Presentation is loading. Please wait.

Presentation is loading. Please wait.

mZig: Enabling Multi-Packet Reception in ZigBee

Similar presentations


Presentation on theme: "mZig: Enabling Multi-Packet Reception in ZigBee"— Presentation transcript:

1 mZig: Enabling Multi-Packet Reception in ZigBee
Thanks for Mo's introdution. This is Linghe from McGill University. I am very glad to introduce our mZig work here, which aims to realize the multi-packet reception in ZigBee.

2 Table of Contents Motivation Preliminary Design Implementation
1 Preliminary 2 Design 3 My talk will be divided into 5 parts. Implementation 4 Performance Evaluation 5

3 Table of Contents Motivation 1 2 3 4 5
Let's start from the first part "Motivation". 4 5

4 ZigBee Communication Standard: IEEE 802.15.4
Focus: low-power, low-cost, low-bitrate Applications: Sensor networks; Smart homes; Internet of things; Industrial control; 3 4 Based on IEEE .15.4, ZigBee is a competitive wireless technology, draw extensive interests. Different from high-power and high-bitrate WiFi, ZigBee focuses on the lowpower, low-cost, and low-bitrate communications, which has been widely used in sensor networks, smart homes, internet of things, and industrial control... 5

5 Concurrent Transmissions in ZigBee
1 Concurrent Transmissions in ZigBee 2 Tree topology Cluster topology 3 4 Since ZigBee based networks usually adopt tree or cluster topology, it is common that multiple transmitters need to send packets to one receiver, leading to a severe collision problem. 5

6 Collision Problem Collision 1 2 3 4 5
For example, in a smart home, when multiple sensors report their data to a control center simultaneously, the collision occurs. Collsion is a typical and fundamental problem in wireless communications. Collision 5

7 retransmission required
1 Related Work Collision Avoidance CSMA/CA[JSAC'04,ToN'08] RTS-CTS[AHNet'03] hidden terminal delay 2 3 Collision Resolution Analog NC[SigComm'07], XORs[SigComm'06], Full Duplex[MobiCom'11] Constructive Interference [IPSN'11, NDSI'13, TPDS'15] ZigZag [SigComm'08] partial known info synchronized same content retransmission required 4 Many and many excellent works have studied the collision problem from different directions. We classify the related work into two categories. The first category is collision avoidance. Conventional ZigBee adopts CSMA/CA and RTS-CTS to avoid collisions. However, CSMA cannot tackle the hidden terminal problem. And RTS-CTS increases the delay becuase of it handshaking process. Methods in the second category try to resolve collisions instead of avoid them. For example, analog network coding, XORs, and full duplex leverage interference cancellation to resolve collisions. But they require partial known information. Constructive interference is used to receive multiple synchronized transmissions. Its major limitation is that all packets must have the same content. ZigZag is the most related work to our work. It resorts to the time difference among multiple retransmissions to resolve the collision. However, the retransmissions reduces the throughput. 5

8 Decompose multiple packets from one collision directly!
1 Our Goal 2 No pre-known info No retransmission requirement Multiple TXs are allowed to send different packets 3 4 Decompose multiple packets from one collision directly! What we want to do is that: even there is no pre-known information, no retransmission requirement, and multiple transmitters are allowed to send different packets, we can still decompose multiple packets from one collision directly. 5

9 Table of Contents Preliminary 1 2 3 4 5
Before introducing our design, let's review some preliminary of the conventional ZigBee. 4 5

10 ZigBee Specification Frequency Coverage Data Rate # of Channels
1 ZigBee Specification 2 Frequency Coverage Data Rate # of Channels Rx Sensitibity Modulation 2.4GHz World 250kbps 16 -85dBm O-QPSK 868MHz Europe 20kbps 1 -92dBm BPSK 915MHz USA 40kbps 10 3 4 ZigBee specifies its operation in three ISM bands. This work focuses on the most widely used 2.4GHz ZigBee. 16 channels are allocated to ZigBee in this band, where the bandwidth of each channel is 2MHz, and the bit rate is 250kbps. 5

