Data Compression for Wireless ECG Devices

Elena Merdjanovska

1,2 a

, Miha Mohor

, Matja

z Depolli

1 b

, Aleksandra Rashkovska

1 c

and Toma

z Javornik

1 d

Department of Communication Systems, Jo

zef Stefan Institute, Ljubljana, Slovenia

zef Stefan International Postgraduate School, Ljubljana, Slovenia

Keywords:

ECG, Wireless, BLE, Compression, Delta Coding.

Abstract:

Wireless ECG devices are the latest novelty in the ﬁeld of electrocardiography. ECG is commonly used in

healthcare systems to observe cardiac activity, however wireless devices bring new challenges to the ﬁeld

of ECG monitoring. These challenges include limited battery capacity, as well as increased data storage

requirements caused by daily uninterrupted ECG measurements. Both of these issues can be mitigated by

introducing an efﬁcient compression technique. This paper explores two direct data compression methods

for ECG data: delta coding and Huffman coding, as well as their variations. We performed experiments

both on measurements from a wireless ECG sensor – the Savvy ECG sensor, as well as on measurements

from a standard public ECG database – the MIT-BIH Arrhythmia Database. We were able to select suitable

parameters for delta coding for efﬁcient compression of multiple ECG recordings from the Savvy ECG sensor,

with a compression ratio of 1.6.

1 INTRODUCTION

Electrocardiogram (ECG) recordings capture the

electric potential on the body surface, which changes

as a result of the electrical activity of the heart (Trobec

et al., 2018b). ECG is the most common and ex-

tensively used vital sign monitoring representation in

modern healthcare systems and various ECG mea-

surement formats exist. The standard 12-lead ECG is

used to provide information on cardiac activity during

a short-term monitoring from 12 different perspec-

tives (leads), whereas Holter ECG records the elec-

trical activity of the heart over longer period of time

(several hours) from 5-7 leads. In addition to these

two methods, which are currently most widely used

in clinical practice, novel small wireless ECG body

sensors are being developed. An example of such a

sensor is the Savvy ECG sensor (Rashkovska et al.,

2020), developed at the Jo

zef Stefan Institute (Fig. 1).

Wireless ECG body sensors, as well as their

broader and more frequent use than traditional ECG-

monitoring methods, pose new challenges. It is well-

https://orcid.org/0000-0003-0794-8438

https://orcid.org/0000-0002-0365-5294

https://orcid.org/0000-0002-2014-8630

https://orcid.org/0000-0002-8676-5658

known that wireless data communication takes up a

large part of the total power consumption in most

portable wireless devices. By compressing the data

prior to wireless transmission, power can be saved as-

suming that the compression operation itself does not

consume too much power. In addition, large volumes

of ECG data are being recorded using these new mea-

surement devices, which must be compressed for ef-

ﬁcient processing and storage. It can be concluded

that compression is a very signiﬁcant topic in com-

putational ECG analysis, most notably in the case of

wireless ECG devices.

The main goal of any compression technique is

to achieve maximum data volume reduction while

preserving the signiﬁcant signal morphology features

upon reconstruction. Data compression techniques

have been utilized in a broad spectrum of commu-

nication areas such as speech, image, and teleme-

try transmission (Jalaleddine et al., 1990). Existing

data compression techniques for ECG signals lie in

two categories: direct data and transformation meth-

ods. Direct data compression techniques for ECG sig-

nals have shown a more efﬁcient performance than

the transformation techniques in regard to processing

speed and compression ratio. Direct data compres-

sors detect redundancies by a direct analysis of the ac-

tual signal samples. In contrast, transformation com-

Merdjanovska, E., Mohor

c, M., Depolli, M., Rashkovska, A. and Javornik, T.

Data Compression for Wireless ECG Devices.

DOI: 10.5220/0010818100003123

In Proceedings of the 15th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2022) - Volume 4: BIOSIGNALS, pages 15-21

ISBN: 978-989-758-552-4; ISSN: 2184-4305

pression methods mainly utilize spectral and energy

distribution analysis for detecting redundancies. Dif-

ferent transforms have been used for this purpose in

the literature, such as Fourier, Wavelet (Rajoub, 2002)

and Cosine transforms (Ranjeet et al., 2011). In addi-

tion, other methods such as compressed sensing (Ma-

maghanian et al., 2011) have been employed for ECG

compression.

