Practical, Real-time,
Full-Duplex Wireless
Mayank Jain, Jung Il Choi, Taemin Kim,
Dinesh Bharadia, Kannan Srinivasan, Siddharth Seth,
Philip Levis, Sachin Katti, Prasun Sinha
September 22, 2011
1
What full-duplex
2
What full-duplex ... and why?
3
Current wireless radios
• Time Division Duplexing
Node 1
Node 2
Node 1
Node 2
TX/RX
TX/RX
TX/RX
TX/RX
Timeslot 1
Timeslot 2
4
Current wireless radios
• Time Division Duplexing
Node 1
Node 2
Node 1
Node 2
TX/RX
TX/RX
TX/RX
TX/RX
Timeslot 1
Timeslot 2
• Frequency Division Duplexing
Node 1
Node 2
TX
RX
RX Frequency 1
TX
Frequency 2
5
Single channel full-duplex
Node 1
Node 2
RX
TX
TX
RX
6
Single channel full-duplex
➔ Very strong self-interference: ~70dB for 802.11
SelfInterference
Node 1
Node 2
RX
TX
TX
RX
7
Single channel full-duplex
➔ Very strong self-interference: ~70dB for 802.11
SelfInterference
Node 1
Node 2
RX
TX
TX
RX
Main idea: cancel self-interference
Combine RF and digital techniques for cancellation
8
The story so far...
Mobicom’10[1]:
Antenna Cancellation + other techniques
TX1
RX
d
TX2
d + λ/2
[1] Choi et al. “Achieving single channel, full duplex wireless communication”,
Mobicom 2010
9
The story so far...
Mobicom’10[1]:
Antenna Cancellation + other techniques
TX1
RX
d
•
•
•
TX2
d + λ/2
Does not adapt to channel changes
Interference pattern can affect intended receivers
Frequency dependent, narrowband cancellation
[1] Choi et al. “Achieving single channel, full duplex wireless communication”,
Mobicom 2010
10
•
•
•
New, better RF and digital cancellation techniques
Adaptive algorithms for auto-tuning cancellation
Real-time full-duplex MAC layer implementation
11
Talk Outline
•
•
•
•
•
RF Cancellation using Signal Inversion
Adaptive RF Cancellation
System Performance
Implications to Wireless Networks
Looking Forward
12
Talk Outline
•
•
•
•
•
RF Cancellation using Signal Inversion
Adaptive RF Cancellation
System Performance
Implications to Wireless Networks
Looking Forward
13
Cancellation using Phase Offset
SelfInterference
∑
Cancellation
Signal
Cancellation using Phase Offset
SelfInterference
∑
Cancellation
Signal
SelfInterference
∑
Cancellation
Signal
Frequency dependent, narrowband
15
Cancellation using Signal Inversion
SelfInterference
∑
Cancellation
Signal
16
Cancellation using Signal Inversion
SelfInterference
∑
Cancellation
Signal
SelfInterference
∑
Cancellation
Signal
Frequency and bandwidth independent
17
BALUN : Balanced to Unbalanced Conversion
Xt
+Xt/2
-Xt/2
BALUN
18
BALUN : Balanced to Unbalanced Conversion
RX
TX
Cancellation Signal
Xt
+Xt/2
+Xt/2
-Xt/2
BALUN
∑
-Xt/2
Xt
TX
RF Frontend
RX
RF Frontend
19
TX
Over the air
attenuation and delay
RX
∑
Xt
+Xt/2
+Xt/2
-Xt/2
BALUN
-Xt/2
Xt
TX
RF Frontend
RX
RF Frontend
20
Signal Inversion Cancellation
TX
Xt
Attenuator and
Delay Line
+Xt/2
+Xt/2
-Xt/2
BALUN
Over the air
attenuation and delay
RX
∑
-Xt/2
Xt
TX
RF Frontend
RX
RF Frontend
21
Signal Inversion Cancellation: Wideband Evaluation
•
•
•
TX
Measure wideband cancellation
Wired experiments
240MHz chirp at 2.4GHz to measure response
Xt
+Xt/2
+Xt/2
∑
RX
TX
Xt
λ/2
Delay
RF
Signal
Splitter
∑
-Xt/2
+Xt/2
Signal Inversion
Cancellation Setup
Phase Offset
Cancellation Setup
RX
22
Higher is
better
Lower is
better
23
Higher is
better
Lower is
better
~50dB cancellation at 20MHz bandwidth with balun vs ~38dB with
phase offset cancellation.
