PaperlayerView exact revision on arXiv

Source-linked AI summary

Design and Characterization of a Full-duplex Multi-antenna System for WiFi networks

Melissa Duarte, Ashutosh Sabharwal, Vaneet Aggarwal, Rittwik Jana, K. K. Ramakrishnan, Christopher Rice, N. K. Shankaranarayanan

arXiv:1210.1639v2cs.NI

TL;DR

The paper asks whether full-duplex can deliver practical gains in IEEE 802.11 networks despite severe self-interference. It designs and evaluates a 20 MHz multi-antenna OFDM PHY with a backward-compatible MAC through experiments, simulations, and analysis. The results support higher rates and substantial throughput benefits, motivating full-duplex modes in future WiFi standards.

  • Problem

    Practical full-duplex communication must address the large power differential between a node’s self-interference and the signal of interest.

  • Method

    The paper combines a multi-antenna wideband self-interference canceller with an integrated PHY and MAC design compatible with IEEE 802.11x.

  • Results

    The design achieves high rate and extended range adequate for most indoor WiFi deployments, with full-duplex WiFi shown to be possible and beneficial in practical propagation environments.

  • Takeaways & Limitations

    The findings support including a full-duplex mode in future WiFi standards and indicate compatibility with accelerated adoption of full-duplex wireless.

Abstract

from arXiv · show

In this paper, we present an experimental and simulation based study to evaluate the use of full-duplex as a mode in practical IEEE 802.11 networks. To enable the study, we designed a 20 MHz multi-antenna OFDM full-duplex physical layer and a full-duplex capable MAC protocol which is backward compatible with current 802.11. Our extensive over-the-air experiments, simulations and analysis demonstrate the following two results. First, the use of multiple antennas at the physical layer leads to a higher ergodic throughput than its hardware-equivalent multi-antenna half-duplex counterparts, for SNRs above the median SNR encountered in practical WiFi deployments. Second, the proposed MAC translates the physical layer rate gain into near doubling of throughput for multi-node single-AP networks. The two combined results allow us to conclude that there are potentially significant benefits gained from including a full-duplex mode in future WiFi standards.

I. INTRODUCTION

The paper investigates whether practical WiFi systems can exploit full-duplex communication, addressing self-interference through a multi-antenna wideband PHY and compatible MAC. Experiments and analysis report improved cancellation, higher PHY rates, and substantial MAC throughput gains across practical settings.

  • Motivation: The central question is whether a practical WiFi system can leverage full-duplex gains for typical communication ranges.The study examines performance over the SNR range typical of WiFi communications.
  • Design: A 20 MHz multi-antenna PHY and MAC design enables full-duplex WiFi while supporting legacy half-duplex operation.The PHY combines antenna placement, per-subcarrier analog cancellation, and digital baseband cancellation; the MAC is designed for compatibility with 802.11.
  • Self-interference cancellation: 15 dB additional suppression is obtained by leveraging antenna polarization, highlighting antenna placement as crucial to full-duplex performance.More passive cancellation also makes the self-interference channel more frequency-selective, motivating per-subcarrier analog cancellation.
  • Self-interference cancellation: 85 dB median cancellation, with 70 dB minimum and 100 dB maximum, is achieved by the three-stage self-cancellation system.The reported median and maximum total cancellation are described as the best reported in the literature at that time.
  • PHY rates: For SNR values greater than 20 dB, 2×1 full-duplex can often outperform the other four evaluated antenna configurations.Its measured multiplexing gain can also exceed that of 2×2 half-duplex, whose measured gain is often less than two.
  • MAC performance: Full-duplex MAC throughput increases by 87% over legacy WiFi half-duplex without RTS/CTS, while gains versus RTS/CTS range from 30%–100%.For equal-node comparisons with RTS/CTS, sum throughput increases by a factor of at-least two; mixed systems can reach a 50% maximum increase.

II. MIMO WIDEBAND CANCELLER DESIGN

The proposed wideband MIMO canceller combines passive suppression with active analog and digital cancellation to reduce OFDM self-interference across subcarriers. Its coefficients depend on estimated self-interference and wired cancellation channels, so imperfect estimation and hardware nonlinearities limit ideal cancellation.

