A practical implementation of 6LoWPAN for metropolitan sized networks. Joseph A. Knapp and Nicolas Sornin Semtech Corporation Abstract This paper will illustrate a practical application of 6LoWPAN networking in a metropolitan sized network at Sub-GHz bands. The application will be based on off-the-shelf components adapted to a wireless metropolitan network of 1000s of sensors. Details such as network capacity and coverage area will be considered. It will be shown that a software defined radio gateway can enable a star network topology and eliminate the issues of mesh routing in such a 6LoWPAN network. Introduction and Problem Statement The deployment of the internet of things will not be limited to consumer devices, home networks, or even local area networks. Municipalities are already seeing the benefits of real time status of their distributed assets such as lighting, parking meters, and mobile equipment. Additionally, utility companies have for many years relied on wireless networks to lower their operational costs in both urban and rural environments. For both of these markets, battery powered sensors are mandatory, and battery life in the range of 15 years is expected. The ability to leverage off of standardized IPv6 addressing and IP protocols will enable device vendors and users of these markets the ability to quickly deploy and reuse applications built upon portable software libraries. However, the continued use of early 20th-century radio modulation technology makes the realization of this metropolitan area network (MAN) of smart devices extremely difficult. This paper will illustrate how an off-the-shelf radio transceiver using modern signal processing techniques, such as spread spectrum can enable a practical application of a wireless MAN. Possible Solutions Mesh Network: The mesh topology provides a reliable and scalable network. Generally, each node of a mesh network can act as a relay to enable 1 messages to “hop” from one node to another. In this way the nodes can be placed out of range of the network gateway. However, this method drains battery power due to its inherent requirement for every node to forward messages not intended for the listening node, increasing the amount of data the node must process and transmit. Additionally, it creates network latency due to the need to “hop” a message thru several nodes. Lastly, for many service providers the lack of a predictable battery life eliminates the possibility to use battery powered sensor nodes in a mesh topology. Specifically, the node’s current consumption depends on the physical location of the node relative to other nodes, the number of nodes in the mesh, and the amount of data to be forwarded as a consequence of other nodes. Star Network: This is a simple topology with nodes only able to communicate to base stations (“gateways”). The star network features are well described by Mr. Garg, Mr. Saroha, and Mrs. Lochab; The advantage of this type of network for wireless sensor networks is in its simplicity and the ability to keep the remote node’s power consumption to a minimum. It also allows for low latency communications between the remote node and the base station. The disadvantage of such a network is that the base station must be within radio transmission range of all the individual nodes and is not as robust as other networks due to its dependency on a single node to manage the network.i However, newly available radios from Semtech address the issue of radio transmission range by providing better link performance, better co-channel interferer rejection, multi-frequency communication, and link rate adaptation due to their use of modern spread spectrum modulation. Additionally, by moving the network management from the gateways to a dedicated network manager controlling all the gateways, we can provide communication redundancy and easily support roaming sensor nodes. Physical Layer Description What might this new physical layer look like when these radio improvements are applied to the star network topology previously mentioned? 2 Figure 1 illustrates the radio portion of this new star network with its gateways and sensor nodes. The gateways provide a link between the radio medium and the internet in order to transfer node data to the final application. Specifically, the gateways forward all the packets received on any of their radio channels to a single network manager that will provide the communications protocol. In this way the gateways are simple bridges translating between Figure 1 - Illustration of network two physical layers. The achievable performance of the proposed star network with the use of the new radio technology can be modeled and visualized. In the following figures the modeled gateways are placed on a regular 500 meter grid and the modeled sensor nodes are randomly dispersed on the grid at a density of 1000 per square kilometer. The link loss of each node to each gateway is computed using the well-established COST-WI (Walsh- Ikegami) model. The model setup simulates a dense urban environment with 12m high buildings and gateways placed on lamp posts of 7m height. One of the outputs of the model is a map indicating the location of connected and unconnected nodes due to link attenuation versus the maximum possible link margin. Figure 2 represents a map of the gateways and nodes when used in a Figure 2 - GFSK network map with unconnected nodes classical GFSK system 3 at a fixed data rate of 2.4Kbps with a sensitivity of -120dBm and radiating +14dBm in the 868MHz ISM band. The model predicts that under the conditions described, more than 6% of nodes will not be able to communicate with