1296

IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 46, NO. 6, JUNE 2011

A 10.2 Mbps Pulse Harmonic Modulation Based Transceiver for Implantable Medical Devices
Farzad Inanlou, Student Member, IEEE, Mehdi Kiani, Student Member, IEEE, and Maysam Ghovanloo, Senior Member, IEEE

AbstractA low power wireless transceiver has been presented for near-eld data transmission across inductive telemetry links, which operates based on pulse harmonic modulation (PHM). This PHM transceiver uses on-off keying (OOK) of a pattern of pulses to suppress inter-symbol interference (ISI), and its characteristics are suitable for low-power high-bandwidth telemetry in implantable neuroprosthetic devices. To transmit each bit across a pair of high-Q LC-tank circuits, the PHM transmitter generates a string of narrow pulses with specic amplitudes and timing. Each pulse generates a decaying oscillation at the harmonic frequency that the receiver LC-tank is tuned at, which is then superimposed with other oscillations across the receiver at the same frequency, to minimize the ISI. This allows for reaching high data rates without reducing the inductive link quality factor (to extend its bandwidth), which signicantly improves the range and selectivity of the link. The PHM receiver architecture is based on non-coherent energy detection with programmable bandwidth and adjustable gain. The PHM transceiver was fabricated in a 0.5- m standard CMOS process, occupying 1.8 mm . The transceiver achieved a measured 10.2 Mbps data rate with a bit error rate at 1 cm distance using planar implant sized (BER) of (1 cm ) gure-8 coils. The PHM transmitter power consumption was 345 pJ/bit and 8.85 pJ/bit at 1 cm and zero link distances, respectively. The receiver dissipates 3 mW at 3.3 V supply voltage. Index TermsCMOS, implantable medical devices, impulse radio ultra wideband, inductive coupling, low power, near-eld, neuroprostheses, pulse harmonic modulation, telemetry, transceivers.

I. INTRODUCTION

RANSCUTANEOUS data telemetry is one of the most important functions in a group of implantable medical devices (IMDs), known as neuroprostheses, which substitute sensory or motor modalities that are lost due to an injury or a disease [1]. Well-known examples are the cochlear implants and visual prostheses, which need to transfer a large volume of data from external articial sensors to the IMD, or invasive

Manuscript received July 20, 2010; revised February 13, 2011; accepted March 03, 2011. Date of publication May 10, 2011; date of current version May 25, 2011. This paper was approved by Associate Editor Ranjit Gharpurey. This work was supported in part by the National Institutes of Health, NIBIB, grant 1R21EB009437-01A1, and the National Science Foundation under award ECCS-824199. F. Inanlou is with the School of Electrical and Computer Engineering, Georgia Institute of Technology, Atlanta, GA 30308 USA (e-mail: farzad.i@gatech.edu). M. Kiani and M. Ghovanloo are with the GT-Bionics Lab, School of Electrical and Computer Engineering, Georgia Institute of Technology, Atlanta, GA 30308 USA (e-mail: mgh@gatech.edu). Color versions of one or more of the gures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identier 10.1109/JSSC.2011.2134570

braincomputer interfaces (iBCI), which collect a massive amount of data from the central neural system (CNS) and transfer the data across the skin to the outside of the body to control the patients environment or prosthetic limbs after signal processing [2][7]. The challenges involved in designing transcutaneous data links relate to the extremely limited space and power available to the IMD for establishing a wideband and robust connection. Because of the signicant electromagnetic eld absorption in the tissue, which exponentially increases with the carrier frequency, high bandwidth must be achieved at the lowest possible carrier frequencies [8], [9]. This requirement rules out the majority of commercially available wideband wireless protocols, such as Bluetooth or WiFi, which operate well in the air at 2.4 GHz. On the other hand, there are specic standards, such as Medical Implant Communication Service (MICS), operating in the 402 405 MHz band, which can only offer a limited bandwidth (300 kHz) [10]. Therefore, there is a need for low power, robust, and wideband wireless links with novel modulation techniques that are specically tailored towards high performance IMD applications. Near-eld inductive coupling is the most common method that has been utilized for establishing wideband data telemetry links with IMDs. In these links, robustness can be measured by bit error rate (BER) in the presence of various sources of external interference, supply ripple, load changes, digital switching noise, and coupling variations due to mechanical vibrations and coils misalignments [9]. It should be noted that similar applications of the near-eld data transmission can be found in radio-frequency identication (RFID), contactless smartcards, and high throughput wireless sensors, where batteries are avoided due to extreme size, cost, and lifetime constraints [11], [12]. The majority of modulation techniques that have been used in near-eld inductive links for IMD applications modify a sinusoidal carrier signal based on the data to be transferred across the link. Amplitude shift-keying (ASK), frequency shift keying (FSK), binary/quadrature phase shift keying (BPSK/QPSK), and load shift keying (LSK), leading to ASK, are examples of such methods [9], [13][17]. The use of carrier signals for data transmission was attractive in the early IMDs because the same low frequency data carrier could be used for transferring power to the IMD. However, the power carrier has to be separated from data carrier in the modern high performance IMDs, in which much wider bandwidth is required for data while the power carrier frequency cannot be increased due to excessive