  • MIMO wideband model: The frequency-domain design represents the self-interference channel per subcarrier and as a K-subcarrier channel vector.The channel coefficient h_i,m,n[k] captures magnitude and phase on subcarrier k, while the vector collects values across K subcarriers.
  • Cancellation architecture: The design combines passive suppression, active analog cancellation before the ADC, and active digital cancellation after analog cancellation.Passive suppression reduces interference through propagation loss, while analog and digital stages subtract estimated residual interference.
  • Active analog cancellation: Analog cancellation subtracts a wired canceling signal whose coefficient is selected to match the self-interference channel on each subcarrier.Perfect cancellation is achieved in the ideal model when b_i,m,n[k] equals h_i,m,n[k]/h^W_i,n[k].
  • Practical constraints: Cancellation is not guaranteed to remain constant because RF and baseband noise affect channel estimation and cancellation weights.Amplifier nonlinearities and their impact on cancellation coefficients are identified as future work.
  • Channel estimation: Channel estimates for cancellation can use RTS/CTS pilots, additional pilot rounds, or correlation between transmitted and received self-interference payloads.The wired channel is static and therefore need not be estimated often, whereas the self-interference channel varies with time and frequency.
  • MAC interaction: RTS/CTS adds overhead but enables full-duplex operation and produces overall rate gains in the reported system.The paper connects the control exchange to the throughput evaluation in later sections.

III. PHY EXPERIMENT DESCRIPTION

The experiments use a testbed, antenna configurations, and physical-layer techniques implemented on WARP and compared across the study.

  • III. PHY EXPERIMENT DESCRIPTION: The experiment section describes the testbed, antenna configurations, and physical-layer techniques evaluated on WARP.Implementation details are included for the compared configurations.

A. Node Locations

The PHY evaluation uses a five-node, ten-link indoor setup spanning line-of-sight and challenging multi-wall channels, with hardware-equivalent full- and half-duplex configurations. Antenna placement and device presence are varied to study passive suppression and fair resource comparisons.

  • A. Node Locations: Five nodes form ten two-node links in an office building, covering line-of-sight and challenging multi-wall propagation environments.Experiments ran both at night and during office work hours, with people moving through the rooms.
  • PHY configurations: The study compares FD1×1, FD2×1, HD2×1, HD3×1, and HD2×2 physical layers using no more than four total chains.FD2×1 is compared with HD3×1 and HD2×2, while FD1×1 is compared with HD2×1.
  • Hardware-equivalent comparisons: Full- and half-duplex systems are compared using the same number of radio resources per node.The full-duplex resource count is M′ + 2N′, versus M + N for half-duplex systems.
  • Multi-antenna processing: The multi-antenna implementations use Alamouti coding for FD2×1 and HD2×1, rate-3/4 OSTBC for HD3×1, and spatial multiplexing for HD2×2.HD2×2 receive processing uses channel inversion.
  • Implementation comparison: The proposed FD2×1 Alamouti implementation requires three antennas per node, compared with at least four for the cited MIMO cancellation technique.The cited transmitter/receiver antenna cancellation technique requires six antennas per node for an FD2×1 Alamouti system.

D. Transmit Power Normalization

Transmit powers are normalized so full- and half-duplex networks radiate comparable energy and satisfy a shared instantaneous power constraint. The evaluation uses 20 MHz, 2.4 GHz over-the-air experiments with controlled traffic and repeated scenarios.

  • Power normalization: The normalization equates the total energy transmitted by a full-duplex node with that of a half-duplex node over a finite bidirectional communication interval.The framework relates transmission powers and durations for the two modes.
  • Instantaneous power constraint: A shared instantaneous network power limit Π constrains simultaneous full-duplex transmissions while preserving the corresponding half-duplex network power.The experiments use Π = 8 dBm and achieve the constraint with equality.
  • Radio operating range: The radios transmit at up to 25 dBm, but their gain relationship is linear only from 0 dBm to 15 dBm for the tested 20 MHz OFDM signals.Experiments therefore use powers near the middle of that linear range; amplifier nonlinearities remain future work.
  • Experimental platform: The testbed uses MATLAB for digital processing and WARP hardware for real-time over-the-air transmission and reception.The experiments run on a 2.4 GHz Wi-Fi channel without concurrent traffic, using a shared carrier-frequency reference.
  • Traffic and waveform: All systems use 20 MHz bandwidth with 64 subcarriers, including 48 payload subcarriers, and QPSK modulation per subcarrier.Each experiment transmits 90 packets from each node, with 68 OFDM symbols per packet.
  • Scenario coverage: The evaluation covers 40 scenarios formed from ten links and four antenna/device configurations.Each node transmits 587,520 bits across the 90 packets in a scenario.