any gateway. As expected, the model reveals that the unconnected nodes are furthest from the gateways. In comparison, Figure 3 illustrates the capabilities of the new radio systems that provide greater link margins and variable data rates. The maximum radiated power is unchanged, at +14dBm. Specifically, in Figure 3 we have applied radio link margins varying from 151dB (This corresponds to a sensitivity of -137dBm at the minimum data rate) to 136dB (-125dBm sensitivity at maximum data rate with +11dBm radiated). Consequently, all the sensor nodes are able to connect to the gateways. Furthermore, Figure 3 - Spread spectrum network map each node uses the highest possible data rate given its link attenuation to the closest gateway maintaining a 10dB demodulation margin. This 10dB margin is called the installation margin and is one of the parameters set by the network manager. The model provides a few more details regarding network. At the left of Figure 4 is a histogram of link losses between all the nodes and the gateway providing the best link. Most nodes’ link loss is dominated by line of site loss. However, a significant portion of nodes appear to be shaded by modeled buildings. At the right of Figure 4 is a histogram of the Figure 4 - Attenuation and Data Rate histograms 4 the new radio enumerated data rates used by the nodes in our model. The model also indicates that 50% of the nodes in our new star network will be received by two (2) or more gateways. This will allow the network to have some redundancy should one of our gateways fail. For safety critical nodes the link rate adaptation layer can be adjusted so that gateway redundancy is always achieved. This can be achieved at the expense of a lower data rate for those nodes (hence longer radio range). Gateways implement a configurable radio front-end able to constantly monitor several radio channels (up to 8) and to demodulate simultaneously several packets on several channels using any of the available data rates. This is accomplished with an FPGA-based Figure 5 - Number of reachable gateways radio system that implements the multi-channel multi-modem receiver and transmitter. The multi-channel SDR enables the nodes to randomly select a new channel for each transmission. This adds protection against RF jammers and other RF channel effects that may limit the gateways ability to receive packets from the sensor nodes without requiring any prior synchronization between the nodes and the gateways. This also implies that nodes are free to move throughout the network from one gateway to another, and that nodes do not need to be associated to a specific gateway while keeping the advantage and robustness of frequency hopping. To further examine the operation of this proposed network, the model will be applied to a wireless parking monitoring use case. Applying the same parameters as before there will be 1000 parking sensor nodes per square kilometer. These parking sensor nodes may want to report battery voltage, health of the node, mode of operation, etc., a few times per day. The nodes will also want to reliably alert the application when a parking space becomes occupied or free. In general these devices would want to update their status only when something has changed. What portion of the network capacity does this use case require? 5 The physical link model previously used can also simulate the capacity of the network. Assuming that each parking sensor node needs to send a message every hour, and that each message is at maximum , 128 octets, the model reveals that a single channel GFSK system would incur a 3% duty cycle. This means that averaged over a long period of time the radio channel is used in average 3% of the time by the application. Clearly there is sufficient capacity, but this GFSK system also has no redundancy and 6.6% of nodes are not connected. Applying the new radio technology, 50% of the nodes are now received by two or more gateways. Additionally the traffic from nodes to gateways is now spread over multiple radio channels and on each channel, multiple data rates. Assuming the SDR system provides four (4) simultaneous receive channels, the resulting duty cycle becomes 1.7% per channel, with 100% of sensor nodes able to communicate with the gateways and more than half of the nodes having a redundant link. The multiple receptions also enable a very basic localization. By utilizing the signal strength of the received transmissions by the multiple gateways, the system can localize a sensor node to within a quadrant, e.g. North, South, East or West of a specific gateway. This enables tracking of sensor nodes attached to mobile assets like portable electronic traffic signs. Link Layer Description So far it has been shown how a star network with variable data rates should have sufficient capacity for upstream status data sent at random times. However, there may be messages that cannot operate without guaranteed delivery. By using a single network manager, as illustrated in figure 1, we can dynamically select the optimal downstream data path for each node and provide just in time acknowledgement when required. Typical Usage: Most of the time, perhaps 16 hours per day, the parking sensor will only need to update the end application on its battery health and affirm that it is still operational. This type of messaging can be implemented by a simple 6LoWPAN compressed UDP type packet transmitted to a multicast address assigned to our network. For every transmission of this status message, the sensor chooses a random RF channel and performs a channel activity detect of the wireless medium. Once the medium is free (typically) the node sends its status and returns to deep sleep. 