Beyond Co-existence: Exploiting WiFi White Space for ZigBee Performance Assurance Jun Huang 1, Guoliang Xing 1, Gang Zhou 2, Ruogu Zhou 1 1 Michigan State University, 2 College of William and Mary
ZigBee Networks Low communication power (10~50 mw) Application domains Smart energy, healthcare IT, Industrial/home automation, remote controls, game consoles…. Ex: 10 million smart meters installed in the US by 2010 Smart thermostat (HAI ) Smart electricity meter (Elster) Industrial sensor networks (Intel fabrication plant)
Challenge & State of the Art Interference in open radio spectrum Numerous devices in 2.4 GHz band: WiFi, bluetooth… AT&T public WiFi usage: 300% up Q1/09~Q1/10 [1] Multi-channel assignment WiFi interferes with 12 of total 16 ZigBee channels Co-existence on same/overlapping channels Carrier sense multiple access (CSMA) [1] http://attpublicpolicy.com/wireless/the-summer%E2%80%99s-hottest-hotspot/
Empirical Study of Coexistence WiFi interferer: 802.11g Change WiFi node location Measure ZigBee sending rate WiFi interference on sender Measure ZigBee packet delivery ratio WiFi interference on receiver Interference link Data link ZigBee sender and recver TelosB with CC2420 Method: change WiFi interferer position to create diversity Measure sending rate for evaluating the impact of WiFi interference on ZigBee Sender, packet delivery ratio for impact on ZigBee recver WiFi sending rate doesn’t change during the experiment WiFi Interferer Position
WiFi Hidden Terminals Don’t trigger backoff at ZigBee sender Corrupt packets at ZigBee receiver WiFi Interferer Position
WiFi Exposed Terminals Defer ZigBee sender’s transmissions Not strong enough to corrupt ZigBee packets WiFi Interferer Position
WiFi Blind Terminals Interfere both ZigBee sender and receivers Severe packet loss on ZigBee link WiFi sending rate not affected
Why Blind Terminals ? Power asymmetry Heterogeneous PHY layers WiFi only senses de- modulatable signals Energy-based sensing? ZigBee tx range ZigBee sender ZigBee recver WiFi interferer WiFi tx range 8
White Space in Real-life WiFi Traffic Large amount of channel idle time WiFi frames are clustered white space: cluster gaps that can be utilized by ZigBee Frame intervals: MAC layer intervals caused by 802.11 DIFS and backoff. Usually less then 50 micro seconds. The PHY and MAC layer header of ZigBee requires nearly 1 milliseconds. Inter-cluster intervals: Application layer intervals
Self-Similarity of Cluster Arrivals Variance is similar at different time scales Rigorously tested via rescaled range statistics and periodogram-based analysis # clusters/5s # clusters/s
Modeling WiFi White Space Length of white space follows iid Pareto distri. Implementation Collect white space samples in a moving time window Generate model by Maximum Likelihood Estimation α = 1ms shorter intervals are not usable for ZigBee Modeling method: Collect white space samples during a time window. Use Maximum Likelihood Estimation for model generation.
Pareto Model: Goodness of Fit Point represents the test result of a trace file. Divide the trace files into equal sized time windows. Drive model for each window. X(Y) shows the proportion of windows that passes the K-S(Independent) test. OSDI ’06 traces SigCOMM’08 traces Pareto model is accurate when modeling window < 100ms Sampling frequency is about 200Hz 20 samples are enough!
Outline Motivation Blind Terminal Problem WiFi White Space Modeling WISE: WhIte Space-aware framE adaptation Experimental Results 13
Basic Idea of WISE Sender splits ZigBee frame into sub-frames Fill the white space with sub-frames Receiver assembles sub-frames into frame WiFi frame cluster ZigBee sub-frames Modeling method: Collect white space samples during a time window. Use Maximum Likelihood Estimation for model generation. ZigBee Time sampling window ZigBee frame pending
Maximum ZigBee frame size Frame Adaptation Collision probability Sub-frame size optimization Sub-Frame size White space age ZigBee data rate 250Kbps Collision Threshold Maximum ZigBee frame size 15
Experiment Setting ZigBee configuration WiFi configuration TelosB with ZigBee-compliant CC2420 radios Good link performance without WiFi interference WiFi configuration 802.11g netbooks with Atheros AR9285 chipset D-ITG for realistic traffic generation Baseline protocols B-MAC and Opportunistic transmission (OppTx) Evaluation metrics Modeling accuracy, sampling frequency, delivery ratio, throughput, overhead 16
Frame Delivery Ratio Broadcast Unicast with 3 retx 17
Conclusions Empirical study of WiFi and ZigBee coexistence Blind terminal problem WiFi white space modeling Rigorous statistic analysis on real WiFi traffic WISE: White space aware frame adaptation Implemented in TinyOS 2.x on TelosB Significant performance gains over B-MAC and OppTx 18
Throughput Overhead 19
Throughput
WiFi Interference Summary Hidden terminal The WiFi node is located within the interference range of ZigBee receiver, but outside the CCA range of ZigBee sender. Exposed terminal The WiFi node is located within the CCA range of ZigBee sender, but outside the interference range of ZigBee receiver. Blind terminal The WiFi node is located within both the CCA range of ZigBee sender and the interference range of ZigBee receiver. Design flaw of CSMA CSMA supposed to work. Why blind terminals? 21
Self-Similarity of WiFi Frame Clusters Arrival process of frame cluster is self-similar Variance is similar at different time scales
WISE Protocol Design Original ZigBee frame Sub-frame layout WISE treat each MAC layer frame as a session MAC protocol independent Protocol overhead? Small sub-frames have low collision probability Large sub-frames are transmission efficient Payload PHY Hdr MAC Hdr CRC PHY Hdr MAC Hdr ID PHY Hdr ID Payload PHY Hdr ID Payload CRC WISE works between MAC and PHY. To make WISE independent with MAC, we didn’t implement retx in WISE. So there is only one CRC for each MAC layer frame 23
Frame Adaptation Optimal sub-frame size λ and ρ are measured on-line Average white space lifetime λ and ρ are measured on-line 24
Measure the White Space Model WiFi white space sampling Sampling the interrupt on CCA pin of CC2420: sampling frequency 4K~8KHz Record white space sample if Signal cannot be decoded Interval between signals is longer than 1ms Impact of ZigBee interference 25
Effect of Sampling Frequency 26
CSMA is NOT White Space Aware Collisions CCA Transmission ZigBee WiFi channel trace Modeling method: Collect white space samples during a time window. Use Maximum Likelihood Estimation for model generation. Time
ZigBee Link Performance Analysis What’s the prob. of colliding w/ WiFi packets? Analytical collision probability model ZigBee carrier sensing model White space model
Choose random waiting time T between [1, CW] Why Blind Terminals ? Heterogeneous PHY layer 802.11 backoff algorithm No 802.11 modulated packet in channel No Choose random waiting time T between [1, CW] Carrier Sense T=0? Count down T 802.11 modulated packet detected Yes Data ready Increase T by the packet duration Send ZigBee In-friendly 29