IV. PHY EVALUATION: CANCELLER PERFORMANCE

The PHY evaluation characterizes passive, analog, and digital self-interference cancellation in a 20 MHz multi-antenna full-duplex design. Antenna placement, frequency-aware cancellation, and device geometry materially affect cancellation performance.

  • Experimental method: The experiments measure self-interference power after each cancellation stage using WARP-radio RSSI readings.Cancellation is computed as the difference between pre- and post-cancellation self-interference power.
  • Passive suppression: A2 improves passive suppression by approximately 5 dB relative to A1.The reported comparison holds both with and without the device.
  • Passive suppression: Device-induced pathloss and orthogonal antenna orientation together improve passive suppression by approximately 15 dB.The receiver antenna main lobe is placed orthogonal to the transmitter antennas’ main lobe in the favorable configuration.
  • Self-interference channel: Greater passive suppression makes the self-interference channel more frequency selective.The reported mechanism is suppression of strong line-of-sight paths, increasing dependence on weaker reflected multipath.
  • Analog cancellation: Per-subcarrier analog cancellation improves analog cancellation by approximately 5 dB over schemes using a common magnitude across subcarriers.It also improves total cancellation by approximately 3 dB in the reported comparison.

D. Performance of Digital Cancellation

Digital cancellation is evaluated after alternative analog cancellers, showing an interaction between cancellation stages. Stronger analog cancellation leaves less digital cancellation to achieve and can increase achievable rates through better frequency adaptation.

  • Digital cancellation: Per-subcarrier analog cancellation achieves less subsequent digital cancellation than FFC1 and FFC2.The paper attributes this behavior to reduced residual self-interference and noisier residual-channel estimation.
  • Total cancellation: Per-subcarrier analog cancellation improves total cancellation by approximately 3 dB over the compared flat-frequency cancellers.The proposed design also achieves larger pre-ADC cancellation than FFC1 and FFC2.
  • Total cancellation: A2 with device achieves total cancellation between 70 dB and 100 dB, with a median of 85 dB.The same configuration also produces the lowest residual self-interference-to-noise ratio.
  • Comparison with prior work: The proposed 20 MHz FD1×1 and FD2×1 canceller achieves total cancellation similar to or larger than prior reported work.The paper identifies the A2-with-device implementation as the best reported wideband multi-antenna result in its comparison.
  • Scope boundary: The cancellation values do not guarantee reduction of self-interference to the noise floor.The paper explicitly states that INR=0 is not guaranteed, although full-duplex can still achieve higher rates under some conditions.

V. PHY EVALUATION: RATE PERFORMANCE

The rate analysis compares empirical ergodic rates of full- and half-duplex systems across WiFi-relevant SNRs. With antennas placed around the device, full-duplex systems outperform hardware-equivalent half-duplex systems over a substantial practical SNR range.

  • Rate metric: The empirical ergodic rate is computed from post-processing SINR averaged over packets and is used as a PHY-layer capacity measure.Half-duplex rates are scaled by transmission time, whereas full-duplex nodes transmit simultaneously.
  • Antenna placement: Without a device between the antennas, full-duplex rates are mostly below half-duplex rates for SNR ≤30 dB.The paper concludes that a median cancellation of 78 dB cannot provide full-duplex gains at WiFi ranges in that setting.
  • Rate comparison: FD1×1 and FD2×1 consistently achieve larger rates than HD2×2, HD3×1, and HD2×1 from 20–30 dB SNR.This range covers nearly half the range of a typical WiFi system.
  • Rate comparison: FD2×1 achieves rates larger than HD2×2 at approximately the higher-SNR crossover and larger rates than HD3×1 above approximately 20 dB.These comparisons apply when antennas are placed around the device.
  • Multiplexing gain: FD1×1 and FD2×1 have larger per-node multiplexing gains than the evaluated half-duplex systems with antennas around the device.The slopes of rate versus SNR are approximately 1.3 and 1.5 times the HD2×2 slope, respectively.
  • Multiplexing gain: The FD1×1 rate slope is approximately 3.7 times larger than HD3×1 and HD2×1, while FD2×1 is approximately 4.2 times larger.These slope comparisons use the linear fits reported for the device-centered configurations.

