This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. IEEE TRANSACTIONS ON BIOMEDICAL CIRCUITS AND SYSTEMS

1

On Integration and Validation of a Very Low Complexity ATC UWB System for Muscle Force Transmission Stefano Sapienza, Marco Crepaldi, Member, IEEE, Paolo Motto Ros, Alberto Bonanno, and Danilo Demarchi, Senior Member, IEEE

Abstract—The thresholding of Surface ElectroMyoGraphic (sEMG) signals, i.e., Average Threshold Crossing (ATC) technique, reduces the amount of data to be processed enabling circuit complexity reduction and low power consumption. This paper investigates the lowest level of complexity reachable by an ATC system through measurements and in-vivo experiments with an embedded prototype for wireless force transmission, based on asynchronous Impulse-Radio Ultra Wide Band (IR-UWB). The prototype is composed by the acquisition unit, a wearable PCB 23 34 mm, which includes a full custom IC integrating a UWB transmitter (chip active silicon area 0.016 mm , 1 mW power consumption), and the receiver. The system is completely asynchronous, it acquires a differential sEMG signal, generates the ATC events and triggers a 3.3 GHz IR-UWB transmission. ATC robustness relaxes filters constraints: two passive first order filters have been implemented, bandwidth from 10 Hz up to 1 kHz. Energy needed for the single pulse generation is 30 pJ while the whole PCB consumes 5.65 mW. The pulses radiated by the acquisition unit TX are received by a short-range and low complexity threshold-based 130 nm CMOS IR-UWB receiver with an Ultra Low Power (ULP) baseband unit capable of robustly receiving generic quasi-digital pulse sequences. The acquisition unit have been tested with 10 series of in vivo isometric and isotonic contractions, while the transmission channel with over-the-air and cable measurements obtained with a couple of planar monopole antennas and an integrated 0.004 mm transmitter, the same used for the acquisition unit, with realistic channel conditions. The entire system, acquisition unit and receiver, consumes 15.49 mW. Index Terms—Average threshold crossing (ATC), impulse-radio ultra wide band (IR-UWB), low-power bio system, surface ElectroMyoGraphic (sEMG).

I. INTRODUCTION

M

OVEMENT recognition through Surface ElectroMyoGraphic signal (sEMG) analysis is one of the current hot topic in bio-potential [1], [2] and Human Machine Interaction research [3]. For this purpose, the force levels of many muscles have to be synchronously acquired and processed, preferably

Manuscript received October 14, 2014; revised January 30, 2015; accepted March 01, 2015. This paper was recommended by Associate Editor J. Georgiou. S. Sapienza, M. Crepaldi, P. Motto Ros, and A. Bonanno are with Istituto Italiano di Tecnologia at Polito, I-10129, Torino, Italy (e-mail: [email protected]). D. Demarchi is with Istituto Italiano di Tecnologia at Polito, I-10129, Torino, Italy, and also with Dipartimento di Elettronica e Telecomunicazioni, Politecnico di Torino, 10129, Torino, Italy (e-mail: [email protected]). Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/TBCAS.2015.2416918

in local to avoid interference, and usually transmitted to an external hardware with powerful computation capacity to get real time gesture information [4]. In this application domain, systems have to be wearable hence requiring one or more batteries, consequently power consumption becomes a critical point especially where continuous recording for long amount of time is required [6]. Power consumption is tightly related with circuit implementation, but also with the number of channels and with the sampling frequency, therefore it is suggested avoiding data overload during the recording. In recent works, different experimental set up were tested varying number and position of sensors: albeit systems with hundreds of acquisition points exist [9], nevertheless the most common approach requires from eight to ten sensors [10], where muscle activation levels are evaluated calculating the Root Mean Square values (RMS) of the sEMG signals with a sampling frequency from 512 to 2048 Hz. ATC is a totally different technique. It involves an event generation (in our system a digital pulse) each time the sEMG potential exceeds a fixed threshold [11], see Fig. 1. It is an “all-ornone” behaviour similar to neuronal action potential. The obtained pulse sequence is a quasi digital signal, exploiting only time dimension and not amplitude. ATC has been demonstrated being highly correlated with the firing rate of motor units present in the muscle, and consequently with the force generated (92%) [12]. Thanks to this property, muscles activation monitoring can be achieved by reducing the amount of data handled by the hardware and consequently circuit complexity, size and power consumption. Furthermore the uniqueness of firing patterns prevents muscles cross talk interferences and ATC robustness permits to tolerate 5–6 dB signal to noise ratio (SNR) and up to 70% of event losses [4], [8]. Transmitting a quasi digital signal requires an high accuracy in time domain because the information is when the event occurred, and Impulse-Radio Ultra Wide Band (IR-UWB), where delay is the key parameter, perfectly matches with this request. Moreover IR-UWB has low power consumption, it is robust to interferences and it can be exploited to measure distance. For these reasons UWB employment in bio application has strongly increased recently [13]. Using ATC events as trigger of an integrated IR-UWB TX, an asynchronous, robust, low power, wireless transmission, correlated with the muscle force, is obtained. In this paper we investigate the lowest level of complexity reachable by an ATC system for wireless force transmission