11 PHY in ZigBee

12 Baseband Signal at TX side
1 Baseband Signal at TX side 2 3 4 An example of a baseband signal at TX side is shown in the figure. In ZigBee, the smallest unit carrying information is a chip. In baseband, the chip 1 is shaped by a positive half-sine and the chip 0 is shaped by a negative half-sine. In addition, every chip has the same duration and the same amplitude. 5

13 Features of Chips at RX side
1 Features of Chips at RX side Oversampling: sampling rate of recent ADCs is much higher than 2MHz. Known shaping: half-sine. Uniform amplitude: O-QPSK, no ASK or QAM. 2 3 4 We observe that the received chips have three unique features. First, oversampling. The sampling rate of recent ADCs is much higher than 2MHz, so every chip can have multiple samples. Second, known shaping. Although there are some noise in wireless channel, the basic shape of a chip is maintained as a half-sine. Third, unlike ASK or QAM, which modulates signals by different amplitudes, ZigBee adopts O-QPSK modulation. As a result, the amplitudes of all chips are equivalent. Inspired by these observations, we think about how to use these features to resolve multi-packet collisions. How to use these features to resolve multi-packet collisions? 5

14 Table of Contents Design 1 2 3 4 5
Then, we introduce the design of mZig. 4 5

15 Core Design of mZig Example: a two-packet collision.
1 Core Design of mZig 2 Example: a two-packet collision. Alice: 11000; Bob: Every chip has 5 samples. 3 4 For ease to understand, we use a toy example to explain how mZig works. Two transmitters Alice and Bob send packets to the receiver simultaneously, causing a collision. Alice sends ‘11000’ and Bob sends ‘10100’. Every chip has five samples. 5

16 Two Categories of Collisions
1 Two Categories of Collisions 2 without chip-level time offset (w/o CTO) with chip-level time offset (w/ CTO) 3 4 Collisions are classified into two categories according to the chip-level time offset. The left figure shows the case that the chips betwee Alice and Bob are not aligned, we call it w/ CTO. On the other hand, the right figure shows the case w/o CTO. Especially, in this case, though two packets have a packet-level time offset, they still have no CTO. 5

17 1 CrossIC 2 Cross Interference Cancellation (CrossIC) for collision w/ CTO. 3 4 CrossIC is one core design in mZig, which leverages the features of ‘oversampling’ and ‘known shaping’ to decompose collision with CTO. The overlapped signal w/ CTO is shown in the figure, where the blue part presents the overlapped signal. Singal in this part cannot be decoded by conventional ZigBee. 5

18 1 CrossIC 2 Cross Interference Cancellation (CrossIC) for collision w/ CTO. 3 4 Using existing techniques, we can detect there are two packets in a collision and their time offset. 5

19 Step I: Extract collision-free samples
1 CrossIC 2 Cross Interference Cancellation (CrossIC) for collision w/ CTO. 3 Collision-free samples 4 The first step of CrossIC is to extract the collision-free samples. In this example, the first three samples are collision-free. 5 Step I: Extract collision-free samples

20 Step II: Estimate samples to form a whole chip
1 CrossIC 2 Cross Interference Cancellation (CrossIC) for collision w/ CTO. 3 Estimated samples 4 The second step is to estimate samples to form a whole chip. In this example, since every chip is oversampled by 5 samples, and the shape of a chip is a half-sine, we can estimate the next 2 samples based on the first 3 collision-free samples. 5 Step II: Estimate samples to form a whole chip

21 Substract the estimated chip from the collision
1 CrossIC 2 Cross Interference Cancellation (CrossIC) for collision w/ CTO. 3 4 Then, substract the estimated chip from the collision, we have two new collision-free samples. New collision-free samples 5 Substract the estimated chip from the collision

22 Repeat the extraction and estimation steps
1 CrossIC 2 Cross Interference Cancellation (CrossIC) for collision w/ CTO. 3 4 Repeat the extraction and estimation steps, a collision can be decomposed chip-by-chip. 5 Repeat the extraction and estimation steps

23 Repeat the extraction and estimation steps
1 CrossIC 2 Cross Interference Cancellation (CrossIC) for collision w/ CTO. 3 4 5 Repeat the extraction and estimation steps

