Link-Level Reliability Control for WirelessElectrocardiogram Monitoring in Indoor Hospital

Juyoung Park, Jaemyoun Lee, and Kyungtae KangDepartment of Computer Science and Engineering, Hanyang University, Korea

Email: {parkjy, poken01, ktkang}@hanyang.ac.kr

Abstract—Reliability is an essential quality of safety-criticalwireless systems for medical applications. However, wireless linksare typically prone to bursts of errors, with characteristics whichvary over time. We propose a wireless system suitable for real-time remote patient monitoring in which the necessary reliabilityis achieved by an efficient error control in the link layer. Wehave paired an example electrocardiography application to thiswireless system. We also developed a tool chain to determinethe reliability, in terms of the packet-delivery ratio, for variouscombinations of system parameters. A realistic case study, basedon data from the MIT-BIT arrhythmia database, shows that theproposed wireless system can achieve an appropriate level ofreliability for electrocardiogram monitoring if link-level errorcontrol is correctly implemented.

Index Terms—Wireless healthcare, remote electrocardiogra-phy, service reliability.

I. INTRODUCTION

The most promising example of patient healthcare usingwireless technologies is telecardiology, which involves thetransmission of electrocardiogram (ECG) signals [1], [2]. Apatient’s heart function can be measured in one location, andthe resulting data can be observed directly at another location,provided that the two locations are linked by a robust wirelessnetwork that provides an adequate quality of service (QoS).

The successful maturing of wireless technologies to theextent necessary to support real-time remote cardiographyrelies on the resolution of several challenging research issues.One of the most crucial is how to guarantee the requiredQoS: unlike conventional network applications in which themain goal has been to improve average throughput, ECGmonitoring applications need a moderate data-rate but requirean extremely high level of reliability, because the corruption,or loss of medical data can have fatal consequences.

In fixed networks, made up of stationary hosts and wiredlinks, reliability is achieved by tuning the transport-layer pro-tocols, such as TCP. The adaptive behavior of these protocolsminimizes the end-to-end delays and packet losses causedby congestion, and this works very well on wired networks.However, wireless links have fundamentally different charac-teristics, with much lower bandwidth and much higher error-rates; furthermore errors occur in bursts and the error-ratechanges drastically over time. Due to these much less favorableerror characteristics, transport protocols designed for wirednetworks have been shown to produce poor throughput andvery long delays in wireless links [3], making these protocolsunsuitable for real-time ECG monitoring services.

The problem posed by errors in wireless links has beenstudied by several researchers, and many solutions have beenproposed. In the split-connection approach [3], the end-to-end connection between the host and a mobile device is splitinto two separate connections: a wired connection betweena fixed host and an access point (AP), and a second wirelessconnection between the AP and the mobile device. The advan-tage of the split-connection approach is that it separates flowand congestion control of the wireless link from that of thefixed network, and hence improves the bandwidth seen by thesender. This approach can achieve robust data transmission andhigh end-to-end reliability because it addresses the problem ofthe frequency of packet errors in the second connection, whichis over a one-hop wireless link. By introducing separate andeffective error control in the link layer, the impact of the errorbehavior of the wireless link can be minimized.

One method of link-level error control is to retransmit failedpackets [4], but this can lead to unpredictable packet deliverytimes, which makes it difficult to bound the resulting end-to-end delay. This is unacceptable in real-time ECG application,because of their strict timing requirements. An alternative link-level error control mechanism that is more suitable for real-time wireless telemetry is therefore necessary to achieve betterreliability and a bounded end-to-end delay.

We propose a wireless system architecture that uses a link-level error control scheme based on Reed-Solomon (RS) [5],[6] coding in conjunction with data interleaving to enablerobust real-time patient monitoring. This is an effective methodof dealing with the error bursts that frequently occur incommunicating with mobile devices, and is capable of achiev-ing homogeneous throughput and a bounded delay. We alsopresent a novel recursive analytic model for predicting theperformance of link-level error control in terms of the av-erage error-correction performance and execution delay. Theproposed model is based on a sophisticated description ofthe behavior of the bursts of block errors which arise incommunicating over fading wireless channels with devicesmoving at the limited speeds associated with hospital patients.

