, low-power sensors are presented in Sections II and III respectively. Modulation implantedB in and placed on the surface of the body and schemes for communication to and from the nodes have been transmitting the data they measure to healthcare providers chosen to lower the power requirements at the nodes and push [1]. Current state-of-the-art biomedical front-ends [2]–[4] can complexity to the base station. These modulation schemes and measure signals using only microwatts of power, making the circuits implementing them are presented in Section IV. it possible to aggressively minimize their volume and the Medium Access and Control (MAC) protocols used in amount of energy stored locally. In order to realize their commercially-deployed standards for wireless BANs, such potential afforded by these advantageous traits, it makes sense as Bluetooth LE and IEEE 802.15.4/Zigbee [14] include to transfer the data using a two step process. Uncompressed significant protocol overhead at the sensor nodes and are not data is first sent to a local base station (like a cellphone) using compatible with microwatt-sensors. Allowing sensors to sleep a link requiring minimal transmission effort by the sensor and requiring them to wake up after a preset amount of time node. The local base station then processes the signals and [15] to check if the channel is unoccupied requires an accurate, Manuscript received August 27, 2013; revised January 16, 2014. This power-hungry clock at the sensor that is not power-gated. A work was funded by the cooperative agreement between the Masdar MAC protocol for e-textile based BANs that relies on Time- Institute of Science and Technology, Abu Dhabi, UAE and the Mas- Division Multiple Access (TDMA) for resource allotment has sachusetts Institute of Technology, Cambridge, MA, USA, Reference No. 196F/002/707/102f/70/9374. been presented in [11]. A custom MAC protocol that moves N. Desai and A. P. Chandrakasan are with the Department of Electrical all decision-making to the base station to reduce power con- Engineering and Computer Science, Massachusetts Institute of Technology, sumption at the sensor nodes has been proposed and integrated Cambridge, MA, 02139 USA e-mail: [email protected] tel: 617-253-0164. J. Yoo is with the Department of Electrical Engineering and Computer on chip. The custom MAC protocol also allows sensors to Science, Masdar Institute of Science and Technology, Abu Dhabi, UAE. be added and removed easily in order to keep the network 2

Inductors forming links with sensor nodes

32 mm Electric routing on textiles

Base Station

Fig. 1. An e-textiles shirt implemented with screen-printed spiral inductors linking to sensor nodes (worn inside out for demonstration). The base station’s Fig. 2. Cross-section of inductive link between e-textiles and sensor node. size is limited by test and debug structures on the board.

inductor and high power consumption when all inductors are reconfigurable and supports TDMA and Contention Access in the same sub-network. Also, by not adhering to a strict (CA) modes for sensors with wide variation in channel access arrangement like the (X,Y) grid in [12] or the linked list and bandwidth requirements. Details of the implemented MAC in [17], the proposed network architecture allows for greater are presented in Section V. freedom in covering sensors around the body based on user- specific sensor locations. An obvious arrangement would be II.SYSTEM ARCHITECTURE to form sub-networks of inductors covering one part of the A. e-Textiles Network body. The only limitations would be the number of sub- Figure 1 shows an example of the proposed network im- networks that can be supported by the base station and the plemented on a shirt with the user holding the local base combined inductance of the parallelly-connected coils in each station, whose size is limited by the test and debug structures sub-network. For example, the shirt in Figure 1 can support on the board. Spiral inductors and their associated routing that 2-channel ECG while a cap made of similar material can connect them to a central base station have been screen-printed function as an EEG helmet. on fabric using the process described in [16]. Restricting the Power is transferred to the sensor nodes by resonant cou- circuits on textiles to screen-printed passives avoids expensive pling between inductors at the base station and the sensor procedures to bond silicon dies to the routing on fabric and nodes. The network has a star topology, with all sensor nodes keeps the costs of the textiles low. These inductors form near- talking directly to the base station. Modulation schemes for field magnetic links with similar inductors on the sensor nodes, data transmitted in both directions have been chosen to push as shown in Figure 2. These nodes are of a band-aid patch form most of the modulation and demodulation effort away from factor and are worn on the body. The local base station can be the sensor nodes towards the base station. integrated in a cellphone linked to the shirt with a snap-button interface or a secondary inductive link (with additional losses), B. Base Station or onto a clip-on bluetooth-like device linked to the user’s cellphone. This hybrid wireless/wired architecture achieves A block diagram of the implemented base station for the better power efficiency compared to fully wireless- or BCC- network-on-shirt of Figure 1 is shown in Figure 4. It can based designs while being more comfortable to use than fully support up to two sub-networks each containing up to two wired systems, such as the one presented in [11]. inductors. The base station has a separate analog interface To maximize coverage and allow different varieties of for each of the two sub-networks it can support. Both sub- sensors to operate in the same network, a large number of in- networks share the same digital back-end. The demodulated ductors must be placed on the e-textiles clothing. Additionally, bitstreams from both the sub-networks are combined asyn- e-textiles need to be designed keeping in mind that clothes are chronously in the digital domain and sent on to the digital subject to harsher environments than most electronics used in back-end that implements the on-chip MAC and manages daily life, for example inside a washing machine. In order to power delivery to the sensor nodes in the network. A regulator prevent network faults from arising due to a break (“opens”), with four discrete output levels controls the amount of power redundant paths must be included in the network routing. transmitted to each sub-network. A digital on/off controller Inductors are grouped to form sub-networks that can be turned switches off a sub-network if it is devoid of active sensor on/off independently, as shown in Figure 3. Inductors in every nodes and doubles up as an on-off keying (OOK) modulator sub-network are connected in parallel and the sub-network as for downlink transmission. The analog interface to each sub- a whole has multiple independent paths connecting it to the network consists of the front-end oscillators, power amplifiers base station. This arrangement allows for a trade-off between and LC-tank filters for power transmission, and modulators routing complexity when each sub-network contains only one and demodulators for data communication. 3

SN0,0 SN0,1 SN0,P SNN,0 SNN,Q Downlink NNI NNI NNI Diverse Set of NNI NNI On-Body SNs Sensor

Clothing Inductor Sub- Uplink Networks on Clothing Power Body-Area Network on Clothing Data (with wireline interface to Base Station)

Fig. 3. Architecture of proposed BAN. Each sub-network can power and talk to a heterogeneous group of sensors. The NNIs can access a sub-network through any one or more of its constituent inductors.

