Beyond the bits – co operative packet recovery

Post on 19-May-2015

185 views 0 download

Tags:

transcript

Beyond The Bits – Co Operative Packet Recovery

Grace Woo,Pouya Kheradpour,Dawei Shen,Dina Katabi @ MIT

Presented By : Venkata Nanda KishoreUBID: 3764-8659

OUTLINE

• Introduction• Related Work• Architecture• Combining Algorithm• DSSS• Estimating the Variance• Reducing Overhead • Implementation• Results• Conclusion

Abstract and Introduction

• Dead Spots• Loss Rate• Addressing these problems-Physical Layer info• SOFT

Related Work

• SPATIAL DIVERSITY: -Similar to Softer Handoff -Less Overhead -Different Modulation Schemes• SOFT DECODING:– Also used in RS codes and LDPC– Focuses on Compatibility,practical

Networks,Experimental Results

Bit-Math : Exploiting Spatial Diversity

• BER: 0.001• Packet Size:1500B=12000bits• Here,Probability that a node receives a

correctly transmitted packet. …..????• Combining physical layer information using

SOFT increases this to 0.99

SOFT SYSTEM ARCHITECTURE

• Sits between MAC, between Physical and Data Link Layer

• Combining Packets occur both on UPLINK and DOWNLINK

• Considers ONLY BINARY MODULATION

UPLINK

• Leverage the high bandwidth of Ethernet• AP to COMBINING AGENT• Invoking COMBINING procedure

-Identify packets-Hash table-Redundancy

• Decoding: Two options: Forward or Combining Agent

DOWNLINK

• EXPLOITS TIME DIVERSITY• Rest of the procedure is similar to UPLINK

ACKs Retransmission & Reordering

• Receiver sends synAck for a correctly received packet and communicates correctly to Combining agent

• If faulty packet arrives,it sends an asynchronous Ack to sender

• No Reordering on DOWNLINK• UPLINK orders the packets using packet_id• Also uses forwarding_time

COMBINING ALGORITHM

• How to combine the s-packets? >= 0Here, is the noise variance in i th AWGN channel

DSSS

• Direct Sequence Spread Spectrum• Eg. 1 - 11100010101 0 - 00011101010STEPS:1. Spreading2. De-spreading

Can this be efficient?

(-)/>=0

Here,y correspond to same codewords.X correspond to either codeword sent for logical 1 or logical 0.

Estimating Noise Variance

=hx+h is attenuationY is the sv-value of bitn is guassian noise

Reducing Overhead

• Currently:10Gb/s Ethernet (10GbE)• Working on 100GbE• Depends on how we express SVApproach: 3 bits – 1 for sign,2 for magnitudeAlso,provide cut-offs.

IMPLEMENTATION

• SDR: Software Defined Radios

• Uses GMSK- Gaussian Minimum Shift Keying

• Also tested with DBPSK-Differential Binary Phase Shift Keying.

Experimental Results:Compared Approaches

• SOFT• Max Confidence• Majority Vote• Current Approach• MRD

Experimental Results:Metrics

• Packet Delivery Rate• No of Retransmissions

Importance of Soft Values

Role Of Combining Rule

Experimental Results Cont.

• Impact of Noise Variance

• Impact of quantization

• BER

Conclusion

• Existing protocols are inadequate• We have seen SOFT (Its approach,

implementation and results)• Novel ideas due to temporal and spatial diversity• Soft increases mean packet deliver and reduces

retransmissions.

Thank You!

Questions/Comments/Suggestions??