Ripple II
Roy and Choudhury, USENIX NSDI ‘16
Design
- Vibra-motor: Linear Resonant Actuator (LRA) driven by a waveform
generator.
- Input signal to the LRA: OFDM(!)
2.2 Microphone as vibration receiver
- Mic: Sound pushes diaphragm, diaphragm vibrates, produces
electrical signal, amplified.
- Ripple II: Notice mic is sensitive to contact vibrations
- Problem: Interference from air vibrations
Interference Cancellation (Sect. 3.1)
Cover sound hole: Figure 5: SINR was -10 dB (@ 10 KHz), increases to
+25 dB (@ 10 KHz). Generally better at higher frequencies.
V (contact vibration), S (interference sound), E (electric noise)
- E comes from common electric supply voltage of mics.
- Goal: Interference Cancellation (subtract S)
- System model shown in Figure 6
- Possible weakness: Physical interfering vibration (i.e. riding in
a Jeep off-road, would it work?)
Failed Attempts (Sect. 3.1)
- E has a spatial signature across mics, MIMO! But, can’t estimate
spatial signature for interference sound.
Symbol Selective Adaptive Noise Filtering
- See slides: https://www.usenix.org/sites/default/files/conference/protected-files/nsdi16slidesroy.pdf
- Slide 24: Sound interference affects only certain subcarriers
- V1, S1, S_2 not defined
- Personal comm. w/authors:
- V1 = V(t)H_{V1}, S1 = S(t)H_{S1}, and so on.
- As vibration from the primary microphone leaks to the secondary microphone, they model the secondary microphone’s signal as a filtered version of the primary.
- If not affected by ambient sound, this channel gain is entirely the function of the solid medium (e.g. the circuit board where these microphones are mounted) and hence it is static.
- Primary and secondary symbols are from Mic1 and Mic2 respectively.
- Avoid lower frequency band interference by starting above 500 Hz
OFDM (Sect. 3.2)
- Characterize the channel in Figure 9
- Multipath components weak, and from motor mass
- 10 dB max excess delay of 400 us, conservative CP of 1 ms
- Coherence B/W 480 Hz, subcarrier chosen 40 Hz (conservative)
MAC Layer (Sect. 4)
- Cool idea: Back EMF lets transmitter sense receiver interference
like the Ethernet
- Interference sound induces a tiny current
- Measure that induced current to motor by voltage drop across
series resistor
- Results in Figure 11 are pretty convincing
Proactive Symbol Recovery (4.3, 4.4)
- Transmitter has better estimate of errored symbols than receiver
(see Figure 14).
- Idea: Transmitter sends on every other OFDM subcarrier, more power.
- Better SNR, half rate, essentially a bit rate adaptation
- Estimates start and end (Fig 14) of interference by Back-EMF
sensing.
- Convolutional coding atop everything adds fall-back layer
Performance Evaluation (S. 5)
- Fig. 17(a) CDF across all noise environments
- PSR retrasnmits erroneous symbols and improves throughput
- Recall is weak, so it misses many symbols that should have been
retransmitted
- Expected/desirable? b/c of coding?
Applications
- Finger Ring
- Tabletop comms
- P2P money transfer