Sub-Network Analog Asynchronous Interfaces MAC Digital Digital Multiplexer Command Demodulator ASK Demodulation Interpreter ASK Demodulation CRC

SN ListSN Inductor Network Checker Sub- Networks VDD 4-Level Power Auto- 4-Level Power MAC Controller Commutative Controller Mgmt. Mgmt. Power Rectifier CSTOR Fabric VDD VDD PAPA Inductor PAPA

PWM PWM PA Serial Patch Packetizer Interface & ADC AFE Modulator

Node Network Interface Application-Specific Fig. 4. Block diagram of base station. (on-chip) Readout Circuitry

Fig. 5. Block diagram of a typical sensor node implementation. Only the C. Sensor Nodes Node Network Interface has been implemented on chip, while the application- dependent sensing and data conversion circuits are off-chip. Sensor nodes are partitioned into an application-specific biomedical data acquisition module and a general-purpose NNI, as shown in Figure 5. The on-chip NNI module supplies I1 I2 power to the sensor readout circuitry and transmits the data + R + + s R R measured by it. The off-chip readout circuits consist primarily 1 2 ..k of a front-end amplifier followed by an ADC and can be any C L L C 1 1 2 2 R state-of-the-art microwatt-level sensor that measures signals on L Vs V1 Z1 Z2 V2 the human body [2]–[4]. An auto-commutative rectifier (ACR) + + sMI sMI minimizes the dead zone for rectification at low-amplitude ac 2 − − 1 inputs, and converts ac power transmitted by the base station - - - to dc for use by all circuits at the sensor node. A low power, fully digital demodulator passes on instructions sent by the Fig. 6. Power transfer through inductively coupled resonators. The√ mutual base station to the digital back-end for interpretation. These inductance M and the coupling coefficient k are related as M = k L1L2. instructions directly manage the data collected by the sensor front-end and control how and when it is packetized and transmitted to the base station. The fabric inductors used have low quality factors (Q ≈ 6 − 8) in the MHz range, hence the overall loaded Q of III.RESONANT POWER DELIVERY INTERFACE the secondary resonator is dominated by the inductor losses 1 A. Theoretical Overview and not the load . Also, the coupling coefficient across the inductive link, k, is approximately 0.1 for a separation as low An RLC circuit model for power transfer across the induc- as 5 mm between the primary and secondary inductors. In tive link is shown in Figure 6. The resonant near-field power this regime of operation, the total efficiency of power transfer transfer scheme implemented is similar to the one described in [18], except that a parallel LC-tank is employed on the primary (base station) side. The resistor RL in Figure 6 models the ac 1Assuming the rectifier supplying dc power to the sensor node consumes input resistance of the rectifier that supplies dc power to the as much as 90 µA (p-p) current at 2.4 V (p-p) to supply 30 µA at 1.8 V dc, sensor node. this yields RL=26.7 kΩ and QL ≈ 100. 4 across the inductive link can be approximated as Class-D PA 2 Static- k Q1Q2 Q2 η ≈ · (1) Configured 1 + k2Q Q Q Vref Cap. Banks 1 2 L - where QL, the quality factor of load, is defined as QL , + ω0C2RL and Q1 and Q2 are the intrinsic quality factors of Lfabric the primary and secondary inductors respectively. Equation (1) Cres C1 can be obtained from the expressions derived in [19]. The voltage transfer function from the primary to the C C res secondary can be approximated as 2 √ r V2(jω0) k Q1Q2 L2 ≈ 2 · (2) Vs(jω0) 1 + k Q1Q2 L1 Fig. 7. Schematic of self-tuned combined oscillator-PA architecture. Increasing the coupling coefficient and the Q of the resonators improves the efficiency of power transfer and the magnitude Vo of the voltage transfer function, as shown by Equations (1) L and (2). From the expressions for inductance in [16], a 6 turn spiral inductor with outer diameter of 30 mm has 1.6 gmVf Ro C Vf µH inductance. This agrees with the measured values of 1.4 1 C2 µH inductance. The measured self-resonant frequency of the inductors is 80 MHz, and the 27.12 MHz ISM band was chosen for operation since it is sufficiently below the self- resonant frequency for the inductors to remain dominantly Fig. 8. Small signal model of oscillator with modified Colpitts’ feedback network. inductive and the on-chip capacitors required are around 20 pF. These values lead to a link (ac-ac) efficiency of 1.7% and voltage transfer function 0.39 at the lowest power setting of the small-signal amplification of the class-D PA, which is dc- the transmitter.2 biased around mid-rail by negative feedback. The small-signal model is obtained by linearizing the circuit at the dc bias point.

