Effectiveness of a Novel Power Control Algorithm in Heart Rate Monitoring of a Mobile Adult: Energy Efficiency Comparison with Fixed Power Transmission

The proliferation of low power wireless sensor networks and their discreet presence have introduced a new paradigm in data collection and analysis of target parameters in both indoor and outdoor environments. Especially in healthcare, body wearable and implantable sensors are used to continuously monitor the vital physiological parameters of patients in hospitals and elderly at home who enjoy independent living [1]. From such “living records”, medical practitioners can draw useful inferences about the health and well-being of an individual. This can be used for self-awareness and analysis to assist in making behaviour changes, and to share with caregivers for early detection of any ailment and allow appropriate intervention. At the same time such procedures are effective and economic ways of monitoring age-related illnesses [2]. An overview of a simple WSN application in healthcare is shown in Figure 1 [3,4]. The wearable and implanted sensors collect vital health parameters like the pulse rate, EEG, blood insulin level, etc. and transmit to the access point. The gateway that is shown in Figure 1 can be with the person or with the access point.


Introduction
The proliferation of low power wireless sensor networks and their discreet presence have introduced a new paradigm in data collection and analysis of target parameters in both indoor and outdoor environments. Especially in healthcare, body wearable and implantable sensors are used to continuously monitor the vital physiological parameters of patients in hospitals and elderly at home who enjoy independent living [1]. From such "living records", medical practitioners can draw useful inferences about the health and well-being of an individual. This can be used for self-awareness and analysis to assist in making behaviour changes, and to share with caregivers for early detection of any ailment and allow appropriate intervention. At the same time such procedures are effective and economic ways of monitoring age-related illnesses [2]. An overview of a simple WSN application in healthcare is shown in Figure 1 [3,4]. The wearable and implanted sensors collect vital health parameters like the pulse rate, EEG, blood insulin level, etc. and transmit to the access point. The gateway that is shown in Figure 1 can be with the person or with the access point.
One unique challenge with body wearable sensors is that these sensors are mobile. Therefore, the communication layer must adapt quickly with the changing environment so that the transmission power can be re-calibrated for reliable transmission [5].