As a case study, we have tightly paired a specific ECGapplication with the proposed wireless system, and developeda tool chain which allows us to predict reliability of this pairedsystem, including the packet delivery ratio and the end-to-enddelay, for different combinations of system parameters. Thistool chain also allows us to investigate the effectiveness ofthe proposed wireless ECG monitoring system and its link-

978-1-4673-2713-8/12/$31.00 ©2012 IEEE 000596

Hospital LAN (local area network)

Smart

Device

Switch

Monitoring

station

Intensive care unit Patient room Operating room

Data centerLoad test

ECG

AP

Fig. 1. Service architecture for ECG monitoring in indoor hospital.

control scheme, by observing the ECG signals reconstructedat the remote monitoring station, and by determining mean-square errors (MSE), which quantify the difference betweenthe reconstructed ECG signal and the original signal obtainedfrom the patient.

The major contributions of our work are as follows:

• We accurately assess the performance of a wireless ECGsystem using the Monte Carlo method. The use of realECG recordings from the MIT-BIH database [7] allowsus to form a realistic assessment of the reliability ofthe proposed wireless ECG system before deployment.This early-phase analysis is crucial because of the criticalnature of cardiography applications.

• Our tool chain allows us to determine an appropriatetradeoff between reliability and end-to-end delay by vary-ing the level of block interleaving, so as to achieve thereliability needed for ECG applications. We show that thereliability that can be achieved with our wireless systemis appropriate for remote ECG monitoring services.

II. A WIRELESS SYSTEM FOR WIRELESS ECGMONITORING IN INDOOR HOSPITAL

Fig. 1 shows the infrastructure-oriented service architecturefor wireless ECG monitoring in indoor hospital that we willuse in this study. The ECG equipment attached to a patientcommunicates with an AP over a single-hop wireless link, andthen the AP relays the ECG data to the monitoring station inindoor hospital, through the hospitals wired network. Fig. 2shows proposed protocol layers and the interactions betweenthem required for reliable wireless telemetry. The data-linklayer, which includes the logical link control (LLC) and MAClayers, provides error control and media access control forreliable communications. In the proposed wireless systemarchitecture, an RS code in the LLC layer is used for reliabledata exchange. Note that the detailed operation of the network-layer and transport-layer protocols will not be considered.

A. Logical link control structure

The RS encoder takes a block of digital data and addsredundant bits which allow it to compensate for errors causedby noise and during transmission. The RS decoder processeseach block and attempts to correct errors and recover the

Physical layer

(coding and modulation)

LLC layer

(error control)

MAC layer

Application layer

Data-link layer

Peer-to-peer protocol

Medical device (sender) Monitoring station (receiver)AP

Transport layer

(RTP, TCP, or UDP)

Application layer

Transport layer

(RTP, TCP, or UDP)

Peer-to-peer protocol

Wireless linkWired infrastructure

Patient

Physical layer

(coding and modulation

Network layerNetwork layer Network layer

LLC layer

(error control)

Data-link layer

Physical layer

MAC layer

Data-link layer

MAC layer

Fig. 2. Proposed protocol layers for data exchange between medical devices.

original data. The number and type of errors that can becorrected depends on the characteristics of the RS code. AnRS code is specified as a tuple (N,K) with s-bit symbols.This means that the encoder takes K data symbols of s bitseach and adds N −K parity symbols of s bits each to makean N -symbol codeword. An RS erasure decoder can correctup to t = R erasures.

The RS coding is used in conjunction with the interleavingtechnique [8], which is an efficient and frequently used methodof combating bursts of errors. A special memory called theerror control block (ECB) is required for this purpose. TheECB is a matrix with N rows, and a sequence of RScodewords, each s bits wide and encoded with the outer RScode of (N,K), spans the columns in M sub-blocks, each ofwhich is Lu bits wide. M is the number of MAC-layer packetsin each ECB row. In block interleaving, the codewords areplaced in the columns of an array, which is then transmittedrow by row. Hence the value of M determines the level ofblock interleaving. As M increases, the error bursts that occurduring data transmission are interleaved more effectively,which favors RS decoding and reliable communication.

