U NIVERSITY OF M ASSACHUSETTS, A MHERST Department of Computer Science Leveraging Interleaved Signal Edges for Concurrent Backscatter by Pan Hu, Pengyu.

UNIVERSITY OF MASSACHUSETTS, AMHERST • Department of Computer Science

Leveraging Interleaved Signal Edges for Concurrent Backscatter

by Pan Hu, Pengyu Zhang, Deepak Ganesan

University of Massachusetts Amherst

UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science

Can we enable concurrent backscatter?

2

Backscatter has tremendous potential as wireless backhaul for IoT and wearables.

Concurrent backscatter canGreatly reduce co-ordination overhead

Enable simpler hardware design for tags

Backscatter reader

Sensor

Fast bit rate

Slow control msg MAC

Processing

Msg

UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 3

Decode bits based on I-Q clustersIdeal 4QAM clusters: 00,01,10,11

Approach 1: QAM-like clustering

UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 4

Phase and Amplitude form QAM-like clustersUsing classification for decoding [Angerer 2010]

Works well for few nodes!

Approach 1: QAM-like clustering

UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 5

Number of clusters grows exponentially (2^N)64 dense clusters for 6 nodes

Not scalable!

Approach 1: QAM-like clustering

UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 6

Approach 2: Belief-propagation decoding

Linear combination of channel coefficients and TX signal [Buzz Sigcomm12]

Received signal : known

UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 7

Approach 2: Belief-propagation decoding

Linear combination of channel coefficients and TX signal [Buzz Sigcomm12]

Channel coefficients can be estimated.

UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 8

Approach 2: Belief-propagation decoding

Linear combination of channel coefficients and TX signal [Buzz Sigcomm12]

TX signal : unknown

UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 9

Approach 2: Belief-propagation decoding

Linear combination of channel coefficients and TX signal [Buzz Sigcomm12]

Channel coefficient is NOT constant!

UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 10

Approach 2: Belief-propagation decoding

Channel coefficient is not always stableCan be affected by object movement

UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 11

Approach 2: Belief-propagation decoding

Channel coefficient is not always stableCan be affected by object movement, tag rotation

UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 12

Approach 2: Belief-propagation decoding

Channel coefficient is not always stableCan be affected by object movement, tag rotation and cross-tag coupling

UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science

Design of BST Backscatter Spike Train

13

Key argument: We can separate concurrent transmissions in the time domain by detecting signal edges corresponding to different nodes.

Node 1 TX signal

Node 2 TX signal

Collided signal

UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science

Design of BST Backscatter Spike Train

14

Why is this approach feasible in backscatter?Edges are sharp: wide spectrum allocated for RFID backscatter (902MHz to 928MHz in US)

Edges are detectable: reader sampling rate >> tag bit rate

(50MHz vs 100kbps)

UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 15

Design of BST Backscatter Spike Train

But, edge amplitude/direction depends on who else is concurrently transmitting!

UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 16

Robust Vector Based Edge Detection - Uses both I and Q channel information to robustly detect edge.

Design of BST Backscatter Spike Train

UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 17

More design details can be found in paper:

Handing edge collision - Efficient back off and bit rate adaptation

Stop error diffusion - Current bit depend current symbol and previous bit

Design of BST Backscatter Spike Train

UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science

Implementation

18

USRP N210 + 16 UMass Moo Nodes

Node:100kbits/s TX speed;

USRP:50MHz Sampling;Separated antennas for TX/RX .

UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science

Preliminary Result

19

Throughput comparison of BST with TDMA and BUZZ

Up to10x improvement over TDMA & Buzz!

UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science

Conclusion

20

Concurrent transmission for backscatter WITHOUT scheduling or encoding.

- Key idea: leverage interleaved signal edges to decode collided signals!

UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science

Thanks

21

Thank you!

UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 22

Backup Slides

UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 23

Handling Edge Collisions - Collision probability can be high

- Either re-start or reduce bit rate

*Sampling Frequency: 25MHz

Tag transmitting Rate: 100kbps

Design of BST Backscatter Spike Train

UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 24

Deleted Slides

UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 33

Multiple node transmitting, multiple edges

Design of BST Backscatter Spike Train

Backscatter reader

Sensor

Fast bit rate

Slow control msg MAC

Processing

Msg