Presentation on theme: "Fine-grained Spectrum Adaptation in WiFi Networks"— Presentation transcript:
1Fine-grained Spectrum Adaptation in WiFi Networks Sangki Yun, Daehyeok Kim and Lili QiuUniversity of Texas at AustinACM MOBICOM 2013, Miami, USA
2Is wide channel always better? Current trend in WiFiWireless applications increasing throughput demandChannel width is increasingBenefit of wide channel: higher throughput802.11a/b/g20MHz802.11n40MHz802.11ac160MHzWireless demand iIs wide channel always better?
3Disadvantage of wideband channel High framing overheadHigh energy consumptionLower spectrum efficiency due to frequency diversitydataACKchannel accesspreambleSIFSwide channelwide channeltransmissionidleperiod
4LessonsStatic spectrum access (wide or narrow spectrum exclusively) is insufficientNeed dynamic spectrum access to get the best of both worldsOur lessons from the previous study on the channel width is that it is not sufficient to statically use either narrow or wide channel because both ones have their own advantages and disadvantages, and in order to achieve the benefit of both cases, the spectrum should be dynamically assigned.
5Ideal case: per-frame adaptation Clients select channel based on their preferenceAP needs per-frame spectrum adaptation to communicates with different clientsPreferred channel may change over time -> further increase the need for per frame adaptation20MHztime5MHz10MHz20MHzSpectrum efficiencyEnergy efficiency
6Challenges Enable per-frame spectrum adaptation Sender and receiver agree on the spectrumDynamically allocate spectrum efficientlyHowever, achieving such a dynamic spectrum access is very challenging. First, changing the spectrum is costly because if we try to do it in hardware level, it causes delay of at least a few microseconds. Also, it requires to send and receive frames from multiple spectrums using one radio hardware. The third thing is how to make the agreement on which spectrum to use. The receiver should know which spectrum is used before starting the symbol decoding, but if we introduce new control frame or control channel for this, its overhead will reduce the gain by adapting the spectrum. Also, we need an efficient algorithm to dynamically allocate the spectrum
7Related work Dynamic spectrum access (WiMAX, LTE, FICA) Requires tight synchronization among clientsSignificant signaling overheadSpectrum adaptation (SampleWidth, FLUID)Focus on spectrum allocation and ignore spectrum agreementSlow to adjust the channel widthWiFi-NCChannel width is fixed to 5MHzRequires longer CP to reduce guard bandwidthIEEE acRTS/CTS for dynamic bandwidth managementNot fine grained (minimum channel width 20MHz)There are some related work that addresses spectrum adaptation. At first, in cellular network, dynamic spectrum access is achieved based on OFDMA channel access mechanism. However, they require tight synchronization among clients and they incur significant signaling overhead, which is not suitable for wireless LAN environment. Also, SampleWidth and FLUID show the benefit of adapting the channel width. It changes the channel width in hardware level, so it pays some overhead by adapting the channel. Also, it does not take the spectrum agreement problem into account and assume the channel width is known. WiFi NC divides a spectrum into multiple 5Mhz channels, which is less flexible than FSA. Though they use very narrow guard band to reduce the overhead by using narrow channel, it requires to use longer cyclic prefix, which increases the overhead in time domain. IEEE ac introduces the concept of dynamic bandwidth management, but it relies on RTS/CTS for that. Also, the minimal channel width supported in it is 20MHz. Fortunately, the detection preamble design of ac is not changed, so FSA can complement ac to utilize narrow channel without modification in the standard.
8FSA: Fine-grained spectrum adaptation Per-frame spectrum accessChange spectrum per-frameCommunicate with multiple nodes on different subbands using one radioIn-band spectrum detection using existing preambleEfficient spectrum allocationIn this work, we propose a new system design that enables fine-grained spectrum adaptation that resolves the challenges in the previous slide. It’s backward compatible with existing standard, so based on narrowband communication supported in a, it adapt the channel width among 5, 10, 20MHz every frame. Also, its baseband design allows to use multiple spectrums simultaneously using one radio. Also, the spectrum detection algorithm does not require any control message overhead and preamble design changes. Finally, the spectrum allocation algorithm introduced here increases the benefit by frequency diversity.
