IETE Journal of Research

ISSN: 0377-2063 (Print) 0974-780X (Online) Journal homepage: https://www.tandfonline.com/loi/tijr20

An Intelligent and Power Efficient Biomedical

Sensor Node for Wireless Cardiovascular Health
Monitoring

Soumyak Chandra, Rajarshi Gupta, Saurav Ghosh & Sanjoy Mondal

To cite this article: Soumyak Chandra, Rajarshi Gupta, Saurav Ghosh & Sanjoy Mondal (2019):
An Intelligent and Power Efficient Biomedical Sensor Node for Wireless Cardiovascular Health
Monitoring, IETE Journal of Research, DOI: 10.1080/03772063.2019.1611489

To link to this article: https://doi.org/10.1080/03772063.2019.1611489

Published online: 16 May 2019.

Submit your article to this journal

View Crossmark data

Full Terms & Conditions of access and use can be found at

https://www.tandfonline.com/action/journalInformation?journalCode=tijr20
IETE JOURNAL OF RESEARCH
https://doi.org/10.1080/03772063.2019.1611489

An Intelligent and Power Efficient Biomedical Sensor Node for Wireless

Cardiovascular Health Monitoring
Soumyak Chandra 1 , Rajarshi Gupta 1 , Saurav Ghosh 2 and Sanjoy Mondal2
1 Electrical Engineering Section, Department of Applied Physics, University of Calcutta, Kolkata 700009, India; 2 AK Choudhury School of
Information Technology, University of Calcutta, Kolkata 700098 India

ABSTRACT KEYWORDS
Wireless human health monitoring using different techniques has been a popular research area in Biomedical sensor node;
the last few decades. In this paper, we describe the development of a biomedical sensor node (BSN) intelligent routing; latency;
for short-range monitoring of static cardiovascular patients using a supervisory computer. In partic- power efficiency; reliability;
ular, the present research focuses on developing (a) an intelligent hybrid routing algorithm which wireless patient monitoring;
ZigBee
takes care of intermediate node(s) failure condition; (b) improved packet delivery ratio with the reli-
ability and efficient use of BSN’s battery power. A data compression algorithm in the BSNs further
enhanced the communication channel efficiency. With 15 BSNs hardware implemented and tested
over a floor area of 93 m2 to collect electrocardiogram and finger pulse signals, we achieved aver-
age packet reception accuracy over 98.45% taken cumulatively over 200 trails, along with a latency
of 1.58 s per data packet with three faulty BSNs and maximum number of allowable transmission
attempts in each hop. A simulation study with the first-order radio model, using 100 BSNs under nine
faulty nodes and worst-case scenario, yielded a node power consumption of 3.49 mW per packet and
a latency of 22 s per data packet. The performance results are superior to those of similar published
works on wireless biomedical health monitoring using wireless sensor networks. The developed sys-
tem can provide a low-cost solution for patient monitoring at indoor hospital wards in developing
nations like India.

1. INTRODUCTION trade-off between the radio part and signal conditioning

Wireless medical health monitoring has been one of the in a sensor node.
key areas of research for electrical and electronic engi-
neering during the last decades. The application can be Over the last decades, extensive research has been car-
broadly divided into three categories, viz., (a) remote pre- ried out for monitoring multiple patients over a wireless
and post-hospital patient monitoring using high-speed sensor network [7]. In such applications, ZigBee proto-
dedicated data link [1]; (b) monitoring of elderly patients col (IEEE 802.15.4) became very popular, due to some
under regular activities [2], and (c) multiple patient mon- of its attractive features like low power ( ∼ mW), mod-
itoring in hospital wards [3]. Typically these systems erate data delivery (250 kbps) and in-built encryption
consist of three modules, viz., patient data acquisition mechanism [8]. The three main constraints in wireless
unit(s), communication unit and receiving end computer. sensor network (WSN) applications are energy efficiency,
A prime research thrust has been focused on the develop- latency and node failure. The medium access control
ment of miniature and power efficient biomedical sensor (MAC) layer of ZigBee can be customized to accomplish
nodes for increased patient comfort and mobility [4–6]. several goals like optimal power consumption, transmis-
The main feature of such developments is integration of sion delay and data reliability [9], which are interrelated
signal conditioning, microcontroller and wireless mod- and often contradictory parameters. Low-energy adap-
ule in a single circuit board with the requirement of tive clustering hierarchy (LEACH) [10] is one pioneering
low power. In [4], a 100 μW ECG acquisition circuit is approach that proposed data aggregation before trans-
proposed. A low-power heart rate and respiratory rate mission to a base station (sink) using a time division
estimation sensor is described in [5] with dual Blue- multiple access (TDMA) scheme decided by the cluster
tooth and ZigBee interface used for driver drowsiness head for each cluster node. An improved scheme, named
detection. The work in [6] provides applications of fur- power efficient gathering in sensor information systems
ther integration of radio part and power consumption (PEGASIS), [11] provided a chain-based protocol that