B. Combined Oscillator-PA Transmitter −1 The oscillator oscillates at f = 2πpLC , where The circuit used for resonant power transfer is shown in 0 eq C = C ||C . For a desired resonant frequency and given Figure 7. The oscillator uses a modified Colpitts’ feedback eq 1 2 inductance value, the area of the on-chip capacitors is mini- structure, which also doubles as the output filter. The filter mized when C = C . This also leads to V and V oscillating is made of inductors on fabric and on-chip MIM capacitors. 1 2 o f between 0 and V in opposite phase, maximizing the voltage A parallel LC tank is used to avoid the large voltage stress DD across the fabric inductor. that would arise from resonating on-chip capacitors in a series LC configuration. A class-D power amplifier is integrated into the oscillator feedback loop. Output power control is achieved C. Auto-Commutative Rectifier by reducing the active device widths of the transistors in the AC power transmitted by the base station is rectified at the PA to make them function as linear regulators in the lower sensor nodes by using a self-synchronous topology similar to power settings. This avoids both requiring two separate tuned the one presented in [20], [21], which trades off efficiency networks in the MHz range and tuning those two networks to for the capability to generate workable dc voltages from resonate at the same frequency. The on-chip capacitors employ ac input amplitudes small enough to limit the transistors’ two-level tuning in order to maintain operation at the desired performance by their threshold voltages. A schematic of the frequency. Coarse tuning selects the number of capacitors to auto-commutative rectifier (ACR) used is shown in Figure 9. be connected in the network depending on the number of The gates of all transistors are biased such that the devices inductors connected in parallel in the sub-network. Fine tuning conduct strongly in the desired half-cycle and are strongly adjusts for variations in the inductance values of the fabric turned off in the other. The use of additional comparators and inductors; a one-time calibration step can tune out up to 20% gate-drive circuits is avoided in order to extract net positive static variation using four bits, which is sufficient based on power in the µW region of operation. Additionally, no charge inductance measurements made on 24 samples. is stored on capacitors, which might be leaky or expensive to A small-signal model of the oscillator is shown in Figure 8. integrate, in order to overcome the threshold voltages of the R The resistance o is the small-signal output resistance of the transistors. class-D PA. The loop transfer function for the oscillator has Simulation waveforms for the operation of the ACR are three poles and no zeros, making the system oscillate above shown in Figure 10. The transistors Mp1 (or Mp2) and Mn1 a minimum dc loop gain. This dc loop gain is provided by (or Mn2) in Figure 9 are supposed to conduct when VACp (or 2 V V V V The value of Rs for the highest and lowest power settings of the transmitter ACn ) is higher than DC+ and ACn (or ACp ) is lower than are approximately 140 Ω and 480 Ω respectively. VDC−. Reverse conduction is possible during the initial phase 5

AC p 1 Auto−Commutative Recitifer Diode−Connected Rectifier 0.8 Mn2 Mp1

0.6 DC- DC+

0.4 Mn1 Mp2

DC Output Voltage (V) 0.2

ACn 0 0.2 0.4 0.6 0.8 1 1.2 Input AC Amplitude (V) Fig. 9. Schematic of auto-commutative rectiﬁer (ACR). Fig. 11. Comparison of simulated output voltages generated by ACR and Ω 800 diode-connected rectiﬁer supplying dc power to a 25 k load. V 600 AC,p V DC+ 400 Cc Cc Cc 200

Voltage (mV) 0 Vac,p Vac,p Vac,p Vac,p

Cres - + - + - + - + ACR ACR ACR 300 I C ACR C C C Mp1 s s s stor A) L µ I patch V V V V 100 Mn1 ac,n ac,n ac,n ac,n 0 −100

Current ( C C C −300 c c c

Fig. 12. Schematic of rectifiers connected in series to generate required dc voltage at sensor node. The schematic for the ACR block is shown in Figure 9. Fig. 10. Simulation waveforms for operation of ACR. of each half cycle and is kept lower than the forward current ulation schemes such as FSK and OQPSK, which are more by the threshold voltages of the transistors. The net dc current complex than OOK-based schemes, or use multiple inductors, supplied to the load in Figure 10 is 11 µA. or are data-only links. Since the sensor nodes have to be of The transistors making up the ACR in Figure 9 need to a small form factor the data link required for e-textile BANs be sized to carry currents much larger than the output load must use the same inductors for power and data transfer in both current supplied, as shown in Figure 10. However, compared directions. A pulse-width modulated (PWM) OOK scheme is to a conventional rectifier where the same transistors are diode- used on the downlink that can be demodulated asynchronously, connected, the ACR supplies 2x higher output voltage than eliminating the need for clock recovery circuits at the sensor the conventional diode-connected rectifier in the low-input node. On the uplink, impedance modulation is used which amplitude regime of operation. This is shown in Figure 11. reduces the modulation circuitry to a switch across the inductor To generate the required dc output voltage, the four rectifiers at the sensor node. in Figure 9 are connected in series in their dc path while their ac inputs are capacitively-coupled to the LC tank as shown in A. Downlink Communication Figure 12, thus avoiding the need for additional voltage boost Binary coded messages to be transmitted to the sensor nodes converters. are pulse-width coded digitally using LOW and HIGH times in a 25:75 ratio for transmitting a “1” and a 75:25 ratio for IV. BI-DIRECTIONAL DATA COMMUNICATION TRX transmitting a “0”. This pulse-width-encoded signal is used The capability to transmit data in both directions is neces- to control a transmit on/off switch which generates the PWM- sary for the star network topology with the base station acting OOK waveform shown in Figure 13. In order to accommodate as the hub. The requirements for the links in both directions the startup/shutdown times of the oscillator, the data rate are highly asymmetric: the downlink from the base station to chosen is 80 kb/s, much lower than the local clock frequencies the sensor nodes needs to carry only small, low-duty cycle at the base station and sensor node. control instructions. On the other hand, the uplink from the Demodulation at the sensor node is done digitally using only sensor nodes to the base station needs to carry potentially the local 1 MHz clock. Two incrementers count the number large amounts of uncompressed data. An added constraint is of clock cycles for which the received waveform is 0 and the limited energy budget at the sensor node and the need to 1, as shown in Figure 14. Since each PWM bit ends with a push complexity to the base station. negative edge, the values of the counters can be compared Various schemes for transferring data to inductively-coupled against each other at the negative edge and the received bit sensors have been proposed in [22]–[25], but they use mod- can be latched. Fully digital demodulation avoids static power 6

