Performance Evaluation of Bluetooth Low Energy: A Systematic Review

Abstract

Small, compact and embedded sensors are a pervasive technology in everyday life for a wide number of applications (e.g., wearable devices, domotics, e-health systems, etc.). In this context, wireless transmission plays a key role, and among available solutions, Bluetooth Low Energy (BLE) is gaining more and more popularity. BLE merges together good performance, low-energy consumption and widespread diffusion. The aim of this work is to review the main methodologies adopted to investigate BLE performance. The first part of this review is an in-depth description of the protocol, highlighting the main characteristics and implementation details. The second part reviews the state of the art on BLE characteristics and performance. In particular, we analyze throughput, maximum number of connectable sensors, power consumption, latency and maximum reachable range, with the aim to identify what are the current limits of BLE technology. The main results can be resumed as follows: throughput may theoretically reach the limit of ~230 kbps, but actual applications analyzed in this review show throughputs limited to ~100 kbps; the maximum reachable range is strictly dependent on the radio power, and it goes up to a few tens of meters; the maximum number of nodes in the network depends on connection parameters, on the network architecture and specific device characteristics, but it is usually lower than 10; power consumption and latency are largely modeled and analyzed and are strictly dependent on a huge number of parameters. Most of these characteristics are based on analytical models, but there is a need for rigorous experimental evaluations to understand the actual limits.

Keywords: Bluetooth Low Energy (BLE); Internet of Things (IoT); performance evaluation; wearable technology; wireless sensor network.

Figure 1

BLE protocol stack. The three main blocks are the Controller (grey), the Host (blue) and the App (green). The HCI (red) is the interface that manages the communication between the Controller and the Host. The rectangular frames represent the different layers of the protocol, and they are ordered in a stack, which starts from the bottom, with the PHY part, and ends at the higher level, that is the App. The arrows show how encapsulation and fragmentation work. Adapted from [48].

Figure 2

Configuration between Bluetooth versions and device types. On the left is the protocol structure of the BR/EDR, while on the right is the BLE. In the middle is the protocol stack of a device compatible with both Bluetooth versions; this type is called Smart Ready or Dual Mode. Adapted from [52].

Figure 3

BLE frequency channels. It can be noticed that channels from 0–36 are assigned to data transmission in connections (blue), while the three remaining channels, from 37–39, are used as advertising channels, shown in orange. How channels are positioned in the frequency band is shown in the x-axis: the first channel, 37, is centered at frequency 2402 MHz, while the last one, the 39th, is centered at 2480 MHz. Adapted from [48].

Figure 4

GATT data hierarchy. Immediately before the connection, the GATT server exposes its services and characteristics. As shown in this figure, services and characteristics are defined in order to form a logical data structure. Moreover, each characteristic exposes its properties, a descriptor that defines what it does, and the data value. Adapted from [52].

Figure 5

GATT data hierarchy, relative to the example described in Section 2.1.7.

Figure 6

This figure shows how the advertising and scanning mechanisms work. In the upper part is shown the scanner; it scans on the three advertising channels (37, 38, 39, colored respectively in blue, red and green), switching the scanning channel within a period called T${}_{scanInterval}$. The effective scanning period, per each channel, lasts for a time called T${}_{scanWindow}$. On the other hand, the advertiser, shown in the bottom part of the Figure, sends a burst of three advertising packets, one for each advertising channel, with a specific period (T${}_{advInterval}$).

Figure 7

This figure shows how two devices (A and B) communicate through BLE. The two vertical blue lines on the left represent the Host layer and the LL of the device A, while the two lines on the right represent the same two layers relative to the device B. Blue arrows represent the communication messages between layers and devices. The black arrow on the left shows that the time increments scrolling the figure to the bottom. Red braces indicate the specific role of the devices A and B during the different parts of the communication. Part (a) of the figure shows the advertising mechanism, in which the advertiser (B) sends a message from the Host to the LL in order to enable the sending of advertising packets. The scanner (A) receives these packets and prepares itself for the connection. In (b) is shown the connection establishment where the initiator (A) sends a message in order to create the connection, firstly from the Host to its own LL and then to the other device, the responder (B). At the end, if the connection is established correctly, the two LLs send a message to the respective Host layers in order to confirm the correct creation of the connection. Starting from (c), the connection is created, and the two devices are called master (A) and slave (B). In (c), the master sends data packets to the slave, writing on a writable characteristic. In this case, data packets are sent from the master Host layer to its own LL and then to the slave. During this process, an Acknowledgment packet (ACK) is sent back to the master in order to communicate to the Host layer if and how many packets were correctly transmitted. This type of communication, with the ACK packet, is called round-trip. A similar process is shown in (d), where the slave writes on its readable characteristics and the master reads the data. Also in this type of communication, there is the transmission of an ACK packet, so this is also a round-trip communication. In (e), a one-way communication is represented, where the slave communicates with the master using a notifiable characteristic, which means there is no ACK packet.