© 2019 IETE
2 S. CHANDRA ET AL.: AN INTELLIGENT AND POWER EFFICIENT BIOMEDICAL SENSOR NODE

enables data aggregation at each hop between neigh- implemented using 15 BSNs within a hospital ward floor
bouring nodes, reducing the overall data traffic to base area of 93 m2 . We also describe the latency and energy
station and energy consumption. By simultaneous trans- consumption model under a simulation platform, con-
missions between nodes with code division multiple sidering a 100 node network.
access (CDMA), the latency can also be reduced using
this scheme. Another important factor is the retrans-
2. METHODOLOGY
mission attempts with acknowledgement between neigh-
bouring nodes. This ensures better quality of service The complete monitoring system consists of a group of
(QoS), but contributes to network delay and energy con- BSNs, each dedicated to one patient, one master sen-
sumption over each retransmission attempt. In [12], an sor node (MSN) and one supervisory computer (Host
anycast routing strategy is proposed that utilized on- PC). The MSN is not attached to any patient and consists
demand distance vector (AODV) principle to automat- of all components as that in BSN except the biomedical
ically detect the neighbouring active nodes in a Zig- sensors. When patient data are requested from a partic-
Bee network. The technique achieved reduced trans- ular BSN, the command is initiated from the Host PC. It
mission delay over multicast/broadcast approaches for can acquire data from one patient (BSN) at a time. The
fall detection of elderly patient(s). An efficient power- Host PC acts as the central repository of the data col-
saving strategy and synchronization with master some- lected from the individual BSN. ECG and PPG data are
times adopted by modulating the duty cycle of the nodes. finally received through the USB port in a compressed
A distributed slot assignment scheme in ZigBee tree format using a ZigBee module a simple 8-bit Universal
network is reported in [13] to simultaneously address Asynchronous Receiver Transmitter (UART) format and
two issues, viz., node latency and power consumption. stored in a text file.
Contention-based protocols, named carrier sense mul-
tiple access/collision avoidance (CSMA/CA), are also
2.1 Biomedical Sensor Node
to enable only one node to communicate at a time
based on the presence of carrier signal [14]. Using tiny- The functional block diagram of a BSN is shown in
sized packet this mechanism can also reduce delay in a Figure 1. It consists of a single-lead electrocardio-
small network. Few notable approaches are described for gram (ECG-Einthoven’s lead I) and a photoplethysmo-
energy optimization using hybrid unequal clustering [15] gram (PPG) sensor. We used AD8232 fully integrated
and flexible data transfer time slots among nodes to adapt single-lead front end from analogue devices [22] and a
dynamic traffic pattern in a network [16]. transmission-type PPG sensor with signal conditioning
from Kyoto Electronics [23]. For ECG, a 40 Hz fourth-
In the context of biomedical health monitoring, most of order Butterworth filter was used additionally.
the reported works addressed node failure challenges,
low latency in emergency condition and energy opti- The laboratory prototype printed circuit board (PCB)
mizations [17–19]. The present research is an extension layout, shown in Figure 1(b) and (c), comes in a size
of our previous works [20] and [21], where we devel- 6 cm × 8 cm. The ECG and PPG signals were sampled at
oped a short- range peer-to-peer connectivity between 250 and 60 Hz, respectively, both at 10-bit resolution. An
the patient and a host computer for ECG acquisition ATmega328 microcontroller (MCU) in the BSN acquires
using a telephone line and a wireless system, respectively. ECG and PPG, storing them in the SD card, compresses
In the present work, we report a networking scheme the data, and also controls the XBee s1 module’s MAC
where multiple patients can be monitored over short layer for data transmission and command reception [24].
range. Towards this objective, a biomedical sensor node The BSN is powered by a standalone 2200 mAh Li-Ion
(BSN) was developed which can acquire and transmit battery.
the cardiovascular signals (in compressed format) using
a wireless network to a local host. The routing is a combi- 2.1.1 Signal Conditioning of ECG and PPG Signals
nation of broadcast and unicast techniques. In particular, The schematic circuits for signal conditioning for ECG
we focus on three key aspects for an efficient system signals are shown in Figure 2. Figure 2(a) details power
design and implementation. First, a “fail-safe” network conditioning circuit of the entire BSN board for the
architecture using an intelligent routing algorithm with OPMAPs, MCU and XBee with a 3.3 V supply. The out-
regard to data collection; second, enhancing node power put of the AD8232 is further amplified and filtered in
efficiency to deliver data packets to neighbouring BSNs; two successive stages, stage 1 and stage 2, each having
and third, low latency of data collection at the super- second-order low-pass Butterworth filters as shown in
visory computer. The hardware testing was successfully Figure 2(b) and (c), respectively. The cut-off frequency
 S. CHANDRA ET AL.: AN INTELLIGENT AND POWER EFFICIENT BIOMEDICAL SENSOR NODE 3

(a)

(b)

(c)

(d)

