Accurate Level-crossing ADC Design for Biomedical Acquisition

Board

Mariam Tlili

1,2

, Manel Ben-Romdhane

, Asma Maalej

, Mohamed Chaker Bali

, François Rivet

Dominique Dallet

and Chiheb Rebai

GRESCOM Research Lab., SUP’COM, University of Carthage, Cité Technologique des Communications,

2083 El Ghazala, Ariana, Tunisia

IMS Research Lab., University of Bordeaux, Bordeaux INP ENSEIRB-MATMECA, 351 Cours de la Libération,

Bâtiment A31, 33405 Talence Cedex, France

Keywords: Biomedical Sensor, Level-crossing Analog-to-Digital Converter, ECG.

Abstract: The aim of this paper is to present a wireless biomedical system for the acquisition and transmission

(Wibio’ACT) of biomedical signals. This work is a part of the Wibio’ACT project which main purpose is to

ensure the minimum power consumption while diagnose patients continuously and in real time. For the

Wibio’ACT system, the bottleneck is the analog-to-digital conversion (ADC) since it controls the power

consumption of the digital signal processing step as well as the amount of the transmitted data. In fact, in

this work case, the ADC continuously measures the electrical activity of the heart to deliver the

electrocardiogram (ECG) signal. Hence, among conventional ADCs, level-crossing analog-to-digital

converters (LC-ADCs) have been investigated for ECG signals processing. Authors propose some design

consideration of the LC-ADC. This reduces the LC-ADC output samples by 13 % to help to save the power

consumption of the wireless data transmitter. The samples with a small variation are reduced by at least

44%. The performance of the proposed design is measured in terms of percentage root mean square

difference (PRD) applied to the reconstructed signal quality. A PRD of 1% is verified using behavioral

simulations on ECG records extracted from different databases. A timer period T

of 0.14 ms ensures an

effective number of bits of 10 bits and a signal to noise ratio of 62 dB.

1 INTRODUCTION

In biomedical engineering field, as the

cardiovascular diseases are at the top of death

causes, many researches focus on the development

of reliable and cost effective system for monitoring

cardiac patients (Mendis, 2011; Gyselinckx, 2006).

One part of these research activities concerns the

acquisition, the processing or the wireless

transmission of the electrical activity of the heart or

any other vital data. Due to the continuous-time

monitoring of the patients, large amounts of data

need to be processed, stored and transmitted. The

key requirements for an efficient high-quality signal

monitoring system are power consumption saving,

data rate enhancement and distortion robustness.

Therefore, in order to reduce the amount of data

after acquisition and before transmission over the

radio channels, electrocardiogram (ECG) signal

compression requires a lot of attention. Many

techniques and algorithms have been proposed

including modulation, coding, cosine and wavelet

transform (Rajoub, 2002; Cox et al., 1968; Miaou et

al., 2005; Ahmed et al., 2009). While many

algorithms are deployed for the data compression,

few works combine the sampling step to the

compression as in level-crossing analog-to-digital

converter (LC-ADC) (Mark et al., 1981; Sayiner et

al., 1996). In the literature, LC-ADCs are essentially

used for audio signal processing particularly speech

and ultrasound signals (Allier et al., 2003; Kozmin

et al., 2009). In fact, since sampling occurs at

irregular times, LC-ADCs are suitable for the

conversion of bursty and sparse signals (Guan and

Singer, 2007). In the most of the applications, it has

been proven that significant performance

improvements can be achieved compared to the

conventional Nyquist converters in terms of power

Tlili, M., Ben-Romdhane, M., Maalej, A., Bali, M., Rivet, F., Dallet, D. and Rebai, C.

Accurate Level-crossing ADC Design for Biomedical Acquisition Board.

DOI: 10.5220/0005855003210326

In Proceedings of the 9th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2016) - Volume 4: BIOSIGNALS, pages 321-326

ISBN: 978-989-758-170-0

321

consumption and surface occupation (Kozmin et al.,

2009; Guan and Singer, 2007; Li et al., 2013).

In this work, an in-depth study of the main

design parameters associated with an LC-ADC is

proposed. The present study consists of a detailed

analysis of two main parameters that affect the

accuracy of the LC-ADC, namely the amplitude

resolution and the timer frequency. Additional

parameters can also affect the LC-ADC output

signal such as the accuracy of the quantization levels

and the comparator delay. As a result of this

analysis, the authors propose a methodology to

select accurate parameters for ECG signals

extraction.