24 Repeat the extraction and estimation steps
1 CrossIC 2 Cross Interference Cancellation (CrossIC) for collision w/ CTO. 3 4 5 Repeat the extraction and estimation steps

25 Repeat the extraction and estimation steps
1 CrossIC 2 Cross Interference Cancellation (CrossIC) for collision w/ CTO. 3 4 5 Repeat the extraction and estimation steps

26 Repeat the extraction and estimation steps
1 CrossIC 2 Cross Interference Cancellation (CrossIC) for collision w/ CTO. 3 4 5 Repeat the extraction and estimation steps

27 Repeat the extraction and estimation steps
1 CrossIC 2 Cross Interference Cancellation (CrossIC) for collision w/ CTO. 3 4 5 Repeat the extraction and estimation steps

28 1 AmpCoD 2 Amplitude Combination based Decomposition (AmpCoD) for collision w/o CTO. 3 4 To decompose a collision w/o CTO, we propose another core design in mZig, named AmpCoD, which leverages the unique feature of ‘uniform amplitude’. The overlapped signal w/o CTO is shown in the figure. After the statistics, we find there are totally 4 different combinations of amplitude, noted by L1 to L4, where L1 is the largest and L4 is the smallest. 5

29 1 AmpCoD 2 Amplitude Combination based Decomposition (AmpCoD) for collision w/o CTO. 3 Assume α>β L1 α+β L2 α-β L3 -α+β L4 -α-β 4 In addition, AmpCoD requires the amplitude difference between Alice and Bob. Assume α is larger than β. We can build a sorted matching relationship as shown in the table. For example, L1 is the largest one in all Ls and α+β is the largest one in all αβ combinations, so they are matched with each other. To this end, chips in the collision can be identified. For example, the second chip has amplitude L2, which matches α-βin the table. So this collided chip is identified by a chip 1 from Alice and a chip 0 from Bob. 5

30 Frequency Offset Compensation m-Packet Collision (m>2)
1 Core Design CrossIC AmpCoD Design Enhancement Time Offset Detection Anti-Noise Multipath Filter Frequency Offset Compensation Scope m-Packet Collision (m>2) Bluetooth WiFi 2 3 4 Besides the core design, in our paper, we also consider some design enhancement including time offset dection, anti-noise design, multipath filter, and frequency offset compensation. Furthermore, we also discuss how to decompose an m-packet collision, where m is larger than 2. And why mZig is able to transplant to Bluetooth and cannot applied in WiFi. I ignore their details due to the time limitation. Please read our paper if you have interests. 5

31 Table of Contents Implementation 1 2 3 4 5
We implement mZig on USRPs and build a testbed. Implementation 4 5

32 1 RX PHY: ZigBee v.s. mZig 2 3 4 mZig requires no change of physical layer at the transmitter side. And only a few changes are needed at the receiver side. Compared with ZigBee, mZig first adds a DmZig module to decompose baseband signals. In addition, we copy the decoding line M times to accelerate the decoding process of m packets. 5

33 1 DmZig Module 2 3 4 This figure illustrates the flow chart of DmZig module. DmZig can decompose the collision w/ CTO, w/o CTO, and also the hybrid case by serial CrossIC and AmpCoD. DmZig has three kinds of outputs. If there is no collision detected, it outputs the original sampls. If collision is detected and mZig can decompose the collision, it outputs the decomposed m sequences of samples. It is possbile that the collision is detected but mZig cannot decompose it. For example, in a too low SNR scenario. In this scenario, DmZig module outputs the original collided samples. Such a output is potential to trigger other methods to address this collision. 5

34 Testbed RX: USRP X310 + PC TX: USRP B210*6 + Laptop*6 + iRobots*6 1 2
4 Our testbed includes 1 receiver and 6 transmitters. The receiver adopts USRP X310, which has strong computational capability. The transmitters adopt USRP B210 because it is portable and power supplied by USB. So we can test mZig in both static and mobile scenarios. 5

35 Table of Contents Performance Evaluation 1 2 3 4 5
Extensive experiments are conducted to evaluate the performance of mZig. 4 Performance Evaluation 5