Figure 1: BSN: (a) functional diagram; (b) top side PCB layout and
(c) bottom side PCB layout
Figure 2: ECG signal conditioning circuit: (a) power conditioning;
of the filter in each stage is given by (b) filter stage 1; (c) filter stage 2; (d) DC biasing circuit

1
fc = (1)
2π(RC)
2.1.2 Mode of Operation of Hybrid ZigBee Routing
where R = R2 = R1 = R5 = R6; C = C1 = C2 = C3 Technique
= C4. The total gain achieved in stage 1 and stage 2 is The activities of the BSN under “active mode” can be
given as divided into “initialization session” and “data transfer
session”. The objective of “initialization session” is to
  
R3 R8 know the actual number of available BSNs, synchronize
G = G1 × G2 = 1 + 1+ (2) them (in case they are powered on a different times) and
R4 R7
setting the routing path of each under the guidance of the
The final stage, as shown in Figure 2(c), provides a DC MSN. The sequence of operation is given as
bias to get unipolar output. The objective of the signal
conditioning circuit is to the present the ECG and PPG (1) the Host PC generates a “sync” pulse directed to the
at the input of the ADC channel of AT328 in the range MSN, which in turn, generates a global “HELLO”
0–3.3 V. message in the broadcast mode to all BSNs;
4 S. CHANDRA ET AL.: AN INTELLIGENT AND POWER EFFICIENT BIOMEDICAL SENSOR NODE

(2) the BSNs wait for this pulse after being powered on.
Upon receiving this “HELLO” message, the BSNs
respond with their relative position to the MSN;
(3) the MSN sets the routing path by checking the
receiving signal strength from the individual BSNs
and forms an “address stack”;
(4) the MSN supplies the respective routing paths for all
the individual BSNs, one by one, through handshak-
ing mode (unicast routing);
(5) finally, this address stack along with the participating
BSN is sent to the Host PC. The Host PC refreshes its
user interface with available BSNs highlighted on the
front panel;
(6) if the data transfer request is not initiated from the
Host PC within a certain time, all BSNs go to the
“cyclic sleep mode”.

The algorithm flowchart for describing BSN activity is

provided in Figure 3. The sequence of events is illustrated
in Figure 4, using a 15-node network, and assuming that
BSN#6 and BSN#11 are switched off. Thus, the remaining
13 nodes will be called participating sensor nodes (PSN).
For illustration, we consider N#3 is the requested BSN,
where from the ECG and PPG are to be collected. The
format of the transmitted packet from an individual BSN
is given as

|Packet Sr. No.| patient ID| compressed 40 bytes ECG

data| compressed 10 bytes PPG data| 2 bytes cyclic redun-
dancy check (CRC) check. The steps for data transfer
session are given with reference to Figure 4:

(1) The Host PC directly initiates a command to the Figure 3: Typical activity of a BSN after being powered on
requested node (here N #3), shown in path C1.
(2) The N#3 informs MSN about this request to prevent At step 4 mentioned above, a compression algorithm was
the Host PC to initiate any further request to any applied to reduce the data volume further so that trans-
other BSN till the current session is completed. This mission time can be reduced. For ECG and PPG data
is shown in path 2. stored at different memory blocks, a zonal complexity-
(3) The MSN puts the N#1, N#2, N#8 and N#9 in “cyclic based modified delta compression technique was applied.
sleep mode” using unicast addressing, as they are not The detail of the algorithm is available in [25,26]. The
in the routing stack of N#3. intelligent routing feature was implemented in each BSN
(4) ECG and PPG data are collected at N#3 and then to cater to the two unpredictable situations. First, the
compressed data packets are routed through the path BSNs which are not in the PSN map during the initial-
N#3 → N#4 → N#5 → N#12 → N#13 → N#7 → ization session can participate in the next session if they
N#14 → N#15 → Host PC (shown in dashed arrows). are available/powered on later. Thus, average hopping
The algorithm steps for routing at any intermediate distances between neighbouring nodes are reduced, and
BSN are shown in Figure 5. At each stage, a 2-byte so the power consumption. Secondly, a sudden node(s)
CRC is carried out to ascertain the data integrity. failure condition during a “data transfer session” can be
(5) After the data transfer is successfully over, The BSN# counteracted by automatic diversion of the hopping path
3 informs MSN (shown in path 5). of the data packet(s). This is illustrated using Figure 4
(6) The MSN again generates a “HELLO” message in with dashed and dotted arrows. We assume that N#4 and
broadcast mode to all MSNs, which puts them in thereafter N#13 fail during the current data transfer ses-
“cyclic sleep mode”. sion from N#3. The data originating node N#3 diverts
 S. CHANDRA ET AL.: AN INTELLIGENT AND POWER EFFICIENT BIOMEDICAL SENSOR NODE 5

Figure 4: A data transfer sequence from BSN #3 illustrated using a 15-node network