PWMEnv

CLK

N1 x 0 1 2 3 0 1 0 1 2 3 0%

N0 x 1 1 2 3 1 1% x OUT

TPWM = 12!s TOOK

Fig. 15. Waveforms for downlink demodulator. The implemented downlink operates at T = 12.T . Fig. 13. Pulse-width modulated data transmitted on downlink using OOK PWM OOK (top) and impedance-modulated data with non-unity modulation index transmitted on the uplink (bottom), on a 27 MHz carrier. 1 DQ Demodulator 0 WT% Output R 4 4 DQ N1 PWMEnv 1 R To CDR

< < Demodulator 1 > > 0 OUT WT% Output 4 4 DQ N0 En Fig. 16. Asynchronous digital multiplexer to combine sub-network demodu- CLK R lator outputs. The output of the multiplexer goes to the Clock-Data Recovery (CDR) block. Negative Edge Trigger domain results in the received voltage signal being measured Fig. 14. Schematic of fully-digital digital demodulator at sensor node. across an impedance that progressively decreases as the number of sub-networks increases. Instead, the received signal from each sub-network is demodulated separately as shown dissipation, which is significant since downlink messages have in Figure 4, and multiplexed in the digital domain. As it a very low duty cycle. Additionally, the demodulator is self- comes before the clock recovery circuit in the receive chain, referenced, i.e. it does not need a reference for the compara- the digital multiplexer needs to be asynchronous. Figure 16 tor, the setting of which would require accurate calibration. shows the schematic of the asynchronous multiplexer. The Sample waveforms for the case TPWM = 4.TOOK are shown in output of each demodulator is “0” when there is no data Figure 15. being transmitted from that sub-network. The circuit shown in Figure 16 can easily be extended to more than two sub- B. Uplink Communication networks. Uplink data is transmitted by short- and open-circuiting the switch across the sensor node LC tank in Figure 5, which V. MEDIUM-ACCESS PROTOCOL changes the impedance reflected back to the base station An application-independent health-monitoring BAN needs inductor and can be sensed as voltage V1 in Figure 6. During to be able to support different kinds of sensor nodes that have uplink transmission, the sensor node primarily relies on stored different channel access and bandwidth requirements. These energy instead of rectifying the received ac power. The signal could range from ECG and single-channel EEG recorders received at the base station, shown in Figure 13, is an ASK- that need a time-averaged data rate of up to 10 kb/s, to modulated waveform with non-unity modulation index. Under temperature, blood pressure or blood oxygenation monitors the conditions outlined in Section III, the expressions for the that need infrequent access and a time-averaged data rate of modulation index of the received waveform m derived in [19] only a few bits/s. In order to accommodate such a variety of can be approximated as sensor nodes, a hybrid MAC protocol has been designed and integrated on silicon along with the power and communication k2Q Q R 1 2 s TRX circuits. The hybrid MAC uses both TDMA access and m ≈ 2 · (3) 1 + k Q1Q2 Rs + Q1ω0L1 CA modes for nodes requiring frequent and infrequent access The ASK modulation index at the base station evaluates to ap- respectively. It also ensures that all decision-making is pushed proximately 0.08 for the parameter values listed in Section III. to the base station, with the sensor nodes only following Demodulation is done differentially by extracting and ampli- instructions sent to them. fying both the positive and negative envelopes of the incoming Due to the architecture of the network, messages on the signal shown in Figure 13. Messages arriving from different downlink are broadcast to all sensor nodes in a sub-network. sub-networks need to be combined to have a single clock Downlink messages are 31b long, and contain a 4b header recovery circuit and digital back-end that handles data integrity that classifies the type of the message. The format of the and network access. Combining the impedance modulated following 27 bits varies based on the 4b header. For example, ASK waveforms from the different sub-networks in the analog it could contain the addressed sensor node’s network ID in 7

Bit Index 100 0 127 135 148 L L+16 128 bytes PLL Lock 7b SN 64 bytes Preamble Segment Segment Seq. ID 90 32 bytes 16 bytes Byte Count 16b CRC 80

70 Fig. 17. Uplink (sensor node to base station) packet structure. 60

Network Start TDMA Mode 50 Configuration m = 0 Normalized Uplink Data Rate (%) n = 0 40 TDMA request −6 −5 −4 −3 m++ 10 10 10 10 Yes to SN # m Channel Bit Error Rate BS sends beacon to n++ subnet n Error Yes Response Fig. 19. Attainable uplink data rate vs channel bit error rate for different Yes ? ? payload segment sizes. No Yes n < No Timeout? No N m < Yes M No No network ID and classiﬁes the sensor as a TDMA node or CA SN responds CA Mode with ACK BS broadcasts node. CA beacon

