PHY + MAC:
The Whole is Greater than the Sum
Romit Roy Choudhury
Associate Professor
1
Two Research Threads
Mobile Computing
(top down)
Application
Security
Transport
Network
MAC / Link
PHY
Wireless Networking
(bottom up)
Our Research
Telescope
Localization
Mobile Computing
(top down)
Activity / Gestures
Application
Smart Content
Security
Transport
Network
CSMA/CN
Freq. Backoff
Rate Control
Wireless Networking
(bottom up)
P2P Gaming
MAC / Link
Energy Management
PHY
Interference Cancellation
Motivation

Significant leaps in PHY
layer link capacity

MIMO, OFDM, Coding,
Beamforming …

Advancement in distributed
algorithms, protocols

Scheduling, coordination,
fairness, synchronization…
4
… Nevertheless
MAC
PHY
5
… Nevertheless
MAC
PHY
Why ?
1. Lack of integrated experimentation platform

Difficult to build cross-layer research prototypes
2. Protocol designers untrained in communications

Past cross layer research mostly MAC and above
6
Software Radios

Software defined radios

Changing landscape of wireless systems
Transparent PHY layer enabling fostering
unconventional ideas …
We intend to contribute here
7
Some Ongoing Projects
1.
Back2F: Backing off in the Frequency Domain
2.
AccuRate: Constellation based Rate Selection
3.
CSMA/CN: Making Wireless MAC like Ethernet
1.
SAWC: Sensor Assisted Wireless Communication
2.
Uncollide: Is SIC worth it?
3.
SleepWell: WiFi Energy Management
8
Some Ongoing Projects
1.
Back2F: Backing off in the Frequency Domain
2.
CSMA/CN: Making Wireless MAC like Ethernet
3.
AccuRate: Constellation based Rate Selection
1.
SAWC: Sensor Assisted Wireless Communication
2.
Uncollide: Is SIC worth it?
3.
SleepWell: WiFi Energy Management
9
Backoff

Distributed contention resolution

ALOHA 1972  introduced notion of randomized backoff
10
Backoff

Distributed contention resolution

ALOHA 1972  introduced notion of randomized backoff
AP1’s
Backoff = 9
AP2’s
Backoff = 15
AP1 Tx
ACK
AP2 Waits
AP1 Waits
AP2 Tx
ACK
…
…
Wastage
11
Fundamentally,
backoff is not a time domain operation …
its implementation has been in the time domain
12
Fundamentally,
backoff is not a time domain operation …
its implementation has been in the time domain
We intend to break away,
and implement backoff on the frequency domain
by taking advantage of the PHY layer
13
Frequency Domain

802.11a/g PHY adopts OFDM



Wideband frequency channel divided into 48 narrow sub-carriers
Copes better with fast, frequency selective fading
Purely a PHY layer motivation
Subcarriers: 1 2 3 4 …
48
Frequency

MAC Opportunity


Pretend OFDM subcarriers are integers
Emulate randomized backoff
14
Back2F: Main Idea


Pick random backoff, say 6
Transmit signal on 6th subcarrier
18
6
0
47
AP1 Backoff = 6
0
47
AP2 Backoff = 18
15
Back2F: Main Idea


Pick random backoff, say 6
Transmit signal on 6th subcarrier
6
0
6
0
47
Listen
Antenna
AP1 Backoff = 6
6
18
18
18
47
Listen
Antenna
AP2 Backoff = 18
AP2 learns some other AP is winner.
AP1 learns AP1 is the winner … hence, AP1 transmits
16
What if Collision?

Introduce a second round of contention

Winners of first go to second
0
Subcarrier
1
0
Subcarrier
1
2
Winner
2
3
First Round
3
4
5
4
5
Second Round
17
Why beneficial?
Avg. temporal backoff ~ 100 micro sec.
Frequency backoff = 1 OFDM symbol = 4 micro sec
2 rounds of backoff = 8 micro sec.
Possible to do better …
18
Creating a Queue
0
Subcarrier
1
2
3
First Round
1
5
Rank 2
Winner
0
Subcarrier
4
2
3
4
5
Second Round
19
Creating a Queue
0
Subcarrier
1
0 2 4 Rank 1
0
Subcarrier
2
3
First Round
0 2 4 Rank 2
1
2
3
4
5
0 2 4 Rank 3
4
Enabling
TDMA
5
Second Round
20
Improved Channel Utilization
Data
Data
Data
Data
WiFi: Contention per packet
TDMA
Data
Data
Data
Data
T2F: OFDM contention per TDMA schedule
21
Several Other Challenges
1.
What happens in multiple collision domains?
2.
What if subcarriers are misdetected?
3.
What happens when new client join, other clients leave?
4.
Can we support legacy APs?
…
22
Performance