Figure 5: Schematic flowchart of the intelligent routing algorithm at kth node

the routing path to N#10, being the next neighbouring number of nodes in address stack of kth.BSN, g = M − k.
node. When N#13 goes off, similarly, N#12 automatically There are two tasks of an intermediate routing node, viz.,
diverts the next hopping path to N#7. receiving a data packet from the previous BSN (shown
in path a, b, c), checking the CRC and then passing the
The routing algorithm to transfer a single data packet same to the next node (shown in path c, d, e . . . .g).
in any intermediate node (say kth) in an M-hop (con- At each node, CRC check (CRC ACK) is done first to
sidering the available nodes in the initialization session ensure data reliability. For ease of discussion, we assume
and 1 < k < M) data path is illustrated in Figure 5. The that the packet is successfully received at kth node from
6 S. CHANDRA ET AL.: AN INTELLIGENT AND POWER EFFICIENT BIOMEDICAL SENSOR NODE

the preceding one. For onward transfer to the following The energy term in Equation (3), ETx−elec (m) repre-
node, if the CRC ACK is not received from the (k + 1)th sents the amount required for processing m-bit data and
node within 100 ms, then the kth node tries to send the ETx−amp (m,d) is required to amplify the signal so that
same packet again (shown as d, e, f, g, and d path). After it is decodable at a distance d in the Rx. At the Rx side
three unsuccessful attempts, it transfers the next avail- the received signal must be processed incurring a loss
able node from the address stack (shown as h, d path). of ERx−elec (m) as shown in equation (4). The constant,
Only when the data packet is successfully received at the namely εfs (typical value of 10 pJ/bit/m2 ), depends on the
(k + 1)th node, the final ACK is sent to (k − 1)th node by environmental parameters and channel encoding tech-
the kth node with a request (shown as g–i path) to trans- niques used. Every time a data packet is transmitted
fer next data packet. This ensures that sudden failure of across a hop, it is processed at the intermediate node. So
kth node will not incur any loss of data packets. This pro- the energy loss (Qd ) to process and transmit a 50-byte
vides a “fail-safe” architecture to the network as far as an data packet across a hop distance d is given by
intermediate node failure is concerned. During the data
transfer session, the kth node (k can be any intermediate Qd = 400Eelec + 400εfs d2 (5)
routing node) always checks the acknowledgement from
the (k + 1)th node three times and forwards the packet to 2.2.2 Latency Estimation
the (k + 2)th node as per its routing address stack. This The system latency gives the response time which is
also ensures the shortest path hopping of data packets, estimated in terms of the following parameters:
leading to minimum power consumption.
data packet size S = 50 bytes,
number of data packets in a patient’s message M = 26
2.2 Energy Consumption and Latency Estimation packets = L (say),
in BSN processing and transmission time of a data packet across
A first-order energy model, as detailed in Section 2.2.1, is a hop = Xt ,
used to account for the transmitter (Tx) and receiver (Rx) number of times a packet is transmitted = Nt ,
energy losses in BSNs. A brief theoretical background successful packet transmission probability across a
is outlined in Sections 2.2.2 and 2.2.3 for estimating hop = p,
data latency and energy loss, respectively. Our analysis maximum number of attempted transmissions across a
accounts for the presence of consecutive faulty BSNs in hop = k,
the scenario. Furthermore, due to the indoor environ- energy dissipated to process and transmit a data packet
ment, the probability p remains fixed for successful data across a hop distance d = Qd ,
transmission between any two nodes irrespective of their channel access time = σ c ,
locations. time required by transmitter to receive ACK from
receiver = σ a ,
2.2.1 Energy Model time out period at transmitter for retransmission = σ T
A first-order radio model is considered in our simula- (σ T > σ a ),
tion in line with [10,27] to account for loss in energy due processing and transmission time of a data packet includ-
to both transmission and reception at a BSN. The model ing σ c and σ A across a hop = X̄t .
accounts for two major sources for this loss namely to run
the radio electronics and power amplifier. Tx and Rx are A transmission failure can occur due to external channel
separated by a distance d, which if less than or equal to a interference, presence of dynamic obstacles or an inter-
threshold distance d0 leads to d2 power loss, as shown in mediate receiving BSN can be down. This can be con-
the following equation. Due to indoor environment it is cluded after k unsuccessful transmission attempts. We
logical to assume the same in our simulation. In order to have the probabilities of successful transmission (Prsucc )
transmit an m-bit message over a distance d, the Tx radio after Nt = i tries and failure (Prfail ) across a hop for a data
consumes the energy given by packet as follows:

ETx (m, d) = ETx - elec (m) + ETx - amp (m, d) Prsucc = (pktsucc |Nt = i) = p(1 − p)i−1
(3)
= Eelec m + εfs md2 if d ≤ d0 ∀i = 1, 2, . . . , k.
Again, to receive this message the Rx radio consumes the Prfail = (pktfail |Nt = k) = (1 − p)k (6)
energy given by
We assume (σ T = σ A ) for simplification of analysis.
ERx (m) = ERx - elec (m) = Eelec m (4) It can be shown that the expected processing and
 S. CHANDRA ET AL.: AN INTELLIGENT AND POWER EFFICIENT BIOMEDICAL SENSOR NODE 7