To properly present this work, this paper is

organized as follows. First, the biomedical smart

sensing system architecture is introduced in Section

II. Section III describes the LC-ADC architecture. In

the same section, the authors present the design

considerations for an efficient ECG signal detection

to perform both signal compression and power

consumption reduction. Section IV presents the

results of behavioral simulations on different ECG

signals. The percentage root mean square difference

(PRD) is used to evaluate signal distortion compared

to uniform sampling. Section V concludes the paper.

2 SYSTEM OVERVIEW

The aim of the Wibio’ACT project is to implement a

smart system for biomedical signals acquisition and

transmission. The two main and innovative topics in

this project concern the digitization with an intrinsic

compression step via the use of LC-ADC and the

reconstruction with a minimum complexity in

implementation and a good recovery of the original

ECG signal.

2.1 Wibio’ACT System Presentation

The Wibio’ACT system is presented in Figure 1. An

acquisition of the biomedical signal is firstly done

through the use of non-invasive sensors that

wirelessly transmit signal so as to form a wireless

body area network (WBAN). The received signal,

often a voltage, is amplified and filtered. A classical

ADC, SAR (Long et al., 2014) or Sigma-Delta

(Giroud et al., 2014) architectures, conventionally

performs the digitization of the acquired analog

signals. In this project, LC-ADC is chosen thanks to

its capacity of compressing the acquired data. This

converter allows bypassing the compression of

digital data usually ensured by an algorithm

implemented on microcontroller. The transmitter is a

combination of several functions. They are the

mixing stage with a local oscillator (second input of

the mixer), as well as the amplification and the

filtering stages. At the receiver side, a front end

stage composed of functions such as filtering and

low noise amplification (LNA) is firstly used. In

order to have the original signals, a reconstruction

step is necessary to allow the doctor to visualize,

analyse and make diagnostics from the biomedical

signals.

Besides, the voltage that appears between the

sensor electrodes is conditioned via a front-end

interface. It includes functions such as amplification

using a programmable gain preamplifier (PGA) and

analog filtering with passive off-chip filters. For the

ECG signal acquisition, a PGA gain of 60 dB is

required (Hartmann, 2003). Passive filters including

a high-pass filter (HPF) and a low-pass filter (LPF)

are composed of capacitors and resistors. The HPF is

mainly used to cancel DC-level shift caused by

human skin and its cut-off frequency is set to 20

mHz. The LPF is used to eliminate interferers at

frequencies above those of ECG signals, so its cut-

off frequency is set to 200 Hz. These filtering steps

highly influence the reconstruction algorithm

performances as any noise would distort their

outputs.

The ECG signal acquisition system requires a

signal-to-noise ratio (SNR) of at least 61 dB in order

to detect heart activities precisely (Li et al., 2013).

The ADC to be chosen must have an equivalent

effective number of bits approximately equal to 10.

As the aim of Wibio’ACT is to enhance

transmission and reception of ECG signal with the

minimum power consumption, the authors propose

the use of LC-ADC. In fact, in this case, the

acquisition system exploits the features of the

biomedical signals with a small variation (or

information) rate to reduce the amount of sampled

data. As excepted results, the compression and

decompression blocks of the radio modules will be

removed and the average speed of the converter will

be reduced.

2.2 LC-ADC Principle

An LC-ADC uses level crossing detection to sample

the ECG signal after filtering. The converter’s

architecture shown in Figure 2 consists of two

comparators, a digital-to-analog converter (DAC),

an up/down counter and a time-to-digital converter

(TDC).

Two thresholds levels 



and 



are set to

Smart-BIODEV 2016 - Special Session on Smart Embedded Biomedical Devices for In Situ Physiological Signal Processing

322

Figure 1: Architecture of Wibio’ACT acquisition board.