Need to control power for wireless mobile sensor nodes
When a sensor is mobile, it means that its communicating distance from the base station is changing with time. The energy loss is primarily dependent on distance. Unwanted obstructions can also lead to signal degradation due to absorption. Beside there are effects of fading and multipath propagation of radio signal in indoor wireless communication [6].
For acceptable performance, the average received E b /N 0 at the base station should be above a threshold value. If that value is maintained on average, then it means that the wireless sensor is performing as expected. Since most of these mobile wireless sensors are battery powered, they have limited energy resources. Therefore, if a mobile node is near to a base station and the received E b /N 0 far exceeds the required threshold, then there is waste of energy. Similarly, when the same sensor node is far from the base station, the node is required to pump in more power than the present value. It may also happen that due to an obstruction between the transmitting node and the base station, the node is transmitting at a power level that is enough to deliver packets. If that obstruction is removed or the node moves to a position in which the obstruction is cleared, then it should adjust its transmitting power down.
Previous work [7] has provided an empirical analysis of the impact of power control for mobile sensor network. The focus of this research paper is on residential health monitoring, in-hospital patient monitoring and sports monitoring that require mobile sensing. Cellular networks deal with mobility by using different types of hand-shake mechanism. However, the emphasis is on the energy constrained sensor nodes. This paper suggested that received signal strength indication (RSSI) data may not be sufficient to evaluate link quality when sensor nodes are mobile. They proposed an active probing scheme for sensor applications that send data periodically and those which are triggered by an event (i.e., event driven). In active probing scheme, the mobile node counts the number of consecutive packets that are successfully transmitted at the current power level. If that is more than a predefined threshold, then the power level is decremented by one level. However for any un-acknowledged packet, the transmission power level is incremented by one level, until the maximum power level is reached. This paper has also modified this approach by using the link quality indication (LQI) values that are provided by CC2420 transceiver modules. In order to make good use of the LQI values of the acknowledgement packets, it allowed the radio to transmit several packets at each of the power levels. It finds the optimal power level region where consistent LQI values higher than 100 is observed. This optimal value is used as the benchmark to set the new transmission power level.
In GSM, a power control algorithm is employed to achieve desired signal strength for faithful communication between the mobile station (MS) and the base transceiver station (BTS). Power control also reduces interference and improves cell capacity. During a connection between the BTS and the MS in a cell, the MS measures the channel's RF link quality after every 480 milliseconds [8]. In this way an acceptable link quality is maintained which can also improve the battery lifetime of the mobile device. However, the research work that is presented here has aimed at saving energy by cutting down on the cost of sensing the RF channel before actual transmission. This is because the wireless sensor devices are battery-powered with capacity in the order of 250-300 mAh [9] and have far less capacity than the batteries used by mobile devices (~1500 -3500 mAh) [10][11][12].
There are a few non-RSSI based power control algorithm that uses matrices like the packet delivery ratio (PDR) or the packet reception rate (PRR) to estimate link quality rather than RSSI or LQI. Among them Practical-TPC [13] and ART [14] are worth mentioning. P-ATPC is a receiver oriented protocol that is considered robust in dynamic wireless environments and uses packet reception rate (PRR) values to compute the transmission power that should be used by the sender in the next attempt. The receiver monitors all incoming packet and counts the successes and failures of the packet transmission within the current sampling window. After the sampling window period is over, the P-TPC protocol computes the next transmission power level and sends to the transmitter. This new power level will be used during the next sampling window. P-ATPC has two main components. One component (fast online model identification FID) estimates the model between the PRR and the transmission power. It initialises or reconfigures the second component proportional-integral with anti-Windup (PI-AW). The PI-AW computes the transmission power level based on the difference between the current PRR and the application specific PRR requirement. P-ATPC runs two feedback loops. The inner loop involves the PI-AW that adapts the transmission power based on the PRR measurement. The outer loop involves FID to adjust the parameters based on the updated power model that defines the relationship between the PRR and transmission power. P-ATPC also initializes the power model parameters before the feedback loops kick in. In this initial phase, each link is set to transmit a sequence of probe packets using highest to lowest power level to build the transmission power model.
ART (Adaptive and Robust Topology control) protocol has been designed for complex and dynamic radio environments. It adapts the transmission power in response to variation in link quality or degree of contention. Empirical studies presented in this paper have shown that the PRR and the signal strength can vary over time. This is primarily due to effect of movements of objects and people in between the transmitter and the receiver during the busy hour of the day and corresponding fading of signal. Analysis of the paper has suggested that RSSI and LQI may not be good or the most reliable indicators of link quality, especially in dynamic indoor radio environment. ART changes the transmission power of a link based on the observed PRR. It has an initialization phase when the ART protocol monitors all the outgoing packets for its successful or failed transmission within a sliding window of predefined size. It compares the number of failures within that window with a minimum and a maximum threshold failure. Based on the comparison, it does anyone of the following: 1) Remains in the same power level or 2) Increases the transmission power level or In these type of non-RSSI based adaptive protocols, there is a running overhead cost as the link quality is sampled after a given interval of time. The sampling period will itself depends on the dynamics of the link quality.
The novel non-RSSI based channel estimation and output power control algorithm is proposed in our earlier papers [15,16]. It does not use RSSI/LQI data as side information for channel link quality estimation. The basis of this lightweight adaptive algorithm is the states where each state represents one cycle of packet transmission. The details of the adaptive power control algorithm are explained in the next section.

