Source-linked AI summary
Pushing the limits of Full-duplex: Design and Real-time Implementation
Achaleshwar Sahai, Gaurav Patel, Ashutosh Sabharwal
TL;DR
The paper addresses the lack of medium access protocols for practical single-channel full-duplex wireless. It implements FD-PHY and FD-MAC, achieving 80 dB self-interference suppression and 70% higher throughput than half-duplex in two-way exchange.
Problem
Prior full-duplex work largely focused on physical-layer feasibility between two nodes, leaving medium access protocol design unaddressed.
Method
The paper implements a real-time 64-subcarrier, 10 MHz OFDM physical layer and a full-duplex random access MAC using shared random backoff, FD headers, and virtual backoffs.
Results
80 dB of self-interference suppression is achieved through device-induced attenuation and analog cancellation, while FD-MAC delivers 70% higher throughput than half-duplex for identical transmit power.
Takeaways & Limitations
Real-time physical and medium access layers create a strong case for practical deployment of full-duplex wireless in networks.
Takeaways & Limitations
Asynchronous full-duplex can incur a 3 dB loss for the same bit error rate, while receiver dynamic-range backoff increases quantization noise and BER.
Abstract
from arXiv · showhide
Recent work has shown the feasibility of single-channel full-duplex wireless physical layer, allowing nodes to send and receive in the same frequency band at the same time. In this report, we first design and implement a real-time 64-subcarrier 10 MHz full-duplex OFDM physical layer, FD-PHY. The proposed FD-PHY not only allows synchronous full-duplex transmissions but also selective asynchronous full-duplex modes. Further, we show that in over-the-air experiments using optimal antenna placement on actual devices, the self-interference can be suppressed upto 80dB, which is 10dB more than prior reported results. Then we propose a full-duplex MAC protocol, FD-MAC, which builds on IEEE 802.11 with three new mechanisms -- shared random backoff, header snooping and virtual backoffs. The new mechanisms allow FD-MAC to discover and exploit full-duplex opportunities in a distributed manner. Our over-the-air tests show over 70% throughput gains from using full-duplex over half-duplex in realistically used cases.
1. INTRODUCTION
The report introduces real-time FD-PHY and FD-MAC designs that address full-duplex implementation and distributed medium access. Experiments report up to 80 dB self-interference suppression and up to 70% throughput gains over half-duplex.
- Motivation: FD-MAC addresses the lack of prior medium-access protocols for exploiting full-duplex opportunities in multi-node networks.Earlier work primarily established physical-layer feasibility for two-node exchanges.
- FD-PHY: FD-PHY is a real-time OFDM physical layer with 64 subcarriers and 10 MHz bandwidth.It implements active analog cancellation on a per-subcarrier basis.
- FD-PHY: 80 dB of self-interference suppression is achieved by combining device-induced attenuation from antenna placement with analog cancellation.The experiments use different antenna configurations on an iPad-sized device and require no baseband cancellation.
- FD-PHY: FD-PHY supports selective asynchronous full-duplex, but receiving while transmitting in the opposite direction is not reliably implementable.The supported asynchronous case incurs a 3 dB loss for the same BER.
- FD-MAC: FD-MAC uses shared random backoff, header snooping, and virtual contention resolution to discover and exploit full-duplex opportunities distributively.These mechanisms balance continued full-duplex use against access for competing flows.
- FD-MAC: 70% higher throughput is reported for full-duplex two-way exchange using FD-MAC than half-duplex at identical transmit power.The reported gain depends on distance, packet arrival pattern, and contention.
2. REVIEW OF FULL-DUPLEX WIRELESS
Full-duplex wireless is constrained by strong self-interference that can overwhelm the desired signal before digitization. The review explains why analog suppression is needed and summarizes prior cancellation approaches and their bandwidth limitations.
- 2.1 Main Bottleneck in Enabling Full-duplex: 15–100 dB higher self-interference can reach the receive antenna than the desired signal, making ADC resolution the main bottleneck.The large power disparity causes the stronger signal to swamp the weaker one during digitization.