VI. MAC DESIGN

The MAC adds full-duplex operation to legacy 802.11 through minimal changes covering packet discovery, acknowledgements, and overhearing-node behavior. Its strategies address simultaneous transmissions while preserving compatibility with half-duplex stations.

  • Discovery and transmission: The design modifies the standard RTS/CTS exchange so the secondary node can transmit immediately after CTS when it has data for the primary node.The secondary node discovers the primary node’s identity from RTS and selects the first queued packet intended for it.
  • Discovery and transmission: Full-duplex reception is not treated as a collision when the received packet comes from the node involved in the RTS/CTS exchange.The packet is accepted if the NAV duration has not expired.
  • ACK management: The MAC allows nodes to send ACKs while waiting for the other endpoint’s ACK, preventing mutual ACK timeouts.After sending the ACK, each node returns to waiting for the ACK packet.
  • ACK management: For asymmetric packet lengths, ACK wait time is adjusted to the end of the NAV duration when necessary.This accounts for the node transmitting the shorter packet waiting while the longer packet finishes.
  • Overhearing nodes: The MAC addresses channel-capture unfairness caused when full-duplex nodes resume after DIFS while other nodes wait for EIFS.Two strategies are proposed: one for all-full-duplex systems and one for mixed full-/half-duplex systems.
  • Overhearing nodes: In mixed deployments, full-duplex exchanges can cause unfairness in half-duplex uplink throughput.The second strategy makes full-duplex nodes use EIFS after exchanges to avoid significantly compromising half-duplex uplink throughput.

VII. MAC EVALUATION

The evaluation uses a commercial 802.11 simulation modified for full-duplex MAC, examining goodput across modulation, packet sizes, network sizes, and coexistence settings.

  • Metrics and scenarios: The MAC evaluation measures goodput under different modulation formats and packet sizes, including QPSK at 18 Mbps.The study also examines scaling to multiple nodes and coexistence dynamics.
  • Implementation: The implementation modifies a standard 802.11 codebase in OPNET Modeler-Wireless to support full-duplex MAC.Full-duplex framing uses RTS/CTS, simultaneous data transmission, and simultaneous ACKs.
  • Evaluation scenarios: The evaluation studies one-STA, multiple-full-duplex-node, and mixed full-/half-duplex-node networks.Mixed-node experiments include both modified nodes that ignore NAV-window collisions and legacy nodes that do not.
  • Simulation setup: Simulation parameters include 14 m AP-to-STA separation, free-space path loss, Gaussian self-interference and thermal noise, and 1500-byte maximum packets.Packets are sent as the payload of one full-duplex MAC frame without segmentation.

A. Goodput characterization for different packet sizes

Full-duplex goodput exceeds half-duplex across symmetric and asymmetric packet sizes, with gains approaching a doubling and favorable scaling across multiple nodes.

  • Symmetric traffic: At QPSK-18 Mbps with 1500-byte packets, total MAC goodput is 25.62 Mbps for full-duplex, versus 12.8 Mbps with half-duplex RTS/CTS and 13.69 Mbps without it.These values include MAC overheads.
  • Symmetric traffic: Full-duplex MAC doubles throughput versus legacy half-duplex with RTS/CTS and increases throughput 87% versus half-duplex without RTS/CTS for symmetric traffic.Full-duplex goodput is higher than half-duplex goodput with or without RTS/CTS.
  • Asymmetric traffic: As uplink packets increase from 40 to 1500 bytes, full-duplex goodput gain ranges from 1.3x (13.14 vs 10.07) to 2x (25.62 vs 12.8) versus half-duplex with RTS/CTS.Shorter uplink packets reduce uplink throughput while downlink throughput remains unchanged.
  • Multiple nodes: Theoretical scaling gives a 1 + 1/n uplink improvement and an n + 1 downlink improvement for full-duplex nodes.The analysis predicts larger downlink than uplink improvement for n > 1.
  • Multiple nodes: For n = 1, 2, 4, and 8 nodes, full-duplex-to-half-duplex sum-goodput ratios are 2, 2.03, 2.03, and 2.02 with RTS/CTS.Full-duplex retransmissions are also lower than in half-duplex with RTS/CTS.