1932-4545 © 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. 2

IEEE TRANSACTIONS ON BIOMEDICAL CIRCUITS AND SYSTEMS

Fig. 1. Acquisition unit block scheme. Every time the sEMG signal exceeds the fixed threshold the circuit generates a pulse. The on chip section includes the components of the full custom IC. Since the purpose is maximum integrability only low complexity circuits have been designed.

through measurements and in-vivo experiments with a miniaturized prototype. The first component of the prototype is the acquisition unit, a mini PCB including a full custom IC with extremely low complexity ATC front end and IR-UWB transmitter. Thanks to its tiny dimensions, 23 mm length and 34 mm width, the PCB is wearable. The pulses are received by the second component of the system, a short range and low complexity threshold-based 130 nm CMOS IR-UWB receiver with an Ultra-Low-Power (ULP) baseband unit capable of robustly receiving generic quasi-digital pulse sequences. The UWB receiver has partially been validated [7] for Received Signal Strength Indicator (RSSI) and over the air and latency measurements. This work evaluates the acquisition unit performances through 10 isometric and isotonic biceps contractions, while the transmission channel is tested with cable sensitivity experiments and new over-the-air and baseband robustness measurements obtained with a couple of planar monopole antennas and the same transmitter integrated in the acquisition unit with realistic channel conditions. The paper is organized as follows: Section II explains the acquisition unit architecture (receiver design has been already discussed in [7]), Section III describes the UWB transmission

channel characterization while Section IV reviews the in vivo experiments: the set up for the sEMG acquisition, the electrical characterization of the board and in vivo measurements, Section V concludes the paper. II. ACQUISITION UNIT ARCHITECTURE The block scheme of the PCB is presented in Fig. 1. The EMG signal is acquired through the electrodes EL1, EL2, using EL REF as reference. The architecture is based on a fully asynchronous design to achieve minimal complexity and area occupation. Low device area plays an essential role for wearable sensors or wireless body area networks (WBAN) because, in a limited region, is required a bio-potential recording and transmission together with maximum comfort of the subject during the measurements [6]. ATC reduces active silicon area of the circuit avoiding the analog to digital converter component, clock generator and complex logic to manage the data. Our circuit area is 0.016 mm and it is integrated into a single chip with standard QFN 48 package (7 7 mm) mounted on a standard FR4 PCB 1.6 mm tick. The acquisition unit is composed by three nodes: the differential amplifier, the comparator and the UWB transmitter.

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. SAPIENZA et al.: ON INTEGRATION AND VALIDATION OF A VERY LOW COMPLEXITY ATC UWB SYSTEM

The supply voltage for our IC has been obtained from three Varta coin micro- batteries providing 3.6 V and then adjusted to 1.2 V by the low-dropout low-power linear voltage regulator TPS71712. The differential amplifier is the first stage of the IC: it comprises a current mirror followed by two gain blocks which have a simulated gain of 60 dB. Differential approach is widely used in bio-potential recording where the acquisition suffers of a really low SNR and recorded signals are affected by power line noise interference. sEMG signals have limited bandwidth (from 30 Hz to 300 Hz) thus two external series capacitors and (1 F nominal value) are used to filter the sEMG signal in input, implementing a first order high-pass filters with cut off frequency of 10 Hz. They remove DC component and low frequencies artefacts which are usually generated by the relative displacement between the electrodes and the muscles during the recording [5]. On the other hand capacitor is used to filter frequencies above 1 kHz. Since the goal is maximum integrability, this amplifier does not include offset removal and chopped techniques, however it is possible to unbalance the differential line by adjusting (see Fig. 1), which flows through the multi-turn trimmer . In this way we avoid the risk of the two gain stages working outside the linearity region compromising the system performances. A notch filter, commonly used to remove power line noise, has not been included in the circuit since it would remove also essential frequency components of muscles activity. It is preferable to eliminate power line noise with digital filter during post processing [14], [22]. The signal generated by the differential amplifier reaches the next circuit section composed by the comparator followed by a monostable multivibrator. The comparator implements the ATC technique: every time the signal exceeds the threshold level , it sets the output to high state, activating a monostable multivibrator. can be tuned by the external trimmer , while the bias of comparator differential stage is related to the current flowing through . The last part of the circuit, connected to monostable multivibrator output, is composed by a set-reset flip-flop followed by three inverters. They trigger the low complexity UWB-TX [17] creating the 3.3 GHz pulses transmitted by a AH086 multilayer Taiyo Yuden antenna. The UWB pulses length and frequency can be adjusted operating on and : we chose an impulse duration of 5 ns. The center frequency of the antenna is 5550 MHz so we do not irradiate the maximum energy during the transmission, nevertheless 3.3 GHz is the point where our RX has the best sensitivity. To achieve minimal complexity, the comparator does not include a Schmitt trigger hence some of the pulses created by the circuit will be generated by the noise. The simulated energy consumption needed for each pulse generation is 30 pJ; multiplying this value for the number of pulses we obtain the total energy transmission consumption. The maximum frequency reached by a sEMG signal is 500 Hz, however we must consider that also noise can generated spikes, consequently assuming the worst case scenario of 1 kHz interference (upper limit of the pass-band filter) the power con-