Figure 1: Savvy ECG sensor in use.

In this paper, the focus is on novel signals obtained

from a wireless single-lead ECG sensor, namely the

Savvy ECG sensor, where compression options are

limited by the properties of the implemented data

transmission method on the ECG sensor (Vilhar and

Depolli, 2018). The sensor transmits the sampled

data in packets and does not re-transmit the packets

that get lost due to insufﬁcient quality of the wire-

less data connection. Therefore, the compression has

to be resilient to missing data and has to work on

chunks of very limited size. In addition, the calcu-

lations, necessary to perform the compression, need

to be done on an embedded low power device. As a

result, a large portion of state-of-the art compression

techniques, mainly transformation-based techniques,

are unsuitable in this case because of their compu-

tational complexity. Thus, we focus on direct data

compression techniques for ECG: delta coding, and

the most popular type of entropy coding – Huffman

coding. These two methods do not require high cal-

culation power and can be applied even in the case of

lost packets.

The paper will provide an overview of how these

data compression techniques work, as well as how

they perform on example signals from the Savvy ECG

sensor, and also on ECG recordings from a standard

public database. The aim is to assess and discuss

whether and how this techniques could be used for

a mobile ECG device, mainly focusing on compres-

sion prior to transmission, but also for efﬁcient stor-

age. This makes this study very signiﬁcant since it

examines compression with a unique combination of

limitations: compression before transmission, deal-

ing with packet loss and testing on novel ECG body

sensor measurements, all of which are rarely found

in other works. In addition, possible future steps to-

wards efﬁcient transmission of differential ECG from

a wireless sensor will be identiﬁed.

2 SINGLE-LEAD ECG SENSOR

DATA

The Savvy ECG sensor is a body gadget with two

self-adhesive electrodes, powered by a rechargeable

battery. It uses a Bluetooth Low Power (BLE) ra-

dio transceiver for communication (Rashkovska et al.,

2020). The measured ECG signal is a difference be-

tween the electrical potentials of the two electrodes.

The analogue signal is converted into a 10-bit digital

sample, and the signal is then streamed to a personal

digital assistant (PDA), like a smartphone or a tablet,

through the BLE connection. The default sampling

rate was chosen to be 125 samples/s – a compromise

between sustainable power consumption and accept-

able measurement quality. If required, the sampling

rate can be increased up to 1000 samples/s. For this

study, we used 4 measurements sampled at 256 sam-

ples/s, which contained segments with high levels of

noise. For future work, other sampling rates should

also been examined, as well as compression for noise-

ﬁltered signals, since better reconstructed signal qual-

ity could potentially be obtained in that case, using the

same methods. For this study however, we decided to

focus on only one frequency, since the methods pre-

sented are frequency speciﬁc, especially key parame-

ters such as the number of bits needed to represent the

difference values.

As deﬁned in the data stream speciﬁcation of the

PCARD wireless protocol(Trobec et al., 2018a), each

packet transmitted via BLE contains 140 bits data

samples (14 10-bit samples without compression). In

order to compare how well the methods perform on a

standard public ECG database, the experiments were

run on 5 recordings from the MIT-BIH Arrhythmia

Database, which contains recordings with 11-bit res-

olution and sampling rate of 360 samples/s. However,

in the description of the methods in the next sections,

the speciﬁcs and the given examples will be shown

only considering the data from the Savvy sensor.

3 METHODS

In the following section, the direct data compression

methods will be presented. First, delta coding will be

BIOSIGNALS 2022 - 15th International Conference on Bio-inspired Systems and Signal Processing

explained, covering both the one-level variant and the

high-and-low-level variant. For each, the proposed

packet for the compressed data before transmission is

also presented. Furthermore, an overview of Huffman

encoding will also be given.