- 2.1 Main Bottleneck in Enabling Full-duplex: Analog-domain suppression is necessary before the ADC, because relative amplitude disparity lowers the weaker signal’s effective digital-domain SINR.SINR is therefore an important metric for evaluating full-duplex methods.
- 2.2 Reported Methods: Prior full-duplex demonstrations reported approximately 60–70 dB of self-interference suppression using antenna cancellation or analog RF cancellation.The antenna-cancellation approach used two transmit antennas and required three physical antennas for SISO full-duplex.
3. REAL-TIME FULL-DUPLEX PHY
The paper implements a real-time OFDM full-duplex PHY and evaluates analog self-interference cancellation, antenna placement, and asynchronous operation on mobile-device form factors. The experiments show that device placement and configuration affect suppression, while asynchronous full-duplex is reliable only in one transmission direction.
- Real-time OFDM Transceiver: Perfect channel estimates would completely suppress self-interference in the analog cancellation model.In practice, estimation errors determine the residual interference after cancellation.
- Real-time OFDM Transceiver: The implementation reuses MIMO physical-layer headers and an open-source 10 MHz, 64-subcarrier design rather than adding special channel-estimation headers.The available MIMO architecture was repurposed for the cancellation scheme.
- Antenna Placement on Mobile Devices: Configuration B has the lowest self-interference power with and without analog cancellation among the tested antenna configurations.The same configuration is also described as suitable for MIMO operation, while larger devices are required because mobile form factors constrain antenna placement and RF-chain accommodation.
- Asynchronous Full-duplex: Asynchronous full-duplex can receive while transmitting with performance loss, but transmitting while receiving cannot be implemented reliably.The reported receiving loss is 3 dB for the same BER, and the unsupported direction constrains the MAC design.
4. MAC PROTOCOL DESIGN
FD-MAC is designed for infrastructure-based networks centered on one access point. Its design addresses identifying full-duplex opportunities, respecting physical-layer constraints, and preserving fair medium access.
- FD-MAC is described for infrastructure-based systems with one access point as the central scenario.
- The protocol must identify which nodes can engage in full-duplex during random packet arrivals.
- FD-MAC must support either synchronous exchanges or asynchronous reception while transmitting to another node.
- The access protocol should provide all nodes opportunities to access the medium rather than unduly favoring full-duplex flows.
4.2 Overview of FD-MAC
FD-MAC extends IEEE 802.11 DCF with selective collision avoidance and three mechanisms for distributed full-duplex access. The design targets both two-node exchanges and hidden-node AP scenarios.
- In the infrastructure network, full-duplex opportunities involve either AP–M1 exchange or simultaneous AP communication with hidden mobile nodes M1 and M2.
- Figure 4 connects nodes that are within one another’s radio range.
- FD-MAC retains dominant IEEE 802.11 DCF elements but applies collision avoidance selectively to exploit full-duplex opportunities.
- Shared random backoff lets paired nodes yield the medium temporarily while preserving the possibility of continued full-duplex transmission.
- Header snooping decodes headers during ongoing transmissions, even when NAV is nonzero, to estimate topology and initiate full-duplex opportunistically.
- Virtual contention resolution balances full-duplex use with network-wide access.
4.3 FD-MAC Packet structure
FD-MAC adopts the IEEE 802.11 packet format and adds a full-duplex header. The added fields signal packet type, queue state, exchange duration, CTS capability, and shared backoff information.
- FD-MAC packets contain PHY, MAC, full-duplex, payload, and CRC fields, with only the FD header differing from IEEE 802.11.
- The FD header’s DUPMODE field distinguishes half-duplex packets from full-duplex packets.
- The HOL field indicates whether the next buffered packet is addressed to the current destination.
- DURNXT and DURFD encode the next head-of-line packet duration and full-duplex exchange duration, respectively.