10 USRP Testbed


Deployed in Duke
Quantify



Reliable subcarrier detection
Collision probability
Net throughput gain over WiFi
23
SNR in dB
Subcarrier Detection
FFT Number
Reliable subcarrier
detection at 12dB
24
Throughput Gain
Throughput Gain increases with higher bitrates
25
Closing Thoughts

Contention is not fundamentally a time domain operation


Back2F shows feasibility in frequency domain
Long standing overheads of backoff can be alleviated
26
Some Ongoing Projects
1.
Back2F: Backing off in the Frequency Domain
1.
CSMA/CN: Making Wireless MAC like Ethernet
2.
AccuRate: Constellation based Rate Selection
1.
SAWC: Sensor Assisted Wireless Communication
2.
Uncollide: Is SIC worth it?
3.
SleepWell: WiFi Energy Management
27
Collision in Wireless Networks
T1
R
T2
Collision
ACK Timeout
time
Retransmit
Collision in Wireless Networks
T1
R
T2
Collision
ACK Timeout
Not Efficient!
Retransmit
Better if T1 stops
right after collision
Ethernet is Efficient
Collision
Transmitter detects a collision,
and immediately aborts transmission.
Called Collision Detection (CSMA/CD)
Unfortunately, CSMA/CD not feasible in wireless networks …
30
We ask: Can we emulate
CSMA/CD in wireless networks
i.e., abort collisions right when they occur
31
Tx
MAC
MAC
PHY
PHY
Rx
Cross Layer
Cross Layer
CSMA/CN: Basic Idea
Tx
MAC
S=S1
MAC
PHY
PHY
Data Transmission (S1)
Rx
Cross Layer
Cross Layer
CSMA/CN: Basic Idea
CSMA/CN: Basic Idea
Tx
MAC
S=S1
MAC
PHY
PHY
Data Transmission (S1)
Rx
Cross Layer
Cross Layer
Check for
Collision
CSMA/CN: Basic Idea
Tx
MAC
Check for
Collision
S=S1
MAC
PHY
PHY
Data Transmission (S1)
Rx
Cross Layer
Cross Layer
Search for
Abort
CSMA/CN: Basic Idea
Tx
MAC
Check for
Collision
S=S1
MAC
PHY
PHY
Data Transmission (S1)
Rx
Cross Layer
Cross Layer
Search for
Abort
CSMA/CN: Basic Idea
Tx
MAC
Collision
Detected
S=S1
MAC
PHY
PHY
Data Transmission (S1)
Rx
Cross Layer
Cross Layer
Search for
Abort
CSMA/CN: Basic Idea
Tx
MAC
Collision.
Send Abort
S=S1+S2
MAC
Abort Signal (S2)
PHY
PHY
Data Transmission (S1)
Rx
Cross Layer
Cross Layer
Search for
Abort
CSMA/CN: Basic Idea
Tx
MAC
Collision.
Send Abort
S=S1
S=S1+S2
MAC
Abort Signal (S2)
PHY
PHY
Data Transmission (S1)
Rx
Cross Layer
Cross Layer
Abort signal
detected
CSMA/CN: Basic Idea
Cross Layer
Tx
MAC
S=S1
S=S1+S2
MAC
Abort Signal (S2)
PHY
PHY
Data Transmission (S1)
Rx
Cross Layer
Collision.
Send Abort
ABORT
Correlation is the Key
Correlation spikes whenever notification arrives
But works when notification is no weaker than 18dB of self-signal
Wireless
Interference Cancellation

Need to detect very weak notification signals

Opportunity
 Pass the Tx signal over wire
 Listen antenna has 2 copies
of the Tx signal


Both copies have same filter
and frequency offset effects

Align the two signals using sampling offset information
 Subtract the wired signal from wireless
Wired
Correlate residue with collision notification
42
MAC
Collision Detection at Rx


Receiver detects collision within 20 bytes
Total turnaround time for CN signature 18us