3

sumed by the circuit for the transmission will be 1000 [1/s] 30 pJ W. The circuit is implemented in a 130 nm RFCMOS technology one-poly-eight-metal (1P8M) top metal 20 k with high speed (HS) core transistors has been used. III. TRANSMISSION CHANNEL CHARACTERIZATION Transmission channel performances are of primary importance in wearable sensor design. In this section we validate the UWB link of ATC acquisition unit exploiting a non standard sensitivity definition. A. Sensitivity Definition The UWB receiver can process a plethora of signals, not only the events of our acquisition unit, consequently a standard sensitivity definition cannot be applied in our context. Without loss of generality, we test the RX using single frequency signals . We report the mean received frequency, implicitly defining a processing gain based on averaging data (in most measurements using 10 kSamples). Since the receiver output toggles when a pulse is received, assuming no event losses, the output signal will be a square wave completing a period every two transmitted pulses, and it will have a frequency equals to . exceeds the range of sEMG frequencies (and the upper limit of the acquisition unit pass-band filter), in this way we have more events, consequently the sample size increases, obtaining more accurate results form a statistical point of view in addition to a more precise channel characterization. Measurements implicitly include mthe operation of our oscilloscope as a high performance “digital backend”, an equivalent time to digital converter which implements frequency measurements. We define , the average estimation of the received frequency information (1) is the number of measurements for estimation the related cycle-to-cycle standard deviation, where , the time interval between two positive edges, to all effects is an analog signal. Sensitivity in standard IR-UWB is rather referred to a logic interpretation of data, ‘0’ or ‘1’, on which an arithmetic error computation is run. Herein, we refer to a steady state measurement of a digital signal frequency [16], and in this work we unconventionally define sensitivity as the minimum input power level for a . The defined processing gain used is assumed large enough to determine our analog signal, particularly parameters and , with sufficient accuracy. The use of large averaging permits an accurate characterization of jitter and allows for successive constraints relaxations for future works, based on the obtained results. By referring SNR to period rather than signal power as, , a sensitivity always corresponds to 60 dB SNR. where and

B. Experimental Set-Up The receiver [7] has been integrated in a 130 nm RFCMOS technology, and implemented using High Speed (HS)

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. 4

IEEE TRANSACTIONS ON BIOMEDICAL CIRCUITS AND SYSTEMS

Fig. 2. (a) Measured received average frequency using an RF cable and attenuators and (b) standard deviation.

MOSFET. The IC is bonded in a 7 mm 7 mm QFN48 package and mounted on a standard FR4 PCB 1.6 mm thick. A single 10 pF DC block capacitor is included at the radio frequency input, before antenna. The baseband power supply has been routed to a separate I/O PAD. The receiver occupies 0.2079 mm , i.e., 270 m 770 m silicon area. The baseband layout occupies 90 40 m plus 55 30 m, i.e., 0.0053 mm , excluding the dedicated level shifters. The transmitter in [17], used in our experiments in a separate PCB, is the same mounted on the acquisition unit. Here it is powered with an Agilent E3631A and triggered by an Agilent 33220A signal generator. We did not use acquisition unit itself as TX since the filters in the circuitry cut frequencies above 1 kHz. The receiver has been powered with an Agilent 3634A and the output frequency is captured with an Agilent DSO9404A 4 GHz 20 GS/s oscilloscope. Each measurement is averaged using a fixed kSample, otherwise explicitly stated. For the wired measurements, the transmitter is connected to the receiver using an SMA metrological cable with a maximum 1 dB loss in a 0–12 GHz band, and a controlled path loss is imposed with dedicated SMA RF attenuators. For the cable interference measurements a Rhode&Schwarz SMB100A signal generator injects a tone to the receiver through an hybrid coupler. For the over-the-air measurements, the transmitter and the receiver are connected with two wideband planar monopole antennas with dB in the band 1.7–3.7 GHz. In all measurements, the off-chip signals setting is V, V, V, V, V and A, otherwise explicitly stated. We measure mean frequency and standard deviation ( and , related to cycle-to-cycle jitter) with the Agilent DSO9404A using the built-in function. The oscilloscope trigger level is kept constant at half the swing, i.e., 0.6 V, positive-to-positive edge frequency measurement. The oscilloscope re-triggers the received waveform at every measurement with its internal time reference,