3.1 Delta Coding

ECG is a waveform signal and successive samples

mostly have small differences between them. Hav-

ing this in mind, delta coding can be applied to re-

duce the dynamic range of original ECG signals. With

delta coding, proposed for the ﬁrst time in (Fang and

Lu, 2018), subsequent encoded samples are gener-

ated from the difference between the current sample

and the previous sample of the original ECG signal.

This difference signal will have a much lower dy-

namic range than the original ECG signal, so it can

be effectively quantized using a smaller number of

bits. An ECG signal contains the P–QRS–T and U

waves, where the minimal dynamic range is the P–Q

or T–U wave, and the maximal dynamic range is the

QRS complex. This compression method has been

extensively used for ECG, commonly combined with

other techniques, such as zero-separation (Fang and

Lu, 2018) and most often Huffman coding (Chang

and Lin, 2010).

In the speciﬁc case of the Savvy ECG signals, it

was shown that the maximal difference can be en-

coded with 9 bits, due to the higher level of noise and

relatively low frequency of 256 samples/s. In the case

of the cleaner MIT-BIH recordings, with lower fre-

quency of 360 samples/s, the largest delta could be

encoded with as few as 8 bits. Having this in mind,

we can conclude that, in general, for each actual dif-

ference in the original analogue signal, for the Savvy

recordings we will always need 1 bit more than for the

MIT-BIH database recordings. The calculation of the

value differences is very simple, requiring only one

pass over all values, which makes it suitable for im-

plementation on an embedded low-power device such

as Savvy.

In the following two sections, different variants of

delta encoding will be presented, with the proposed

packet format for the Savvy sensor.

3.1.1 Single Category

In single-level delta coding, each difference is en-

coded with the same number of bits, i.e., the number

needed to code the largest difference that could ap-

pear in the signal. In the case of the Savvy example

data, this is 9 bits. With higher frequencies, this num-

ber decreases. In the case of simple delta coding, each

packet is 140-bit long. The ﬁrst 10 bits are the anchor

value – the value in the original signal. After that, 130

bits remain, in which around 14 9-bit differences can

ﬁt, making a total of 15 signal samples in one packet,

as opposed to the original 14. This means that the

compression ratio we can expect from this method is

around 1.07.

The anchor value can also be transmitted only

once in the ﬁrst packet, with the following ones trans-

mitting only delta values. This case does not allow for

lost packets, since one lost packet will make all of the

following incorrect. Due to this, we decided to send

an anchor with each packet.

3.1.2 Low and High Categories

In this paper, additional experiments were done with

a variant of delta coding, where two levels of differ-

ences are introduced, as proposed in (Hatim et al.,

2016). The main idea of this scheme is to utilize the

parts of the ECG signal with smaller dynamic range,

by encoding those differences with less bits. This

way two coding categories are created: low encod-

ing, where lower differences in the signal are encoded

with as few bits as possible, and high encoding for the

remaining differences. For each data sample, the dif-

ference is calculated and it is determined whether it

belongs to the low category (lower than some thresh-

old value) or to the high one. The number of bits for

the high category is the same as the one used in single-

category delta encoding – 9-bits in the case of Savvy.

To determine the low-encoding bits, an analysis of

the delta values present in the signal needs to be per-

formed. In Figure 2, the distribution of the absolute

values of the differences present in one Savvy record-

ing is shown. Each bar corresponds to the values that

can be coded with the same minimal number of bits,

in order to see which difference bit-widths are most

common. For example, the sixth bar ranges from 32

to 64, meaning the actual differences falling in that

interval are [-64,-32) and [32,64) and can be repre-

sented with a minimum of 7 bits (2

= 32 + one sign

bit). From the distribution graph, we can see that pos-

sible low encoding bit-widths are 3, 4 and 5. In addi-

tion, we can see that the large delta values very rarely

appear in the signal, which could mean they are a re-

sult of noise. Due to this, it is possible to decrease

the number of bits representing the high category, by

introducing a very small error (loss) in the decom-

pressed signal. This is also true for single-level delta

encoding and this possibility for lossy delta compres-

sion will be explored in Section 5.