Significant improvement in wideband cancellation
24
Other advantages
RX
TX
Attenuator and
Delay Line
+Xt/2
∑
-Xt/2
Xt
TX
RF Frontend
•
•
RX
RF Frontend
From 3 antennas per node to 2 antennas
Parameters adjustable with changing conditions
25
RX
TX
Attenuator and
Delay Line
+Xt/2
∑
-Xt/2
Xt
TX
RF Frontend
RX
RF Frontend
Passive components better than active components
• No gain required
• Saturation can lead to non-linearity
• Passive components are more frequency flat
26
Talk Outline
•
•
•
•
•
•
RF Cancellation using Signal Inversion: ~50dB for 20Mhz
Adaptive RF Cancellation
Adaptive Digital Cancellation
System Performance
Implications to Wireless Networks
Looking Forward
27
Adaptive RF Cancellation
RX
TX
RF Reference
Balun
Σ
RF Cancellation
Wireless
Receiver
Wireless
Transmitter
TX Signal Path
•
•
Attenuation
& Delay
RX Signal Path
Need to match self-interference power and delay
Can’t use digital samples: Saturated ADC
28
Adaptive RF Cancellation
RX
TX
RF Reference
Balun
Σ
RSSI
RF Cancellation
Wireless
Receiver
Wireless
Transmitter
TX Signal Path
•
•
Attenuation
& Delay
RX Signal Path
Need to match self-interference power and delay
Can’t use digital samples: Saturated ADC
RSSI : Received Signal Strength Indicator
29
Adaptive RF Cancellation
RX
TX
RF Reference
Balun
Σ
RSSI
RF Cancellation
Wireless
Receiver
Wireless
Transmitter
TX Signal Path
•
•
Attenuation
& Delay
Control
Feedback
RX Signal Path
Need to match self-interference power and delay
Can’t use digital samples: Saturated ADC
Use RSSI as an indicator of self-interference
30
RX
TX
RF Reference
Balun
Attenuation
& Delay
Control
Feedback
Σ
RSSI
RF Cancellation
Wireless
Receiver
Wireless
Transmitter
TX Signal Path
RX Signal Path
Objective: Minimize received power
Control variables: Delay and Attenuation
31
Objective: Minimize received power
Control variables: Delay and Attenuation
32
Objective: Minimize received power
Control variables: Delay and Attenuation
➔ Simple gradient descent approach to optimize
33
Off-the-shelf electronically tunable hardware
approximation: QHx220 noise canceler
Signal +
Interference
Interference
Sample
Gain I
Σ
λ/4
Delay
Σ
Gain Q
Cancellation
Signal
Clean
Signal
34
Off-the-shelf electronically tunable hardware
approximation: QHx220 noise canceler
RX
TX
Gain I
Balun
Σ
λ/4
Delay
Wireless
Transmitter
Σ
Gain Q
Control
Feedback
RSSI
Wireless
Receiver
TX Signal Path
RX Signal Path
35
Off-the-shelf electronically tunable hardware
approximation: QHx220 noise canceler
But ...