6 This type of fixed interval, fixed length communication in a star network enables simple battery life estimation that is not possible with a mesh network topology. Alternatively, the parking space sensor may need to alert the application when the parking space becomes occupied or free. These alert messages indicate a transient state of the parking sensor and require a reliable delivery of the message to ensure that the end application has the correct state of the parking space, i.e. free or occupied. For this alert message scenario, the sensor selects a random RF channel to use for its upstream message to the gateway as normal. The radio of the sensor does a quick channel activity detect and upon determining the medium to be free, transmits its 6LoWPAN compressed UDP multicast packet. As shown earlier, the transmission may actually be received by two or more gateways. Packets are tagged by each gateway with received data rate, channel, signal to noise ratio, RSSI, and Time-ofArrival. Both gateways relay their message to the network manager. The network manager recognizes the duplicate messages received by the separate gateways and provides one copy of the message to the application layer, while updating its routing table based on the received RF information provided by each gateway. In this manner there is only a transitory association of a sensor node with a specific gateway based on the last RF channel measurements. The network manager may also determine that a change in the sensor node’s data rate or output power is appropriate to optimize network operation and may append commands to any downstream messages to be sent to the node. Meanwhile the node has returned to the sleep state. It will sleep in a low power state for a predetermined duration before awakening and listening for an acknowledgment. Before the 5 seconds from the initial transmission expires, the application layer has determined that an acknowledgement message is required. It prepares a response and provides that response to the network manager with the IPv6 address of the destination node. The network manager compresses the address to the 6LoWPAN format, and then appends any commands for the node to the UDP packet provided in addition to the precise time at which the acknowledgement should be sent. This concatenated message is sent to the gateway with the best radio link for the targeted node, along with the requirements for data rate, channel and output RF power level. The gateway receives the message with the commands and at the prescribed time begins transmitting with the requested power and 7 data rate on the requested channel. Meanwhile the sensor node will switch from sleep to listening state and receive the unicast 6LoWPAN compressed UDP message sent from the application along with any network manager commands appended to the UDP message. Should the node awake and not receive an acknowledgement it will retry the full procedure again after some random back-off. Network Management: While it has been shown how rate and power adaptation can be handled in unicast downstream messages, it may also be desirable to provide broadcast type messaging to minimize the RF channel utilization by the gateways. The difficulty in providing broadcast messaging in this scenario is a consequence of our need to have battery powered sensor nodes that are normally sleeping. By simply providing a pre-determined time for downstream broadcasts of aggregated packets, the network can provide this functionality. However, to address the large quantities of sensor nodes, the system must utilize some additional features of the new radio system and SDR gateways. Specifically, the new radio system provides for orthogonal data rates. The consequence is that the gateways can transmit multiple packets simultaneously on a single RF channel. In this manner the gateways can send multiple high speed downstream packets to nearby nodes while sending a low speed packet to the sensor nodes with the greatest link attenuation. Specifically, if the system uses four different data rates, then 5000 bytes of data can be sent in just 6.5 seconds, i.e. each node serviced by a gateway could parse 20 bytes of directed data from the aggregated packet transmissions while using less than 0.2% of the medium per hour. Figure 6 - Illustration of orthogonal variable spreading factor Conclusion It has been shown that with the development of long range, interferer-tolerant advanced modulation radio systems, and with the 8 deployment of a single network manager that the disadvantages of a star network can be overcome. The practical system described supports many desirable features in a metropolitan area network including mobile nodes, link rate adaptation, and a predictable battery life. Unlike a cellular system, this network does not require any frequency planning and behaves as a single cell for all nodes. It should also be clear that the examples illustrated can be applied to other applications such as AMR, lighting, alarm systems, etc. It has also been shown that a simple localization service can be provided by these new radio systems and centralized management. In fact, the modulation used by these new radio transceivers can provide nano-second accurate time-of-arrival at the gateways. In the future, the network manager will be able to use the accuracy of this timing to accurately pin point the origin of wireless packets. References i Garg, Punet, et al. “Review of Wireless Sensor Networks – Architecture and Applications.” International Journal of Computer Science & Management Studies May 2011:35-36. 9