In order to differentiate between different coding

categories, the transmitted packet must be structured

with speciﬁc ﬁelds. In Figure 3, the proposed packet

format for 2-level delta encoding is shown. Each

Data Compression for Wireless ECG Devices

Figure 2: Distribution of the delta values in an example

Savvy signal.

packet starts with an anchor value, which is 10-bit

value from the original signal, followed by a window

size, indicating how many consecutive differences of

the current category are given. In this case, we set the

window size to 6-bits because the maximal number

of low-category deltas that can ﬁt in one packet (140

bits) is never more than 41 (when low category is 3

bits). In addition, the interval between neighboring

heartbeats is around 0.8 seconds on average, which

in the case of Savvy data equals to around 200 signal

samples. This means that on average, between two

clean heartbeats, we need to change the category at

least 5 times, each time we construct a new packet,

when in reality the delta category of the actual signal

does not change. The overhead introduced with each

category change is around 5% (7 bits) of the entire

packet, and this overhead is added many more times

than actual changes of category in the signal.

The anchor value in each packet was included as a

safety mechanism in case of a data stream packet loss,

so that the damage does not affect received samples

from other packets. However, this comes at a cost

of not being able to entirely utilize the advantages of

two-level delta encoding, as well as delta encoding in

general.

Figure 3: Packet format of two-category delta coding.

On Figure 4, we can see an example of the distri-

bution of the length of the windows in one Savvy sig-

nal, when two-category delta coding is applied (low:

5 bits, high: 7 bits). We can see that the longest win-

dows (consecutive deltas in the same category) are al-

most always from the low category, while most of the

high-category windows are among the shortest. This

behavior is desired and conﬁrms the potential of two-

level delta coding.

Figure 4: Distribution of the length of the windows in low

and high categories separately.

3.2 Huffman Coding

Huffman coding is an entropy encoding algorithm

used for lossless data compression (Jalaleddine et al.,

1990). It creates variable-length codes with shorter

code words for higher probabilities, and each is rep-

resented by an integer bit number. A Huffman code

is generated by constructing a binary tree. The path

from the root to each leaf in the tree gives the code-

word for the bit sequence of the edges passing through

the path. This type of coding is also resistant against

packet loss, since each value is assigned a predeﬁned

code, independent of the values in the other packets.

4 PERFORMANCE METRICS

In general, compression results in a compromise be-

tween efﬁciency and quality. Due to this, it is neces-

sary to calculate both of them (N

emcov

a et al., 2018).

The main metric used for the evaluation of compres-

sion efﬁciency is the compression ratio (CR). The

compression ratio is deﬁned as:

CR =

original

compressed

(1)

In our case, the original and compressed data can

be measured in the total number of bits, or the number

of BLE packets, as deﬁned in Section 3.1, needed to

transmit the same data. The two compression ratios

should be close in value, which is why only one was

chosen, and the CRs given in Section 5 are the packet-

level compression ratios.

To evaluate the quality of the decompressed sig-

nal, Percentage Root Mean Square Difference (PRD)

was used. PRD takes into account the mean of the

signal and the offset. It is deﬁned as:

PRD =

∑

n=1

[x(n) − x(n)]

∑

n=1

[x(n)]

∗ 100 (2)

BIOSIGNALS 2022 - 15th International Conference on Bio-inspired Systems and Signal Processing

, where x(n) is the original signal and x(n) is the de-

constructed signal.

The main goal of compression is to increase the

compression ratio as much as possible, while keeping

the diagnostic value of the ECG. Thus, small enough

values of PRD do not affect the main ECG character-

istics and, in those cases, the compression does not

reduce the ability to detect heart abnormalities and

arrhythmia from the ECG waveform. For atrial ﬁb-

rillation (AF) detection, it has been shown that PRD

up to 30% is acceptable and does not degrade RR-

interval based AF detection performance (Cervig

et al., 2021). For other tasks such as QRS detection

(Elgendi et al., 2017) and arrhythmia classiﬁcation

(Yildirim et al., 2019), it has been shown that PRD

of up to 0.53% and 0.75% respectively does not inﬂu-

ence classiﬁcation and detection performance at all.