•
Uses λ/4 delay to generate quadrature
component: Not precise for all frequencies
•
Active components for gain: saturation
leading to non-linearities
36
Off-the-shelf electronically tunable hardware
approximation: QHx220 noise canceler
37
Typical convergence within 8-15 iterations (~1ms total)
38
39
Saddle
Point
Recovery from local minimas and saddle
points possible, needs more iterations
40
•
•
~65% converge without going through a local minima
98% converge in <20 iterations
41
Talk Outline
•
•
•
•
•
RF Cancellation using Signal Inversion: ~50dB for 20Mhz
Adaptive RF Cancellation: ~1ms convergence
System Performance
Implications to Wireless Networks
Looking Forward
42
Digital Cancellation
•
Measure residual self-interference after RF
cancellation
•
Subtract self-interference from received digital
signal
43
TX Signal
Analog
Conversion
and Shaping
TX
RX
Filtering and
Digital
Conversion
Residual
Self-interference
Digital
Receiver
44
TX Signal
Analog
Conversion
and Shaping
TX
RX
Channel
Model
Digital Cancellation
Filtering and
Digital
Conversion
Residual
Self-interference
+
Cancellation
Signal
-∑
Digital
Receiver
45
Bringing It All Together
RX
TX
RF Reference
Balun
Control
Feedback
Attenuation
& Delay
Σ
RSSI
RF Cancellation
Baseband ➔ RF
RF ➔ Baseband
DAC
ADC
Digital Interference Cancellation
Digital Interference
Reference
Encoder
FIR filter
Channel
Estimate
TX Signal Path
Decoder
RX Signal Path
46
Performance
•
•
•
~73 dB cancellation
WLAN full-duplex:Yes, with reasonable antenna separation
Not enough for cellular full-duplex
47
Channel Coherence
~3dB reduction in cancellation in 1-2 seconds
~6dB reduction in <10 seconds
48
Talk Outline
•
•
•
•
•
RF Cancellation using Signal Inversion: ~50dB for 20Mhz
Adaptive RF Cancellation: ~1ms convergence
System Performance: ~73dB cancellation
Implications to Wireless Networks
Looking Forward
49
Implications to Wireless Networks
•
•
Breaks a basic assumption in wireless
Can solve some fundamental problems with
wireless networks today[1,2]
•
•
Hidden terminals
Network congestion and WLAN fairness
[1] Choi et al. “Achieving single channel, full duplex wireless communication”,
in Mobicom 2010
[2] Singh et al. “Efficient and Fair MAC for Wireless Networks with Selfinterference Cancellation”, in WiOpt 2011
50
Implementation
•
•
WARPv2 boards with 2 radios
OFDM reference code from Rice
University
•
•
•
10MHz bandwidth OFDM signaling
CSMA MAC on embedded processor
Modified for Full-duplex
51
Mitigating Hidden Terminals
Current networks have
hidden terminals
N1
•
•
AP
N2
CSMA/CA can’t solve this
Schemes like RTS/CTS introduce significant overhead
52
Mitigating Hidden Terminals
Current networks have
hidden terminals
N1
•
•
AP
N2
CSMA/CA can’t solve this
Schemes like RTS/CTS introduce significant overhead
Full Duplex solves
hidden terminals
N1
AP
N2
Since both sides transmit at the same time, no
hidden terminals exist
53
Mitigating Hidden Terminals
Current networks have
hidden terminals
N1
•
•
AP
N2
CSMA/CA can’t solve this
Schemes like RTS/CTS introduce significant overhead
Full Duplex solves
hidden terminals
N1
AP
N2
Since both sides transmit at the same time, no
hidden terminals exist
Reduces hidden terminal losses by up to 88%
54
Network Congestion and WLAN Fairness
Without full-duplex:
1/n bandwidth for each node in network, including AP
Downlink Throughput = 1/n Uplink Throughput = (n-1)/n
•
55
Network Congestion and WLAN Fairness
Without full-duplex:
1/n bandwidth for each node in network, including AP
Downlink Throughput = 1/n Uplink Throughput = (n-1)/n
•
With full-duplex:
AP sends and receives at the same time
Downlink Throughput = 1 Uplink Throughput = 1
•
56
Network Congestion and WLAN Fairness
1 AP with 4 stations without any hidden terminals
Throughput (Mbps)
Fairness (JFI)
Upstream
Downstream
Half-Duplex
5.18
2.36
0.845
Full-Duplex
5.97
4.99
0.977
Full-duplex distributes its performance gain to improve fairness
57
Talk Outline
•
•
•
•
•
•
RF Cancellation using Signal Inversion: ~50dB for 20Mhz
Adaptive RF Cancellation: ~1ms convergence
Adaptive Digital Cancellation: ~30dB cancellation
System Performance: ~73dB cancellation
Implications to Wireless Networks: Collisions, Fairness
Looking Forward
58
•
Other cancellation techniques
Digital estimation for analog cancellation[1]
RX
TX
Σ
Baseband ➔ RF
RF ➔ Baseband
DAC
DAC
ADC
TX Signal
Cancellation Signal
RX Signal
Baseband ➔ RF
[1] Duarte et al. “Full-Duplex Wireless Communications Using Off-The-Shelf
Radios: Feasibility and First Results.”, in Asilomar 2010.