B. MAC structure

The MAC protocol specifies the reverse data channel usedby a mobile device to transmit data to the AP. We assume thatthe reverse link has a slot structure in which each slot lasts1.67ms. Multiple streams of data from a single device canbe simultaneously transmitted over a single communicationchannel by interleaving pulses representing bits from differentsub-channels or time-slots. The time domain is divided intoa sequence of time-slots of equal length, one for each sub-channel. A byte or data-block on sub-channel 1 is transmittedduring time-slot 1, a data-block on sub-channel 2 during time-slot 2, and so on. After transmission of the last sub-channelhas taken place, the cycle starts all over again. This formof multiplexing can guarantee the delay bound required forclinical data by assigning a dedicated sub-channel to eachapplication.

978-1-4673-2713-8/12/$31.00 ©2012 IEEE 000597

III. PAIRING ECG TO PROPOSED WIRELESS SYSTEM

In an electrocardiogram (ECG) application, electrodes (sen-sors) are attached to a patient’s body. The signals, whichtrack the polarization and depolarization of muscle cells duringa heartbeat, and are continuous waveforms, which must beperiodically sampled and digitized for transmission to a mon-itoring station. The choice of sampling frequency, digitizationmethod, and compression technique plays a critical role indetermining the traffic characteristics during transport. If theelectrical signals from the heart are sampled by Nv channels,and the signal from each channel is digitized at a rate of rsamples per second with a sample size of S bits, then theresulting ECG application data traffic requirement μe willbe (NvrS) bits per second. For example, a 12-lead ECG,which uses eight channels, has a data-rate of 64kb/s whenthe sampling rate is 500Hz and the size of each sample is 16bits. More detailed characteristics of an ECG application aregiven in Table I of [9], and we will use this application as acase study to analyze the data traffic required to achieve anacceptable reliability.

IV. DEVELOPMENT OF A TOOL CHAIN FOR RELIABILITY

ANALYSIS

A. Error statistics in data transmission over fading channels

Zorzi et al. [10] investigated the behavior of the blockerrors which arise in data transmission over fading channels,and showed that a Markov approximation is a good modelfor a range of modulation schemes, block lengths, and errorcorrection capabilities. It has been shown [10] that Markovtransition probabilities are largely insensitive to parameterssuch as block length, rate of inner coding and modulationformat, and depend largely on the steady-state packet error-rate(PER), ε, and the Doppler frequency normalized to the blocktransmission time (i.e., V = fDLphy/μp, where fD is theDoppler frequency, Lphy is the size of a physical-layer frame,and μp is the reference data-rate of the physical channel).

B. A tool chain for reliability analysis

We designed the tool chain, shown in Fig. 3, to analyze thereliability of a wireless medical system. It is written in Java,and provides a convenient way of deriving reliability for ECGapplications from system parameters and outputs.

The tool chain accepts system parameters input from theuser, through a graphical interface, and provides appropriatedefaults. The input control module distributes those parametersto each part of the tool chain. To accommodate Nv-channelECG, we require Nv ECBs to perform RS encoding at thesending device worn by the patient, and these Nv ECBs aremultiplexed by a MAC simulator for transmission on thewireless channel using the multiplexing structure describedin Section II-B. Then the input control module passes theparameters needed to obtain the Markov state-transition prob-abilities to the Mathematica kernel through J/Link which is atoolkit that connects Mathematica and Java. The Markov state-transition probabilities are used to approximate the behaviorof the frame errors in the physical layer which are caused

Lphy, s, N, K, Ns

Markov model

simulation

RS erasure encoding

and interleaving

Reconstructed

ECG signals

Nv-channel ambulatory

ECG recordings

De-multiplexing

Channel 1

Channel 2

Channel N l

Input

control

Mathematica

kernel

(1) Pattern of

frame errors

Output

control

LLC simulation

(2) LLC simulation

Lphy, s, N, K, Ns

(4) Post-processing

(MIT/BIH

arrhythmia database)

RS erasure decoding

and de-interleaving

(3) Residual pattern of ECG

packet errors per channel

QoS results

(reliability and

latency)

Input

Output

ECBs

ECBs

ECBs

Construct ECBs

Multiplexing

MAC simulationµu

η

ECBs for channel 1

ECBs for channel 2

ECBs for channel Nv

Construct ECBs

ε, v, fc, µp

Nv, r, S .