10Generating narrowband signals Convert 5 or 10MHz signal based on 20MHz signal through upsampling and low pass filteringLPFupsamplingfrequency20MHzNarrowband signal20MHzfrequency20MHz signalUpsampling generates images outside tx bandfrequency20MHz
11Sending signals together Center frequency shifting is performed and the signals in different spectrum are added20HzNarrowband signal𝑠 10 [𝑛]20HzShifted signal𝑠 10 𝑓𝑠 𝑛Center frequency shifting𝑠 10 𝑓𝑠 𝑛 = 𝑠 10 [𝑛] 𝑒 𝑗2𝜋∆𝑠 𝑛 = 𝑠 10 𝑓𝑠 𝑛 + 𝑠 5 𝑓𝑠 𝑛adding another narrowband signalDeliver to RFRF20Hz20HzMixed signal𝑠[𝑛]
13Spectrum detector is key component Receiver designCF shiftLPFdown-samplerPHYdecoderRFSpectrum detector. . .. . .. . .CF shiftLPFdown-samplerPHYdecoderSpectrum detector is key component
14Spectrum detectorGoal: Receiver identifies the spectrum used by the transmitterPossible solutionsUse control channel or frameToo much overheadTarget for attackControl channel may not be always available further increase overheadDesign special preamble [Eugene,12]Deployment issueThe goal of the spectrum detector is to identify which spectrum is used before the receiver starts to decode the signals. For the spectrum detection, the conventional solution is relying on additional control channel or control frame such as RTS/CTS. This incurs too much overhead, and it can be easily targeted for attack by malicious users. Also, the congestion in the control channel can further increase the overhead. Another solution is designing a special preamble for the spectrum detection purpose, but it will lose the backward compatibility so it will be difficult to difficult to deploy the system.
15Spectrum detection using STF It is ideal to detect spectrum using existing frame detection preamble (STF)One solution: Spectral and Temporal analysis of the detection preamble (STD)Power spectral density to detect the total spectrum widthTemporal analysis to identify exact spectrum allocationCostly and inaccurate especially in noisy channelOur approachExploit special characteristics of STF for spectrum detectionOne possible way to detect the spectrum without control overhead and special preamble is performing the spectral density analysis in the frequency domain to identify the received signal spectrum, and perform temporal analysis to check how many frames are transmitted together in the spectrum. Our evaluation result shows it performs so bad in noisy channel condition. So instead, we propose a novel detection algorithm that exploits the special property of the detection preamble.
16We exploit the subcarrier interval for the spectrum detection! Characteristic of STFTime domain: 10 repetitions of 16 signalsFrequency domain: 12 spikes out of 64 subcarriers with 4 subcarrier intervalst1t2t3t4t5t6t7t8t9t10We exploit the subcarrier interval for the spectrum detection!