0018-9200/$26.00 2011 IEEE

INANLOU et al.: A 10.2 Mbps PULSE HARMONIC MODULATION BASED TRANSCEIVER FOR IMPLANTABLE MEDICAL DEVICES

1297

loss in the tissue at higher carrier frequencies [18]. On the other hand, the use of high frequency data carriers for robust wideband communication with IMDs require complex and power consuming frequency stabilization RF circuits, such as phase-locked loops (PLL), which is not desired. This has motivated us to devise a new data transmission technique for near-eld wideband applications, called pulse harmonic modulation (PHM), which has similarities to the impulse radio ultra-wideband (IR-UWB) in the far-eld domain [19], [20]. It should be noted that there are other pulse-based near-eld data transmission methods, developed for chip-to-chip communication and body area networks [21], [22]. However, they require an inductive link with a low quality factor, , to achieve wide bandwidth, which is not suitable for the IMD applications, where higher transmission distance and better selectivity are needed. The PHM transmitter generates a string of narrow pulses with specic amplitudes and timing to transmit each bit across a pair of high-Q LC-tank circuits. Each pulse generates a decaying oscillatory response at the harmonic frequency that the receiver LC-tank is tuned at, which adds to the oscillations at the same frequency that are resulted from subsequent pulses within that bit period to minimize the inter-symbol interference (ISI) at the receiver input. This allows reaching high data rates without reducing the inductive link quality factor by adding dissipative elements to extend its bandwidth. This method also improves the signal-to-noise ratio (SNR) at the receiver, leading to improvements in the BER, transmission range, selectivity, and robustness against misalignments of the link [19]. We recently described the PHM concept and its theoretical analysis in [19] along with measurements from a discrete proof-of-concept prototype. In this paper we present a fully integrated CMOS transceiver based on the PHM technique, which block diagram is shown in Fig. 1. On the transmitter side is charged up to a voltage set by a digital-to-analog (Tx), is charged up to . is then disconverter (DAC) and charged into the primary coil followed by via an LC-Driver circuit according to a specic timing that is dictated by an FPGA that accepts the serial data bit stream to be transmitted due (Tx-Data).The sharp current impulses passing through to discharges, couple onto the secondary tank, resonance and initiate/terminate an oscillation at the frequency, , across the receiver (Rx) input. Inside Rx, which operates based on a non-coherent energy detection (ncED) scheme, the received signal is amplied, squared, and low-pass ltered, and nally, a comparator recovers the serial data bit stream. A brief overview of inductive links for data telemetry as it applies to the PHM is included in Section II, followed by detailed PHM transceiver circuit

Fig. 1. Block diagram of the pulse harmonic modulation based transceiver. represents the parasitic capacitance of and switches at the output of the LC Driver.

descriptions in Section III. Section IV presents the transceiver characterization and measured results, followed by discussions in Section V. Finally, the concluding remarks are included in Section VI. II. PULSE HARMONIC MODULATION FOR NEAR-FIELD INDUCTIVE DATA TELEMETRY A. Inductive Link Impulse Response Fig. 2 shows a lumped equivalent circuit model of an inductransmitter) and secondary ( tive link. The primary ( receiver) sides are composed of LC-tank circuits and their associated parasitic components ( and ). tank is tuned at , while the tank, depending on the data rate and transmission range requirements, can be either tuned to or left at its self resonance frequency (SRF), in which case in Figs. 1 and 2 simply represents the parasitic capacitance of . Since and are loosely coupled (small ) and the current in tank is very small, unlike inductive power transmission links [23], we can safely neglect the effect of loading on the Tx circuitry to simplify our equations. Hence, the inductive link transfer function in the S-domain can be described as in equation (1), shown at the bottom of the page, where , , , and are the Tx output voltage, Rx input voltage, current passing through , and induced voltage across , respectively. Other parameters are lumped circuit elements in Fig. 2. is composed of two second-order systems, one originating from the primary and the other from the secondary LC-tanks, each of which can be expressed as

(2)

(1)

1298

IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 46, NO. 6, JUNE 2011

Fig. 2. Lumped model for near-eld data telemetry through coupled coils.

is the damping ratio, is the natural frequency, and is the natural damping frequency of the system. From (1) and (2), these second-order system parameters can be expressed in terms of the lumped circuit elements in Fig. 2,

where

robustness of the link against interference without increasing the transmitted power. and can be calculated from (5) and (6) by substituting them in , which indicates that should be small and thus, . As a result, in (12) can be simplied into two exponentially decaying oscillations, one with a long time-constant of on the secondary and the other with a short time-constant of on the primary. The sum of these two terms will result in an oscillation across in Fig. 2, which builds up rapidly but decays slowly [19]. The envelope of can be expressed as (13) B. Pulse Harmonic Modulation Using the inductive link impulse response from Section II-A, for any arbitrary input waveform by conone can calculate volving it with . However, if we only apply narrow pulses across , with pulse width and amplitude to reduce Tx power consumption, then we can expect to appear across [19]. On the other hand, in high-speed pulse-based data transmission, it is important for the received signal (i.e., the impulse response), resulted from transmitting a single bit, to rise sharply (small ) to enhance detection and decline rapidly (small ) to minimize ISI with the following bit, and reduce BER. Both of these requirements suggest that and should be lowered to improve bandwidth, which is the method that has been adopted in previous carrierless near-eld data links [21], [22]. However, this is not a suitable choice for robust transcutaneous communications. Considering that has a sharp rise time and a slow decay, in PHM we generate a rapid oscillatory response across by applying a rst pulse, called the initiation pulse, across at the onset of every digital bit 1 . Then we transmit a second pulse (or pattern of pulses), called the suppression pulse(s), before the end of the bit period, , to create a second oscillation. We choose the amplitude, , and delay, , of the suppression pulse such that the second oscillation has the same amplitude as the rst one by the time that it appears across , but 180 out of phase. The result is the cancellation of both oscillatory responses after the arrival of the suppression pulse, leading to a rapid decline in the ISI, which is the amplitude of the oscillation across by the time that a new bit is due to begin. In this implementation, for a digital bit 0, we simply do not transmit any pulses [19]. In order to minimize BER, we should rst allow the oscillation resulted from the initiation pulse to build up. According to (13), this can take . Hence, to reduce the buildup time, according to (5), and should be as small as possible.In fact the parasitic capacitance of is often sufcient to form . Then we should apply the suppression pulse at an odd multiple of a half cycle to ensure that the two responses are going to be out of phase (14) is the number of cycles that where before it is suppressed. is allowed to have