Basics of an Adaptive Power Control Algorithm
The packet success rate performance of a mobile sensor node depends primarily on the distance between the base station and the sensor, obstacles in the communication path, and fading due to movement of objects in between. In particular, when a node operates on the fringes of the communicable distance, the power amplifier pumps in maximum power so that it is discernible at the receiving station. It uses its allocated retry limit to achieve the threshold packet success rate. On the other hand, the radio conditions can be most favourable so that successful data communication is possible at the lowest available output power. At intermediate stages of communication distances, the transmitter may use higher power levels. In a non-RSSI based power control approach, the key is to keep track of the success or failure of packets at a particular transmission power. The outcome of the last packet transmission gives an indication as to what is the expected outcome when a new packet transmission starts. This adaptive power control can be most successfully applied when a node transmits quite frequently. Nodes that transmit once every hour or a day do not come under the purview of this research. The reason is that the transmitter does not send probe packets for channel estimation neither it can avail RSSI information of the last data transmission. Therefore, the last power level is an indicator of the channel condition. Now, this channel condition can be transient or semi-permanent. Transient condition can occur because of momentary dip in the signal level due to fading or due to change of distance or any obstruction in between the transmitter and receiver.
In case the channel change is transient, the adaptive algorithm must drop-off fast to the lower state. When the change in the link quality or the channel condition is not transient, then the adaptive algorithm must continue to transmit at a higher power level to ensure that packets are delivered successfully. Even in this case, when the distance between the transmitting node and the base station improves or the obstruction is removed, then the transmitter should be able to back off to a lower state.

Non-RSSI/LQI based channel estimation and power control algorithms for energy efficiency
The basis of this lightweight adaptive algorithm is the states where each state represents one cycle of packet transmission. In each state there are output power levels in increasing order which can be used by the transmitter. State transition occurs depending on the power level at which the transmission is successful or failed. State 4 uses only the maximum power level and is allowed to transmit 4 times. There is no direct transition from state 4 to state 1 or 2. Similar conditions hold true when transiting from 3. The most energy efficient state is 1. The more it stays in state 1, the more it saves energy. State 4 is where the maximum energy may be used to transmit the packet. The adaptive algorithm is designed in such a way that it takes into account of performances in each state. It also has a unique drop-off algorithm that allows it to drop down to a lower state when deemed necessary. It is guided by the drop-off factor R. In this paper, R values of 0.01, 0.05, 0.1, 0.5 and 1 are used. Higher value of R means higher rate of dropoff. Figure 2 shows the state transition diagram of the adaptive power control algorithm. State transition occurs depending on the power level at which the transmission is successful or has failed.
The objective of the adaptive power control algorithm is to respond to the packet error rate and move to a new state with different retry limits. The adaptive algorithm is designed in such a way that it takes into account the performance in each state. Each state has a different retry limit. Increasing state number indicates poorer channel quality. The proposed adaptive algorithm does not allow retransmission in the same power level except when it is in state 4 and transmitting at 0 dBm. When the system is in state 4, it is considered the worst channel condition and three retries are allowed. The retry limit of state 1 is three. However, the retry limit of states 2 and 3 have been set at 2 and 1. The asymmetry is because the increase in the retry limit in states 2 and 3 can increase the current consumption while only marginally improving the packet success rate.     Table 1 shows the available power levels based on the states. Transmission starts at the lowest available power level of that particular state. The transmitter can be in any one of the states during the start of transmission of a packet. There are two separate algorithms that determine the state transitions, one from a lower state to higher state and the other from a higher to lower states. The logic to transit to lower states also includes situations when it remains in the same state or transit to a lower state. Table 2 describes the state transition matrix when state level goes up. All the state transition decisions depend on the success or failure of the packet being transmitted to the destination hub. Table 3 describes the state transition logic when state level goes down. The primary objective of the adaptive algorithm is to save energy by transmitting at a power level that is enough to send the packet successfully through the channel. For example, when the system is in state 4, it is transmitting at the maximum power. With time, the channel condition can improve and packet can be successfully transmitted at a lower power level. If the system drops down to state 3, the transmission starts at a lower power level. This drop-off from a higher state to a lower state is determined by a drop-off algorithm which is probabilistic in nature.
In the proposed adaptive algorithm, the drop-off or the back-off process is dependent on the number of successes (S) in the higher power level and a drop-off factor (R). By default, the drop-off factor is 1. The probability of the system to drop-off to a lower power level is represented by Equation (1).
Here, P drop-off =probability of drop-off; S=the number of successes in that power level of the higher state; R=drop-off factor.
The plots in Figure 3 show the state transition probability based on different values of R. When there is a state change, the value of S is reset to 0. Overall, the value of R indicates how fast the system will fall from a higher state to a lower state. When there is no success, the probability of state transition is 0, meaning that there will be no state transition. At the same time, when the number of successes is too high, it converges to 0.
Back-off algorithms are extensively used in data communication (both wired and wireless) by MAC protocols to resolve contention among transmitting nodes to acquire channel access. In a MAC protocol, the back-off algorithm chooses a random value from the range [0, CW], where CW is the contention window size. The contention window is usually represented in terms of time slots.
The number of time slots to delay before the nth retransmission attempt is chosen as a uniformly distributed random integer r in the range 0<r<2k.
Where k=min (n, 10), 10 is the maximum number of retries allowed.
The n th retransmission attempt also means that there have been n collisions. For example, after the first collision, it has to retransmit. Based on the back-off algorithm, the sender will choose between 0     and one time slot for the retransmission. After the second collision, the sender will wait anywhere from 0 to three time slots (inclusive). After the third collision, the senders will wait anywhere from 0 to seven time slots (inclusive), and so forth. As the number of retransmission attempts increases, the number of possibilities for delay increases exponentially [17,18].
Similarly, an exponential operator is used in this novel adaptive algorithm to decide to switch from a higher state to a lower state. The drop-off algorithm is dynamic as it re-evaluates at every successful transmission. It gets reset to 0 when it leaves the state and jumps to a lower state and starts a new packet transmission at a lower power level [17,18].