Fig. 3. One-dimensional plot of the average and relative std. deviation , for 20 and 30 dB RF attenuation. frequency

therefore and ( and in time domain) are derived from a set of independent data, instead from subsequent periods. The results obtained implicitly include missing triggers due to reduced sensitivity. The transmitter has been biased to generate UWB pulses at 3.3 GHz, the center frequency that maximizes the receiver sensitivity. Compared to simulation results, we have observed a significant center frequency selectivity deviation, probably due to the presence of a significant parasitic capacitance in the front-end section that causes mismatch at PCB-level. Using a Tektronix RSA3308B spectrum analyzer with built-in channel power functions, the measured transmitter average power at 1 MHz Pulse Repetition Frequency (PRF) is dBm in a 1 GHz bandwidth. At different , the TX power has been scaled with the term . C. Cable Sensitivity Measurements Fig. 2 shows the deviation of the measured frequency with reas a function of path loss spect to the expected value (dB) and transmitted signal. Average frequency matches half the transmitted frequency while it degrades for high path loss to indicate a receiver sensitivity decrease. At higher PRF, deviates from the expected value. Based on simulations we

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. SAPIENZA et al.: ON INTEGRATION AND VALIDATION OF A VERY LOW COMPLEXITY ATC UWB SYSTEM

5

Fig. 4. Cable interference measurements with tones centered at 3.3, 1.8 and 2.4 GHz.

conclude that this effect is due to the maximum PRF limit of the baseband. The relative results of Fig. 2(b) show that a lower than 0.5% relative is achievable until 30 dB attenuation and 1 MHz received signal. Notwithstanding an increased relative std. deviation, these results show that with such system transmission up to 8 MHz is ideally possible. Fig. 3 shows a subset of the data presented in the mesh. A relative std. deviation is obtained for kHz, 30 dB attenuation. Sensitivity is then dBm. At 700 kHz PRF, 20 dB attenuation curve, sensitivity is dBm. D. Cable Interference Measurements Fig. 4 shows interference robustness measurements obtained kHz, average received signal power for dBm, as a function of interferer power and Signalto-Interference Ratio (SIR). The measurements include hybrid coupler and cable losses., 1 kS points. Due to the use of a very simple decoupling capacitor at the baseband [18], the receiver can tolerate dBm in-band interference for a relative std. deviation of 0.4%. The tolerable out-of-band interference levels are and dBm at 1.8 and 2.4 GHz for % relative std. deviation. We report very high interference rejection at 2.4 GHz, higher than at 1.8 GHz. This can be possibly explained by parasitic effects at PCB level. E. Over-the-Air Measurements Fig. 5 shows the most significant interference sources and the corresponding 50 MHz channel powers, measured with a Tektronix RSA3308B spectrum analyzer connected to the receiver’s antenna, but measured channel powers are reduced to dBm and dBm for 1.85 and 2.15 GHz interferers, respectively. The instantaneous power is not fixed and depends on the wireless networks load. During experiments we have observed fluctuations of 10 dB compared to the given average values. Fig. 6 shows the measured frequency as a function of antennas separation, for a distance range 10–28 cm, frequency between 100 Hz and 8 MHz. The added feature here is the presence of more pronounced channel non-ideality such as realistic out-of-band interference and multipath propagation, although with limited spread for such short distance link. The key frequencies for an integrated ATC prototype managing sEMG signal are those lower than 1 kHz.

Fig. 5. Measured over-the-air received narrowband (left) and wideband (right) interference with a copper square monopole antenna.

Below 27 cm distance, the periodic quasi-digital frequency output, can be transmitted with relative std. deviation below 0.1% until 300 kHz, and below 1% for frequency MHz. Within this distance range, does not show appreciable variations compared to . The test on mean frequency deviation at different operating distances [7] demonstrated that the described wireless link, permits the transmission of the digital output in [19], without impacting on the overall measurement chain. A m wireless link has been obtained with higher output power, to demonstrate feasibility at larger operation distance. A train of 10 ns duration pulses, center frequency 3.3 GHz, PRF 1 MHz, 5 dBm peak power ( dBm average power) are radiated through the same planar monopole antennas used to validate the integrated receiver. With the depicted objects in the environment, therefore with larger multipath propagation compared to a short distance link, the 1 MHz periodic signal is received with % relative std. deviation. From these tests, we have noticed a high dependency on antenna orientation as well as receiver board positioning on link quality, at long distance. This can be explained by the poor pulses amplitude control of the wireless channel and antenna orientation. F. Bit Error Rate In an asynchronous event driven transmission like ATC, the standard theoretical Bit Error Rate (BER) is not defined since the pulses generated are not correlated to a binary value but only to the presence of an event. As already proved by extensive simulations in [4], ATC technique is robust enough to estimate

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. 6

IEEE TRANSACTIONS ON BIOMEDICAL CIRCUITS AND SYSTEMS

Fig. 6. (a) Measured over-the-air average frequency using two planar monopole antennas and (b) relative standard deviation.

muscle force although 70% of lost events, consequently losing pulses is not a critical point. However, asynchronous threshold UWB detector proved to be robust to BER [21]. For the present work the low probability of missing a pulse has been demonstrated by the over-the-air measurements with our transmission channel (see Fig. 6). For distance below 27 cm and up to 10 kHz (far above sEMG upper band), the received signal does not show appreciable variations and relative st. deviation is below the 0.1%. Some in depth analysis of IR-UWB pulse transmission over on-human body channel showed that also pulse shape and modulation play a role in determine the BER performance [20].

Fig. 7. Impact of measurements),

on the receiver performance (over-the-air kHz.