Yes SN classified, B. TDM Access Phase No Yes added to list Response Error ? ? No Once the network is configured, it enters the TDMA phase. BS sends ACK During this, the base station polls through each TDMA sensor node one by one and asks them to transmit their buffered data. Fig. 18. Flowchart of implemented medium-access protocol. The network The total duration of this phase is limited by the number of has N sub-networks with M TDMA sensor nodes combined. TDMA nodes multiplied with the maximum network access time given to each node, which corresponds to 4 KB data a TDMA data request. The uplink packet structure for all transferred at 1 Mb/s. packets after network configuration is shown in Figure 17. The uplink packet header contains an RX PLL lock sequence, C. Contention Access Phase followed by a preamble and packet metadata. The payload After completion of the TDMA phase, the network moves data is divided into variable-sized segments, each appended to the CA phase. The start of CA is denoted by a beacon with a 16b CRC checksum for error detection, and prevents broadcast by the base station to all sub-networks. Since uplink re-transmission of the entire packet in case of burst-mode data must cross two inductive hops to be heard by other channel noise. Smaller segment sizes increase MAC overhead sensor nodes, a CSMA-based protocol cannot be used for time but reduce the number of re-transmissions required. A plot multiplexing. Instead, CA nodes with data to transmit back of the attainable uplink data rate vs channel bit error rate for off in time by a pseudorandom 16b value upon receiving the different possible segment sizes (Figure 19) demonstrates this. beacon. The end of the 4 second period (1 TDMA phase + 1 Based on the prevailing error rate determined by the number CA phase) is again broadcast to all sub-networks and the base of re-transmissions required, the base station decides whether station goes back to looping through all the TDMA sensors. the segment sizes in succeeding packets needs to increase, decrease or remain unchanged. The network starts off with VI.MEASUREMENT RESULTS the configuration phase and subsequently moves on to alternate between TDMA and CA access. The sum of a TDMA and CA The base station and the NNI for the e-textiles network period is limited to 4 seconds by the requirement to service have been fabricated in a 0.18 µm CMOS process. Die the sensor nodes with the highest data acquisition rates at least photographs are shown in Figure 20. A brief summary of the during that period. A flowchart of the MAC protocol is shown design characteristics are presented in Table I and performance in Figure 18. comparisons with other recently reported solutions in the same application space are presented in Table II. The power consumption of the base station during A. Network Configuration transmit/receive-mode is 2.9 mW, which is comparable to the Upon startup, the base station automatically allocates unique narrowband RF- and BCC-based systems in Table II. The identifiers to all sensor nodes connected to the network. The energy per bit on the uplink is more important since the base station starts out by sending a configuration beacon to downlink only carries small, low duty-cycle packets. At 1 each sub-network port one at a time. Upon receiving the Mb/s, the base station expends 2.9 nJ per uplink bit. The 14 configuration beacon, each sensor node backs off in time by a µW maximum power consumption by the communication and 16b pseudorandom value and responds with details about its digital baseband circuits at 1 Mb/s translates to the sensor network access requirements. The base station then assigns a node’s expended energy per uplink bit approaching the pJ/bit 8

2.1mm

1.8mm 2.1mm 1.6mm

(a) (b) Fig. 20. Die photographs of (a) base station and (b) Sensor NNI.

TABLE I 3 SUMMARY OF E-TEXTILESNETWORKCHARACTERISTICS 1.4 DC−to−DC Eff. Rectifier V 2.5 Sensor Node 1.2 out Base Station Network Interface 1 2 Technology 0.18 µm CMOS (V)

2 2 (%) 0.8 Area 2.95 mm 1.46 mm 1.5 dc−dc 2.0 V (PA only) 0.6 Rect,out η

VDD,min 1.2 V V 1.4 V (others) 1 0.4 fCLK 1 MHz 1 MHz 0.5 2.7 mW (transmitter) 0.2 Power Consumption 14 µW 158 µW (others) 0 0 27.12 MHz ISM 1 10 100 Frequency Band Load Current (µA) Impedance modulation (uplink, 80 kb/s) Modulation Scheme PWM-OOK (downlink, 1 Mb/s) Fig. 21. Efﬁciency (dc-to-dc) of power transfer across resonant inductive link Max. Sensor Count 128 and rectiﬁer output voltage for varying load current. The separation between the transmitter and receiver coils in this case was 5 mm and the transmitter Hybrid TDMA & CA Multi Access PA was at its lowest power setting with 2 V supply.

range seen in fully wired systems [11], [17]. The higher ignore the effects of the source resistance Rs in Figure 6. Of power consumption at the base station, where energy is more a maximum 34 µW power recovered at the sensor node, a readily available, trades off against greater comfort due to maximum of 14 µW is consumed by the communication and the wireless/wired hybrid architecture when compared against digital baseband circuits, leaving the remaining 20 µW for fully wired systems. the biomedical signal acquisition front-end which is sufficient Figure 21 shows the efficiency of power transfer and the to power state-of-the-art biomedical data-acquisition circuits voltage at the output of the sensor node’s rectifier for varying [2]–[4]. A plot of the measured efficiency and output power load currents. As the load current increases beyond 20 µA, normalized to its maximum value with varying misalignment the rectifier operation is limited by the series ac-coupling is presented in Figure 22. The base station increases trans- capacitor reactances and the output voltage falls rapidly. At mitted power with increasing misalignment and is capable of its lowest power setting, the power transmitter on the base supplying the node’s minimum 14 µW power requirement for station supplying one inductor consumes 2.7 mW with the misalignments up to half the radius of the primary inductor. PA operating from a 2 V supply. Owing to the weak inductive OOK and envelope-detected waveforms for a downlink link, the power drawn by the transmitter varies negligibly with PWM message are shown in Figure 23. The waveforms changes in the load current of the rectifier. The received power correspond to the initial beacon transmitted by the base station peaks at 20 µA load current and the peak dc-to-dc power asking sensor nodes to join the network. Waveforms pertaining transfer efficiency across the link is 1.2% with voltage transfer to the packet metadata from an uplink packet with simulated function of 0.29, which are smaller by approximately 25% than ECG data are shown in Figure 24. The measured modulation the values calculated in Section III since Equations (1) and (2) index of 0.11 agrees with the value calculated theoretically 9

