Comparison between Preamble Sampling and Wake-Up Receivers in Wireless Sensor Networks

Transcription

1 Comparison between Preamble Sampling and Wake-Up Receivers in Wireless Sensor Networks Richard Su, Thomas Watteyne, Kristofer S. J. Pister BSAC, University of California, Berkeley, USA Abstract Having a wake-up receiver constantly listening is often seen as a replacement for running a duty cycled medium access protocol on a commercial low-power radio chip. Wakeup receivers do offer a better latency while consuming negligible power. Recent wake-up receivers show an impressively low power consumption of 5µW, but at the cost of a sensitivity of -7dBm, -3dB higher than the sensitivity of a commercial radio chip. This difference in sensitivity causes the the wake-up receiver to have a much smaller communication range than the commercially available low power radio. In practice, this translates into requiring a denser deployment, or having to add an external power amplifier. This paper discusses the applicability of wake-up receivers in low-power wireless multihop networks. We show how, at available sensitivity levels, a wake-up receiver helps reduce the power consumption, but also requires a dramatically higher design or deployment cost. I. INTRODUCTION Wake-up receivers with extremely low power consumption and moderate sensitivity have been recently demonstrated. With such low power consumption, it is interesting to consider whether it is worthwhile to replace preamble sampling with wake-up receivers. When preamble sampling is used, a sender needs to transmit a preamble before sending the real data. Every node in the network will wake up periodically and listen for a short period time to check the preamble signal. The receiver, after receiving the preamble, will stay on to continue the communication with the sender, while the rest of nodes will go back to sleep. The amount of time between a node wakes up is called check interval (CI). In order for this mechanism to work, the length of the preamble signal needs to be longer than the check interval. The idea of using a wakeup receiver is to eliminate the need and power consumption to periodically wake up the main receiver. Instead, the ultra low power wake-up receiver will stay on to monitor the traffic at all time. The wake-up receiver sits right next to the main receiver. During the course of traffic monitoring, if a wake-up receiver hears a wake-up signal with the destination address of itself, it will relay this wake-up signal to the main receiver through a wired connection. The main receiver will wake up and start communicating with the sender. No periodic checking is necessary from the main receiver and the main receiver stays asleep unless it receives a wake-up signal from the wake-up receiver. The two settings that are being compared are the following. The first one is a commercially available low power radio with a wake-up receiver. The second is a commercially available low power radio without any wake-up receiver, and preamble sampling is used instead. Commercially available low power receivers have sensitivity much better than the stateof-the-art wake-up receivers, but they have much higher power consumption compared to wake-up receivers. Section II of this paper introduces two recently published state-of-the-art wakeup receivers and five commercially available IEEE radios will also be discussed. A fair way to compare the power consumption of a network using wake-up receivers with one using preamble sampling is to compare the optimum power consumption operating point of a network using preamble sampling with that of a network using wake-up receivers. Section III of this paper covers how to locate the optimal operating point, in terms of the power consumption, of a network using preamble sampling. Since the wake-up range is smaller than the communication range [1], one method to compensate for the inferior sensitivity of a wake-up receiver is to allow the transmitter to output higher power. An external power amplifier can be attached to the transmitter for this purpose. Note that this solution will incur inconvenient hardware design problems in reality, and higher design cost. In certain situations, such as regulatory requirements, it might not even be feasible for the transmitter to exceed certain transmitter output power. However, to get an insight on the power consumption of the two different settings mentioned earlier, section IV of this paper performs comparison with the assumption that transmitter output power does not exceed regulatory requirements. In an event that output power is limited, other methods need to be applied to alleviate the fact that wake-up