G. Power Consumption The receiver front-end consumes 9.84 mW, at first approximation independent from pulse rate. With no input signal, therefore with no switching activity of receiver output, the baseband consumes 22.08 W, lower than the required power consumption of non duty-cycled clock circuitry of recent state-of-the-art receivers [18]. At kHz, input signal level dBm, power increases to 22.24 W. The active energy per pulse is then , i.e., 533 fJ/pulse. H. Baseband

Robustness

variation on and Fig. 7 shows the impact of , for a TX-RX distance cm. Bias voltage and are unvaried. The baseband can tolerate % variation, still maintaining relative std. deviation . IV. IN VIVO MEASUREMENT A. Experimental Setup A photo of the experimental set up is shown in Fig. 8. The measurements have been done recording the right and left short head biceps brachii sEMG signal of two healthy male subjects (age 26 and 25) during isotonic and isometric contractions. This

Fig. 8. Experimental set up for the isometric and isotonic measurements.

muscle was used because it gives an easy access to electrodes placement together with a good signal amplitude. We used the H124SG EMG Kendall Ag/AgCl electrodes, 24 mm diameter. The reference electrode has been placed on the elbow which is the region with low electrical activity closest to the signal recording point. The connection wires have been maintained as short as possible to avoid interferences from cables movement during the experiments.

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. SAPIENZA et al.: ON INTEGRATION AND VALIDATION OF A VERY LOW COMPLEXITY ATC UWB SYSTEM

NUMBER

OF

7

TABLE I SPIKES OF NON CONSISTENT TRIALS BEFORE FILTERING TOO CLOSE EVENTS

AND

AFTER

Fig. 9. Amplifier block transfer characteristic.

The subject was seated with his arm and forearm aligned on the same sagittal plane; the elbow was flexed at 90 degrees. Five level of force have been measured giving to the subject different weights to hold in his hand: 2 kg, 4 kg 6 kg, 8 kg and 10 kg. All measurements last 2 seconds and between two recording at least three minutes of rest have been observed to avoid fatigue effects. Skin preparation has been adopted by rubbing a textile, wetted with ethyl alcohol, on the area of signal acquisition. In this way it is possible to remove dead cells on the surface improving skin impedance condition. has been set in a range from 650 mV to 750 mV, slightly above the amplitude of environmental noise. With this variable noise margin, minimal muscle activities can be sensed because the sum of sEMG and noise immediately makes the input signal cross . The IR-UWB-TX pulses and the differential sEMG amplified signal have been simultaneously recorded using an Agilent oscilloscope DSO9404A, sampling frequency of 50 kSamples/s. The highest frequencies reached by a surface EMG signal are lower than 1 kHz, but the IR-UWB-TX spikes last 50 s or less, so sampling frequency higher than kHz is needed to avoid loss of pulses during recording.

a growth of the number of the ATC events when the weight sustained by the subject is increased. Three of the experiments do not show this behaviour. It is clear that some post processing work is needed to discriminate pulses generated by the sEMG signal from those produced by the noise oscillations. We analysed the Power Spectral Density (PSD) of the amplifier output signal x(t). Considering sEMG as Wide Sense Stationarity (WSS) signal it is possible applying the Wiener-Khinchin-Einstein theorem and calculate Energy Spectral Density as the Fourier transform of the autocorrelation function of the signal

If we introduce

as

We obtain the PSD of the signal calculating the limit

B. Measurements The measured power consumption of the acquisition unit, when there is no transmission, is 5.65 mW. Fig. 9 shows the transfer characteristic of the amplifier block. We spanned the bandwidth from 1 Hz to 10 kHz with a differential sinusoidal wave, 40 V peak to peak, as input and measured the amplitude of the amplifier output. Measurement results have shown that the nominal amplification of one thousand cannot be obtained due to offset issues. The system can amplify the input signal 375 times at maximum. The pass-band width is from 10 Hz to 1 kHz as expected and no changes in the morphology or any kind of distortions have been observed. In this frequency interval the measured CMRR of the amplifier can be considered constant in first approximation with a measured value of 55 dB. The input referred noise of the amplifier block is below 0.26 V. A negligible jitter, probably generated by the comparator unit, has been measured with the oscilloscope at the output of the pulse trigger: it does not exceed 500 ns. A simple Matlab script has been written to count the number of spikes in the IR-TX-UWB signal and to perform some post processing analysis. Ten series of experiments, measuring the five different levels of force, have been done. We expect, because of the correlation between ATC and muscle force,

In this way it has been possible to verify that the maximum frequency reached by sEMG was less than 200 Hz. This means that two pulses, if generated by the muscle activity, must be spaced at least of 5 ms. Our experiments last two seconds, consequently there cannot be more than 1 thousand of spike generated by the EMG signal. We implemented therefore a Matlab filter which removes events temporally too close each other. The numbers of spikes measured before and after filtering of the three non consistent measurements are presented in Table I. The filter strongly decreases the number of pulses and all three filtered trials show an increase of events recorded when the sustained weight augments. We have applied the filter to all the measurements, the results are graphically represented in Figs. 10 and 11. To align the values on a common scale, in Fig. 10 each quantity has been normalized with the number of spikes of 10 kg series. C. Discussion In this work we realized a wireless force transmission prototype with the lowest complexity circuit possible to investigate

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. 8

IEEE TRANSACTIONS ON BIOMEDICAL CIRCUITS AND SYSTEMS

TABLE II COMPARISON IN TERM OF POWER CONSUMPTION OF OUR INTEGRATED ATC PROTOTYPE (BOLD) WITH COMMERCIAL CHIPSET

Fig. 10. Normalized number of spikes versus weight sustained. The measurements have been normalized with the number of spikes at 10 kg. The dark red coloured bars represent the standard deviations of the means values.