TABLE II COMPARISON OF BODY-AREANETWORKCOMMUNICATIONSYSTEMS

Wong et al.a Roh et al.a Yoo et al. Lee et al. Mercier et al. This Work [6] [7] [12] [17] [11] Type Narrowband RF BCC Wired+Wireless Wired e-textiles Wired e-textiles Wired+Wireless e-textiles e-textiles Coverage Wide Local Local Local Wide Wide Network Topol- Star Master-slave (X,Y) grid Bus Star Star ogy Base Station 2.3 mW b 3.2 mW b 5.2 mW 75 µW 180 µW 2.9 mW Power Sensor Node 2.7 mW b 1.6 mW b 12 µW 25 µW 20 µW 14 µW c Power Data Rate 50 kb/s 1.25 Mb/s 120 kb/s 20 Mb/s 10 Mb/s 1 Mb/s (uplink), 80 kb/s (downlink) Power Transfer - - Inductive Wired Wired Resonant induc- Mechanism tive Power Transfer - - 0.2% d - 96% 1.2% Efﬁciency a No remote power delivery to sensor nodes b Uplink communication power only c Power consumption of communication circuits and digital baseband circuits only d Based on power consumption values of network controller and supported sensor node

0.8 1.4 DC−to−DC Eff. 1.2 Output Power 30 0.4

1 Voltage (V) 0 W) 0.8 20 µ ( 0 20 40 60 80 100 120 140 160 180 200 (%) Network Configuration Beacon dc−dc 0.6 η Rect,out P 0.4 10

0.2

0 0 0 5 10 15 Center−to−Center Coil Misalignment (mm) 0 20 40 60 80 100 120 140 160 180 200 Time (µs) Fig. 22. Peak efficiency (dc-to-dc) of power transfer and peak output power for varying misalignment, measured as the center-center distance between the Fig. 23. Measured PWM-OOK and envelope-detected downlink waveforms coils. The diameter of the primary inductor on clothing is 32 mm. at sensor node. The packet corresponds shown corresponds to the network configuration beacon transmitted by the base station upon startup. in Equation (3). The uplink message shown consists of an initial sequence of alternating 1s and 0s for clock recovery, at the end of 2 seconds. followed by the random number ACK described in Section V. The trailing bits of the message consist of the CRC checksum VII.CONCLUSION that ensures data integrity. In order to address the problems of high path loss that BANs Figure 25 shows the network self-configuring upon startup. using radios and BCC face, a bi-directional communication The example network consists of two sub-networks, each with TRX and wireless power delivery system based on e-textiles one and two sensor nodes respectively. The base station sends has been presented. The proposed e-textiles BAN has been the beacon shown in Figure 23 to each sub-network one at a designed across multiple layers to support a wide variety of time. Sensor nodes respond with messages similar to the one sensors on the body. By delivering power wirelessly to state- shown in Figure 24 after a random delay to avoid collisions.3 of-the-art microwatt-level health monitors through resonant After waiting for 1 second for all sensor nodes in the sub- coupling, the problem of long-term energy storage can be network to respond, the base station moves on to the next sub- addressed and their form factors can be reduced. Implemented network. In Figure 25, both sub-networks are fully configured in 0.18 µm CMOS, the base station, which acts as the hub in 3Only the random number ACK changes between the responses sent by a star-connected network, consumes 2.9 mW per inductor it different sensor nodes. supplies power to. The network architecture has been chosen 10