(3) (4) (5) (6) Assuming both second order systems are under damped, i.e., and , which is often the case for the LC-tanks used in data telemetry links, (1) can be rearranged as

(7) Now we can break ponents up into the sum of its rst order com-

(8) where

(9)

(10) and apply the inverse Laplace transform to nd the impulse response for the inductive link (11)

(12) High-Q is desired for the inductive data transmission links used in IMD applications to improve transmission range and

INANLOU et al.: A 10.2 Mbps PULSE HARMONIC MODULATION BASED TRANSCEIVER FOR IMPLANTABLE MEDICAL DEVICES

1299

Fig. 3. (a) Schematic diagram of the LC-Driver block in Fig. 1(b) LC-Driver timing diagram ( is the signal controlling the MUX in the PHM Tx in Fig. 1, to either or the comparator output). which connects

For a perfect cancellation, when choosing the suppression pulse amplitude, , the decay of from the initiation pulse at should be considered. The amplitude of can be found from its envelope in (13). Since , the rst term, which represents the buildup period, can be neglected and the ratio between the suppression and initiation pulses can be found from (15) Using (2)(15), and can be easily calculated from the inductive link lumped circuit values and other parameters, such as the desired data rate and pulse width , to design a PHM-based near-eld data telemetry link. In practice, perfect cancellation between initiation and suppression pulses is not possible due to timing jitter, parasitic components, and process variations [19]. Nonetheless, as shown in Section IV, PHM is an effective method for reducing the ISI and achieving wide bandwidth at low BER without lowering the link quality factor or consuming too much power. III. PULSE HARMONIC MODULATION BASED TRANSCEIVER A. Transmitter Design The PHM transmitter, which is enclosed in the upper dashed box in Fig. 1, operates by charging a pair of capacitors, and , up to and , respectively, and discharging them into to generate sharp initiation and suppression pulses. is provided by a 5-bit DAC, which controls the voltage at the lower terminal of . The DAC was designed based on folded resistor string architecture using 50 k resistors [24]. Fig. 3(a) shows a more detailed schematic of the LC-Driver block in Fig. 1. This circuit, which is controlled by an off-chip FPGA through and signals, can operate either in singleended or differential modes, which switching diagram can be seen in Fig. 3(b). In the single-ended mode, 0 and the right node of is always connected to by . In order to generate the

initiation pulse, in Fig. 1 is set to 0, causing the multiplexer in Fig. 1 (MUX) to connect to , to turn on and charge to full scale, i.e., . During this charging period 0 and gate is open. Therefore, no current passes through . At the onset of a bit 1 period, toggles to 1 for a very short time period, (only a few ns in this design), during which the inductor current, , starts increasing at a rate proportional to . This current variation in is responsible for inducing a voltage across , which is depicted as in Fig. 2. Of course there will also be voltage drops across and , depending on their sizes, which need to be considered in a more detailed analysis. To end the initiation pulse, is toggled back to 0, at which time shorts the left node of to and provides a path for to return back to zero without causing undesired oscillations. After terminating the initiation pulse, the FPGA applies the PHM time delay in (14), , and at the same time prepares the transmitter for generating the suppression pulse by setting 0, similar to the initiation pulse. The difference, however, is that this time, 1 , causing the MUX to connect to the output of a comparator, which compares and . The comparator output is initially high because . Therefore, is turned on and charges up to the point where . At this time, the comparator output goes low and turns off to maintain the voltage across at . The rest of the steps for generating the suppression pulse is similar to the initiation pulse, and includes toggling 1 for to connect in parallel with and induce across . Obviously the 5-bit DAC input, S[1:4], can be adjusted such that , which was calculated in (15). In the differential mode, the initiation pulse is generated by the right side of the LC-Driver in Fig. 3(a) by connecting the left node of to ( 0) and charging to full scale, i.e., , through . At the onset of a bit 1 period, toggles to 1 for , during which increases towards right. Then both and are lowered, during , to connect both ends of to and reduce its current. At the same time, is charged to and the suppression pulse is generated by the left side of the LC-Driver in the exact same way as in the single-ended mode. It should be noted that in this mode, since the suppression pulse is already out-of-phase with respect to the initiation pulse ( owing in opposite directions), unlike (14), should be an even multiple of a half cycle, or simply (16) of and should be small to charge and rapidly and meet the timing requirements of the desired data rate. In other words, and for and should be large. On the other hand, driving larger devices results in higher power consumption per bit. In this design we chose as a compromise between and power consumption requirements. Similar constraints should be considered in the design of , , , and , which are much larger devices responsible for driving , and as such should be designed considering the chip area constraints also. In this design, and