5 EXPERIMENTS AND RESULTS

In this section, the results for a few experimental se-

tups of delta and Huffman coding are presented. In

addition, data from two sources are included: record-

ings from the single-lead Savvy sensor, which is the

main concern of this paper, and recordings from the

standard MIT-BIH Arrhythmia Database, in order to

examine how slightly different recording conditions

affect the compression performance.

In Tables 1 and 2, the compression results from

the standard version of both delta and Huffman cod-

ing are shown. Single-category delta coding is ex-

amined with a few different parameters, more specif-

ically with 6, 7, 8 and 9 bits to represent each differ-

ence value. From the analysis described in Section

3.1.1, we were able to conclude that the maximum

number of bits required to code any delta value in the

Savvy recordings is 9, while in the case of the MIT-

BIH database, it is 8 bits. This can be seen again in

Table 1 for Savvy and in Table 2 for the MIT-BIH

database, where the PRD for the higher bit values is

0. What is interesting to see is that even for the lower

numbers of bits (6 and 7), the PRD is not very high

(up to 5%), while the compression ratio increases.

With single-category coding with 6 bits, a maximum

compression ratio of around 1.6 can be achieved on

Savvy recordings, while 1.77 on MIT-BIH record-

ings. In addition, by examining the location of the

errors between the original and decompressed Savvy

signals, it could be concluded that they largely appear

in the parts of the signal with a lot of noise. This

conﬁrms that the obtained maximum delta value of

9 bits is not representative of the changes in the ECG

waveform, but it was caused by the noisier parts of the

recording, which is why single-category coding with

a lower number of bits for delta is a good solution for

encoding normal ECG.

In addition, in Tables 1 and 2 also Huffman coding

results are shown. It can be seen that Huffman results

are comparable with single-category delta coding re-

sults, especially considering that it enables lossless

compression, which was not entirely the case in delta

coding. However, it is worth noting that the Huffman

coding examined in this paper does not consider the

memory of the device and building of frames. In ad-

dition, each Savvy sensor can have a slightly different

signal baseline, which means that a different Huffman

tree needs to be constructed for each device when us-

ing Huffman to the raw signal values. However, if

we choose to do Huffman encoding on top of delta

(one or two level), this problem could be avoided. The

analysis of the combination of Huffman and delta (ap-

plying Huffman coding to the delta values) is out of

the scope of this paper, but it is a direction worth ex-

ploring in the future.

In Tables 3 and 4, the experimental results for 2-

category delta coding are presented. Different com-

binations of the following parameters of delta coding

were examined: 3, 4 and 5 bits for the low category;

and 7, 8 and 9 bits for the high category. We can no-

tice that the compression ratios depend on the signal

characteristics much more, i.e. less bits does not al-

ways mean better compression. In general, we can see

again that the PRD(%) are very low, since this is an

extension of the single-category delta previously dis-

cussed. The best compression ratios for each record-

ing are highlighted. We can see that the best CRs in

general are obtained with 7 bits for the high category

and either 4 or 5 bits for the low category. In addition,

we were able to conﬁrm that, on MIT-BIH recordings,

better compression ratios could be achieved, due to

the better quality of the signals and higher sampling

rate.

6 CONCLUSIONS

In this paper, an analysis of ECG data compression

methods was given. The main focus was to examine

whether and how two groups of direct data compres-

sion can be applied to ECG signals, more speciﬁcally

recordings from a wireless ECG sensor. Different pa-

rameters for one- and two-category delta coding, as

well as Huffman coding were compared, both accord-

ing to the compression ratios and the quality of the de-

compressed signal. The compression obtained on the

single-lead wireless sensor recordings was compara-

ble using both delta and Huffman coding, in the best

Data Compression for Wireless ECG Devices

Table 1: Compression results on 3 Savvy measurements using 1-level delta coding (6, 7, 8 and 9 bits for delta) and Huffman

coding.