Choice of hardware
The adaptive power control algorithm has a unique channel estimation method without RSSI side information. The hardware used in the research includes the nRF24L01p transceiver module that acts as a transmitting sensor. For the receiver at the hub, another nRF24L01p transceiver module is used that has an additional PA and LNA. The receiver has a maximum output power level of 20 dBm. The reason to choose a high power transmitter at the hub is to make the path between the hub and the sensor practically error free. The primary features of the receiver and the transmitter receiver are presented in Tables 4 and  5 respectively [19][20][21].
The transceiver can transmit at four power levels: -18 dBm, -12 dBm, -6 dBm and 0 dBm. In general a wireless transceiver has different modes of operation. All the software programming was done in C in the open-source Arduino (version 1.0.5-r2) software (IDE) [22]. The programs or sketches in Arduino are used to interface with the nRF24L01p modules to do the necessary changes.
With regards to the heart rate data, a set of heart rate data are preloaded in the transmitter module and made to transmit after every 5 seconds. These data has been borrowed from PhysioNet which offers free web access to large collections of recorded physiologic signals [23,24]. Here we are testing the energy efficiency of the protocol, so we have used heart rate data from PhysioNet while the subjects provide the sensor mobility.

Location and subject
The experimental setup is a typical house with the base station powered by Mains while the transmitting sensor is piggybacked on the subjects. Two subjects of different age groups ([30-35] years and [65-70] years) are chosen to observe the effect of age on the adaptive protocol. The subjects are allowed to roam freely inside the house and follow their routine activities. In this paper, subject 1 refers to the person in age group 65-70 and subject 2 refers to age group 30-35. Data were collected for a period of approximately five hours on different days of the week, starting from afternoon till late evening. In the 1 st set of experiments, the mean distance between the subject and the base station was approximately 5 meters. During the 2 nd set of experiments, the mean distance was changed to roughly 10 meters. The different distances are expected to have an effect on the evaluation parameters which are presented in subsection 3.3.