1300

IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 46, NO. 6, JUNE 2011

were selected based on specications and post-layout simulation results. It should also be noted that the parasitic capacitances added by these switches to and terminals are dwarfed by the parasitic capacitance of and values of and , which are off-chip components. B. Receiver Design The PHM signal at the Rx input, , is similar to the low-frequency equivalents of the IR-UWB on/off keying (OOK) signals, in which the presence or absence of a number of oscillations within a certain band represent bits 1 or 0, respectively [25], [26]. Among different architectures for such receivers, ncED is popular due to its low complexity [27][29]. This architecture does not require high power consuming blocks such as local oscillator (LO) or PLL for down-conversion either, which makes it suitable for low power applications. A recent comparison in terms of energy per bit and data rate between coherent and non-coherent UWB and narrowband receivers in [30] showed an order of magnitude lower power consumption in non-coherent receivers for data rates up to 10 Mbps. Moreover, applying this method to high-Q inductive links, which have high selectivity, can diminish some of the drawbacks of the ncED architecture, such as susceptibility to noise and interference. The block diagram of our ncED-based PHM Rx is shown in the lower dashed box in Fig. 1. The PHM signal across is amplied and squared before passing through a Chebyshev LPF for its energy to be compared with a reference value, which is determined by the noise oor and level of interference in the system. All blocks are designed fully differential to have better immunity against common mode interference, supply ripple, and noise. 1) LNA: The LNA block, shown in Fig. 4(a), which has common-source cascode architecture with untuned load, provides 27 dB gain at the input of the PHM Rx. In order to achieve data rates above 10 Mbps, should be tuned at a high frequency to provide enough oscillations per bit at the Rx input. We chose MHz for resonance frequency and 100 MHz for the LNA bandwidth. Resistive load was used to avoid off-chip inductors and capacitors, which are needed otherwise due to the low frequency of the RF signal. The highresonance of the -tank at the LNA input eliminates the need for any additional bandpass ltering. The LNA gain can be reduced to 13.5 dB by closing in Fig. 4(a) when the input RF signal is large due to strong coupling between and . In this case the LNA power consumption is also reduced in half by cutting the tail current to save power. 2) Passive Self-Mixer: The amplied RF signal is applied to a passive self-mixer, shown in Fig. 4(b), to be converted to a baseband signal proportional to its energy. The RF and LO ports of the mixer are tied together to implement a squaring function. Transistors in this circuit act as switches with the gate-source voltage close to the threshold voltage to provide the maximum possible gain. 3) Third-Order Chebyshev Low-Pass Filter: The squared signal at the output of the mixer passes through a third-order Chebyshev LPF with 0.5 dB ripple in the pass band, which is shown in Fig. 5(a) [31]. The normalized lter transfer

Fig. 4. Receiver front-end circuits. (a) Adjustable gain LNA. (b) Passive selfmixer circuit.

Fig. 5. (a) Block diagram of the third-order Chebyshev low-pass (b) Schematic diagram of the fully differential OTA with CMFB.

lter.

function, which has been realized by a bi-quad stage following a rst-order low-pass stage, can be dened by

(17) and determine the DC gain and the 3 dB cut-off where frequency rad/s . is the capacitance and dB is the transconductance of each operational transconductance amplier (OTA) stage. dB is adjustable by 3 bits, a [0:2], which change the value to of in (17). This method was preferred over changing keep the low frequency gain of the LPF constant regardless of dB . The OTA was implemented by a fully differential amplier, shown in Fig. 5(b). Cascode stages and source degeneration

INANLOU et al.: A 10.2 Mbps PULSE HARMONIC MODULATION BASED TRANSCEIVER FOR IMPLANTABLE MEDICAL DEVICES

1301

TABLE I THIRD-ORDER CHEBYSHEV LOW-PASS FILTER SPECIFICATIONS

*From post-layout simulations.

Fig. 7. Fully differential comparator circuit.

Fig. 8. PHM transceiver chip microphotograph.

Fig. 6. Schematic diagram of the variable gain amplier (VGA).

resistor were employed to increase the output impedance and linearity of the OTA, respectively. In this circuit, if transconductance of the input transistors, and , is sufciently large, i.e., , then , independent of the tail currents and input voltage swing. A common-mode feedback (CMFB) circuit was also added to set the common mode voltage at the OTA outputs to a predetermined value, . is used in the CMFB circuit to increase the stability of the CMFB loop by reducing its gain. Table I shows the specications of the Chebyshev LPF. 4) Variable Gain Amplier (VGA): In order to improve the sensitivity of the PHM Rx for different coupling distances, the Rx gain can be adjusted both in the RF front-end (1-bit coarse control) and baseband (3-bits ne control) stages by the LNA and VGA, respectively. A fully differential open-loop VGA providing 10 to 30 dB gain in 8 steps (G[0:2]) was implemented using the circuit shown in Fig. 6. In order to reduce power consumption, the VGA tail current was also changed with the gain control bits to obtain the necessary gain-bandwidth product. 5) Comparator: Fig. 7 shows the schematic of the fully differential high speed comparator, which has positive feedback at the input. This comparator has an adjustable reference voltage, , that sets the threshold at which the comparator switches. The voltage difference at the VGA output is compared to to generate the received Rx-Data bit stream. The effects of the common mode voltage variations at the comparator input are reduced because of the fully differential design of both VGA and comparator.

Fig. 9. PHM transceiver measurement setup. Inset: Inductive telemetry link in Fig. 1) on FR4 PCB, made of a pair of planar gure-8 coils [32] ( and on which the transceiver chips have been directly wire-bonded.