DAC

Up/Down

Counter

Timer

Comparators

INC

DEC

Figure 2: LC-ADC converter architecture.

identify the signal value 



. These signals are

generated via the digital-to-analog converter with a

difference equal to the quantization step. Thereby, in

digital representation, the difference between 



and





is equal to 1 LSB. The sampling occurs when the

signal value 



is either higher than 



or lower

than 



. In fact, according to the comparison result,

the increment signal (INC) or decrement signal

(DEC) are exclusively set to “1” and consequently

activate the up/down counter. The counter output is

either incremented or decremented by 1 LSB. This

signal controls the DAC whose generated signals,





and 



, will be updated to keep tracking the

input value. LC-ADC samples are equivalent to the

up/down counter output. As long as the signal’s

value 



is between 



and 



, INC and DEC are

both set to “0” and no changes occurs on the counter

or DAC outputs. Thus, no new sample is taken.

The LC-ADC converter adopts 2



1

levels

equally spaced along the input signal range where 

is the converter resolution. Samples













are

generated only when the input signal 



crosses the

defined levels. The time interval, 



, between two

consecutive samples is measured by a timer of

period 



. The conversion results of this LC-ADC

are thus composed of digital codes, 



, for the

voltage magnitude and the time intervals, 



. The

accuracy of data conversion using LC-ADC is

dependent on two main factors; the reconstruction

condition and the timer frequency.

3 LC-ADC DESIGN

CONSIDERATIONS

The LC-ADC samples are non uniformly taken.

Thus, the sampling frequency is variable. The

number of quantization levels and the signal

variations directly influence the number of samples

and so the average sampling frequency. Figure 3

presents results of average sampling frequency for

different ECG signals when varing the resolution

value . The set of ECG signals used in this

simulation is collected through Apnea-ECG

Database (apneaecg), combined measurement of

ECG Database-Breathing and Seismocardiograms

(cebsdb), Long Term ST Database (lgdb), Non-

Invasive Fetal ECG Database (nifecdb) and PTB

Diagnotic ECG Database (ptbdb) (Moody et al.,

2001). The maximum frequency of the ECG signal

max

is fixed to 200 Hz to ensure a good quality of

ECG parameters detection. The given signals are

oversampled and used in digital format in order to

emulate an analog signal. The amplitude signal is

scaled to the LC-ADC full-scale amplitude. The

ECG signal is sampled using the LC-ADC with

different resolution values, . The irregular

sampling instants are stored to compute the average

sampling frequency.

Figure 3: Average sampling frequency versus resolution

for ECG signals.

It can be noticed from Figure 3 that the average

sampling frequency depends of the LC-ADC

resolution . To satisfy Shannon theorem that

imposes the use of a sampling frequency at least

equal to the double of the input frequency, the

Biomedical

sensor

Amplifier

Digital

reconstruction

Filter

DAC

Filter

Power

amplifier

Mixer

Transmitter

Human-Machine Interface

for drawing the biomedical

signal

Receiver

LNA

Mixer

Microcontroller

(µC)

Central Processing

Unit(CPU)

ADC

LC-ADC

2 4 6 8 10 12

Resolution (Bits)

Average Sampling Frequency (Hz)

apneaecg

cebsdb

ltdb

nifecgdb

ptbdb

Accurate Level-crossing ADC Design for Biomedical Acquisition Board

323

minimum required average sampling frequency is

equal to 400 Hz. At this frequency, and according to

the input signal, a minimum of 7 bits is needed to

recover the original signal after LC-ADC

digitization.

The second factor in the LC-ADC design

consideration is the timer frequency. Quantization

noise is added to the signal due to the inexact time of

the sampling instants resulting in jitter noise as

presented by Figure 4.

Figure 4: Quantization error due to inexact sampling time.

This error in time, , causes an error to the

output voltage as in (1), where the derivative of 



with respect to  is the slope of the input signal.





(1)

In this case, the SNR of such system is defined as

the ratio of the signal power, computed according to





value, to the noise power depending of . The

theoretical value for the LC-ADC case can be

computed as in (2), where 



is the timer clock

period which determines  (Allier et al., 2003).































10dB

log

log 10=SNR

(2)

According to (2), the SNR no longer depends on the

bit resolution of the ADC but instead depends on the

timer period 



and the statistical properties of the

input signal 



. In Figure 5, the SNR and the

effective number of bits (ENOB) are plotted versus





for ECG signal.