Delta (1-level) Huffman

6 bits 7 bits 8 bits 9 bits

Recording CR PRD(%) CR PRD(%) CR PRD(%) CR PRD(%) CR PRD(%)

Savvy1 1.6176 0.4643 1.3971 0.1830 1.2319 0.0146 1.1029 0 1.5032 0

Savvy2 1.6176 2.0115 1.3971 0.3526 1.2319 0 1.1029 0 1.3746 0

Savvy3 1.6176 0.2368 1.3971 0.0266 1.2319 0 1.1029 0 1.7369 0

Table 2: Compression results on 5 MIT-BIH Arrhythmia measurements using 1-level delta coding (6, 7, 8 and 9 bits for delta)

and Huffman coding.

Delta (1-level) Huffman

6 bits 7 bits 8 bits 9 bits

Recording CR PRD(%) CR PRD(%) CR PRD(%) CR PRD(%) CR PRD(%)

100 1.7718 2.2288 1.5298 0.6799 1.3469 0 1.2055 0 1.706 0

101 1.7718 2.1405 1.5298 0.3453 1.3469 0 1.2055 0 1.5676 0

103 1.7718 4.3631 1.5298 2.0187 1.3469 0 1.2055 0 1.5231 0

105 1.7718 1.0757 1.5298 0.2086 1.3469 0 1.2055 0 1.4179 0

107 1.7718 5.2711 1.5298 2.6601 1.3469 0 1.2055 0 1.1753 0

Table 3: Compression results on 3 Savvy measurements using 2-level delta coding, with different low and high bit-levels

(low: 3, 4 and 5 bits; high: 7, 8 and 9 bits).

low: 3, high: 7 low: 4, high: 7 low: 5, high: 7 low: 3, high: 8 low: 4, high: 8 low: 5, high: 8 low: 3, high: 9 low: 4, high: 9 low: 5, high: 9

Recording CR PRD(%) CR PRD(%) CR PRD(%) CR PRD(%) CR PRD(%) CR PRD(%) CR PRD(%) CR PRD(%) CR PRD(%)

Savvy1 1.17 0.12 1.47 0.09 1.47 0.06 1.11 0.01 1.43 0 1.46 0 1.07 0 1.4 0 1.44 0

Savvy2 1.06 0.23 1.31 0.18 1.4 0.14 1.01 0 1.27 0 1.38 0 0.96 0 1.24 0 1.36 0

Savvy3 1.44 0 1.64 0.01 1.58 0.01 1.4 0 1.62 0 1.57 0 1.35 0 1.59 0 1.56 0

Table 4: Compression results on 5 MIT-BIH Arrhythmia measurements using 2-level delta coding, with different low and

high bit-levels (low: 3, 4 and 5 bits; high: 7, 8 and 9 bits).

low: 3, high: 7 low: 4, high: 7 low: 5, high: 7 low: 3, high: 8 low: 4, high: 8 low: 5, high: 8 low: 3, high: 9 low: 4, high: 9 low: 5, high: 9

Recording CR PRD(%) CR PRD(%) CR PRD(%) CR PRD(%) CR PRD(%) CR PRD(%) CR PRD(%) CR PRD(%) CR PRD(%)

100 1.45 0.46 2.08 0.38 1.81 0.36 1.42 0 2.05 0 1.80 0 1.38 0 2.02 0 1.78 0

101 1.29 0.31 2.07 0.28 1.80 0.23 1.26 0 2.04 0 1.78 0 1.22 0 2.01 0 1.77 0

103 1.25 1.48 2.05 1.25 1.78 1.27 1.22 0 2.02 0 1.76 0 1.18 0 1.98 0 1.75 0

105 1.12 0.19 1.78 0.18 1.70 0.16 1.08 0 1.74 0 1.68 0 1.04 0 1.68 0 1.66 0

107 1.08 2.36 1.40 2.11 1.57 1.53 1.02 0 1.34 0 1.54 0 0.95 0 1.27 0 1.50 0

case around 1.6. This compression ratio corresponds

to PRD of less than 0.5%, which has been shown to

be sufﬁciently low for ECG-based diagnosis.