receivers have inferior sensitivity. Deploying more nodes into the fields is the method chosen for the comparison in section V of this paper. We will give concluding remarks in section VI. II. WAKE-UP RECEIVERS AND LOW-POWER RADIOS There have been published wake-up receivers with extremely low power consumption in recent years. The first one is from Berkeley Wireless Research Center, and the wakeup receiver has -7dBm sensitivity with only 5µW of power consumption []. This receiver uses On-off keying modulation scheme and runs at GHz. The second one is a.4ghz receiver that has -69dBm sensitivity with 51µW of power consumption [3]. The summary of these two wake-up receivers are in Table I. There are several commercially available low power radios operating in the.4ghz band. The power consumption ranges

2 Operating Frequency RX Power Consumption GHz 5µW -7dBm [].4GHz 51µW -69dBm [3] TABLE I STATE-OF-THE-ART WAKE-UP RECEIVERS Model Power Consumption ATMEL AT86RF31 36mW -11dBm [4] DUST DN58mW -9dBm [5] FREESCALE MC13 1mW -9dBm [6] ST M3W 77mW -99dBm [7] TI CC5 5mW -98dBm [8] JENNIC JN mW -95dBm [9] TABLE II COMMERCIALLY AVAILABLE LOW POWER RADIOS from 18mW to 1mW, and the sensitivity ranges from - 11dBm to -9dBm. A summary of six low power radios in given in Table II. ATMEL AT86RF31 is chosen as the main radio for all comparisons done in this paper because AT86RF31 has the best sensitivity among the five low power radios listed. The sensitivity of the AT86RF31 is 9dB better than the best reported wake-up receiver and the power consumption of AT86RF31 is close to 36mW for either the transmitter or the receiver. III. PREAMBLE SAMPLING OPTIMUM OPERATING POINT Table III contains a list of variables used in this paper. In order to find the optimum operating point of a network using preamble sampling, we need to see how duty cycle depends on the check interval. The operating point with the lowest duty cycle, but still result in successful network communication, is the operating point with lowest power consumption. Note that this optimum operating point will not result in smallest latency, but lowest power consumption. It can be shown [1]: DutyCycle = (CI + (N 1)CI ) + D N CI N For optimum power consumption, there is a CI that results in lowest duty cycle in a preamble sampling scheme given a fixed. For example, with N = 1, D = 4µs, and = 6/1, the duty cycle versus check interval plot is shown at Figure 1, and the CI that has optimal power consumption N CI D Number of motes in a 1-hop neighborhood that can communicate Number of messages sent among N nodes per second Check interval for the preamble sampling Sampling time every CI TABLE III VARIABLES IN THIS PAPER (1) Duty Cycle in Percentage Check Interval in milliseconds Fig.. Optimum Point Fig. 1. Check Interval in milliseconds Optimum Operating Point of Preamble Sampling Check Interval Duty Cycle 1 Duty Cycle in Percentage Optimum Check Interval and Corresponding Duty Cycle versus is 1ms. As the parameter changes, the optimal CI will change accordingly, as shown in Figure. IV. WAKE-UP RECEIVER VERSUS PREAMBLE SAMPLING As mentioned previously, one method to compensate for the 9dB inferior sensitivity of a wake-up receiver is to add an external power amplifier (PA) at the transmitter (TX) to boost the TX output power. Assuming that the ATMEL AT86RF31 is transmitting at its maximum output power of +3dBm, the external PA will need to boost up the output power to +3dBm. In this output power range and output frequency of.4ghz, it is fair to assume the PA efficiency is approximate 3% from looking at the datasheet of commercially avaiable PA. With this output power and PA efficiency, we can back calculate the power consumption of the external PA to be 5.3W. The latency of a receiver with preamble sampling can be estimated to be the check interval (CI). When a wakeup receiver is used, the latency will be the amount of time

3 Total power consumption of all the motes from preamble sampling, including TX sends preamble, and RX periodically wakes up 1.69mW Power consumed when a TX sends 4ms (1 packet) data and a RX receives 4ms of data.88mw Summation of previous two rows 1.98mW 1-hop latency 7ms TABLE IV POWER CONSUMPTION AND LATENCY WITH PREAMBLE SAMPLING Total power consumption of all wake-up receivers 5µW Power consumed when a TX sending a wake-up signal 3 bit (source address + destination address) long 1.71mW Power consumed when a TX sends 4ms (1 packet) data and a RX receives 4ms of data.88mw Summation of previous three rows.5mw 1-hop latency 1.3ms TABLE V POWER CONSUMPTION AND LATENCY WITH WAKE-UP RECEIVERS Latency in milliseconds 1 Fig. 3. Preamble Sampling Wake Up Receiver Latency: Preamble Sampling vs Wake-Up Receivers required to wake up the main receiver (RX) after the wakeup receiver acquires the wake-up signal. The wake-up time of ATMEL AT86RF31 is 1ms [4]. At this point, it could be beneficial to look at some real numbers. A wake-up receiver with sensitivity of -7dBm consumes 5µW with 1% duty cycle. A main radio, AT86RF31, consumes 13.mA from a.7v supply, i.e. 36mW. AT86RF31 has similar power consumption in either TX or RX mode. In the case of being 1 and N being 1, a network using preamble sampling has optimum CI of 7ms and duty cycle of.97%. The total power consumption to wake up a node in preamble sampling is 1 * 36mW *.97% = 1.69mW with latency of 7ms. The actual data transmission, assuming a 4ms long packet, will consume * 36mW * 4ms * 1/s =.88mW. This gives total power consumption of 1.98mW, summarized in Table IV. In the case of using wake-up receivers, the transmitter will need to send out a total of 5.3W + 36mW = 5.336W as discussed earlier in this section. With being 1 and N being 1, assuming that a transmitter only needs to send out the source address and destination address, total of 3 bits, in order to wake up a receiver, the TX power consumption is: 5.336W 3bit 1/s = 1.71mW () 1kbps The total power consumption of all wake-up receivers is 1*5µW = 5µW. The 1-hop latency in this case will be the wake-up time of the AT86RF31,1ms, plus the time it takes TX to transmit the wake-up signal, which is 3bit/1kbps=.3ms. Table V summarizes these results. Figure 3 is a comparison between networks using preamble sampling and ones using wake-up receivers in terms of the latency. Networks using wake-up receivers clearly have the advantage of lower latency. Power Consumption (mw) Fig. 4. Preamble Sampling Wake Up Receiver 1 Power Consumption: Preamble Sampling vs Wake-Up Receivers Figure 4 is a comparison between networks using preamble sampling and ones using wake-up receivers in terms of the power consumption. It is interesting to see that when network traffic is low, total power consumption of networks using preamble sampling is higher. However, when network traffic increases, the network using wake-up receivers will end up consuming more power overall. This is because the transmitters in those kind of networks need to have additional 9dB output power as discussed earlier. In a network using wake-up receivers, if most of the traffic is coming from a small number of nodes in the network, those nodes are more likely to run out of energy compared to if preamble sampling was used. In summary, networks using wake-up receivers will have lower latency. However, it bears higher probability of nodes running out of energy, especially if the information is mostly sent out from a small number of nodes.

4 V. WHAT HAPPENS IF OUTPUT POWER IS LIMITED Analysis in previous section relies on one assumption: the transmitter output power can be scaled up by 9dB to make up for the sensitivity difference between AT86RF31 and the wake-up receiver in []. In some situations, becaseu of regulatory requirements, scaling up TX output power is not a feasible solution. If wake-up receivers are to be used, more nodes need to be scattered in the field to result in the same degree of connectivity. As a result, when transmitter output power cannot be scaled, one fair approach is to randomly add nodes in the field until all nodes achieve connectivity. A simulation is run in MATLAB, in which nodes are randomly scattered in a 1 meter by 1 meter field. TX output power is dbm. Based on the discussion in previous sections, we know that a wake-up receiver has sensitivity of -7dBm and AT86RF31 has a sensitivity of - 11dBm. We use the path-loss model in [11]: P r = P t G t G r λ U(,4)dB (3) (4πr) The P r is the received power, P t is the transmitted power, G t is the transmitter antenna gain, G r is the receiver antenna gain, λ is the wavelength, r is the distance between TX and RX and U(,4)dB represents an additional path loss uniformly distributed between db and 4dB. We ran simulations by picking a field size, 1 meter by 1 meter, and adding in nodes at random locations inside the field. Without loss of generality, the first node randomly generated will be chosen as the gateway. Additional notes are populated in the 1m X 1m fields until every single note in the field can communicate with the gateway either directly, or through multiple hops. When a certain number of nodes randomly scattered within the field can all communicate with the gateway, the run is considered fully connected. Only when there is 95% success rate (95 runs out of 1 runs achieve full connectivity) in deployment do we consider the deployment has sufficient number of