36 Experiment Setting Configuration Sampling rate: 32Msps TX power: 0dB
1 Experiment Setting 2 Configuration Sampling rate: 32Msps TX power: 0dB Channel selection: 26 Metrics Bit Error Rate (BER) Throughput 3 4 Compared with ZigBee ZigZag In the experiment, the default sampling rate is set 32Msps, which is the available largest rate in our testbed. The TX power is set 0dB. We select channel 26 in order to avoid the co-existence problem because this ZigBee channel has no overlap with WiFi channels. Our experiment is conducted in a general office area. The main metrics we evaluate are bit error rate and throughput, two common metrics in wireless communications. We compare the proposed mZig with the conventional ZigBee and the closest work ZigZag. Field 7.5m×6.8m office 5

37 BER: Different Sampling Rates
1 BER: Different Sampling Rates 2 3 Reference 4 This figure shows the BER performance of mZig in different sampling rates. The X-axis is the number of concurrent transmissions and the Y-axis is the log-scale of BER. We find that the sampling rate is the higher the better. We adopt the general reference that a packet is successfully decoded when its BER is smaller than ten to the minus three. This figure shows that 32Msps can achieve up to four concurrent transmissions. 5

38 BER: Different Techniques
1 BER: Different Techniques 2 3 Reference 4 This figure compares the BER among different techniques. ZigBee can decode only single transmission. Once the number of transmissions increases to 2, ZigBee's BER is larger than the reference. And ZigZag's BER is better than our mZig. 5

39 Throughput: Different Techniques
1 Throughput: Different Techniques 2 3 4 Although ZigZag's BER is better, mZig's throughput is much higher than ZigZag. The reason is: ZigZag requires retransmissions, which increases the delay and reduces the throughput. However, mZig realizes the multi-packet recpetion, so the throughput is significantly improved. 5

40 Throughput: Different Techniques
1 Throughput: Different Techniques 2 3 4 From this figure, we find the best throughput performance of mZig is at the four concourrent transmissions case, which also matches the BER result. 5

41 Throughput: Different Techniques
1 Throughput: Different Techniques 2 4.5X 3 4 At this case, the throughput of mZig is 4.5-fold of the conventional ZigBee. 1X 5

42 Throughput: Static v.s. Mobile
1 Throughput: Static v.s. Mobile 2 3 4 Moreover, we compare the throughput between static and mobile cases. The throughput performance in mobile case is only a little lower than the static case, which demonstrates mZig works in both static and mobile cases. 5

43 1 Conclusion 2 We design mZig, a novel RX design to enable multi-packet reception in ZigBee. Theoritcally, the maximal concurrent transmissions is m=S/2C. We implement mZig on USRPs. In our testbed, the throughput of mZig achieves 4.5x of ZigBee with four or more TXs. 3 4 To conclude this paper, we design mZig, a novel RX design to enable multi-packet reception in ZigBee. We prove that the ideal maximal concurrent transmissions is m=S/2C, where S is the sampling rate of ADCs and C is the chip rate. In addition, we implement mZig on USRPs. In our testbed, the throughput of mZig achieves 4.5-fold of ZigBee with four or more TXs. 5

44 Q & A linghe.kong@mail.mcgill.ca 44 Thanks for your attention!
Comments and questions are welcome! 44 44

45 Backup

46 Time Offset Detection

47 Anti-Noise Design For CrossIC For AmpCoD

48 Multipath Filter Channel estimation is required to estimate the impulse responses of multipath. Multipath effect is filtered chip-by-chip.

49 Frequency Offset Compensation
Channel estimation is also required to estimate the frequency offset. Compensate the frequency offset chip-by-chip.

50 Scope m-Packet Collision in ZigBee Bluetooth WiFi

51 MAC for mZig The conventional MAC for ZigBee cannot be applied directly: CSMA/CA ACK

52 Simulation Impact of CTO Impact of SNR

53 Simulation (cont) Multipath Filter Freq. Compensation


Download ppt "mZig: Enabling Multi-Packet Reception in ZigBee"

Similar presentations


Ads by Google