Accurate WiFi Packet Delivery Rate Estimation and Applications Owais Khan and Lili Qiu. The University of Texas at Austin 1 Infocom 2016, San Francisco.

Slides:



Advertisements
Similar presentations
Feedback Reliability Calculation for an Iterative Block Decision Feedback Equalizer (IB-DFE) Gillian Huang, Andrew Nix and Simon Armour Centre for Communications.
Advertisements

VSMC MIMO: A Spectral Efficient Scheme for Cooperative Relay in Cognitive Radio Networks 1.
The Impact of Channel Estimation Errors on Space-Time Block Codes Presentation for Virginia Tech Symposium on Wireless Personal Communications M. C. Valenti.
VTrack: Accurate, Energy-Aware Road Traffic Delay Estimation Using Mobile Phones Arvind Thiagarajan, Lenin Ravindranath, Katrina LaCurts, Sivan Toledo,
Model-Driven Energy-Aware Rate Adaptation M. Owais Khan, Vacha Dave, Yi-Chao Chen Oliver Jensen, Lili Qiu, Apurv Bhartia Swati Rallapalli 1 MobiHoc 2013,
Strider : Automatic Rate Adaptation & Collision Handling Aditya Gudipati & Sachin Katti Stanford University 1.
Channel Estimation for Mobile OFDM
Oct 21, 2008IMC n Under the Microscope Vivek Shrivastava Shravan Rayanchu Jongwon Yoon Suman Banerjee Department Of Computer Sciences University.
ICTON 2007, Rome, Italy The Performance of PPM using Neural Network and Symbol Decoding for Diffused Indoor Optical Wireless Links 1 S. Rajbhandari, Z.
Doc.: IEEE /568r0 Submission Frequency Selective Scheduling (FSS) for TGax OFDMA May 2015 Slide 1 Date: Authors: Kome Oteri (InterDigital)
ISSPIT Ajman University of Science & Technology, UAE
Diversity Combining Technique for Soft Handoff in OFDMA Cellular Systems Xiu-Sheng Li and Yuh-Ren Tsai Presented by Xiu-Sheng Li ( 李修聖 ) Wireless.
4. Cellular Systems: Multiple Access and Interference Management Fundamentals of Wireless Communication, Tse&Viswanath 1 4. Cellular Systems: Multiple.
1 Enhancement of Wi-Fi Communication Systems through Symbol Shaping and Interference Mitigation Presented by Tanim M. Taher Date: Monday, November 26 th,
Submission doc.: IEEE /1452r0 November 2014 Leif Wilhelmsson, EricssonSlide 1 Frequency selective scheduling in OFDMA Date: Authors:
Improving the Performance of Turbo Codes by Repetition and Puncturing Youhan Kim March 4, 2005.
A Low-Power Low-Memory Real-Time ASR System. Outline Overview of Automatic Speech Recognition (ASR) systems Sub-vector clustering and parameter quantization.
Muhammad Imadur Rahman1, Klaus Witrisal2,
CEFRIEL Deliverable R4.1.5 MAIS adaptive and reconfigurable modem Giovanni Paltenghi Roma – 24 Novembre 2005.
The University of Texas at Austin
Improvements in throughput in n The design goal of the n is “HT” for High Throughput. The throughput is high indeed: up to 600 Mbps in raw.
Joint PHY-MAC Designs and Smart Antennas for Wireless Ad-Hoc Networks CS Mobile and Wireless Networking (Fall 2006)
1 PROPAGATION ASPECTS FOR SMART ANTENNAS IN WIRELESS SYSTEMS JACK H. WINTERS AT&T Labs - Research Red Bank, NJ July 17,
A 4G System Proposal Based on Adaptive OFDM Mikael Sternad.
1 Adaptable applications Towards Balancing Network and Terminal Resources to Improve Video Quality D. Jarnikov.
Doc.: n-proposal-statistical-channel-error-model.ppt Submission Jan 2004 UCLA - STMicroelectronics, Inc.Slide 1 Proposal for Statistical.
Sunghwa Son Introduction Time-varying wireless channel  Large-scale attenuation Due to changing distance  Small-scale fading Due to multipath.
Soft-in/ Soft-out Noncoherent Sequence Detection for Bluetooth: Capacity, Error Rate and Throughput Analysis Rohit Iyer Seshadri and Matthew C. Valenti.
Space-Time and Space-Frequency Coded Orthogonal Frequency Division Multiplexing Transmitter Diversity Techniques King F. Lee.
Harnessing Frequency Diversity in Wi-Fi Networks Apurv Bhartia Yi-Chao Chen Swati Rallapalli Lili Qiu MobiCom 2011, Las Vegas, NV The University of Texas.