59
•
Other cancellation techniques
Digital estimation for analog cancellation[1]
•
Non-linear channel response
60
•
Other cancellation techniques
Digital estimation for analog cancellation[1]
•
Non-linear channel response
Reduce distortion: feedforward amplifiers
TX Signal
High Power
Amplifier
61
•
Other cancellation techniques
Digital estimation for analog cancellation[1]
•
Non-linear channel response
Reduce distortion: feedforward amplifiers
TX Signal
High Power
Amplifier
+ ∑Estimate
Distortion
62
•
Other cancellation techniques
Digital estimation for analog cancellation[1]
•
Non-linear channel response
Reduce distortion: feedforward amplifiers
Compensate: non-linear digital cancellation
63
•
Other cancellation techniques
Digital estimation for analog cancellation[1]
•
Non-linear channel response
Reduce distortion: feedforward amplifiers
Compensate: non-linear digital cancellation
•
Single antenna solution: circulators
TX Signal
RX Signal
Circulator
64
•
Other cancellation techniques
Digital estimation for analog cancellation[1]
•
Non-linear channel response
Reduce distortion: feedforward amplifiers
Compensate: non-linear digital cancellation
•
•
Single antenna solution: circulators
Device precision: 1 ps resolution for delay line
65
•
Other cancellation techniques
Digital estimation for analog cancellation[1]
•
Non-linear channel response
Reduce distortion: feedforward amplifiers
Compensate: non-linear digital cancellation
•
•
•
Single antenna solution: circulators
Device precision: 1 ps resolution for delay line
Going mobile: Higher cancellation, faster adaptation
66
•
Other cancellation techniques
Digital estimation for analog cancellation[1]
•
Non-linear channel response
Reduce distortion: feedforward amplifiers
Compensate: non-linear digital cancellation
•
•
•
•
Single antenna solution: circulators
Device precision: 1 ps resolution for delay line
Going mobile: Higher cancellation, faster adaptation
MIMO full-duplex
67
Full-duplex Networking
Access Point networks
68
Full-duplex Networking
Access Point networks
Cell
Basestation
Relay
Cellular networks
69
Full-duplex Networking
Access Point networks
Cell
Basestation
Relay
Cellular networks
Multi-hop Networks
70
Full-duplex Networking
Access Point networks
Secure Networks[1,2]
Cell
Basestation
Relay
Cellular networks
Multi-hop Networks
[1] Gollakota et al. “They Can Hear Your Heartbeats: Non-Invasive Security
for Implantable Medical Devices.”, in Sigcomm 2011.
[2] Lee et al. “Secured Bilateral Rendezvous using Self-interference
Cancellation in Wireless Networks”, in IFIP 2011.
71
Full-duplex Networking
Access Point networks
Secure Networks[1,2]
?