Table 1: Latency and energy dissipated across a hop

No. of bytes
Number of Total transmitted Energy
transmissions (Nt ) Outcome Event probability consumed time (ignoring ACKs) dissipated
1 Success P 1(σ c + Xt) +σ A S Qd
2 Success p(1 − p) 2(σ c + Xt) + σ T + σ A 2S 2Qd
3 Success p(1 − p)2 3(σ c + Xt) + 2σ T + σ A 3S 3Qd
... ... ... ... ... ...
k Success p(1 − p)k − 1 k(σ c + Xt) + (k-1) σ T + σ A kS kQd
k Failure (1 − p)k k(σ c + Xt +σ T) kS kQd

Figure 6: Hardware testing of BSN: (a) signal conditioning of ECG and (b) signal conditioning of PPG; (c) screenshot of acquired ECG and
PPG signals from node #3 in the host PC

transmission delay of 50-byte data packet across a sin- There is theoretically no change in latency for a sin-
gle hop including channel access and acknowledgement gle BSN fault. Similar argument holds for multiple BSN
times is given by faults.

1 2.2.3 Energy Consumption Estimation

E[Xt ] = (σc + Xt + σT ) (7)
p It can be shown that the energy consumed in routing a
message Me comprising L packets of 50 bytes each from
Similarly, the expected processing and transmission delay the originating node to the Host PC along the routing
of a message across a single hop including channel access path decided by MSN, having i intermediate nodes out
and acknowledgement times is given by of which u are faulty in consecutive locations, is given by
L
E[Mt ] = LE[Xt ] = (σc + XT + σT ) (8) 
u+1
p E [Me , i, u ] = (i − u) L E[Qd ] + L E[Qkd ] (10)
k=1
The expected end-to-end latency for message Mt,i origi-
nating from a node with i intermediate nodes to the Host The receiving energy of a node is neglected. If we put
PC, ignoring the receiving delay, is given by u = 0 we get E [Me , i, 0] , i.e. the no error case. In the
case of u consecutive errors we have u out of i inter-
E[Mt,i ] = (i + 1)E[Xt ]L (9) mediate nodes that does not transmit resulting in a
8 S. CHANDRA ET AL.: AN INTELLIGENT AND POWER EFFICIENT BIOMEDICAL SENSOR NODE

Table 2: Hardware latency for a 15-node network with faulty

nodes
Success at
transmission Latency tlat (in
Faulty nodes (f ) attempt (k) s)
1 1 1.28
2 2.26
3 3.25
2 1 1.43
2 2.34
3 3.25
3 1 1.58
2 2.42
3 3.25
Figure 7: Packet reception accuracy against nodes disengaged
distance between neighbouring BSNs. Finally, at all faulty
nodes, the situation becomes the same as that of a peer-
saving
 of uLE[Qd ] Joules but at the cost of additional to-peer transmission. For each trial, up to three BSNs
L u+1
k=2 E[Qkd ]Joules. Table 1 shows the energy dissi- were put under the switched-off condition to simulate
pation and consumed time across a hop for different node failure condition. The actual transmission and pro-
transmissions of S byte packet. cessing time per packet per BSN was 75.6 ms. Hardware
latency per packet as found under different faulty nodes
3. RESULTS AND DISCUSSIONS is shown in Table 2. Here k indicates success at kth trans-
mission attempt. It can be observed that the latency for
3.1 Hardware Experimentation Results transmission attempts with k = 3 for faulty nodes f = 1,
Figure 6(a) and (b) shows the signal conditioning per- 2 and 3 are equal. This is because with the increase of
formance of ECG and PPG signals, respectively, during faulty nodes, the number of hops is also reduced. The
BSN development stage. In Figure 6(c) a screen shot of generalized latency expression for N number of nodes is
the acquired ECG and PPG signals from N#3 is shown in given as
Host PC. The Host PC displays the available and unavail- tlat = 75.6 × [3f + {k(N − f − 1) + 1}] (11)
able BSNs (shown in crossed circle). The user can select
the BSN to be requested, one at a time, and short or
long duration acquisition option by the radio button.
3.2 Latency and Energy Calculation
The acquired data can be saved in text files also using
the “File” menu. Each BSN supports two types of data For the simulation study, we consider a network con-
collection, viz., long (30 min) and short (10 s). For long sisting of 100 nodes (BSN), fixed in location, deployed
duration acquisition, the ECG and PPG data are stored over an area of 15.24 m × 37 m, with no static obsta-
in a compressed form in the secure digital (SD) card cles in between the nodes, but the dynamic ones may be
and not transferred over sensor network to Host PC. A present due to the movement of the hospital personnel.
pilot study was done at an indoor patient ward of Medi- The BSNs are powered with 3.7 V and 2200 mAh bat-
cal College and Hospital, Kolkata, India with 15 patients. tery. The energy and latency calculation are based on
In the first stage, the data reception accuracy was esti- the following assumptions: (a) a total of 26 packets of 50
mated using a 15-node network, hardware implemented bytes each are generated during each collection from a
in a hospital ward of 93 m2 area. For each trial, the per- BSN; (b) from each BSN data are collected for 20 s per
formance parameters of farthest BSN from the Host PC day (twice, 10 s each); (c) each node also participates in
were estimated. An experiment was performed with 1–14 routing data for other nodes, and is engaged 300 s for
faulty BSNs (created intentionally). The average packet each collection cycle; (d) when no data are being gen-
reception accuracy over 200 independent trials, spanned erated the nodes enter the sleep mode. Thus, during a
over a period of 3–4 months for 10 s data, is provided in day the nodes collectively remain awake for 3 h 22 min.
Figure 7. The worst-case latency per packet (of 50 bytes) is shown
in Figure 8 for the farthest node against the number of
During each trial, dynamic obstacles were imposed in faulty BSNs. The receiving delay at a node is ignored. It
the form of personnel movements over the floor area. is observed that for maximum three attempted transmis-
The average packet reception accuracy was 98.45%. As sions (which are very unusual) at each hop, the latency
the number of PSNs was increased, the packet recep- per data packet of data is around 23 s. Under a similar
tion accuracy slightly dropped from 100% due to large condition, the energy consumption (in J/packet) in each
 S. CHANDRA ET AL.: AN INTELLIGENT AND POWER EFFICIENT BIOMEDICAL SENSOR NODE 9