2 REFERENCES 1.5 [1] T. Norgall, R. Schmidt, and T. von der Grun,¨ “Body area network 1 - A key infrastructure element for patient-centered telemedicine,” in Wearable eHealth Systems for Personalised Health Management: State 0.5 Voltage (V) of the Art and Future Challenges, ser. Studies in Health Technology and 0 Informatics. IOS Press, 2004, pp. 142–148. 0 10 20 30 40 50 60 70 80 [2] M. Yip, J. L. Bohorquez, and A. P. Chandrakasan, “A 0.6V 2.9µW PLL Lock Random Dig. Symp. VLSI Sequence Preamble Number ACK mixed-signal front-end for ECG monitoring,” in Circuits, Jun. 2012, pp. 66–67. [3] J. Yoo, L. Yan, D. El-Damak, M. A. Bin Altaf, A. H. Shoeb, and A. P. Chandrakasan, “An 8-channel scalable EEG acquisition SoC with patient-specific seizure classification and recording processor,” IEEE J. Solid-State Circuits, vol. 48, no. 1, pp. 214–228, Jan. 2013. [4] N. Verma, A. Shoeb, J. Bohorquez, J. Dawson, J. Guttag, and A. P. Chandrakasan, “A micro-power EEG acquisition SoC with integrated 0 10 20 30 40 50 60 70 80 feature extraction processor for a chronic seizure detection system,” Time (µs) IEEE J. Solid-State Circuits, vol. 45, no. 4, pp. 804–816, Apr. 2010. [5] P. P. Mercier, D. C. Daly, and A. P. Chandrakasan, “An energy-efficient Fig. 24. Measured waveforms for uplink packet header at input and output all-digital UWB transmitter employing dual capacitively-coupled pulse- IEEE J. Solid-State Circuits of base station demodulator. shaping drivers,” , vol. 44, no. 6, pp. 1679– 1688, Jun. 2009. [6] A. C. W. Wong, G. Kathiresan, C. K. T. Chan, O. Eljamaly, O. Omeni, D. McDonagh, A. J. Burdett, and C. Toumazou, “A 1 V wireless transceiver for an ultra-low-power SoC for biotelemetry applications,” IEEE J. Solid-State Circuits, vol. 43, no. 7, pp. 1511–1521, Jul. 2008. [7] T. Roh, S. Hong, H. Cho, and H.-J. Yoo, “A 259.6µW nonlinear HRV-EEG chaos processor with body channel communication interface for mental health monitoring,” in IEEE Int. Solid-State Circuits Conf. (ISSCC) Dig. Tech. Papers, Feb. 2012, pp. 294–296. [8] J. Bae, K. Song, H. Lee, H. Cho, L. Yan, and H.-J. Yoo, “A 0.24nJ/b wireless body-area-network transceiver with scalable double-FSK modulation,” in IEEE Int. Solid-State Circuits Conf. (ISSCC) Dig. Tech. Papers, Feb. 2011, pp. 34–36. [9] S. J. Song, N. Cho, and H.-J. Yoo, “A 0.2-mW 2-Mb/s digital transceiver based on wideband signaling for human body communications,” IEEE J. Solid-State Circuits, vol. 42, no. 9, pp. 2021–2033, Sep. 2007. [10] C. Gopalsamy, S. Park, R. Rajamanickam, and S. Jayaraman, “The Wearable MotherboardTM: The first generation of adaptive and respon- sive textile structures (ARTS) for medical applications,” Virtual Reality, vol. 4, no. 3, pp. 152–168, 1999. [11] P. P. Mercier and A. P. Chandrakasan, “A supply-rail-coupled eTex- tiles transceiver for body-area networks,” IEEE J. Solid-State Circuits, Fig. 25. Self-configuration of e-textiles BAN with two sub-networks. The vol. 46, no. 6, pp. 1284–1295, Jun. 2011. bottom three traces track a bit that goes HIGH once the sensor node is [12] J. Yoo, L. Yan, S. Lee, Y. Kim, and H.-J. Yoo, “A 5.2 mW self-configured configured. wearable body sensor network controller and a 12µW wirelessly powered sensor for a continuous health monitoring system,” IEEE J. Solid- State Circuits, vol. 45, no. 1, pp. 178–188, Jan. 2010. to offer a good trade-off between flexibility in choosing any [13] N. V. Desai, J. Yoo, and A. P. Chandrakasan, “A scalable 2.9mW 1Mb/s eTextiles body area network transceiver with remotely powered sensors subset of inductors to power while reducing complexity on the and bi-directional data communication,” in IEEE Int. Solid-State Circuits clothing. A NNI that recovers power transmitted by the base Conf. (ISSCC) Dig. Tech. Papers, Feb. 2013, pp. 206–207. station and handles data modulation and demodulation has also [14] IEEE Standard for Local and metropolitan area networks: Low-Rate Wireless Personal Area Networks (LR-WPANs), IEEE Std. 802.15.4, been implemented in 0.18 µm CMOS. It consumes 14 µW and 2011. harvests a maximum of 34 µW, leaving the remaining 20 µW [15] O. Omeni, A. Wong, A. J. Burdett, and C. Toumazou, “Energy effi- for the biomedical front-end. Modulation schemes have been cient medium access protocol for wireless medical body area sensor networks,” IEEE Trans. Biomed. Circuits Syst., vol. 2, no. 4, pp. 251– chosen keeping in mind the asymmetric nature of resource 259, Dec. 2008. requirements for data flow in both directions, and such that [16] Y. Kim, H. Kim, and H.-J. Yoo, “Electrical characterization of screen- complexity is pushed to the base station as much as possible. printed circuits on the fabric,” IEEE Trans. Adv. Packag., vol. 33, no. 1, pp. 196–205, Feb. 2010. Finally, logic implementing a hybrid MAC protocol has also [17] S. Lee, L. Yan, T. Roh, S. Hong, and H.-J. Yoo, “A 75µW real-time been integrated on chip, with all decisions being taken by the scalable network controller and a 25µW ExG sensor IC for compact base station and the sensor nodes acting as slaves. The hybrid sleep-monitoring applications,” in IEEE Int. Solid-State Circuits Conf. (ISSCC) Dig. Tech. Papers, Feb. 2011, pp. 36–38. MAC uses different network access schemes for sensor nodes [18] M. W. Baker and R. Sarpeshkar, “Feedback analysis and design of with different access and average data transfer rates. RF power links for low-power bionic systems,” IEEE Trans. Biomed. Circuits Syst., vol. 1, no. 1, pp. 28–38, Mar. 2007. [19] N. V. Desai, “A low-power, reconfigurable fabric body area network ACKNOWLEDGMENT for healthcare applications,” Master’s thesis, Massachusetts Institute of Technology, Cambridge, MA, 2012. The authors would like to thank members of the Semicon- [20] W. A. Tabisz, F. C. Lee, and D. Y. Chen, “A MOSFET resonant synchronous rectifier for high-frequency DC/DC converters,” in Proceedings ductor System Lab at KAIST, Daejeon, Korea for assistance of the IEEE Power Electronics Specialists Conference (PESC), Oct. with e-textiles fabrication and MOSIS for fabrication support. 1990, pp. 769–779. 11