Cell
Basestation
Relay
Cellular networks
Multi-hop Networks
[1] Gollakota et al. “They Can Hear Your Heartbeats: Non-Invasive Security
for Implantable Medical Devices.”, in Sigcomm 2011.
[2] Lee et al. “Secured Bilateral Rendezvous using Self-interference
Cancellation in Wireless Networks”, in IFIP 2011.
72
Thank You
Questions?
73
Backup
74
Talk Outline
•
•
•
•
•
•
RF Cancellation using Signal Inversion: ~50dB for 20Mhz
Adaptive RF Cancellation: ~1ms convergence
Adaptive Digital Cancellation
System Performance
Implications to Wireless Networks
Looking Forward
75
Digital Cancellation
•
Create a precise “digital replica” of the selfinterference signal using TX digital samples
•
Subtract self-interference replica from received
digital signal
Requires ADC not saturated: RF cancellation
76
OFDM processing
Signal
Band
77
OFDM processing
Sub-bands
78
OFDM processing
Channel
Distortion
79
OFDM processing
Channel
Distortion
Equalization
80
OFDM processing
Channel
Distortion
RX
Equalization
RF Mixer
ADC
Carrier Frequency
Offset Correction
FFT Engine
Packet Detect
Carrier
Frequency
Demapping
Equalization
Channel
Estimation
81
Step 1: Estimation
Self-interference
Sounding
FIR Filter
Training
Preamble
Sequence
IFFT
RX
Self-interference
Estimate
RF Mixer
ADC
Carrier Frequency
Offset Correction
FFT Engine
Packet Detect
Carrier
Frequency
Demapping
Equalization
Channel
Estimation
Estimation includes effect of RF cancellation
82
Step 2: Cancellation
Cancellation
Signal
TX Signal
FIR Filter
IFFT
RX
RF Mixer
ADC
+
Carrier Frequency
Offset Correction
Self-interference
Estimate
FFT Engine
Packet Detect
Carrier
Frequency
Demapping
Equalization
Channel
Estimation
83
Step 2: Cancellation
Cancellation
Signal
TX Signal
FIR Filter
IFFT
RX
RF Mixer
ADC
+
Carrier Frequency
Offset Correction
Self-interference
Estimate
FFT Engine
Packet Detect
Carrier
Frequency
Demapping
Equalization
Channel
Estimation
30dB Cancellation
84
Talk Outline
•
•
•
•
•
•
RF Cancellation using Signal Inversion: ~50dB for 20Mhz
Adaptive RF Cancellation: ~1ms convergence
Adaptive Digital Cancellation: ~30dB cancellation
System Performance
Implications to Wireless Networks
Looking Forward
85
Phase Offset Cancellation: Block Diagram
TX1
d
RX
d + λ/2
TX2
Attenuator
Power Splitter
RX
RF Frontend
TX
RF Frontend
Digital Processor
86
Phase Offset Cancellation: Performance
TX1
TX2
-25
RSSI (dBm)
-30
Only TX1 Active
-35
-40
-45
-50
-55
-60
0
5
10
15
20
25
Position of Receive Antenna (cm)
87
Phase Offset Cancellation: Performance
TX1
TX2
-25
RSSI (dBm)
-30
Only TX1 Active
Only TX2 Active
-35
-40
-45
-50
-55
-60
0
5
10
15
20
25
Position of Receive Antenna (cm)
88
Phase Offset Cancellation: Performance
TX1
TX2
-25
RSSI (dBm)
-30
Both TX1 &
TX2 Active
Only TX1 Active
Only TX2 Active
-35
-40
-45
-50
Null
Position
-55
-60
0
5
10
15
20
25
Position of Receive Antenna (cm)
89
Phase Offset Cancellation: Performance
TX1
TX2
-25
RSSI (dBm)
-30
Both TX1 &
TX2 Active
Only TX1 Active
Only TX2 Active
-35
-40
-45
-50
Null
Position
-55
-60
0
5
~25-30dB
10
15
20
25
Position of Receive Antenna (cm)
90
What about attenuation at intended receivers?