In the irregular sampling scheme, the SNR can

be computed from the power density spectrum

(PSD) of the reconstructed ECG signal. In this

paper, the cubic spline interpolation is the

reconstruction algorithm. Doubling the counter

frequency results in 6 dB increase of SNR. In fact, to

reach an ENOB of 10 bits, a timer frequency of 71

MHz, equivalent to a period 



of 0.14 s is

considered. Hence, an SNR of 62 dB is ensured by

this LC-ADC. From Figure3, with an ENOB equal

to 10 bits, the average sampling frequency for the

selected database signals is between 3 to 15 kHz.

Figure 5: SNR and ENOB versus timer frequency for ECG

signal.

Additional factors can influence the performance

of the LC-ADC such as the the variation of the

comparator delay, the accuracy of the quantization

levels and the reconstruction errors due to spline

interpolation. Those errors might be taken into

account in future work to avoid SNR degradation

(Ravanshad et al., 2013).

4 SIMULATIONS RESULTS ON

ECG SIGNALS

The LC-ADC can compress the sampled data for the

same level of performance measured by SNR in case

of uniform sampling. It is highly important to

accurately choose the design specifications of the

converter that are evaluated through the quality of

the reconstructed signals. To demonstrate the

efficiency of the proposed design in detecting ECG

signals, the LC-ADC converter was simulated using

a Simulink model for various ECG records obtained

from different databases (Moody et al., 2001). The

simulation parameters are 0.14 s for 



and 10 bits

for the resolution M. Since level-crossing is an

irregular sampling and in order to use the standard

distorsion mesure which is the percentage root mean

square difference (PRD), the signal reconstruction

based on spline interpolation was performed.

Figure 6 shows the temporal representaion of

PTB Diagnotic ECG Database (ptbdb) signal at the

input and the output of the LC-ADC. A

reconstruction scheme is applied on the output, and

presented in the second curve of Figure 6.

In order to measure the signal quality, the PRD

of the reconstrcuted signal is measured. The LC-

Timer Fre

uenc

(

)

Signal-to-Noise Ratio (dB)

ENOB

SNR

ENOB

Smart-BIODEV 2016 - Special Session on Smart Embedded Biomedical Devices for In Situ Physiological Signal Processing

324

ADC ensures a PRD of 1%. For a lifestyle

application, a PRD value lower than 2% is

considered to be adequate as it corresponds to a

‘very good’ reconstruction quality (Zigel et al.,

2000). For the same level of SNR and compared to

uniform sampling, the LC-ADC compresses the

sampled data. As it is show in Figure 7, the LC-

ADC reduces the number of low changing voltage

sample. In fact, the number of samples representing

the small variation between 0.2 and 0 mV are

reduced by almost 44%. The total number of

samples are reduced by 13%.

(a)

(b)

Figure 6: LC-ADC input and output signals (a) and the

reconstructed signal after LC-ADC digitization (b).

Figure 7: Number of samples in the case of uniform

sampling and irregular sampling using LC-ADC.

5 CONCLUSIONS

As demonstrated in this paper, LC-ADC compresses

the signal in case of low changing amplitudes. This

work investigates the use of LC-ADC to digitize the

ECG signals while ensuring an SNR of 62 dB. In

fact, it defines the design specification of the LC-

ADC converter in order to guarentee a good

representation of such vital biomedical signal. To

measure the reconstructed signal quality, the PRD

parameter is selected. The computed PDR is equal of

1%. Such value is considered to be adequate in

biomedical signal processing field. As future work,

the LC-ADC circuit implementation can be realized

in CMOS process for the proposed values of design

specifications. The comparators are hysteretic and

can also be implemented as Schmitt triggers. The

DACs can be implemented by using the capacitors

or the resistor-capacitor hybrid. The digital blocks

are simple and can also be designed off-chip.

In addition, future work is necessary to design

and implement an adaptive LC-ADC architecture

that aims to obtain higher compression rates and

more significant power reduction compared to the

conventional LC-ADC.

ACKNOWLEDGEMENTS

The authors would like to thank the CMCU (Comité

Mixte de Coopération Universitaire) to financially

support this project. The funding program is

Partenariat Hubert-Curien Utique (PHC-Utique)

2015 N°15G1407.

REFERENCES

Ahmed, S. M., Al-Ajlouni, A. F., Abo-Zahhad, M., Harb,

B., 2009. ECG signal compression using combined

modified discrete cosine and discrete wavelet

transforms, Journal of Medical Engineering and

Technology.