Table 3: Comparison of power consumption in BSN with pub-

lished works
Deployment
Power scenario
consumption (mW) Data rates (topology/number
Work per packet (bytes/s) of nodes)
Cha [3] 132 6 8 nodes
Boquete [28] 0.6309 9600 P–P
Auteri [29] 96.7 NA P–P
Chen [12] – 200 30–50 nodes
Wang [30] 23.25 NA P–P
Present work 3.49 (worst case) 661 100 nodes

Figure 8: Latency against number of faulty nodes f = 0, 1, power consumption, the proposed research clearly out-
2, . . . ,9 with total number of deployed BSNs set to 100 requiring performs [3, 6, 27]. In [28], a ZigBee-based peer-to-
k transmissions across a link
peer communication is reported to acquire eight different
biopotentials in a PC. The system is for continuous usage
and requires manual operation. Although the power con-
sumption is less than that in the proposed work, the
device is bulkier and its performance if connected in a
wireless network is still not tested.

4. CONCLUSIONS
In this paper, we describe the development of a power
efficient and intelligent BSN which provides low latency
in monitoring a group of patients distributed in a small
hospital ward. A significant research outcome is the intel-
ligent routing algorithm which provides high reliability
and low latency in received data packets in the host PC.
The hardware experimental validation using 15 BSNs
provided two important parameters related for network
performance. The first, hardware latency of 1.58 s per
data packet for three faulty BSNs with maximum allow-
able transmission attempts. The second, packet reception
Figure 9: Power consumption against number of faulty nodes
accuracy was of 96.7% under the worst-case scenario (all
(f ) = 0, 1, 2 . . . 9 (number of BSN = 100) relaying BSN fault). The hardware experimentation was
supplemented by a software simulation considering 100
BSNs under various faulty nodes with maximum allow-
BSN along a routing path involving all the BSNs is plot- able transmission attempts between successive nodes.
ted in Figure 9(a) against the number of faulty BSNs, Here, the latency per data packet is around 23 s, which is
f = 0, 1, 2, . . . , 9 respectively. The power consumption lower than that reported in similar works. In a develop-
per packet, shown in Figure 9(b), is computed by divid- ing nation like India, there is an acute shortage of medical
ing the energy consumption by the respective latency. practitioners and nursing personnel [31]. Thus, the BSN
Since the latency and energy consumption is highest, the can offer a low-cost alternative solution against tradi-
power consumption is least at k = 3, and the maximum tional health monitoring system in a small-sized hospital
for k = 1. As the number of faulty nodes (f ) increases, the ward.
power consumption remains almost constant for k = 1
and 2, since the faulty nodes do not transmit. However,
for k = 3, the power consumption marginally increases ACKNOWLEDGEMENT
as the power and latency increases with failed attempts The authors sincerely thank Dr Arunansu Talukdar, Profes-
(up to k = 2). sor, Medical College and Hospital, Kolkata for his coopera-
tion in performance testing of the BSNs in the hospital ward.
The authors also thank the SAP DRS-II program (2015–2020)
Table 3 shows the comparison of power consumption of from University Grants Commission (UGC) at Department
our model with the published literature in similar area. of Applied Physics, University of Calcutta for the technical
The packet size is only mentioned in [12]. In terms of support.
10 S. CHANDRA ET AL.: AN INTELLIGENT AND POWER EFFICIENT BIOMEDICAL SENSOR NODE