Figure 8

Example of a round-trip data communication in a connection with the transmission of data packets and ACKs. In this figure are described all the connection parameters.

Figure 9

BLE packet structure. The packet has one or two bytes of PRE, depending on the radio data rate, four bytes of AA, from two to 257 bytes of PDU and three bytes of CRC.

Figure 10

PDU structure of an advertising packet. In (a) the first two bytes represent the Header, while the other bytes are the effective payload of the packet. The Header, showed in detail in (b), is composed of 4 bits of the PDU Type, 1 bit of RFU, 1 bit of ChSel, 1 bit of TxAdd, 1 bit of RxAdd and 8 bits of length. Then, in the case of advertising packets, only 37 bytes of the remaining 255 of payload are filled. The first six bytes are the AdvA and the last 31 form the AdvData, as shown in (c).

Figure 11

PDU structure of a BLE connection packet. In (a) the first two bytes represent the Header, while the other bytes are the effective payload of the packet; in addition to this there are four optional bytes at the end, existing only in LL encrypted connections, which are the MIC. The Header, showed in detail in (b), is composed of 2 bits of LLID, 1 bit of NESN, 1 bit of SN, 1 bit of MD, 3 bits of RFU and 8 bits of Length. Then, in the case of BLE connection packets, only 20 bytes of the remaining 255 of payload are filled, and these bytes represent the ConnData, as shown in (c).

Figure 12

Example of BLE topology [59]. In the figure, solid arrows point from master to slave; dashed arrows indicate a connection initiation and point from initiator to responder. Each device is represented with a capital letter; devices that are connected are represented with a circle, while devices that are advertising are indicated using stars. Group (a) in Figure 12 is a simple broadcasting topology, where A is an advertiser, while B and C are scanners, which are using a BLE advertising physical channel. Group (b) is a basic piconet, with only one physical channel, where D acts as master and E as slave. In group (c), the master is F, and it is using two piconet physical channels with slaves G and H. Device F is also the initiator of the connection with device I, which is advertising with connectable advertising packets on the advertising physical channel; device F can start the connection and add slave I to its piconet. A network topology like this one, with only one master and several slaves, is called a star network. In scatternet (d), device J is using one LE physical channel with K and one with L. J is the master in the piconet with L and the slave in the one with K. In scatternet (e), device M is the slave of two physical channels, whose masters are N and O. Device P is advertising using a connectable advertising event on the advertising physical channel, and the device M is the initiator; when the connection is formed, M will result in being the master of this link.

Figure 13

Maximum throughput of a Bluetooth Low Energy link for various connInterval (ranging between 7.5 ms and 4000 ms) and BER values (ranged from zero to 10${}^{-3}$): simulation (symbols) vs. analysis (lines). The simulation has been performed using 1,000,000 connEvents per each parameter set. Adapted from [65].

Figure 14

Multi slave communication in a network with a star topology (see Section 2.4). There is a master sending data with three slaves. This is the particular case of non-overlapping, i.e., when different slaves do not share the same interval of time for the communication with the master. In fact, as can be seen, all the connEvents are distributed without overlapping. The black arrows indicate how the time goes. The blue arrows outline the direction of the communication.

Figure 15

Theoretical maximum number of slaves per piconet for various types of interaction between devices, scheduling schemes and connInterval. The different types of interaction examined are the one-way and round-trip communications (described in Section 2.2.2). Moreover, the analysis has been done in the ideal case of non-overlapping communications, as well as in the case of overlapping, which denotes the upper bound limit. Adapted from [50].

Figure 16

Average discovery latency and energy consumption of the advertiser according to $T_a$ ($T_s$ = 1.28 s and T = 10.24 s) [78]. The green line shows the model proposed by [76], while the black and blue lines positioned in the lower part show the models analyzed in [78]. The crosses represent the simulation based on the theoretical model. As can be noticed, when $T_a$ > $T_s$, the two results do not agree anymore. On the other side, the red line [79] and the two blue and black lines positioned in the upper part of the figure represent the data relative to the latency, analyzed in Section 3.4. M is the number of pairs of scanners and advertisers existing in the communications, while $p_0$ is the failure probability due to the interference with other devices. Adapted from [78].