Destructive interference can affect this signal too!
•
Different transmit powers for two TX helps
-58 dBm
20
10
0
-10
-20
-30
-30 -20 -10 0 10 20 30
x axis (meters)
Single Transmit Antenna
30
-52 dBm
y axis (meters)
y axis (meters)
30
20
10
0
-52 dBm
-10
-20
-30
-30 -20 -10 0 10 20 30
x axis (meters)
Two Transmit Antennas
91
Higher is
better
Cancellation (dB)
Cancellation (dB)
Sensitivity of Phase Offset Cancellation
Higher is
better
dB
Error (mm)
Amplitude Mismatch
between TX1 and TX2
Placement Error
for RX
92
Cancellation (dB)
Cancellation (dB)
Sensitivity of Phase Offset Cancellation
dB
Error (mm)
Amplitude Mismatch
between TX1 and TX2
Placement Error
for RX
30dB cancellation < 5% (~0.5dB) amplitude mismatch
< 1mm distance mismatch
93
•
•
Cancellation (dB)
Cancellation (dB)
Sensitivity of Phase Offset Cancellation
dB
Error (mm)
Amplitude Mismatch
between TX1 and TX2
Placement Error
for RX
Rough prototype good for 802.15.4
More precision needed for higher power systems (802.11)
94
Bandwidth Constraint
A λ/2 offset is precise for one frequency
TX1
TX2
RX
d
d + λ/2
fc
95
Bandwidth Constraint
A λ/2 offset is precise for one frequency
not for the whole bandwidth
TX1
d
fc -B
fc
TX2
RX
d + λ/2
fc+B
96
Bandwidth Constraint
A λ/2 offset is precise for one frequency
not for the whole bandwidth
TX1
TX2
RX
d1 + λ-B/2
d1
TX1
d + λ/2
d
fc -B
fc
fc+B
TX2
RX
TX1
TX2
RX
d2
d2 + λ+B/2
97
Bandwidth Constraint
A λ/2 offset is precise for one frequency
not for the whole bandwidth
TX1
TX2
RX
d1 + λ-B/2
d1
TX1
d + λ/2
d
fc -B
fc
fc+B
TX2
RX
TX1
TX2
RX
d2
d2 + λ+B/2
WiFi (2.4G, 20MHz) => ~0.26mm precision error
98
Bandwidth Constraint
5.1 G
Hz
2.4 G
Hz
300 M
Hz
Edge frequency
fc
99
Bandwidth Constraint
5.1 G
Hz
2.4 G
Hz
300 M
Hz
•
•
•
WiFi (2.4GHz, 20MHz): Max 47dB reduction
Bandwidth⬆ => Cancellation⬇
Carrier Frequency⬆ => Cancellation⬆
100
Mitigating Hidden Terminals
Packet Reception Ratio
0.9
0.8
Full Duplex
0.7
Half Duplex
0.6
0.5
0
2000
4000
6000
8000
Data Load (Kbps)
101
Mitigating Hidden Terminals
Packet Reception Ratio
0.9
0.8
Full Duplex
0.7
Half Duplex
0.6
0.5
0
2000
4000
6000
8000
Data Load (Kbps)
•
Full-duplex reduces hidden terminal related losses by 88% at 2 Mbps
102
0.9
1.000
0.8
0.925
Full Duplex
0.7
Half Duplex
0.6
0.5
0
2000
4000
6000
Data Load (Kbps)
•
•
Fairness (JFI)
Packet Reception Ratio
Mitigating Hidden Terminals
8000
Full Duplex
0.850
Half Duplex
0.775
0.700
0
2000
4000
6000
8000
Data Load (Kbps)
Full-duplex reduces hidden terminal related losses by 88% at 2 Mbps
At higher loads, half-duplex improves PRR at the expense of fairness
103