03-09-2012, 12:52 PM
Achieving Single Channel, Full Duplex Wireless Communication
[attachment=33190]
Abstract
This paper discusses the design of a single channel fullduplex
wireless transceiver. The design uses a combination
of RF and baseband techniques to achieve full-duplexing
with minimal effect on link reliability. Experiments on real
nodes show the full-duplex scheme achieves a median gain
of 84% in aggregate throughput as compared to traditional
half-duplexing wireless for a single hop network.
This paper presents using Antenna Cancellation, a novel
technique for self-interference cancellation. In conjunction
with existing RF interference cancellation and digital baseband
intereference cancellation, antenna cancellation achieves
the amount of self-interference cancellation required for fullduplex
operation.
The paper also discusses potentialMAC and network gains
with full-duplexing. It suggests ways in which a full-duplex
system can solve some important problems with existing wireless
systems including hidden terminals, loss of throughput
due to congestion, and large end-to-end delays.
INTRODUCTION
A basic precept of wireless communication is that a radio
cannot transmit and receive on the same frequency at the
same time, i.e. operate in a full duplex fashion. As wireless
signals attenuate quickly over distance, the signal from a local
transmitting antenna is hundreds of thousands of times
stronger than transmissions from other nodes. Hence it has
been generally assumed that one cannot decode a received
signal at a radio while it is simultaneously transmitting.
This paper challenges that assumption, and shows via analysis
and practical implementations on 802.15.4 radios that it
is possible to build full duplex radios. The implementation is
fairly simple, and can be built using off-the-shelf hardware
with software radios.
WIRELESS FULL DUPLEXING
This section examines why existing cancellation techniques,
RF and digital, are not enough to achieve full-duplex.
To understand the challenges in implementing wireless
full-duplex, we need to understand the way signals are received
at wireless nodes. The received signal from the antenna
is amplified through an automatic gain control stage
(AGC) and downconverted to either baseband or intermediate
frequency, filtered and then sampled through an Analogto-
Digital Converter (ADC) to create digital samples.
The accuracy of digital samples depends on the resolution
of the ADC. The AGC adjusts the gain of the received signal
to match the maximum level of the ADC to get maximum
resolution in the received signal. For the receiver to decode
a weaker signal using digital cancellation, the signal needs
to be strong enough to be captured within the resolution of
the ADC. Typical ADCs are 8-12 bit, representing a range
of 48-72dB. For an 8-bit ADC, if the weaker signal is 40dB
lower in power than the stronger signal, it only gets 1-bit
resolution.
ANTENNA CANCELLATION
This section analyzes the possible reduction in self-interference
by using antenna cancellation. It also evaluates its limits
with respect to bandwidth of the signal being transmitted and
the sensitivity of antenna cancellation to engineering errors.
It shows, using actual measurements, that antenna cancellation
achieves 20dB reduction in self-interference. This section
also evaluates the effects of using two transmit antennas
for antenna cancellation on the communication range. It
shows that antenna cancellation degrades the received signal
at other nodes in the network by at most 6dB compared to
the single antenna setup.
Performance of Antenna Cancellation
In an ideal scenario, the amplitudes from the two transmit
antennas would be perfectly matched at the receiver and the
phase of the two signals would differ by exactly . However,
we find that the bandwidth of the transmitted signal places
a fundamental bound on the performance of antenna cancellation.
Further, real world systems are prone to engineering
errors which limit system performance. The sensitivity of
the antenna cancellation to amplitude mismatch at the receive
antenna and to the error in receive antenna placement
is important to consider.
Digital Interference Cancellation
There is extensive existing work that describes digital cancellation
techniques [7, 8, 9]. Traditionally, digital cancellation
is used by a receiver to extract a packet from a desired
transmitter after the packet has collided with a packet
from an unwanted transmitter. To do this, the receiver first
decodes the unwanted packet, remodulates it and then subtracts
it from the originally received collided signal. In case
of canceling self-interference for full-duplex, the transmitted
symbols are already known, and thus decoding is not necessary
in order to reconstruct a clean signal.
Instead of decoding, coherent detection is used to detect
the self-interfering signal. The detector correlates the incoming
signal with the clean transmitted signal, which is
available at the output of the transmitter. The main challenge
in subtracting the known signal is in estimating the
delay and phase shift between the transmitted and the received
signals. As the detector has the complete knowledge
of originally transmitted signal, it uses this signal to correlate
with the incoming signal to detect where the correlation
peaks. The correlation peak technique gives both the delay
and the phase shift needed to subtract the known signal.