Fig. 11. Filtered number of spikes versus weight sustained.

how much ATC technique relaxes circuit constraints. Active silicon area of the acquisition unit IC is only 0.016 mm , with 1 mW power consumption. Despite the minimal complexity of the prototype the results of the isometric and isotonic experiments show a correlation between the number of spikes recorded by the ATC and muscle activity. ATC pulses grows when the weight sustained by the subject increases (see Fig. 10), nevertheless huge standard deviations are present especially during the lightweight measurements. Four are the sources of the high variability observed: • The number of spikes is tightly correlated with the threshold, which is set each time differently, conditioned

by the environmental noise level that vary for each measurement; • The amplifier gain of 375 instead than the estimated 1000 which lowers the SNR; • The amplifier CMRR of 55 dB in the sEMG bandwidth; • The intrinsic variability in the myoelectric signals of two different subjects. For these reasons it is not possible to clearly discriminate all the levels of force, see Fig. 11. The system recognizes, with a 0% error rate, the difference in muscle activation when the weights held by the subject differ at least of 6 kg. Therefore the prototype can discern three qualitative level of strain: a resting state (0 kg), a low activation state (2 kg or 4 kg) and high activation state (8 kg and 10 kg). An external, digital post processing filter was needed to discriminate the pulses generated by the EMG and by the background noise. This function will be easily implemented in future works, adding a Schmitt trigger in the amplifier block that will prevent unwanted threshold crossing. Furthermore, chopped technique and offset removal have to be integrated to increase signal amplification and to improve SNR. These improvements will slightly increase the circuit complexity in exchange of better measurement performance and consistency. The jitter affecting the energy detection of the UWB link does not impact on the experiments: 10 ns jitter, is indeed negligible compared to sEMG time constants [7]. Table II shows the power consumption of our system with some commercial chipset [22]–[24]. Even with our proof-of-concept prototype ATC technique associated with UWB transmission allows a huge reduction of power consumption. The whole system (RX and TX) consumes 15.49 mW, while our single channels differential acquisition unit consumes 92% less than the Trigno device [24] which has, to the best of our knowledge, the best performance in term of energy saving. Nevertheless ATC systems measure force levels only. If signal morphology information are necessary a standard approach is required. V. CONCLUSION This paper investigates the lowest level of complexity reachable by an ATC system through a miniaturized prototype, which realizes one step toward a fully integrated ATC-IR-UWB device. The core of the prototype is a wearable, very low complexity PCB realizing a two channel differential EMG acquisition, managing ATC technique and transmitting the events through UWB pulses to an external receiver. The acquisition unit and the transmission channel have been separately tested. In-vivo experiments with the prototype have shown that the

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. SAPIENZA et al.: ON INTEGRATION AND VALIDATION OF A VERY LOW COMPLEXITY ATC UWB SYSTEM