Fig. 3. Structure of the proposed tool chain for reliability evaluation.

by a fading channel ((1) in Fig. 3). The part of the LLCsimulation which models the RS decoder uses this pattern offrame errors. Next, the LLC simulator mimics the iterations ofthe RS decoding process and interleaving needed to deal withan ECB, using a Monte Carlo method ((2) in Fig. 3). Thisyields the number of residual packet errors and their patternin that ECB after RS decoding.

C. Post-processing: reconstruction and evaluation of ECGgraphs

The result of LLC simulation is a number of residualpacket errors and their locations after RS decoding, and thesewere subsequently injected into the MIT-BIH arrhythmia datathat we used ((3) in Fig. 3). Finally, the reconstructed ECGsignals were evaluated ((4) in Fig. 3) by determining themean-square errors, which quantify the difference between thereconstructed ECG signals and the original signals obtainedfrom the patient. The MSE for the time interval t can beestimated as 1

ξt

∑ξti=1(si− si)

2, where si is the digitized valueof the ith ECG signal obtained from the patient and si is thedigitized value of ith ECG signal reconstructed at the remotemonitoring station after wireless transmission. Recall that ξ isthe number of ECG samples per channel per second.

The MIT-BIH arrhythmia database [7] contains 48 half-hour excerpts of two-channel ECG recordings, obtained from47 ambulatory subjects studied by the BIH Arrhythmia Lab-oratory. Although it was originally created as standard testmaterial for evaluation of arrhythmia detectors, it is by far themost widely used data for testing and comparing proposedstrategies for real-time ECG services. We selected the data-stream ‘100.dat’ from the database and we assume that it istransmitted without compression.

V. RELIABILITY ANALYSIS WITH EXAMPLE SYSTEM

PARAMETERS

Using the parameters that are summarized in Table I,Table II shows how the reliability depends on the level of blockinterleaving used in the data-link layer. More interleavingincreases reliability and Table II suggests that values of M

978-1-4673-2713-8/12/$31.00 ©2012 IEEE 000598

TABLE ISYSTEM PARAMETERS USED IN THE CASE STUDY.

Symbol Value(s) DescriptionNetwork parameters

Tslot 1.67ms Slot durationNumber of sub-channels in a reverse dataη 8channel

Lphy 128 bits Length of a physical-layer frameLu 120 bits Length of an ECG payload in a frame

Physical channel parametersfc 1.8GHz Carrier frequencyμp 76.8kb/s Reference channel data-rateε 0 ∼ 0.03 Steady-state PERV 0.0444 ∼ 0.0888 Degree of fading

Data processing parameterss 8 bits Length of a symbol

(N,K) (16,12) Example RS codeM 1 ∼ 16 Number of MAC-layer packets in an ECB row

Wireless ECG parametersNv 8 Number of channelsr 400Hz Samples per channel per secondS 16 bits Sample sizeμe 51.2kb/s Data-rate for the ECG application

TABLE IITHE RATE OF RESIDUAL PER IN THE WORST CASE.

Level of block interleaving (M)1 2 4 8

Maximum rate ofresidual PER

0.00384 0.00059 0.00005 0.00003

greater than 7 satisfy an allowable PER of 10−4 in an ECGapplication. Fig. 4, showing results for Lead V, demonstratehow the residual rate of ECG packet errors directly affects thequality of ECG signals reconstructed at the remote monitoringdevice. Lead V5 is one of the precordial leads that is placeddirectly on the chest.