Thus, this technique, unlike some of the digital interference
techniques, does not require any special preamble or postamble
and is backwards compatible. Moreover, this technique
is modulation-independent as long as the clean signal can be
constructed.
RELATED WORK
Digital cancellation has been extensively used in many existing
schemes [8, 9, 7]. ZigZag [7] uses multiple transmissions
of colliding packets to decode the underlying packets.
This helps with solving the hidden terminal problem,
requiring n time slots to resolve collisions among n packets.
Analog network coding [9] uses access points as analog
symbol repeaters which also repeat symbols of colliding
packets. These repeated symbols are decoded at the respective
destinations. Such a technique gives throughput gains
when two flows are flowing in opposite directions through a
single route. For setups such as the one in Figure 12 where
the multiple flows are intersecting at a single node, analog
network coding will not give any throughput gain over traditional
routing unless the transmitting nodes can overhear
each other. Full-duplex will work in both the setups.
Successive interference cancellation [8] decodes and then
cancels strong interference signals, as long as the interference
is only about 20dB stronger than the signal being received.
ZigZag [7] extends this approach to decode multiple
colliding packets from multiple collisions. Our digital cancellation
scheme does not require decoding symbols, since
the decoder has the knowledge of the transmitted symbols.
DISCUSSION AND CONCLUSIONS
This paper has described the design of a practical single
channel wireless full-duplex system for 802.15.4. The
throughput gains achievable for a single hop wireless channel
are 84% in median. This paper also discusses additional
gains possible with wireless full-duplexing for multihop networks.
The main restrictions in implementing wireless fullduplex
systems are the design of wider band noise cancellation
circuits and making the digital cancellation algorithm
work in real time.
The paper shows that a combination of antenna cancellation,
RF interference cancellation and digital interference
cancellation can bring self-interference to within a few dB
of the noise floor. There still is a loss of a few dB in SINR,
which can lead to a difference in performance for multirate
systems. Existing rate selection algorithms take two approaches,
namely packet error rate based [3, 13], and signal
to noise ratio based [17, 19]. Packet error rate based
schemes would work directly for full-duplex radios. SNR/SINR
based schemes would have to take into account the loss
in SINR due to self-interference.
[attachment=33190]
Abstract
This paper discusses the design of a single channel fullduplex
wireless transceiver. The design uses a combination
of RF and baseband techniques to achieve full-duplexing
with minimal effect on link reliability. Experiments on real
nodes show the full-duplex scheme achieves a median gain
of 84% in aggregate throughput as compared to traditional
half-duplexing wireless for a single hop network.
This paper presents using Antenna Cancellation, a novel
technique for self-interference cancellation. In conjunction
with existing RF interference cancellation and digital baseband
intereference cancellation, antenna cancellation achieves
the amount of self-interference cancellation required for fullduplex
operation.
The paper also discusses potentialMAC and network gains
with full-duplexing. It suggests ways in which a full-duplex
system can solve some important problems with existing wireless
systems including hidden terminals, loss of throughput
due to congestion, and large end-to-end delays.
INTRODUCTION
A basic precept of wireless communication is that a radio
cannot transmit and receive on the same frequency at the
same time, i.e. operate in a full duplex fashion. As wireless
signals attenuate quickly over distance, the signal from a local
transmitting antenna is hundreds of thousands of times
stronger than transmissions from other nodes. Hence it has
been generally assumed that one cannot decode a received
signal at a radio while it is simultaneously transmitting.
This paper challenges that assumption, and shows via analysis
and practical implementations on 802.15.4 radios that it
is possible to build full duplex radios. The implementation is
fairly simple, and can be built using off-the-shelf hardware
with software radios.
WIRELESS FULL DUPLEXING
This section examines why existing cancellation techniques,
RF and digital, are not enough to achieve full-duplex.
To understand the challenges in implementing wireless
full-duplex, we need to understand the way signals are received
at wireless nodes. The received signal from the antenna
is amplified through an automatic gain control stage
(AGC) and downconverted to either baseband or intermediate
frequency, filtered and then sampled through an Analogto-
Digital Converter (ADC) to create digital samples.
The accuracy of digital samples depends on the resolution
of the ADC. The AGC adjusts the gain of the received signal
to match the maximum level of the ADC to get maximum
resolution in the received signal. For the receiver to decode
a weaker signal using digital cancellation, the signal needs
to be strong enough to be captured within the resolution of
the ADC. Typical ADCs are 8-12 bit, representing a range
of 48-72dB. For an 8-bit ADC, if the weaker signal is 40dB
lower in power than the stronger signal, it only gets 1-bit
resolution.