IV. MEASUREMENT RESULTS The PHM transceiver prototype, shown in Fig. 8, was fabricated in a 0.5- m 3M2P standard CMOS process, occupying 0.61 mm of chip area. Two chips were used in our experimental setup, shown in Fig. 9. Each chip was glued and wirebonded directly on a 2-layer custom designed PCB (1 oz copper on 1.5-mm-thick FR4) and covered with epoxy. Each PCB included a 2-turn planar gure-8 coil (see Fig. 9 inset), which specications are summarized in Table II [32]. The coils were held in parallel and perfectly aligned using only non-conducting materials, such as PVC and Plexiglas. At cm between the two coils and perfect alignment, they had a simulated value of [32]. In addition, based on

1302

IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 46, NO. 6, JUNE 2011

TABLE II FIGURE-8 COILS SPECIFICATIONS

*From HFSS nite element analysis simulations.

at MHz, the bandwidth of the inductive link can be calculated as MHz. In order to minimize the effect of interconnects and SMA connectors on the coils magnetic eld we added 3.4 cm and 4.25 cm clearance on the PCB between the Tx and Rx coils and their associated PHM transceivers, respectively. In practice, the data coil and transceiver microassembly will be packed inside an IMD with shorter interconnects and smaller parasitic effects, while the tissue volume conductor can add to the coils parasitic components and reduce their SRF [33]. Evaluation of the PHM transceiver performance in a more realistic setup, inside tissue volume conductor, is among our future research directions. Generation of the random serial data bit stream to be transmitted, Tx-Data, and BER analysis of the recovered data, Rx-Data, in Fig. 1 were conducted by a pair of Tektronix GB1400 (GigaBERT), shown in Fig. 9. Once the two GigaBERTs are synchronized via a separate hardwired clock signal, Tx-clk in Fig. 1, they can measure the wireless link BER in real time. Instead of the receiver block in the PHM transceiver chip, we initially used a custom-designed discrete RF receiver [19], in which we had easier access to the internal nodes, to observe and evaluate the effects of PHM on the received signal. We tried to avoid probing the inductive link directly, which could change its characteristics. In order to achieve Mbps, which was our target for this PHM transceiver prototype, the oscillation across must be initiated and suppressed within ns. If we dedicate complete cycles for the oscillation to build up after the initiation pulse, and three more cycles for the oscillation to die out after the suppression pulse, before the next bit arrives, then needs to be 60 MHz. It should also be pointed out that in a second order system with , such as in Fig. 2 (18) is the peak magnitude of the transfer function, where i.e., the second term in the denominator of (1). Therefore, increasing by reducing while maintaining constant can lead to higher and higher gain for the inductive link, both of which can lead to better SNR at the receiver input. This is a major advantage of the PHM over previous wideband carrierless data transmission methods, which can only operate when the link Q-factor is low [19]. High gain and Q-factor are key in inductive data links used in IMDs because the coils distance, , is in the cm range, orders of magnitude larger than applications such as chip-to-chip communication [21], and there might be a strong interference from a nearby inductive power carrier [18]. For the above reasons, we chose MHz. Using the coils specied in Table II and , the main PHM param-

Fig. 10. Transmitter waveforms at the LC-Driver input and receiver waveforms at the LNA and LPF outputs of the custom designed receiver, (a) with and (b) MHz, Mbps, , ns, without PHM, when , ns, and mm. Elimination of the suppression pulse in (b) renders the received data completely undetectable due to strong ISI.

eters were found to be ns and based on (14) and (15), respectively. In Table II, it can be seen that no tuning capacitor has been used for the primary coil to reduce Tx power consumption. We also noticed that at this , since s, the performance of the PHM transceiver is not very sensitive to , and the wireless link also works ne with . Therefore, the PHM Tx was set to operate in the single-ended mode in order to further reduce power consumption. Fig. 10(a) and (b) from top to bottom show one of the LC-Driver control signals, , which was applied to the PHM Tx as shown in Fig. 1, the received RF signal at the output of the LNA, which is proportional to at the input of the receiver, and the LPF output, which is proportional to the energy content of the received pulse around , with and without PHM, respectively. It can be seen that the elimination of the suppression pulse in Fig. 10(b) renders the received data completely undetectable at Mbps, while Fig. 10(a) clearly shows the effectiveness of the PHM in reducing the ISI. Considering and the model shown in Fig. 2, the channel loss in these measurements was dB. Fig. 11 shows a snapshot of the key measured waveforms in the PHM transceiver setup of Fig. 9. From top, the trans-

INANLOU et al.: A 10.2 Mbps PULSE HARMONIC MODULATION BASED TRANSCEIVER FOR IMPLANTABLE MEDICAL DEVICES

1303

Fig. 12. BER versus coils distance, , at different data rates, when PHM-Tx . supply voltage is constant at

TABLE III PHM-BASED TRANSCEIVER SPECIFICATIONS

Fig. 11. PHM transceiver waveforms from top: Transmitted serial data bit stream at 10.2 Mbps, one of the LC-Driver control signals, receiver VGA output, and recovered data bit stream, (a) with and (b) without PHM.It can be seen that while everything between these two measurements are the same, without PHM the Rx Data in (b) is completely erroneous.

mitted serial data bit stream at Mbps, of the LC-Driver, receiver VGA output, and recovered serial data bit stream are shown with the gure-8 coils aligned at mm. In Fig. 11(b), because of the strong ISI in the absence of the suppression pulse, the receiver has failed to recover the serial data bit stream at this data rate. On the other hand, utilizing the PHM technique in Fig. 11(a) has resulted in correct detection of the serial data bit stream in an otherwise similar condition. To better characterize the performance of the PHM-based transceiver, we measured the BER at various data rates, while changing from 5 to 10 mm. The results, shown in Fig. 12, indicate that if the acceptable BER limit is considered , then the best data rate that can be achieved with the current PHM transceiver at mm is 10.2 Mbps. The PHM Tx supply voltage, , in this case was at 3.3 V, consuming 3.52 mW, which is the equivalent of 345 pJ/bit. The PHM Rx power consumption in these conditions was 3 mW. Consuming the same amount of power at mm resulted in Mbps at BER . It should be noted that the receiver bandwidth limitation is the main reason for high BER at 11.4 Mbps and 12 Mbps. The cutoff frequency of the Chebyshev LPF in the Rx should be increased in order to achieve higher data rates. Also the signal to interference ratio decreases due to the ISI when the PHM pulses are too close to each other at high data rates, which