number of ATC pulses increase with the force applied in an isometric and isotonic contraction. It has been possible discriminate, with a 0% error rate, a rest state (with no spikes detected) besides two level of force differing at least 6 kg. Transmission channel characterization showed that, below 27 cm distance, a periodic quasi-digital frequency output, can be transmitted with relative std. deviation below 0.1% until 300 kHz, and below 1% for frequency MHz. Within this distance range the number of event losses is negligible. With higher output power a m wireless link has been obtained. Compared to commercial chipset, our prototype showed good performance in term of power consumption. Our acquisition unit consumes 5.65 mW, 92% less than the Trigno device (4–8 channels) which has the best performance in term of energy saving. However, the power consumption of our acquisition unit can be strongly reduced, indeed the IC managing ATC technique and UWB transmission (active silicon area 0.016 mm ), consumes only 1 mW. The 4.65 mW dissipated by the off chip components, like trimmer and voltage regulator, can be easily decreased using other low power COTS components. The whole system (RX and TX) consumes 15.49 mW. Schmitt trigger, to remove unwanted threshold crossing, and chopped and offset removal techniques, to improve SNR and reduce variability, should be implemented in future version of the ATC system. These improvements will increase prototype performance and consistency. REFERENCES [1] R. Castillo-Lozano, A. Cuesta-Vargas, and C. P. Gabel, “Analysis of arm elevation muscle activity through different movement planes and speeds during in-water and dry-land exercise,” J. Shoulder Elbow Surgery, vol. 23, pp. 159–165, Feb. 2014. [2] F. Al Omari, G. Liu, and Y. Ding, “Classification of hand motions using RBF neural network based on extracted surface electromyography signal,” Sensors & Transducers, vol. 167, pp. 19–22, Mar. 2014. [3] S. M. Patil and C. G. Patil, “An approach for human machine interaction using electromyography,” J. Med. Imag. Health Inform., vol. 4, pp. 71–75, Mar. 2014. [4] P. M. Ros, M. Paleari, N. Celadon, A. Sanginario, A. Bonanno, and M. Crepaldi, “A wireless address-event representation system for ATC-based multi-channel force wireless transmission,” in Proc. 5th IEEE Int. Workshop Advances in Sensors and Interfaces, Jun. 2013, pp. 51–56. [5] C. J. DeLuca, L. D. Gilmore, M. Kuznetsov, and S. H. Roy, “Filtering the surface EMG signal: Movement artifact and baseline noise contamination,” J. Biomech., vol. 43, pp. 1573–1579, May 2010. [6] L. Huang, M. Ashouei, F. Yaziciogl, J. Penders, R. Vullers, G. Dolmans, P. Merken, J. Huisken, H. de Groot, C. Van Hoof, and B. Gyselinckx, “Ultra-low power sensor design for wireless body area networks: Challenges, potential solutions, and applications,” Int. J. Digital Content Technol. Appl., vol. 3, no. 3, pp. 136–148, Sep. 2009. [7] M. Crepaldi, P. Motto Ros, A. Bonanno, M. Morello, and D. Demarchi, “A non-coherent IR-UWB receiver for high sensitivity short distance estimation,” in Proc. IEEE Int. Symp. Circuits and Systems, Jun. 2014, pp. 1905–1908. [8] A. Phinyomark, C. Limsakul, and P. Phukpattaranont, “A novel feature extraction for robust EMG pattern recognition,” J. Comput., vol. 1, no. 1, pp. 2151–9617, Dec. 2009. [9] J. Hahne, B. Graimann, and K. Muller, “Spatial filtering for robust myoelectric control,” IEEE Trans Biomed. Eng., vol. 59, no. 5, pp. 1436–1443, May 2012. [10] B. Peerdeman, D. Boere, H. Witteveen, R. Huis in t Veld, H. Hermens, S. Stramigioli, H. Rietman, P. Veltink, and S. Misra, “Myoelectric forearm prostheses: State of the art from a usercentered perspective,” J. Rehab. Res. Devel., vol. 48, pp. 719–738, Aug. 2011. [11] A. Phinyomark, C. Limsakul, and P. Phukpattaranont, “EMG feature extraction for tolerance of white Gaussian noise,” in Proc. Int. Workshop and Symp. Science and Technology, 2008.

9

[12] M. Crepaldi, M. Paleari, A. Bonanno, A. Sanginario, P. Ariano, D. Tran, and D. Demarchi, “A quasi-digital radio system for muscle force transmission based on event-driven IR-UWB,” in Proc. IEEE Biomedical Circuits and Systems Conf., 2012, pp. 116–119. [13] M. Baboli, A. Sharafi, A. Ahmadian, and M. S. Nambakhsh, “An accurate and robust algorithm for detection of heart and respiration rates using an impulse based UWB signal,” in Proc. Int. Conf. Biomedical and Pharmaceutical Engineering, 2009, pp. 1–4. [14] D. T. Mewett, H. Nazeran, and K. J. Reynolds, “Removing power line noise from recorded EMG,” in Proc. 23rd Annu. Conf. IEEE Engineering in Medicine and Biology Soc., Istanbul, Turkey, Oct. 25–28, 2001. [15] Revision of Part 15 of the Commissions Rules Regarding Ultra-Wideband Transmission Systems, Feb. 14, 2002, Report and order, adopted, FCC. [16] M. Jableka, M. Misandkowicz, and D. Kos andcielnik, “Uncertainty of asynchronous analog-to-digital converter output state,” in Proc. IEEE Int. Symp. Industrial Electronics, Jul. 2010, pp. 1692–1697. [17] M. Crepaldi, D. Dapra, A. Bonanno, I. Aulika, D. Demarchi, and P. Civera, “A very low-complexity 0.34.4 GHz 0.004 mm2 all-digital ultra-wide-band pulsed transmitter for energy detection receivers,” IEEE Trans. Circuits Syst. I, Reg. Papers, vol. 59, no. 10, pp. 2443–2455, 2012. [18] S. Gambini, J. Crossley, E. Alon, and J. H. Rabaey, “Fully integrated, 290 pJ/bit UWB dual-mode transceiver for cm-range wireless interconnects,” IEEE J. Solid-State Circuits, vol. 47, no. 3, pp. 586–598, 2012. [19] A. Bonanno, M. Crepaldi, I. Rattalino, P. Motto, D. Demarchi, and P. Civera, “A 0.13 m CMOS operational Schmitt trigger R-to-F converter for nanogap-based nanosensors read-out,” IEEE Trans. Circuits Syst. I, Reg. Papers, vol. 60, no. 4, pp. 975–988, 2013. [20] Y. P. Zhang and Q. Li, “Performance of UWB impulse radio with planar monopoles over on-human-body propagation channel for wireless body area networks,” IEEE Trans. Antennas Propag., vol. 55, no. 10, pp. 2907–2914, Oct. 2007. [21] M. Crepaldi and P. R. Kinget, “Error ratio model for synchronised-OOK IR-UWB receivers in AWGN channels,” Electron. Lett., vol. 49, no. 1, pp. 25–27, Jan. 2013. [22] TeleMyo Guide [Online]. Available: http://www.seedtech.co.kr/ product/noraxon/TeleMyo%202400T%20G2_v2.pdf [23] EMG2GO Guide [Online]. Available: http://www.ti.com/corp/docs/ university/pdf/adc2012_medical_university_of_vienna_project.pdf [24] Trigno User Guide [Online]. Available: http://www.delsys.com/Attachments_pdf/Trigno%20Wireless%20System%20Users%20Guide%20%28MAN-012-2-3%29.pdf Stefano Sapienza received the M.S. degree in biomedical engineering from Politecnico di Torino (Polito), Turin, Italy, in 2012. His thesis was titled, “Design and Realization of an Hardware Software Interface for EEG Signals.” Currently, he is working toward the Ph.D. degree at the Istituto Italiano di Tecnologia (IIT) at Polito. His research activities cover biopotential acquisition system, bio-inspired electronics, neural networks, electroencephalographic, and electromyographic signals post processing.