17Spectrum detector design (Cont.) Depending on the transmitter spectrum width, the received STF has various subcarrier intervals20MHzSubcarrier interval: 4Our main observation about the subcarrier interval is the interval changes depending on the channel width difference between transmitter and receiver. Given the 20MHz receiver,10MHzSubcarrier interval: 25MHzSubcarrier interval: 1
18Spectrum detection using STF 20MHz transmitter to 20MHz receiver20MHz receiver20MHz transmitter20MHzSTF in the frequency domain at the 20MHz receiver
19Spectrum detection using STF 10MHz transmitter to 20MHz receiverTwo subcarriers of 10MHz transmitter is merged into one subcarrier of 20MHz receiver20MHz20MHz receiver10MHz transmitterSTF in the frequency domain at the 20MHz receiver
20Spectrum detection using STF 5MHz transmitter to 20MHz receiver20MHz20MHz receiver5MHz transmitterSTF in the frequency domain at the 20MHz receiver
21Spectrum detection using STF The subcarrier interval difference let us easily identify the spectrum20MHz receiver20MHzSTF in the frequency domain at the 20MHz receiver20MHz receiver20MHz transmitter20MHz
22Spectrum detector design (Cont.) 10MHzTransform spectrum detection into pattern matching.5MHzOur main observation about the subcarrier interval is the interval changes depending on the channel width difference between transmitter and receiver. Given the 20MHz receiver,10MHz10MHz10MHz5MHz5MHz
23Spectrum detector design Cross-correlation checkMaximum likelihoodpattern matchingRF-frontendpreamble detectionFFT-64spectrum detectionReceived signal sampled in 20MHz rateMagnitude of 64 subcarriersOptimal Euclidean distance based spectrum detectionBinary detection𝐗 =arg min 𝑖 𝑘= 𝑦 𝑘 − 𝑥 𝑖 𝑘𝐗 =arg min 𝑖 𝐗 𝑖 ⊕𝐘
24Spectrum Allocation AP AP AP Controller client client client client bufferAPAPAPNow that we have the capability of fine-grained spectrum access, the next important question to ask is how to allocate spectrum to maximize efficiency?In this paper, we focus on optimize spectrum allocation for the downlink traffic, which is the dominant traffic.We consider the following architecture where there is a controller that performs optimization for all APs.clientclientclientclient
25Spectrum Allocation (Cont.) InputDestinations of buffered framesCSI between APs and clientsConflict graphGoal: Minimize finish timeAvoid interferenceHarness frequency diversityKnobsSpectrumScheduleAP used for transmissionThe controller takes the input of buffered frame destination, CSI between Aps and clients, and conflict graphs between different links in the graph, and tries to minimize the finish time of sending all buffered frames.In order to achieve this goal, we need to avoid interference and harness frequency diversity.We optimize finish time by selecting appropriate spectrum, schedule, and AP to use for transmission.
26Spectrum allocation (Cont.) Break a frame into mini-framesBreak the entire spectrum into mini-channelsGreedily assign a mini-frame to a mini-channel that minimizes the overall finish time while avoiding interferenceFind a swapping with an assigned mini-frame that leads to the largest improvement, go to step 31) To take advantage of fine-grained spectrum access, we … [just read the slide]
27Evaluation methodology Implemented testbed in Sora2.4GHz20MHz maximum bandwidthEvaluates detection accuracy and latency, spectrum allocation performance in testbedTrace based simulation for spectrum allocation in large-scale network
29Spectrum detection delay Median detection delay 4.2 us < detection delay budget
30Throughput evaluation – no interference FSA improves throughput by exploiting frequency diversity
31Throughput evaluation – interference In the another experiment, we added an interferer that sends signal in 2MHz narrowband. When the channel width is fixed, the client cannot avoid this narrowband interference, so the throughput is seriously reduced. When FSA is applied, the clients can effectively avoid the interference, so the impact of interference is marginal. In this evaluation, our scheme shows 110% higher throughput than the fixed channel case.With narrowband interference, the gain grows larger
32Summary FSA – a step towards enabling dynamic spectrum access Flexible baseband designFast and accurate channel detection methodSpectrum adaptation
34Comparison with WiFi-NC We performed extensive simulation to compare the performance of FSA with WiFi-NC. Here, we simulated a fading channel with RMS of delay spread is 100ns. In this simulation, FSA gives 20% higher throughput than WiFi-NC. There are two reasons for this. First, WiFi-NC uses longer CP which has larger constant overhead. Also, its signal quality was more degraded in fading channel because of the side effect of using sharp filter.Simulation in fading channel width RMS of delay spread = 100 nsWiFi NC incurs lower SNR due to sharp filtering
35Discussion Detection accuracy Antenna gain control Bi-directional traffic