leads to more BER degradation. Without PHM, on the other hand, the data rate could only reach 1.5 Mbps at mm with BER . Table III summarizes the key specications of the PHM transceiver ASIC. Even though the present transceiver has not been optimized for minimum power consumption in this large feature sized and relatively high threshold voltage process, we tried to lower the PHM Tx power by reducing its from 3.3 V down to 2.4 V, while measuring the BER at mm. Fig. 13 depicts the results of this experiment, which demonstrates the strong relationship between BER, Tx power, and Rx sensitivity. In order to accurately measure the minimum required energy to transfer high speed data and compare the PHM transceiver performance with other recently published near-eld data telemetry links, we reduced down to 0.1 mm by separating the two coils with a sheet of paper and lowering down to 1.2 V. Since the FPGA was still operating at 3.3 V, the PHM-Tx control signals in Fig. 1 were also divided down to 1.2 V. Finally, was reduced from 7 ns to 3 ns using the FPGA gate delays [34]. The average energy consumption in this case was measured by differentially recording the voltage across a resistor in series with the PHM-Tx supply pin, and integrating it over a certain period of time as random data was being transferred across the link at 10.2 Mbps. The results showed that for a BER the PHM-Tx power consumption was only 8.85 pJ/bit, which can be further reduced by migrating the PHM transceiver to smaller feature sized processes.

1304

IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 46, NO. 6, JUNE 2011

TABLE IV BENCHMARKING OF RECENT INDUCTIVE DATA TELEMETRY LINKS

*Only the links with potential IMD application are listed. **Calculated from component values given in the paper. Load shift keying consumes low power for data transfer at the cost of reducing the delivered power to the IMD by up to 50%. Biphase modulation.

Fig. 14. Simulated Rx amplitude ( in Fig. 2) for both data and ISI versus in Fig. 1, while keeping all other paramprimary parasitic capacitance, i.e., eters constant according to Tables II and III. Fig. 13. BER versus transmitter power consumption at mm and Mbps, when of the PHM-Tx has been reduced from 3.3 V to 2.4 V.

V. DISCUSSION It can be concluded from the measurement results using the coils in Table II that matching the primary and secondary resonance frequencies, while desired, is not a precondition to reliable data recovery with the PHM technique. It was observed in measurements that tuning the primary coil at MHz increases the received PHM signal amplitude ( in Fig. 2) at the cost of higher power consumption in Tx. To better understand the effects of mismatch between the primary and secondary resonance frequencies on the ISI and received signal amplitude, which are important in the PHM performance, we swept in Fig. 1 in a simulation while keeping all other parameters constant according to Tables II and III. The result in Fig. 14 shows that sweeping in a wide range does not have a signicant effect on the ISI. The key to successful ISI cancellation in the PHM is proper adjustment of and to a lesser extent on the Tx based on the resonance frequency and time constant of the LC-tank on the Rx according to (14)(16). Therefore, whether PHM is being used for uplink (i.e., from inside to outside of the body) or downlink (from outside into the body), it is always possible to accurately adjust either (e.g., by using an array of tuning capacitors)

or (e.g., by a programming delay), respectively, whichever is accessible from outside of the body, to create optimal conditions for the PHM operation. If the same pair of LC tanks is to be used for both transmitting and receiving data in a half-duplex fashion, a closed loop scheme can be employed to adjust and based on the ISI or BER, in the downlink or uplink modes, respectively, both from outside of the body. Finally, it should be noted that for a complete stand-alone receiver, a clock and data recovery (CDR) scheme is required. In high performance IMDs, such as cochlear implants and visual prostheses, the power needed for the operation of the implanted device is transmitted via a separate inductive link, responsible for wireless power transfer [23], [32]. In our CDR scheme, we plan to utilize the power carrier, which frequency is adjusted by a crystal oscillator, to generate synchronized Tx and Rx clocks on both sides of the skin barrier. Oversampling or simple encoding schemes, such as Manchester, are among other possible alternatives to recover synchronized clock from the data bit stream. Adding CDR to the PHM transceiver is the subject of our ongoing research. VI. CONCLUSION We have presented the integrated circuit implementation of a low power transceiver based on PHM technique in a standard CMOS process. The PHM technique, which was intro-