CSpy: Finding the Best Quality Channel without Probing Souvik Sen, Bozidar Radunovic, Jeongkeun Lee, Kyn-Han Kim MobiCom’13 1.
Scrutinizing bit-and symbol-errors of IEEE Communication in Industrial Environments Filip Barac, Student Member, IEEE, Mikael Gidlund, Member,
Wireless Multiple Access Schemes in a Class of Frequency Selective Channels with Uncertain Channel State Information Christopher Steger February 2, 2004.
Recitation: Rehearsing Wireless Packet Reception in Software
Replicating Memory Behavior for Performance Skeletons Aditya Toomula PC-Doctor Inc. Reno, NV Jaspal Subhlok University of Houston Houston, TX By.
ITERATIVE CHANNEL ESTIMATION AND DECODING OF TURBO/CONVOLUTIONALLY CODED STBC-OFDM SYSTEMS Hakan Doğan 1, Hakan Ali Çırpan 1, Erdal Panayırcı 2 1 Istanbul.
TCP-Cognizant Adaptive Forward Error Correction in Wireless Networks
Improving Loss Resilience with Multi- Radio Diversity in Wireless Networks by Allen Miu, Hari Balakrishnan and C.E. Koksal Appeared in ACM MOBICOM 2005,
Performance of Adaptive Beam Nulling in Multihop Ad Hoc Networks Under Jamming Suman Bhunia, Vahid Behzadan, Paulo Alexandre Regis, Shamik Sengupta.
Space Time Codes. 2 Attenuation in Wireless Channels Path loss: Signals attenuate due to distance Shadowing loss : absorption of radio waves by scattering.
EE359 – Lecture 12 Outline Combining Techniques
A Simple Transmit Diversity Technique for Wireless Communications -M
MIMO: Challenges and Opportunities Lili Qiu UT Austin New Directions for Mobile System Design Mini-Workshop.
Sujan Rajbhandari LCS Convolutional Coded DPIM for Indoor Optical Wireless Links S. Rajbhandari, N. M. Aldibbiat and Z. Ghassemlooy Optical Communications.
Model-Driven Energy-Aware Rate Adaptation
An Analytical Study of Wireless Error Models for Bluetooth Networks Hao-Hsiang Hung and Ling-Jyh Chen Academia Sinica.
Doc.: IEEE g Submission December 2009 Tim Schmidl, Texas Instruments Inc.Slide 1 Project: IEEE P Working Group for Wireless Personal.
Partially Overlapped Channels Not Considered Harmful Arunesh Mishra, Vivek Shrivastava, Suman Banerjee, William Arbaugh (ACM SIGMetrics 2006) Slides adapted.
Amplifier Nonlinearities in OFDM Multiple Antenna Systems FERNANDO GREGORIO Signal Processing Laboratory HUT.
Courtesy Piggybacking: Supporting Differentiated Services in Multihop Mobile Ad Hoc Networks Wei LiuXiang Chen Yuguang Fang WING Dept. of ECE University.
Data and Computer Communications Tenth Edition by William Stallings Data and Computer Communications, Tenth Edition by William Stallings, (c) Pearson Education.
Doc.: IEEE 11-04/0304r0 Submission March 2004 John S. Sadowsky, Intel PER Prediction for n MAC Simulation John S. Sadowsky (
GPGPU Performance and Power Estimation Using Machine Learning Gene Wu – UT Austin Joseph Greathouse – AMD Research Alexander Lyashevsky – AMD Research.
Diversity.
Wireless Communication
Space-Time and Space-Frequency Coded Orthogonal Frequency Division Multiplexing Transmitter Diversity Techniques King F. Lee.
Stateful Inter-Packet Signal Processing for Wireless Networking
Recent Advances in Iterative Parameter Estimation
Coding and Interleaving
Howard Huang, Sivarama Venkatesan, and Harish Viswanathan
Su Yi Babak Azimi-Sadjad Shivkumar Kalyanaraman
Ian C. Wong, Zukang Shen, Jeffrey G. Andrews, and Brian L. Evans
mEEC: A Novel Error Estimation Code with Multi-Dimensional Feature
Comments on PHY Abstraction
Simulation Methodology Proposal
Packet Error Probability Prediction for MAC Simulation
Joint Channel Estimation and Prediction for OFDM Systems
Sofia Pediaditaki and Mahesh Marina University of Edinburgh
PER Prediction for n MAC Simulation
PHY Signaling for Adaptive Repetition of 11p PPDU
Presentation transcript:

Accurate WiFi Packet Delivery Rate Estimation and Applications Owais Khan and Lili Qiu. The University of Texas at Austin 1 Infocom 2016, San Francisco.

2 Frequency Diversity Wireless channel are frequency selective Bandwidth is increasing with new standards BW (MHz) Channel Subcarriers SNR (dB) ~22dBs Wider Channels experience more frequency diversity

3 Challenges Frequency diversity results in bursty errors – Lower SNR subcarriers have more errors Delivery Rate hard to predict for bursty errors even with Channel State Information (CSI) Existing techniques do not perform well in bursty errors

Limitation of Average SNR Avg. SNR is widely used for delivery rate estimation SNR is not a good predictor in Freq. Selective Channels Delivery Rate Ref. Halperin10 12 dB Gap 4

Effective SNR Effective SNR shows better results than Avg. SNR Approach 1.Map the SNR per subcarrier to BER 2.Map BER eff,k back to Effective SNR 3.Use Effective SNR to select the appropriate rate How Accurate is Effective SNR? 5

Effective SNR Accuracy Effective SNR is also not very accurate Scatter plot of Estimated delivery rate vs. gnd. truth 95% 39% Gap of 56% 6

Error Burstiness across Frame code rate: 1/ interleaver does not uniformly distribute bits 7 WiFi interleaver has a skewed distribution Error still bursty in WiFi interleaver 65% Delivery rate estimation must incorporate error burstiness

Contributions Develop two new techniques to estimate delivery rate in bursty errors Propose a new Interleaver to reduce burstiness 8

Goal – To accurately predict delivery ratio using CSI information Problem Formulation Options 1.Analytical modeling Intractable for freq. selective channels. 2.Simulate online Prohibitively expensive 3.Lookup table based approach Pre compute delivery rate for error patterns Error patterns capture the burstiness 4.Machine learning based approach Use supervised learning to estimate delivery rate 9 Our Approach

10 Lookup Table: Idea Decoding success depend on error pattern to FEC Decoder – Same errors have different outcome depending on location Receive Chain De-ModulationDe-InterleaveFEC Decode Data Analog Signal j, j,… 1, 0, 1, 1, 0, … Idea: “Calculate the error probability of all possible patterns offline and use them to estimate delivery rate for given CSI”

Run Viterbi decoding offline and build a lookup table Use error pattern as input Lookup Table: Approach Independent of wireless channel or hardware Using an error pattern for frame prohibitively expensive … … 11

Errors in frame decouple after a certain period Lookup Table: Observation Error seq. 1Error seq. 2 The de-coupling length is dependent on code rate Code RateWindow Size 1/275 bits 2/350 bits 3/450 bits 5/640 bits 2^40 is still a very large number! Independent! 12

We find upper and lower error thresholds – Errors less than lowerThresh, decoding always successful – Errors greater than upperThresh, decoding always fails Can we reduce size further? Code Rate LowerThreshUpperThresh 1/2510 2/335 3/424 5/624 Complexity reduces to nChoosek 13

To Summarize 1.Each code rate has a fixed window size 2.The table is bounded by number of errors in a window Implement the table as a hash map – The error sequence is converted to integer and used as key Actual Lookup Table … … … How to use these tables to estimate delivery rate?