Marco Crepaldi (M’09) received the engineering degree and Ph.D. degree in electronic engineering from Politecnico di Torino (Polito), Turin, Italy, in 2005 and 2009, respectively. During 2008, he was a Visiting Scholar in the Electrical Engineering Department, Columbia University, New York, NY, USA. After completing the Ph.D. degree, he worked as a postdoctoral student at the VLSI Lab in the Department of Electronics and Telecommunications (DET), Polito. Currently, he is a postdoctoral student at the Istituto Italiano di Tecnologia at Polito. His research interests and expertise include analog and digital CMOS IC design, modeling of analog and digital circuits and systems, in particular ultra-low-power UWB transceivers design, all-digital and logic-based radio design, capacitive and resistive interfaces design for robotics, CMOS integrated sensing elements for nanodevices, and professional digital audio-video broadcasting systems development. He has authored and coauthored more than 30

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. 10

papers in IEEE journals or conferences. He is now the scientific responsible in charge for IIT of the SMAC project. Dr. Crepaldi served as a reviewer for the IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—I: REGULAR PAPERS and IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—II: EXPRESS BRIEFS, IEEE TRANSACTIONS ON WIRELESS COMMUNICATION, and IEEE JOURNAL OF SOLID STATE CIRCUITS.

Paolo Motto Ros received the electrical engineering degree and the Ph.D. degree in electrical engineering from Politecnico di Torino (Polito), Turin, Italy, in 2005 and 2009, respectively. From 2009 to 2012, he was with Neuronica Laboratory (Dipartimento di Elettronica, Politecnico di Torino) as a Research Associate, working on assistive technologies, computer vision, and learning machines projects. He joined the Center for Space Human Robotics (CSHR), Istituto Italiano di Tecnologia (IIT), Turin, Italy, in 2012. His current interests include design of full-custom ultra-low-power asynchronous digital integrated circuits, event-based smart-sensors, sensor networks, bio-inspired electronics, and neuromorphic engineering.

IEEE TRANSACTIONS ON BIOMEDICAL CIRCUITS AND SYSTEMS

Alberto Bonanno received the double degree in electronic engineering from Politecnico di Torino, Italy, and Ecole Polytechnique Federale de Lausanne, Switzerland, in 2005, and the Ph.D. degree in electronic engineering from the Politecnico di Torino in 2014. From 2005 to 2010, he worked on low-power digital IC design for a private company and Politecnico di Torino. In 2011, he joined the Istituto Italiano di Tecnologia, Center for Space Human Robotics (CSHR at Polito) to work on the Micro-for-Nano project. In 2013, he was a visiting Ph.D. student at Technische Universiteit Eindhoven. Currently, he is is a Research Fellow at CSHR at Polito. His research activities cover low-power digital IC design, printed organic technology, techniques for nanomaterial integration on CMOS technology, and mixed-signal IC design for nanostructured materials read-out.

Danilo Demarchi (M’10–SM’13) received the M.S. and Ph.D. degrees in electronics engineering from Politecnico di Torino, Italy, in 1991 and 1995, respectively. Currently, he is an Assistant Professor for the “Bio-Micro and Nano Systems” and “Nanoelectronics ” classes at Politecnico di Torino, and leads the Micro and Nano Electronic Systems (MiNES) Group in the Department of Electronics and Telecommunications. He also coordinates the microelectronics research line at the Istituto Italiano di Tecnologia, IIT at Polito, Center for Space Human Robotics (CSHR). He has authored or coauthored two patents and articles in more than 150 scientific publications in journals and conference proceedings related to micro and nano electronic systems.