INANLOU et al.: A 10.2 Mbps PULSE HARMONIC MODULATION BASED TRANSCEIVER FOR IMPLANTABLE MEDICAL DEVICES

1305

duced for the rst time in [19] with a discrete prototype, has enabled us to achieve a high data rate (10.2 Mbps) despite utilizing a narrowband (1.4 MHz) high- inductive link, which is needed in IMD applications to extend the transmission range , increase the SNR, improve the link selectivity, and minimize the ISI at the receiver input. The PHM-Tx block is composed of an LC-Driver circuit to generate sharp initiation and suppression pulses with adjustable amplitudes and timing. The PHM-Rx block is a highly congurable ncED receiver with low power consumption. It can be seen in Table IV that the PHM transceiver can achieve a data rate of 10.2 Mbps with a BER of across a 10 mm inductive link, set up between 1 and 2.25 cm gure-8 coils. This performance has been achieved with an unprecedented of 541.9 and 48 at low frequency and 67.5 MHz, respectively. The differential design of the transceiver building blocks and utilization of high- inductors, made possible by the PHM, are expected to immune this data link, to a high extent, against external interference, particularly from the strong power carrier in a multiband wireless link for IMD applications [18]. ACKNOWLEDGMENT The authors would like to thank U. M. Jow from the GT-Bionics Lab for design and fabrication of the data telemetry coils. REFERENCES
[1] R. A. Normann, Technology insight: Future neuroprosthetic therapies for disorders of the nervous system, Nature Clinical Practice, vol. 3, no. 8, pp. 444452, Aug. 2007. [2] N. Lovell, J. Morely, S. Chen, L. Hallum, and G. Suaning, Biologicalmachine systems integration: Engineering the neural interface, Proc. IEEE, vol. 98, no. 3, pp. 418431, Mar. 2010. [3] J. Fayad, S. Otto, R. Shannon, and D. Brackmann, Cochlear and brainstem auditory prostheses, neural interface for hearing restoration: Cochlear and brain stem implants, Proc. IEEE, vol. 96, no. 7, pp. 10851095, Jul. 2008. [4] B. Thurgood, D. Warren, N. Ledbetter, G. Clark, and R. Harrison, A wireless integrated circuit for 100-channel charge-balanced neural stimulation, IEEE Trans. Biomed. Circuit Syst., vol. 3, no. 6, pp. 405414, Dec. 2009. [5] A. Nurmikko, J. P. Donoghue, L. R. Hochberg, W. R. Patterson, Y. Song, C. W. Bull, D. A. Borton, F. Laiwalla, S. Park, Y. Ming, and J. Acerosand, Listening to brain microcircuits for interfacing with external world-progress in wireless implantable microelectronic neuroengineering devices, Proc. IEEE, vol. 98, no. 3, pp. 375388, Mar. 2010. [6] S. B. Lee, H. M. Lee, M. Kiani, U. M. Jow, and M. Ghovanloo, An inductively-powered scalable 32-channel wireless neural recording system-on-a-chip for neuroscience applications, IEEE Trans. Biomed. Circuits Syst., vol. 4, no. 6, pp. 360371, Dec. 2010. [7] A. Sodagar, G. Perlin, Y. Yao, K. Naja, and K. D. Wise, An implantable 64-channel wireless microsystem for single-unit neural recording, IEEE J. Solid-State Circuits, vol. 44, no. 9, pp. 25912604, Sep. 2009. [8] J. C. Lin, Computer methods for eld intensity predictions, in CRC Handbook of Biological Effects of Electromagnetic Fields, C. Polk and E. Postow, Eds. Boca Raton, FL: CRC, 1986, ch. 2, pp. 273313. [9] M. Ghovanloo and K. Naja, A wideband frequency-shift keying wireless link for inductively powered biomedical implants, IEEE Trans. Circuits Syst. I, vol. 51, no. 12, pp. 23742383, Dec. 2004. [10] FCC Rules and Regulations, MICS Band Plan, pt. 95, 2003.

[11] A. Shameli, A. Safarian, A. Rofougaran, M. Rofougaran, J. Castaneda, and F. De Flaviis, A UHF near-eld RFID system with fully integrated transponder, IEEE Trans. Microw. Theory Tech., vol. 56, no. 5, pp. 12671277, May 2008. [12] D. Paret, RFID and Contactless Smart Card Applications. Hoboken, NJ: Wiley, 2005. [13] R. R. Harrison, R. Kier, C. Chestek, V. Gilja, P. Nuyujukian, S. Ryu, B. Greger, F. Solzbacher, and K. Shenoy, Wireless neural recording with single low-power integrated circuit, IEEE Trans. Neural Syst. Rehab. Eng., vol. 17, no. 4, pp. 322329, Aug. 2009. [14] Y. Huand and M. Sawan, A fully integrated low-power BPSK demodulator for implantable medical devices, IEEE Trans. Circuits Syst. I, vol. 52, no. 12, pp. 25522562, Dec. 2005. [15] M. Zhou, M. Yuce, and W. Liu, A non-coherent DPSK data receiver with interference cancellation for dual-band transcutaneous telemetries, IEEE J. Solid-State Circuits, vol. 43, no. 9, pp. 20032012, Sep. 2008. [16] G. Simard, M. Sawan, and D. Massiocotte, High-speed OQPSK and efcient power transfer through inductive link for biomedical implants, IEEE Trans. Biomed. Circuits Syst., vol. 4, no. 3, pp. 192200, Jun. 2010. [17] S. Mandal and R. Sarpeshkar, Power-efcient impedance-modulation wireless data links for biomedical implants, IEEE Trans. Biomed. Circuits Syst., vol. 2, no. 4, pp. 301315, Dec. 2008. [18] M. Ghovanloo and S. Atluri, A wideband power-efcient inductive wireless link for implantable microelectronic devices using multiple carriers, IEEE Trans. Circuits Syst. I, vol. 54, no. 10, pp. 22112221, Oct. 2007. [19] F. Inanlou and M. Ghovanloo, Wideband near-eld data transmission using pulse harmonic modulation, IEEE Trans. Circuits Syst. I, vol. 58, no. 1, pp. 186195, Jan. 2011. [20] M. Z. Win and R. A. Scholtz, Impulse radio: How it works, IEEE Commun. Lett., vol. 2, pp. 1012, Jan. 1998. [21] N. Miura, D. Mizoguchi, M. Inoue, T. Sakurai, and T. Kuroda, A 195-Gb/s 1.2-W inductive inter-chip wireless superconnect with transmitter power control scheme for 3-D-stacked system in a package, IEEE J. Solid-State Circuits, vol. 41, no. 1, pp. 2333, Jan. 2006. [22] J. Yoo, S. Lee, and H. J. Yoo, A 1.12 pJ/b inductive transceiver with a fault tolerant network switch for multi-layer wearable body area network applications, IEEE J. Solid-State Circuits, vol. 44, no. 11, pp. 29993010, Nov. 2009. [23] U. Jow and M. Ghovanloo, Design and optimization of printed spiral coils for efcient transcutaneous inductive power transmission, IEEE Trans. Biomed. Circuits Syst., vol. 1, no. 3, pp. 193202, Sep. 2007. [24] D. Johns and K. Martin, Analog Integrated Circuit Design. New York: Wiley, 1996. [25] M. Crepaldi, C. Li, K. Dronson, J. Fernandes, and P. Kinget, An ultra-low-power interference-robust IR-UWB transceiver chipset using self-synchronizing OOK modulation, in IEEE Int. Solid-State Circuits Conf. Dig. Tech. Papers, 2007, pp. 226227. [26] R. Dokania, X. Wang, S. Tallur, C. Dorta-Quinones, and A. Apsel, An ultralow-power dual-band UWB impulse radio, IEEE Trans. Circuits Syst. II, vol. 57, pp. 541545, Jul. 2010. [27] F. Lee and A. Chandrakasan, A 2.5 nJ/b 0.65 V 3-to-5 GHz subbanded UWB receiver in 90 nm CMOS, in IEEE Int. Solid-State Circuits Conf. Dig. Tech. Papers, 2007, pp. 116590. [28] T.-A. Phan, V. Krizhanovskii, and S.-G. Lee, Low-power CMOS energy detection transceiver for UWB impulse radio system, in Proc. IEEE Custom Integrated Circuits Conf. (CICC), 2007, pp. 675678. [29] Y. Zheng, Y. Tong, J. Yan, Y.-P. Xu, W. G. Yeoh, and F. Lin, A low power noncoherent CMOS UWB transceiver ICs, in Proc. IEEE RFIC Symp., 2005, pp. 347350. [30] A. Chandrakasan, F. S. Lee, D. D. Wentzloff, V. Sze, B. P. Ginsburg, P. P. Mercier, D. C. Daly, and R. Blazquez, Low-power impulse UWB architectures and circuits, Proc. IEEE, vol. 97, pp. 332352, Feb. 2010. [31] J. E. Kardontichik, Introduction to the Design of Transconductance Capacitor Filters. Norwell, MA: Kluwer Academic, 1992. [32] U. Jow and M. Ghovanloo, Optimization of data coils in a multiband wireless link for neuroprosthetic implantable devices, IEEE Trans. Biomed. Circuits Syst., vol. 4, no. 5, pp. 301310, Oct. 2010. [33] U. Jow and M. Ghovanloo, Modeling and optimization of printed spiral coils in air, saline, and muscle tissue environments, IEEE Trans. Biomed. Circuits Syst., vol. 3, no. 5, pp. 339347, Oct. 2009.