PHY
Quicker turnaround  Faster Tx abortion
Throughput gain over PPR
Median gain = 25%
43
Some Ongoing Projects
1.
Back2F: Backing off in the Frequency Domain
2.
CSMA/CN: Making Wireless MAC like Ethernet
3.
AccuRate: Constellation based Rate Selection
1.
SAWC: Sensor Assisted Wireless Communication
2.
Uncollide: Is SIC worth it?
3.
SleepWell: WiFi Energy Management
44
Current Wireless Rate Selection
Frame Based
History
Info.
SNR Based
Data
Data
ACK
SampleRate, RRAA
SNR
RBAR, CHARM
Recently PHY-based:
✦ SoftRate [SIGCOMM ’09]
•
•
•
Uses a BER heuristic to estimate bit rate
BER accurately identifies when to increase/decrease rate
However, may not be able to jump to optimal rate
We dive deeper into PHY … jump to the optimal rate
45
In General
Weak
Channel
0
Moderate
Channel
Strong
Channel
01
0111
46
In General
Weak
Channel
0
6 Mbps
Moderate
Channel
01
24 Mbps
Strong
Channel
36 Mbps
0111
Smaller dispersion permits higher rate
47
AccuRate
Hypothesis:
Symbol dispersion is independent of modulation
Observation:
Dispersion conveys the optimal rate that should
have been used for that packet
48
Hypothesis Verification
11
01
11
01
00
10
00
10
Tx 4QAM
Tx 16QAM
Channel
Rx QPSK
Rx16QAM
49
Hypothesis Verification
McKinley et. al., 2004, “EVM calculation for broadband modulated signals” 50
Given that symbol dispersion is
independent of modulation
Observe symbol dispersion
and select optimal modulation
51
BPSK
Data
4QAM
16QAM
52
BPSK
Data
4QAM
16QAM
53
BPSK
Data
4QAM
16QAM
We call this Virtual Channel
Replay
54
Channel Replay Vector
d1
Vector V = {d1, d2, ...., dn}
d2
55
Receiver
Demodulator
Packet
Channel
Replay
Demodulator
CRC
Check
4QAM
Channel
Replay
Demodulator
CRC
Check
16QAM
Channel
Replay
Demodulator
CRC
Check
BPSK
Best Rate
56
Optimal modulation ≠ Optimal rate
Bit-rate is a function of
both modulation and coding
Need to find the optimal <modulation,coding>
for a received packet?
57
Receiver
Demodulator
Decoder
Data
BPSK
1/2
Channel
Replay
Demodulator
Decoder
CRC
Check
BPSK
3/4
Channel
Replay
Demodulator
Decoder
CRC 9 Mbps
Check
1/2
Channel
Replay
Demodulator
Decoder
CRC 12 Mbps
Check
QAM4
3/4
Channel
Replay
Demodulator
Decoder
CRC 18 Mbps
Check
QAM64
3/4
Channel
Replay
Demodulator
Decoder
CRC 54 Mbps
Check 58
6 Mbps
Best Rate
QAM4
Throughput at Walking Speeds
Testbed
AccuRate achieves 87% of the optimal throughput
59
Some Ongoing Projects
1.
Back2F: Backing off in the Frequency Domain
2.
CSMA/CN: Making Wireless MAC like Ethernet
3.
AccuRate: Constellation based Rate Selection
1.
SAWC: Sensor Assisted Wireless Communication
2.
Uncollide: Is SIC worth it?
3.
SleepWell: WiFi Energy Management
60
Sensor Assisted Wireless Communication
Synergy between sensing and wireless
- Out-of-band contexts can provide useful cues
- Useful for optimizing wireless PHY/MAC
(e.g., switch WiFi channel on sensing microwave hum …
modify rate control based on accelerometer …
turn off WiFi when in subway train …)
Some Ongoing Projects
1.
Back2F: Backing off in the Frequency Domain
2.
CSMA/CN: Making Wireless MAC like Ethernet
3.
AccuRate: Constellation based Rate Selection
1.
SAWC: Sensor Assisted Wireless Communication
2.
Uncollide: Is SIC worth it?
3.
SleepWell: WiFi Energy Management
62
Thank You
Visit our
Systems Networking Research Group (SyNRG)
http://synrg.ee.duke.edu
63
64
Some Ongoing Projects
1.
Back2F: Backing off in the Frequency Domain
2.
CSMA/CN: Making Wireless MAC like Ethernet
3.
AccuRate: Constellation based Rate Selection
1.
SAWC: Sensor Assisted Wireless Communication
2.
Uncollide: Is SIC worth it?
3.
SleepWell: WiFi Energy Management
65
2. Distinct receivers
Case: Two links with distinct receivers
Situations much less favorable to SIC
Case: Two links with distinct receivers
R1
R2
T2
T1
Main Concern:
T1 will transmit at best possible bit rate to R1
R2 has to decode T1’s signal at this bit rate …
despite the presence of T2’s signal
Case: Two links with distinct receivers
R1
R2
T2
T1
Thus, necessary (but not sufficient)
conditions:
1. R2’s interferer (T1) must be closer than its
own transmitter (T2)
2. T1’s own receiver (R1) must be further than
interered receiver (R2)
Gains available when all conditions hold:
R1
R2
T2
T1
How often do these SIC permissible
topologies occur?
R1
R2
T2
T1
Enterprise WLANs:
• Clients likely to associate with stronger AP
• Such scenarios unlikely
Residential WLANs:
• Neighbors AP may be stronger
• Some SIC scenarios possible
Monte Carlo Simulations
(AP Transmit Range)
Gain with SIC in less than 10% of the cases
73
Collision Probability
Benefit of
second round
Small collision probability in dense networks
74