Allier, E., Sicard, G., Fesquet, L., Renaudin, M., 2003. A

new class of asynchronous A/D converters based on

time quantization, International Symposium on

Asynchronous Circuits and Systems, IEEE.

Cox, J. R., Nolle, F. M., Fozzard, H. A., Oliver, C. G.,

1968. AZTEC, a preprocessing program for real time

ECG rhythm analysis, IEEE Transactions on

Biomedical Engineering.

Giroud, F., Persechini, P., Ruedi, P.-F., Viemon, F.,

Masson, F., 2014. ECG sensor interface in a low

power SoC for wireless portable ECG monitoring,

Biomedical Circuits and Systems Conference, IEEE.

Guan, K., Singer, A., 2007. Opportunistic sampling of

bursty signals by level-crossing—An information

theoretical approach, 41st Annual Conference on

Information Sciences and Systems.

Gyselinckx, B., Vullers, R.J.M., Van Hoof, C., Ryckaert,

J., Yazicioglu, R.F., 2006. Human++: Emerging

Technology for Body Area Networks, VLSI-SoC.

Hartmann, E., 2003. ECG front-end design is simplified

0,4 0,5 0,6 0,7 0,8

-0.6

-0.4

-0.2

0.2

0.4

Time (s)

Amplitude (mV)

LC-ADC input

LC-ADC output

0.4 0.5 0.6 0.7 0.8

-0.6

-0.4

-0.2

0.2

0.4

Time

(

)

Amplitude (mV)

Reconstructed signal

LC-ADC output

-0.6 -0.4 -0.2 0 0.2 0.4 0.6

1000

2000

3000

4000

5000

Variation (mV)

Number of samples

Irregular sampling

Uniform sampling

Accurate Level-crossing ADC Design for Biomedical Acquisition Board

325

with microconverter, Analog Dialogue.

Kozmin, K., Johansson, J., Delsing, J., 2009. Level

crossing ADC performance evaluation toward

ultrasound application, IEEE Transactions on. Circuits

and Systems I: Regular Papers.

Li, Y., Zhao, D., Serdijn, W. A., 2013. A sub-microwatt

asynchronous level-crossing ADC for biomedical

applications, IEEE Transactions on Biomedical

Circuits and Systems.

Long, Y., Harpe, P., Pamula, V.R., Osawa, M., Harada,

Y., Tamiya, K., Van Hoof, C., Yazicioglu, R.F., 2014.

A 680 nA ECG Acquisition IC for Leadless

Pacemaker Applications, IEEE Transactions on

Biomedical Circuits and Systems.

Mark, J. W., Todd, T. D., 1981. A nonuniform sampling

approach to data compression, IEEE Transactions on

Communications.

Mendis, S., Puska, P., Norrving, B., 2011. Global atlas on

cardiovascular disease prevention and control, World

Health Organization.

Miaou, S.G., Chao, S.N., 2005. Wavelet-based lossy-to-

lossless ECG compression in a unified vector

quantization framework, IEEE Transactions on

Biomedical Engineering.

Moody, G. B., Mark, R. G., Goldberger, A. L., 2001.

PhysioNet: a web-based resource for the study of

physiologic signals, IEEE Engineering in Medicine

and Biology Magazine.

Rajoub, B., 2002. An efficient coding algorithm for the

compression of ECG signals using the wavelet

transform, IEEE Transactions on Biomedical

Engineering.

Ravanshad, N., Rezaee-Dehsorkh, H., Lotfi, R., 2013.

Detailed study of the time estimation in level-crossing

analog-to-digital converters, Iranian Conference

Electrical Engineering.

Sayiner, N., Sorensen, H. V., Viswanathan, T. R., 1996. A

level-crossing sampling scheme for A/D conversion,

IEEE Transactions on Circuits and Systems II, Analog

and Digital Signal Processing.

Zigel, Y., Cohen, A., Katz, A., 2000. The weighted

diagnostic distortion (WDD) measure for ECG signal

compression, IEEE Transactions on Biomedical

Engineering.

Smart-BIODEV 2016 - Special Session on Smart Embedded Biomedical Devices for In Situ Physiological Signal Processing

326