We can infer from these results that the ECG signals thatwould be seen by physician will be less distorted by theoccurrence of frequent error bursts during transmission as thelevel of block interleaving increases, which also leads to steepreductions in the resulting value of MSE. It is noticeable that,when sufficient interleaving is applied (e.g., when M = 4),the MSE of the reconstructed ECG signal is near zero, andthus it is almost identical to the original ECG signal obtainedfrom the patient.

VI. CONCLUSIONS AND FUTURE WORK

We have proposed a wireless system for remote cardiog-raphy, in which reliability is achieved by FEC-based errorcontrol with interleaving in the data-link layer. We havepredicted the behavior of the proposed wireless electrocardiog-raphy system in terms of reliability, for various combinationsof system parameters, using a tool chain that we developedfor this purpose. Realistic simulations using ECG recordingsobtained from cardiology patients highlight the potential ofthe proposed wireless system to provide a reliable ECGmonitoring service.

-1

-0.5

0

0.5

1mV

(a) Original ECG signal obtained from patient.

-1

-0.5

0

0.5

1

MSE=0.002035

mV

(b) Reconstructed ECG signal without error control in theLLC layer.

-1

-0.5

0

0.5

1

MSE=0.000827M=1

mV

-1

-0.5

0

0.5

1

MSE=0.000234M=4

mV

(c) Reconstructed ECG signal with the proposed error controlin the LLC layer for varying values of M .

Fig. 4. Analysis of an ECG signal (lead V5) over 10 seconds (t=10), for aselected interval that contains many error bursts (v=2km/h, PER=0.03).

REFERENCES

[1] K. Kang, K.-J. Park, J.-J. Spng, C.-H. Yoon, and L. Sha, “A Medical-Grade Wireless Architecture for Remote Electrocardiography,” IEEETransactions on Information Technology in Biomedicine, vol. 15, no. 2,pp. 260–267, Mar. 2011.

[2] D. Cypher, N. Chevrollier, N. Montavont, and N. Golmie, “Prevailingover Wires in Healthcare Environments: Benefits and Challenges,” IEEECommunications Magazine, vol. 44, no. 4, pp. 56–63, Apr. 2006.

[3] C. Parsa and J.J. Garcia-Luna-Aceves, “Improving TCP Performanceover Wireless Networks at the Link Layer,” Mobile Networks andApplications, vol. 5, no. 1, pp. 57–71, Apr. 2004.

[4] J. Sarkar , S. Sengupta , M. Chatterjee , and S. Ganguly, “DifferentialFEC and ARQ for Radio Link Protocols,” IEEE Transactions onComputers, vol. 55, no. 11, pp. 1458–1472, Nov. 2006.

[5] K. Kang, “Probabilistic Analysis of Data Interleaving for Reed-SolomonCoding in BCMCS,” IEEE Transactions on Wireless Communications,vol. 7, no. 10, pp. 3878–3888, Oct. 2008.

[6] K. Kang, J. Park, H. Shin, “On the Effect of Reed-Solomon Codingwith Maximum Block Interleaving on MPEG-4 FGS Video Quality in3G Broadcasting,” Simulation Modelling Practice and Theory, vol. 17,no. 3, pp. 504-512, Mar. 2009.

[7] G.B. Moody and R.G. Mark, “The impact of the MIT-BIH Arrhyth-mia Database,” IEEE Engineering in Medicine and Biology Magazine,vol. 20, no. 3, pp. 45–50, 2001.

[8] K. Kang and L. Sha, “An Interleaving Structure for Guaranteed QoS inReal-Time Broadcasting Systems,” IEEE Transactions on Computers,vol. 59, no. 5, pp. 666-678, May 2010.

[9] N. Golmie, D. Cypher, and O. Rebala, “Performance Analysis ofLow Rate Wireless Technologies for Medical Applications,” ComputerCommunications, vol. 28, no. 10, pp. 1266–1275, June 2005.

[10] M. Zorzi, R.R. Rao, and L.B. Milstein, “Error Statistics in Data Trans-mission over Fading Channels,” IEEE Transactions on Communications,vol. 46, no. 11, pp. 1468–1477, Nov. 1998.

978-1-4673-2713-8/12/$31.00 ©2012 IEEE 000599