ANTENNA CANCELLATION
This section analyzes the possible reduction in self-interference
by using antenna cancellation. It also evaluates its limits
with respect to bandwidth of the signal being transmitted and
the sensitivity of antenna cancellation to engineering errors.
It shows, using actual measurements, that antenna cancellation
achieves 20dB reduction in self-interference. This section
also evaluates the effects of using two transmit antennas
for antenna cancellation on the communication range. It
shows that antenna cancellation degrades the received signal
at other nodes in the network by at most 6dB compared to
the single antenna setup.
Performance of Antenna Cancellation
In an ideal scenario, the amplitudes from the two transmit
antennas would be perfectly matched at the receiver and the
phase of the two signals would differ by exactly . However,
we find that the bandwidth of the transmitted signal places
a fundamental bound on the performance of antenna cancellation.
Further, real world systems are prone to engineering
errors which limit system performance. The sensitivity of
the antenna cancellation to amplitude mismatch at the receive
antenna and to the error in receive antenna placement
is important to consider.
Digital Interference Cancellation
There is extensive existing work that describes digital cancellation
techniques [7, 8, 9]. Traditionally, digital cancellation
is used by a receiver to extract a packet from a desired
transmitter after the packet has collided with a packet
from an unwanted transmitter. To do this, the receiver first
decodes the unwanted packet, remodulates it and then subtracts
it from the originally received collided signal. In case
of canceling self-interference for full-duplex, the transmitted
symbols are already known, and thus decoding is not necessary
in order to reconstruct a clean signal.
Instead of decoding, coherent detection is used to detect
the self-interfering signal. The detector correlates the incoming
signal with the clean transmitted signal, which is
available at the output of the transmitter. The main challenge
in subtracting the known signal is in estimating the
delay and phase shift between the transmitted and the received
signals. As the detector has the complete knowledge
of originally transmitted signal, it uses this signal to correlate
with the incoming signal to detect where the correlation
peaks. The correlation peak technique gives both the delay
and the phase shift needed to subtract the known signal.
Thus, this technique, unlike some of the digital interference
techniques, does not require any special preamble or postamble
and is backwards compatible. Moreover, this technique
is modulation-independent as long as the clean signal can be
constructed.
RELATED WORK
Digital cancellation has been extensively used in many existing
schemes [8, 9, 7]. ZigZag [7] uses multiple transmissions
of colliding packets to decode the underlying packets.
This helps with solving the hidden terminal problem,
requiring n time slots to resolve collisions among n packets.
Analog network coding [9] uses access points as analog
symbol repeaters which also repeat symbols of colliding
packets. These repeated symbols are decoded at the respective
destinations. Such a technique gives throughput gains
when two flows are flowing in opposite directions through a
single route. For setups such as the one in Figure 12 where
the multiple flows are intersecting at a single node, analog
network coding will not give any throughput gain over traditional
routing unless the transmitting nodes can overhear
each other. Full-duplex will work in both the setups.
Successive interference cancellation [8] decodes and then
cancels strong interference signals, as long as the interference
is only about 20dB stronger than the signal being received.
ZigZag [7] extends this approach to decode multiple
colliding packets from multiple collisions. Our digital cancellation
scheme does not require decoding symbols, since
the decoder has the knowledge of the transmitted symbols.
DISCUSSION AND CONCLUSIONS
This paper has described the design of a practical single
channel wireless full-duplex system for 802.15.4. The
throughput gains achievable for a single hop wireless channel
are 84% in median. This paper also discusses additional
gains possible with wireless full-duplexing for multihop networks.
The main restrictions in implementing wireless fullduplex
systems are the design of wider band noise cancellation
circuits and making the digital cancellation algorithm
work in real time.
The paper shows that a combination of antenna cancellation,
RF interference cancellation and digital interference
cancellation can bring self-interference to within a few dB
of the noise floor. There still is a loss of a few dB in SINR,
which can lead to a difference in performance for multirate
systems. Existing rate selection algorithms take two approaches,
namely packet error rate based [3, 13], and signal
to noise ratio based [17, 19]. Packet error rate based
schemes would work directly for full-duplex radios. SNR/SINR
based schemes would have to take into account the loss
in SINR due to self-interference.