FUNDING 10. W. B. Heinzelman, A. P. Chandrakasan, and H. Balakr-

ishnan, “An application-specific protocol architecture for
The work is funded by Department of Higher Education,
wireless microsensor networks,” IEEE Trans. Wirel. Com-
Science & Technology and Biotechnology (DHESTB), Govt.
mun., Vol. 1, no. 4, pp. 660–670, 2002.
of West Bengal [sanction No.851(sanc.)/ST/P/S&T/6G-2/2013
dtd: 11/01/2016].
11. S. Lindsey, C. Raghavendra, and K. M. Sivalingam, “Data
gathering algorithms in sensor networks using energy met-
ORCID rics,” IEEE Trans. Parallel Distrib. Syst., Vol. 13, no. 9, pp.
924–935, 2002.
Soumyak Chandra http://orcid.org/0000-0001-8369-3495
Rajarshi Gupta http://orcid.org/0000-0001-5964-7683 12. S. K. Chen, et al., “A reliable transmission protocol for
Saurav Ghosh http://orcid.org/0000-0001-9015-9004 ZigBee-based wireless patient monitoring,” IEEE Trans.
Inf. Technol. Biomed., Vol. 16, no. 1, pp. 6–16, 2012.

REFERENCES 13. M. S. Pan, and P. L. Liu, “Low latency scheduling for

1. K. Sailunaz, M. Alhussein, M. Shahiduzzaman, F. Anowar, converge cast in ZigBee tree-based wireless sensor net-
and K. A. Al Mamun, “CMED: Cloud based medi- works,” J. Netw. Comput. Appl., Vol. 46, pp. 252–263, 2014.
cal system framework for rural health monitoring in
developing countries,” Comput. Electr. Eng., Vol. 53, 14. N. Golmie, D. Cypher, and O. Rebala, “Performance anal-
pp. 469–481, 2016. ysis of low rate wireless technologies for medical applica-
tions,” Comput. Commun., Vol. 28, no. 10, pp. 1266–1275,
2. A. Huang, et al., “WE-CARE: An intelligent mobile tele- 2005.
cardiology system to enable mHealth applications,” IEEE
J. Biomed. Health Informatics, Vol. 18, no. 2, pp. 693–702, 15. L. Malathi, R. K. Gnanamurthy, and K. Chandrasekaran,
2014. “Energy efficient data collection through hybrid unequal
clustering for wireless sensor networks,” Comput. Electr.
3. Y. D. Cha, and G. Yoon, “Ubiquitous health monitoring Eng., Vol. 48, pp. 358–370, 2015.
system for multiple users using a ZigBee and WLAN dual-
network,” Telemed. J. E Health, Vol. 15, no. 9, pp. 891–897, 16. S. Gajjar, M. Sarkar, and K. Dasgupta, “Self organized, flexi-
2009. ble, latency and energy efficient protocol for wireless sensor
networks,” Int. J. Wirel. Inf. Networks, Vol. 21, no. 4, pp.
4. L. Yseboodt, et al., “Design of 100 μW Wireless Sensor 290–305, 2014.
Nodes for Biomedical Monitoring,” J. Signal Process. Syst.,
Vol. 57, no. 1, pp. 107–119, 2009. 17. C. Bachmann, M. Ashouei, V. Pop, M. Vidojkovic, H.
De Groot, and B. Gyselinckx, “Low-power wireless sensor
5. M. Magno, C. Spagnol, L. Benini, and E. Popovici, “A low nodes for ubiquitous long-term biomedical signal moni-
power wireless node for contact and contactless heart mon- toring,” IEEE Commun. Mag., Vol. 50, no. 1, pp. 20–27,
itoring,” Microelectronics J., Vol. 45, no. 12, pp. 1656–1664, 2012.
2014.
18. M. R. Yuce, “Implementation of wireless body area net-
6. P. Vijayakumar, P. Pandiaraja, M. Karuppiah, and L. Jegatha works for healthcare systems,” Sensors Actuators, A Phys.,
Deborah, “An efficient secure communication for health- Vol. 162, no. 1, pp. 116–129, 2010.
care system using wearable devices,” Comput. Electr. Eng.,
Vol. 63, pp. 232–245, 2017. 19. T. Gao, et al., “Wireless medical sensor networks in emer-
gency response: implementation and pilot results,” in 2008
7. H. Alemdar, and C. Ersoy, “Wireless sensor networks for IEEE International Conference on Technologies for Home-
healthcare: A survey,” Comput. Networks, Vol. 54, no. 15, land Security, HST’08, 187–192, 2008.
pp. 2688–2710, 2010.
20. R. Gupta, J. Bera, and M. Mitra, “A bi-phase enabled
8. P. Baronti, P. Pillai, V. W. C. Chook, S. Chessa, A. Gotta, and serial acquisition system for remote processing of digitized
Y. F. Hu, “Wireless sensor networks: A survey on the state ECG,” Comput. Electr. Eng., Vol. 38, no. 1, pp. 68–74, 2012.
of the art and the 802.15.4 and ZigBee standards,” Comput.
Commun., Vol. 30, no. 7, pp. 1655–1695, 2007. 21. R. Gupta, J. N. Bera, and M. Mitra, “An intelligent telecar-
diology system for offline wireless transmission and remote
9. I. Lamprinos, A. Prentza, E. Sakka, and D. Koutsouris, analysis of ECG,” J. Med. Eng. Technol., Vol. 36, no. 7, pp.
“Energy-efficient MAC Protocol for Patient Personal area 358–365, 2012.
networks,” in Conference Proceedings: Annual Interna-
tional Conference of the IEEE Engineering in Medicine 22. “AD8232 front end.” [Online]. Available: http://www.
and Biology Society. IEEE Engineering in Medicine and analog.com/media/en/technical-documentation/data-
Biology Society. Conference, Vol. 4, pp. 3799–3802, 2005. sheets/AD8232.pdf.
 S. CHANDRA ET AL.: AN INTELLIGENT AND POWER EFFICIENT BIOMEDICAL SENSOR NODE 11