Figure 17

Average discovery latency and energy consumption of the advertiser according to $T_s$ ($T_a$ = 0.64 s and T = 2.56 s) [78]. The green line shows the model proposed by [76], while the black and blue lines positioned in the lower part show the models analyzed in [78]. The crosses represent the simulation based on the theoretical model. As can be noticed, when $T_s$ < $T_a$, the two results do not agree anymore. On the other side, the red line [79] and the two blue and black lines positioned in the upper part of the figure represent the data relative to the latency, analyzed in Section 3.4. M is the number of pairs of scanners and advertisers that can communicate with each other, while $p_0$ is the failure probability due to the interference with other devices. Adapted from [78].

Figure 18

Average current consumption, per variation of connInterval, measured in a CC2540 slave. The operating voltage of this IC is 3 V. It is a one-way communication with connSlaveLatenxy = 0. The Texas Instruments BLE module CC2540 [83] has a USB interface that implies a more relevant drawn current compared to I2C or SPI interfaces (e.g., the CC2541 [77] module has an I2C interface). A BLE module with a USB interface could be a good reference for a central node rather than a peripheral one. Adapted from [50].

Figure 19

Current consumption as a function of throughput for different BLE ICs. This value is computed using the maximum and the minimum number of packets per connEvent (PPCE) each device may support. Adapted from [85].

Figure 20

Average latency comparison with varied advertising duty ratio ($\alpha$) and a fixed scanning duty ratio ($\beta$ = 0.75). Adapted from [79].

Figure 21

Average latency comparison with varied scanning duty ratio ($\beta$) and a fixed advertising duty ratio ($\alpha$ = 0.3). Adapted from [79].

Figure 22

Average discovery latency and energy consumption of the advertiser according to $T_a$ ($T_s$ = 1.28 s and T = 10.24 s) [78]. The green line shows the model proposed by [76], while the black and blue lines positioned in the lower part show the models analyzed in [78]. The crosses represent the simulation based on the theoretical model. As can be noticed, when $T_a$ > $T_s$, the two results do not agree. On the other side, the red line [79] and the two blue and black lines positioned in the upper part of the figure represent the data relative to the latency, analyzed in Section 3.4. Adapted from [78].

Figure 23

Average discovery latency of the advertiser according to different values of Scan Interval. Blue and black curves indicate the theoretical model and the simulation proposed in [78], while the red lines represent the latency model of [79]. M is the number of pairs of scanners and advertisers, which can communicate with each other, while $p_0$ is the failure probability due to the interference with other devices. Adapted from [78].

Figure 24

Mean discovery latency with various parameter settings. (a) The mean discovery latency as the number of scanners is increased (${\tau }_{SI}$ = 10,240, ${\tau }_{SW}$ = 2560, ${\tau }_{WA}$ = 10, ${\tau }_{AI}$ = 1280); (b) the mean discovery latency as the number of advertisers increases (${\tau }_{SI}$ = 10,240, ${\tau }_{SW}$ = 2560, ${\tau }_{WA}$ = 10, ${\tau }_{AI}$ = 1280); (c) the mean discovery latency as ScanWindow (${\tau }_{SW}$) is varied (${\tau }_{SI}$ = 10,240, ${\tau }_{WA}$ = 10, ${\tau }_{AI}$ = 1280, M = 5, N = 5); (d) the mean discovery latency as ${\tau }_{WA}$ is varied (${\tau }_{SI}$ = 10,240, ${\tau }_{SW}$ = 2560, ${\tau }_{AI}$ = 1280, M = 5, N = 5); (e) the mean discovery latency as AdvInterval (${\tau }_{AI}$) is varied (${\tau }_{SI}$ = 10,240, ${\tau }_{SW}$ = 2560, ${\tau }_{WA}$ = 10, M = 5, N = 5). ${\tau }_{SI}$ is the scanInterval; ${\tau }_{SW}$ is the ScanWindow; ${\tau }_{AI}$ is the advInterval; M is the number of advertiser; N is the number of scanners. Adapted from [88].

Figure 25

Average latency for one-way and round-trip message exchanges, for various connInterval and BER values. Adapted from [50].

Figure 26

A graphic representation of path loss , obtained from Equation (6). Adapted from [55].