[21] S. Mandal and R. Sarpeshkar, “Low-power CMOS rectifier design for Anantha P. Chandrakasan (M’95-SM’01-F’04) re- RFID applications,” IEEE Trans. Circuits Syst. I, vol. 54, no. 6, pp. ceived the B.S, M.S. and Ph.D. degrees in Electri- 1177–1188, Jun. 2007. cal Engineering and Computer Sciences from the [22] M. Ghovanloo and K. Najafi, “A wideband frequency-shift keying University of California, Berkeley, in 1989, 1990, wireless link for inductively powered biomedical implants,” IEEE Trans. and 1994 respectively. Since September 1994, he has Circuits Syst. I, vol. 51, no. 12, pp. 2374–2383, Dec. 2004. been with the Massachusetts Institute of Technology, [23] M. Ghovanloo and S. Atluri, “A wide-band power-efficient inductive Cambridge, where he is currently the Joseph F. and wireless link for implantable microelectronic devices using multiple Nancy P. Keithley Professor of Electrical Engineer- carriers,” IEEE Trans. Circuits Syst. I, vol. 54, no. 10, pp. 2211–2221, ing. Oct. 2007. He was a co-recipient of several awards including [24] S. Mandal and R. Sarpeshkar, “Power-efficient impedance-modulation the 1993 IEEE Communications Society’s Best Tu- wireless data links for biomedical implants,” IEEE Trans. Biomed. torial Paper Award, the IEEE Electron Devices Society’s 1997 Paul Rappaport Circuits Syst., vol. 2, no. 4, pp. 301–315, Dec. 2008. Award for the Best Paper in an EDS publication during 1997, the 1999 DAC [25] G. Simard, M. Sawan, and D. Massicotte, “High-speed OQPSK and Design Contest Award, the 2004 DAC/ISSCC Student Design Contest Award, efficient power transfer through inductive link for biomedical implants,” the 2007 ISSCC Beatrice Winner Award for Editorial Excellence and the IEEE Trans. Biomed. Circuits Syst., vol. 4, no. 3, pp. 192–200, Jun. ISSCC Jack Kilby Award for Outstanding Student Paper (2007, 2008, 2009). 2010. He received the 2009 Semiconductor Industry Association (SIA) University Researcher Award. He is the recipient of the 2013 IEEE Donald O. Pederson Award in Solid-State Circuits. His research interests include micro-power digital and mixed-signal integrated circuit design, wireless microsensor system design, portable multimedia devices, energy efficient radios and emerging technologies. He is a co-author of Low Power Digital CMOS Design (Kluwer Academic Publishers, 1995), Digital Integrated Circuits (Pearson Prentice-Hall, 2003, 2nd edition), and Sub-threshold Design for Ultra-Low Power Systems (Springer 2006). He is also a co-editor of Low Power CMOS Design (IEEE Press, 1998), Design of Nachiket Desai (S’10) received the Bachelor of High-Performance Microprocessor Circuits (IEEE Press, 2000), and Leakage Technology (B. Tech.) degree in electronics & elec- in Nanometer CMOS Technologies (Springer, 2005). trical eommunication engineering from the Indian He has served as a technical program co-chair for the 1997 International Institute of Technology, Kharagpur, India in 2010 Symposium on Low Power Electronics and Design (ISLPED), VLSI Design and the S. M. degree in electrical engineering and ’98, and the 1998 IEEE Workshop on Signal Processing Systems. He computer science from the Massachusetts Institute of was the Signal Processing Sub-committee Chair for ISSCC 1999-2001, the Technology, Cambridge, MA, USA in 2012, where Program Vice-Chair for ISSCC 2002, the Program Chair for ISSCC 2003, he is currently working toward the Ph. D. degree. the Technology Directions Sub-committee Chair for ISSCC 2004-2009, and Between June 2012 and August 2012, he was the Conference Chair for ISSCC 2010-2014. He is the Conference Chair for at Texas Instruments, Dallas, TX, USA, designing ISSCC 2015. He was an Associate Editor for the IEEE Journal of Solid-State power converters for energy harvesting applications. Circuits from 1998 to 2001. He served on SSCS AdCom from 2000 to 2007 His research interests include wireless power transfer for consumer devices and he was the meetings committee chair from 2004 to 2007. He was the and personal-area networks (PANs), circuits for near-field communication Director of the MIT Microsystems Technology Laboratories from 2006 to (NFC), and power converters. 2011. Since July 2011, he is the Head of the MIT EECS Department.

Jerald Yoo (S05-M10) received the B.S., M.S., and Ph.D. degrees in electrical engineering from the Korea Advanced Institute of Science and Technology (KAIST), Daejeon, Korea, in 2002, 2007, and 2010, respectively. In May 2010, he joined the faculty of Microsys- tems Engineering, Masdar Institute of Science and Technology, Abu Dhabi, United Arab Emirates, where he is currently an associate professor. During June 2010 to June 2011, he was also with Mi- crosystems Technology Laboratory, Massachusetts Institute of Technology (MIT), Cambridge, MA, USA, as a visiting scholar. He developed low-energy Body Area Network (BAN) transceivers and wearable body sensor network using Planar-Fashionable Circuit Board (P-FCB) for continuous health monitoring system. His research focuses on low energy circuit technology for wearable bio signal sensors, wireless power transmission, SoC design to system realization for wearable healthcare applications, and energy-efﬁcient biomedical circuits. He is an author of the book chapter in Biomedical CMOS ICs (Springer, 2010). Dr. Yoo is a co-recipient of the Asian Solid-State Circuits Conference (A- SSCC) Outstanding Design Awards in 2005. Currently he serves as a technical program committee member of Asian Solid-State Circuits Conference (A- SSCC).