23. “Easy Pulse PPG sensor.” [Online]. Available: http:// IEEE Internet Things J., Vol. 1, no. 4, pp. 311–318,
embedded-lab.com/blog/easy-pulse-version-1-1-sensor- 2014.
overview-part-2/.
28. L. Boquete, et al., “A portable wireless biometric multi-
24. “Ds_Xbee Multipoint Modules Datasheet.” [Online]. Avail- channel system,” Meas. J. Int. Meas. Confed., Vol. 45, no.
able: https://www.digi.com/pdf/ds_xbeemultipoint 6, pp. 1587–1598, 2012.
modules.pdf.
29. V. Auteri, L. Roffia, C. Lamberti, and T. Salmon Cinotti,
25. P. Bera, and R. Gupta, “Real-time compression of electro- “ZigBee-based wireless ECG monitor,” Comput. Cardiol.,
cardiogram using dynamic bit allocation strategy,” in 2016 Vol. 34, pp. 133–136, 2007.
IEEE 1st International Conference on Control, Measure-
ment and Instrumentation, CMI 2016, 21–25, 2016. 30. L. H. Wang, T. Y. Chen, K. H. Lin, Q. Fang, and S. Y.
Lee, “Implementation of a wireless ECG acquisition SoC
26. S. Alam, and R. Gupta, “Zonal complexity based measure for IEEE 802.15.4 (ZigBee) applications,” IEEE J. Biomed.
for lossy compression of photoplethysmogram using delta Health Informatics, Vol. 19, no. 1, pp. 247–255, 2015.
encoding,” in 2015 IEEE International Conference on Sig-
nal Processing, Informatics, Communication and Energy 31. “An approach to 12th Five year plan (2012–2017), Govt. of
Systems, SPICES 2015, Vol. 2015, pp. 1–5. India, Planning Commission report.” [Online]. Available:
http://planningcommission.gov.in/plans/planrel/fiveyr/
27. Y. Zhang, L. Sun, H. Song, and X. Cao, “Ubiquitous WSN 12th/pdf/12fyp_vol3.pdf.
for healthcare: recent advances and future prospects,”

Authors Saurav Ghosh is currently an Assistant

Professor at A.K Choudhury School of
Soumyak Chandra is currently a Junior Information Technology, University of
Research Fellow (JRF) attached to the Calcutta, India. His research interests are
project funded by Department of Higher Mobile Crowd Sensing, Sensor Networks
Education, Science & Technology and and Machine Learning. He has a number
Biotechnology (DHESTB), Govt. of West of publications in reputed International
Bengal at Department of Applied Physics, Journals and conferences. He is currently
University of Calcutta. He obtained B. guiding two PhD scholars and one scholar has been awarded
Tech in Electronics and Communication PhD.
Engineering in 2013. He has one Indian copyright to his credit.
His research interests are intelligent systems, networking and Email: sauravghoshcu@gmail.com
medical instrumentation.
Sanjoy Mondal is currently a Rajiv Gandhi
Corresponding author. Email: scaphy_rs@caluniv.ac.in Senior Research Fellow at A.K Choudhury
School of Information Technology, Uni-
Rajarshi Gupta is currently an Asso- versity of Calcutta, India. He has com-
ciate Professor at Electrical Engineering pleted MCA and M. Tech in Information
Section, Dept. of Applied Physics at Uni- Technology, and currently pursuing PhD
versity of Calcutta, India. His research in the area of smart sensing. He has a
interests include cardiovascular signal number of international conference pub-
measurements and intelligent health mon- lications to his credit.
itoring. He has 49 publications in National
and International peer-reviewed Journals Email: sanjoymondal1988@gmail.com
and conferences, one Indian copyright, two book chapters and
one book. Currently he is guiding three PhD students and one
scholar has been awarded PhD.
Email: rgaphy@caluniv.ac.in