Evaluation parameters
The evaluation parameters are • Average cost per successful transmission • Expected success rate or protocol efficiency [25] One of the parameters for the optimization is the energy consumed per useful bit transmitted over a wireless link [26,27]. Similarly in this paper, the cost per successful transmission has been considered.  Table 6: Average cost, PSR and protocol efficiency with subject 1 and mean distance equals to 5 meters. All cost values are measured in mJoules. The total cost of transmission includes the expenditure for the first transmission attempt of a packet and the subsequent retries if the first attempt fails. The total packet to send count does not include the retry packets. Therefore the denominator in equation 3 is only the count of successfully transmitted packets.
The expected success rate or efficiency is defined as the expected number of successes and takes into account the average number of retries [3]. It can also be defined as the expected number of successes per 100 transmissions. Mathematically,

Where
Succ rate =expected success rate

Ret T =total number of retries
Here P S -P L =total number of successes (P succ ). If both the numerator and denominator are divided by Ps, then in percentage term, where Ret avg =average number of retries per packet and is defined as Here, This parameter indicates the total number of transmissions (on average) to achieve a given packet success rate (PSR).

Experimental Results and Analysis
The results and analysis of the experiments are presented in this section.       Table 6 presented the average values of three runs of the experiments with subject 1 when the distance between the base station and the transmitting mobile node is varying with mean distance approximately equal to 5 meters. Based on Figure 4 it can be observed that the PSR in all the cases are ~100%. The protocol efficiency values are all above 95%. Since the mean distance is very small (~ 5 meters), the adaptive protocol could only marginally perform better than the fixed power transmission. It is still able to save 6% energy. A low value of R means that the adaptive system will back-off at a slow pace. This is the reason that the minimum cost is achieved at R=1 because it has the ability to back-off fast to a lower state and transmit at a lower power as compared to when R is set at 0.01.

Scenario 1: Mean distance is 5 meters
Not much change is observed when the experimental data for subject 2 is analysed. Table 7 and Figure 5 present the experimental results in details.
It can be observed from the Figures 4 and 5 that since the mean distance between the subjects and the base station are very small, power control may not be able to save significant amount of energy.

Scenario 2: Mean distance is 10 meters
In this 2 nd set of experiments, the mean distance is changed to approximately 10 meters. The results are tabulated in Tables 8 and  9 and comparison are presented in Figures 6 and 7 respectively for subjects 1 and 2.
The minimum amount of energy per successful transmission is consumed in fixed transmission mode when the power level is set at -12 dBm. Figure 6 shows that if the drop-off factor is set at 1, the adaptive protocol can save 13% energy as compared to fixed transmission at -12 dBm. Table 9 and Figure 7 present the analysis and graphical comparison of the evaluation parameters when data from subject 2 is used with a mean distance of 10 meters.
The minimum amount of energy per successful transmission is consumed in fixed transmission mode when the power level is set at -12 dBm. Figure 7 shows that if the drop-off factor is set at 1, the adaptive protocol can save 21% energy as compared to fixed transmission at -12 dBm.
Overall, it can be observed that the use of adaptive protocol in typical home environment for sensor monitoring purposes can save energy and extend the operational lifetime before batteries are replaced. The younger adult as subject 2 of the experiment is expected to move faster as compared to the elder adult, denoted by subject 1. This variability in the mobility is not reflected in the 1 st set of results as the distance is small. As the mean distance changed, the experimental results show that the savings for subject 2 is more than subject 1. This is because subject 2 is more active than subject 1 and the adaptive protocol finds enough space to modulate the power level, thereby saving more energy.

Conclusion
The results in this paper demonstrate the advantages and limitations of using power control under different channel conditions to achieve energy efficiency. When the link quality is good (mean distance ~5 meters and very few obstructions in between the transmitter and the receiver), the adaptive power control algorithm is able to save energy marginally as compared to fixed power transmission. This is because there was no scope of output power manoeuvring to achieve energy efficiency. When the mean distance is roughly doubled to 10 meters, the adaptive power control approach has proved to be energy-saving as compared to fixed power. It is able to save upto 21% energy. It can be observed from Table 5 that the output power levels scale poorly with  the corresponding current rating. A drop in output power by 63 times only halves the current consumption approximately. Considering this values, the energy savings by the adaptive protocol is significant. More experiments will be conducted as part of future research scopes in other types of radio environments.