Delivery Rate Estimation Calculate BER per subcarrier from SNRGenerate Error Patterns from BERLookup Error Prob. for error patternLookup next pattern in sequenceTake product of all Error Probs.Repeat for multiple error sequences 15 Error Pattern Generation

Sliding window operation b1b1 b3b3 b4b4 b2b2 b5b5 b6b6 b8b8 b7b7 b9b9 b 10 b 11 b 12 b 13 b0b0 b 14 b p1p1 p2p2 p3p3 P e,1 = p 1 x p 2 x p 3 Example: Window Size = 4 Delivery Rate is the average of Prob. of Error of N seqs 16

Machine Learning Motivation – Avoids the time and space complexity of lookup tables – Machine learning can provide faster online solution Machine learning algorithm – We chose Neural Networks Reason – Supports non-linear continuous functions – Appropriate for delivery ratio 17

Feature Set – Use Bit error Rate per bit Advantage – Easy to obtain from CSI information – Allows de-coupling from the interleaver – Feature size is limited by number of bits in OFDM symbol 18

Neural Network: Setup 19 Input Layer Hidden Layers Output Layer Multi Layer Perceptron Bit Error Rate 2 Layers 20 Neurons per layer Neuron Delivery Rate

Neural Network Operation Neural Network Network weights Training Intel Channel Traces TGn Channel models Bit-Error-Rate Delivery Ratio Input Output Testing Trained Network Delivery Rate Bit-Error-Rate Input Output 20

Contributions Develop two new techniques to estimate delivery rate Propose a new Interleaver 21

Standard Interleaver Spreads adjacent bits by four subcarriers 2728… … …26 12… … Subcarriers 4 spaces apart can experience similar fate …

Proposed Interleaver Key Ideas 1.Maximize the separation between likely corrupted bits 1.Use SNR information to identify more error prone bits Interleaving pattern dependent on SNR 23

Proposed Interleaver 10 Subcarriers s1s2s3s4 s5s6s7s8s0 s9 Sorted based on SNR Maximize separation between bad locations s0 s1 maximum separation next ofdm symbol 24

Proposed Interleaver s0 s1 s2 s3 s6 s4s5s7s9s8 b1b1 b3b3 b4b4 b2b2 b5b5 b6b6 b8b8 b7b7 b9b9 b0b0 maximum separation 25 s1s2s3s4 s5s6s7s8s0 s9 10 Subcarriers Sorted based on SNR Maximize separation between bad locations Interleaver maximizes the separation between bad bits

26 Evaluation and Results

Evaluation Methodology Trace driven simulation to evaluate performance Traces collected using Intel 5300 WLAN card Trace Features – Static and mobile traces – Both 20Mhz and 40Mhz bandwidths Traces generated using IEEE-TGn Channel models – Traces simulate a range of indoor and outdoor environments – Include both 20Mhz and 40Mhz bandwidths 27

Evaluated Schemes SchemesDescription EffSNR Average BER is used to estimate effective SNR Lookup Our lookup table based approach ML Trained neural network is used to estimate delivery ratio 28

Delivery Rate Accuracy (40MHz) EffSNR Wifi: 11% Our: 27% Lookup Wifi: 4.5% Our: 3% ML Wifi: 6% Our: 3% 29 Lookup and ML show improvement over Effective SNR

Interleaver Performance 20 MHz40 MHz 17% impv. 22% impv. 30 Proposed interleaver outperforms interleaver

Rate Adaptation 20 MHz40 MHz 31 Lookup Wifi: up to 62-75% Our: up to 72-79% ML Wifi: up to 65-75% Our: up to 72-79% Significant throughput benefit at transition regions

Energy Savings Transmitter 20 MHz40 MHz 32 Lookup Wifi: up to 26-35% Our: up to 29-36% ML Wifi: up to 27-35% Our: up to 29-36% Significant Energy savings at transition regions

Summary Identify the reasons for poor performance Propose Lookup and ML based approach to estimate delivery ratio – Reduce delivery rate error by 60% for interleaver and 88% for proposed interleaver Future Work – Explore other applications of machine learning in wireless network management 33

THANK YOU! 34