1306

IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 46, NO. 6, JUNE 2011

[34] Altera, Cyclone II Field Programmable Gate Array [Online]. Available: http://www.altera.com/products/devices/cyclone2/overview/ cy2-overview.html Farzad Inanlou (S06) received the B.S. degree from Ferdowsi University, Mashhad, Iran, in 2006, and the M.S. degree from Boston University, Boston, MA, in 2008, both in electrical engineering. During summer 2008, he was with Boston Scientic Neuromodulation, Valencia, CA, where he worked on the development of implantable spinal cord stimulation (SCS) devices for the treatment of chronic pain. Since 2008, he has been at the Georgia Institute of Technology, Atlanta, GA, where he is working toward the Ph.D. degree. His research interest is the design of Analog/RF integrated circuits for biomedical applications. Mr. Inanlou is the recipient of the 2010 EAPSI Fellowship from the National Science Foundation, and the Georgia Institute of Technology Presidents Fellowship.

Mehdi Kiani (S09) received the B.S. degree from Shiraz University, Shiraz, Iran, and the M.S. degree from Sharif University of Technology, Tehran, Iran, in 2005 and 2008, respectively. He joined GT-Bionics Lab at the Georgia Institute of Technology, Atlanta, GA, in 2009 where he is working towards the Ph.D. degree.

Maysam Ghovanloo (S00M04SM10) was born in 1973 in Tehran, Iran. He received the B.S. degree in electrical engineering from the University of Tehran, Tehran, Iran, in 1994, the M.S. degree in biomedical engineering from the Amirkabir University of Technology, Tehran, Iran, in 1997, and the M.S. and Ph.D. degrees in electrical engineering from the University of Michigan, Ann Arbor, in 2003 and 2004, respectively. From 2004 to 2007, he was an Assistant Professor in the Department of Electrical and Computer Engineering, North Carolina (NC) State University, Raleigh. In June 2007, he joined the faculty of Georgia Institute of Technology, Atlanta, where he is currently an Assistant Professor and the Founding Director of the GT-Bionics Laboratory in the School of Electrical and Computer Engineering. He has authored or coauthored more than 70 peer-reviewed conference and journal publications. Dr. Ghovanloo is an Associate Editor of the IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS II, IEEE TRANSACTIONS ON BIOMEDICAL CIRCUITS AND SYSTEMS, and a member of the Imagers, MEMS, Medical, and Displays (IMMD) subcommittee at the International Solid-State Circuits Conference (ISSCC). He is the 2010 recipient of a CAREER award from the National Science Foundation. He has also received awards in the 40th and 41st Design Automation Conference (DAC)/ISSCC Student Design Contest in 2003 and 2004, respectively. He has organized several special sessions and was a member of Technical Review Committees for major conferences in the areas of circuits, systems, sensors, and biomedical engineering. He is a member of the Tau Beta Pi, AAAS, Sigma Xi, and the IEEE Solid-State Circuits Society, IEEE Circuits and Systems Society, and IEEE Engineering in Medicine and Biology Society.