nodes in the field. For receiver with sensitivity of -11dBm, only 6 nodes need to exist in the field and an average of 1.3 hops to result in fully connected network. However, for wakeup receiver, it requires 66 notes and an average of 3. hops in order to achieve full connectivity. Figure 5 shows the number of nodes needed to get full connectivity along with average number of hops when the wake-up receivers have different sensitivity. Within a 1m X 1m field, when the sensitivity gets above -8dBm, the number of nodes needed to result in full connectivity shoots up, which is an indicator that a receiver with sensitivity above -8dBm will result in significant increase on deployment cost. To gain more solid proof of how sensitivity impacts the number of nodes needed in a deployment, we also ran simulation in different field sizes, 5 meter by 5 meter (Figure 6) as well as meter by meter (Figure 7). From all three sets of simulation results, we see that the number of nodes increases slowly when the sensitivity is at or better than - 9dBm range. However, as it approaches -8dBm, the number 3 1 1m X 1m field Fig. 5. Needed for Full Connectivity and Hops versus 6 4 5m X 5m field Fig. 6. Needed for Full Connectivity and Hops versus of nodes needed to cover any of the three field sizes increases dramatically. In summary, networks using wake-up receivers with sensitivity worse than -8dBm will need a significant larger number of nodes to achieve full connectivity and hence have much higher deployment cost. In summary, with the current sensitivity, networks using wake-up receivers may need to deploy a significantly larger number of nodes into the field in order to achieve full connectivity. VI. CONCLUSIONS AND FUTURE WORK This paper compares two different approaches in achieving low-power communication in wireless multi-hop networks, namely, preamble sampling medium access control protocols and wake-up receivers. Assuming the transmit power can be scaled up without any limitation, wake-up receivers can achieve lower latency. In

5 1 5 m X m field Fig. 7. Needed for Full Connectivity and Hops versus reality, however, adding an external power amplifier to output 3dBm significantly increases the hardware design complexity. In a network with low traffic, wake-up receivers can also help reduce power consumption. However, as the network traffic increases, because of the additional TX output power needed to compensate for the inferior sensitivity of wake-up receivers, the overall power consumption of a network using wake-up receivers might go beyond that of one using preamble sampling. Finally, because of the additional power consumed by the external amplifier, nodes can drain their battery faster compared to if preamble sampling was used. 4 In an event that TX output power cannot be scaled up, and more nodes need to be deployed to achieve network connectivity, networks using wake-up receivers will need a larger number of nodes to achieve full connectivity. When the sensitivity of wake-up receivers gets worse than -8dBm, the number of nodes needed increase dramatically, and at that point, it is hard to justify the significant increase in the amount of additional deployment cost needed to set up the network. REFERENCES [1] I. Demirkol, C. Ersoy, and E. Onur, Wake-Up Receivers for Wireless Sensor Networks: Benefits and Challenges, IEEE Wireless Communications, vol. 16, no. 4, pp , August 9. [] N. M. Pletcher, S. Gambini, and J. M. Rabaey, A GHz 5 microwatt Wake-Up Receiver with -7dBm Using Uncertain-IF Architecture, in International Solid State Circuits Conference (ISSCC). San Francisco, CA, USA: IEEE, 3-7 February 8. [3] W. Huang, S. Rampu, X. Wang, G. Dolmans, and H. d. Groot, A.4GHz/915MHz 51 microwatt Wake-Up Receiver with Offset and Noise Suppression, in International Solid State Circuits Conference (ISSCC). San Francisco, CA, USA: IEEE, 7-11 February 1. [4] AT86RF31 Low Power.4GHz Transceiver for ZigBee, IEEE , and ISM Applications, [data sheet, [5].4 GHz Mote-on-Chip SMARTMESH IA-51 Dust Networks DN51, [data sheet, [6] Freescale Semiconductor MC13 Technical Data, Rev. 1.5, [data sheet, [7] STM3W 3-bit ARM CortexTM-M3 IEEE SoCs, [data sheet, [8] CC5.4 GHZ IEEE /ZigBee RF Transceiver, [data sheet, [9] Jennic JN5148 IEEE Wireless Microcontroller, [data sheet, [1] T. Watteyne, Energy-Efficient Self-Organization for Wireless Sensor Networks, Ph.D. dissertation, Institut National des Sciences Appliquees de Lyon, 8. [11] L. Doherty, W. Lindsay, and J. Simon, Channel-Specific Wireless Sensor Network Path Data, in 16th International Conference on Computer Communications and Networks (ICCCN). Turtle Bay Resort, Honolulu, Hawaii, USA: IEEE, August , pp