C. Coexistence for full-duplex and modified half-duplex nodes that ignore collision during NAV

With modified half-duplex nodes that ignore NAV-window collisions, coexistence with full-duplex nodes increases total throughput while preserving uplink throughput from individual nodes.

  • Downlink behavior: Downlink throughput to a half-duplex node improves by a factor of 2, while downlink throughput to a full-duplex node improves by a factor of n.The corresponding theoretical transitions are 1/(n(n+1)) to 1/(m(n+1)) and 1/(n+1).
  • Throughput: In a mix of m full- and m half-duplex nodes, total throughput increases by 1 + m/(2m + 1) over a half-duplex-only system.Simulated factors are 1.39, 1.42, and 1.45 for m = 1, 2, and 3, respectively.
  • Uplink behavior: The uplink throughput from any node remains almost the same as in the half-duplex system.This matches the theoretical coexistence analysis.
  • Downlink behavior: Simulated throughput improvement for half-duplex nodes is 1.8 for m = 1, 2.05 for m = 2, and 2.5 for m = 4.These results are reported as comparable to the theoretical values.

D. Coexistence for full-duplex and legacy half-duplex nodes

Legacy-node coexistence requires a more polite full-duplex MAC: extra waiting reduces overall throughput but limits losses to legacy nodes.

  • Scope boundary: With limited AP buffering, missing packets for full-duplex nodes can reduce overall throughput toward half-duplex performance.The reported coexistence results were generated with a large AP buffer.
  • Legacy coexistence: The modified coexistence MAC avoids greatly impacting legacy half-duplex uplink throughput.Full-duplex nodes wait EIFS after full-duplex exchanges rather than DIFS before backoff.
  • Throughput trade-off: With eight full-duplex nodes, overall throughput decreases from 25.6 Mbps to 25.05 Mbps under the extra-wait scheme.For four full-duplex and four half-duplex nodes, throughput decreases from 18.37 Mbps to 17.12 Mbps.
  • Legacy coexistence: Full-duplex nodes waiting EIFS after full-duplex exchanges reduces overall throughput compared with coexistence using modified half-duplex nodes.The extra wait is intended to prevent repeated channel access by full-duplex nodes.
  • Extensions: The paper proposes extensions including rate-mode selection, packet aggregation, full-duplex signaling, node pairing, and three-node full-duplex.These are presented as possible MAC extensions beyond the evaluated design.

APPENDIX

The appendix documents the full-duplex OFDM node, experimental configurations, cancellation measurements, ergodic-rate evaluation, and FD-MAC procedures and scenarios.

  • MAC operation: The FD-MAC exchange lets an AP and station transmit data frames in opposite directions simultaneously after RTS/CTS coordination.The station sends FDDATA1 and the AP sends FDDATA2 after SIFS, followed by reciprocal acknowledgments.
  • System design: The 2×1 full-duplex OFDM node uses two transmitter antennas, one receiver antenna, passive suppression, and active analog self-interference cancellation.Passive suppression uses propagation loss through the transmitter-to-receiver paths; transmitter radios up-convert BB to RF, while receiver radios down-convert RF to BB.
  • Experimental setup: The experiments evaluate antenna placements, node locations, links, and hardware-equivalent full- and half-duplex antenna and radio configurations.The setup includes alternative antenna placements and ten considered links, with configurations summarized by antenna and radio counts.
  • Cancellation characterization: Cancellation evaluation covers passive, analog, and digital stages, including cancellation distributions, frequency responses, and residual self-interference INR.The results examine different cancelers and antenna placements, while the residual INR is measured after all cancellation stages.
  • Rate evaluation: Ergodic-rate experiments plot rate against SNR for antenna placements with and without the device between antennas.The appendix also includes condition-number measurements and comparisons across analog cancelers and modulation or constellation choices.
  • Simulation and protocol analysis: The appendix specifies MAC simulation parameters, packet-size goodput, multi-node scenarios, and coexistence cases involving modified or legacy half-duplex nodes.The FD-MAC flowchart includes defer/backoff, scheduling FDDATA2, NAV handling, and FDACK state transitions.
Loading 1210.1639v2…