From the experiments in this study, we were able

to choose one set of parameters which work well on

multiple recordings. This conclusion and parameters

would later need to be conﬁrmed on a larger number

of recordings. Other frequencies and noise levels of

the ECG signal should also be examined. Further-

more, a combination of Huffman and delta coding is

also worth exploring.

REFERENCES

Cervig

on, R., McGinley, B., Craven, D., Glavin, M., and

Jones, E. (2021). The Effects of Compression on the

Detection of Atrial Fibrillation in ECG Signals. Ap-

plied Sciences, 11(13).

Chang, G. C. and Lin, Y. D. (2010). An Efﬁcient Lossless

ECG Compression Method Using Delta Coding and

Optimal Selective Huffman Coding. In Lim, C. T.

and Goh, J. C. H., editors, 6th World Congress of

Biomechanics (WCB 2010). August 1-6, 2010 Singa-

pore, pages 1327–1330, Berlin, Heidelberg. Springer

Berlin Heidelberg.

Elgendi, M., Mohamed, A., and Ward, R. (2017). Efﬁcient

ecg compression and qrs detection for e-health appli-

cations. Scientiﬁc Reports, 7.

Fang, H.-W. and Lu, C.-C. (2018). A real time and lossless

encoding scheme for patch electrocardiogram moni-

tors. Applied Sciences, 8(12).

Hatim, A., Latif, R., El Bachari, O., and Arioua, M. (2016).

Efﬁcient electrocardiogram (ECG) lossy compression

scheme. In 2016 5th International Conference on

Multimedia Computing and Systems (ICMCS), pages

151–156.

Jalaleddine, S., Hutchens, C., Strattan, R., and Coberly, W.

(1990). ECG data compression techniques-a uniﬁed

approach. IEEE Transactions on Biomedical Engi-

neering, 37(4):329–343.

Mamaghanian, H., Khaled, N., Atienza, D., and Van-

dergheynst, P. (2011). Compressed sensing for real-

time energy-efﬁcient ecg compression on wireless

body sensor nodes. IEEE Transactions on Biomedi-

cal Engineering, 58(9):2456–2466.

emcov

a, A., Smisek, R., Mar

anov

a, L., Smital, L., and

Vitek, M. (2018). A comparative analysis of methods

for evaluation of ECG signal quality after compres-

sion. BioMed Research International, 2018.

BIOSIGNALS 2022 - 15th International Conference on Bio-inspired Systems and Signal Processing

Rajoub, B. A. (2002). An efﬁcient coding algorithm for the

compression of ECG signals using the wavelet trans-

form. IEEE Transactions on Biomedical Engineering,

49(4):355–362.

Ranjeet, K., Kumar, A., and Pandey, R. K. (2011). ECG

signal compression using different techniques. In Un-

nikrishnan, S., Surve, S., and Bhoir, D., editors, Ad-

vances in Computing, Communication and Control,

pages 231–241, Berlin, Heidelberg. Springer Berlin

Heidelberg.

Rashkovska, A., Depolli, M., Toma

c, I., Avbelj, V., and

Trobec, R. (2020). Medical-Grade ECG Sensor for

Long-Term Monitoring. Sensors, 20(6):1695.

Trobec, R., Bregar, K., Depolli, M., and Mohor

c, M.

(2018a). PCARD Wireless Protocol. Technical Re-

port ISR2-PWP, Department of Communication Sys-

tems, Jo

zef Stefan Institute.

Trobec, R., Toma

c, I., Rashkovska, A., Depolli, M., and

Avbelj, V. (2018b). Body Sensors and Electrocardio-

graphy. Springer.

Vilhar, A. and Depolli, M. (2018). Time synchroniza-

tion problem in a multiple wireless ECG sensor mea-

surement. In 14th Annual Conference on Wireless

On-demand Network Systems and Services (WONS),

pages 83–86.

Yildirim, O., Baloglu, U. B., Tan, R.-S., Ciaccio, E. J.,

and Acharya, U. R. (2019). A new approach for ar-

rhythmia classiﬁcation using deep coded features and

lstm networks. Computer Methods and Programs in

Biomedicine, 176:121–133.

Data Compression for Wireless ECG Devices