- The CTS bit indicates whether the current packet’s destination can send to its source, while SRB carries a 10-bit shared random backoff number.
- DURNXT and DURFD are optional for infrastructure hidden-node handling; without them, the FD header is 13 bits.
4.4 Shared Random Backoff
Shared random backoff synchronizes paired nodes after they discover reciprocal queued packets, while allowing other nodes to contend. The procedure also handles fairness, countdown behavior, and failed exchanges.
- FD-MAC begins by determining whether both contending nodes have packets for each other, because asynchronous full-duplex is unavailable in the two-node case.
- AP advertises HOL and initiates the exchange, after which M1’s ACK reports its queued packet duration and CTS capability.
- Both nodes become ready for full-duplex after exchanging ACK information needed for channel training and queue-state discovery.
- Shared random backoff prevents paired nodes from continuously occupying the medium and starving other nodes.
- Each node places a randomly selected contention-window backoff in the SRB field, and both wait for the larger value before resuming.
- Nodes do not pause countdowns during medium energy; instead, they perform a final idle-for-DIFS check after the shared interval.
- When failures prevent synchronized backoffs, nodes purge queue-state information and return to contention.
4.5 Snooping to Leverage FD Mode
FD-MAC uses overheard headers and topology-dependent rules to identify when nodes can exploit full-duplex transmissions without creating collisions.
- Topology estimation: FD-MAC distinguishes clique and hidden-node topologies by whether a mobile node overhears another mobile’s ACK.The estimate is made by overhearing packets rather than through explicit topology discovery.
- Clique topology: In a clique, only AP ⇄M1 and AP ⇄M2 are feasible because cross-direction combinations create simultaneous incoming-packet collisions.The topology prevents both mobile nodes from transmitting to the AP while the AP simultaneously transmits to the other mobile.
- Hidden-node topology: In a hidden-node topology, FD-MAC permits four full-duplex combinations, including simultaneous AP-to-mobile and mobile-to-AP transmissions.The additional combinations are enabled because the mobile nodes are outside each other’s radio range.
- Snooping-based access: For AP →M1 with M2 transmitting to AP, AP must begin first because a node cannot initiate transmission while receiving another packet.M2 snoops the AP’s FD header and may transmit if M1 is out of range and does not claim a two-way exchange.
- Snooping-based access: M2 waits through an ACK duration and uses the observed ACK to infer whether M1 is within its radio range.This waiting rule supports hidden-node identification before M2 attempts the complementary transmission.
- Snooping-based access: M2 fragments its packet to finish no later than AP’s transmission, avoiding collision with M1’s ACK.It decodes AP’s duration and marks the packet with FRAG =1.
4.6 Virtual Contention Resolution
FD-MAC uses buffer reordering and virtual contention to increase full-duplex opportunities while controlling unfairness and delay at the access point.
- Buffer reordering: The AP can reorder packets within a buffer of depth Bufdepth to place a packet for the active mobile node at the head of line.Bufdepth = 1 preserves FIFO operation, while larger values provide more matching opportunities.
- Virtual contention: Larger buffer depth increases full-duplex opportunities but can starve transmissions to other mobile nodes.Virtual contention lets the AP break away from an ongoing full-duplex handshake and serve other destinations.
- Virtual contention: Virtual contention probabilistically chooses between the current head packet’s destination and the destination active in the full-duplex exchange.The mechanism is arranged after an exchange to balance throughput with access to other queued destinations.
- Virtual contention: The probability of selecting consecutive out-of-order packets decays geometrically as p_pick^k.This makes long sequences of head-of-line bypasses increasingly unlikely.
- Tradeoff: Increasing Bufdepth increases throughput at the cost of increased packet delay, with throughput and average delay linearly related in the simulation.Similar throughput-delay pairs can often be obtained using a smaller Bufdepth and a higher reordering probability.
4.7 State Transitions in FD-MAC
FD-MAC coordinates transitions among half-duplex and full-duplex modes, with topology and contention mechanisms determining which transitions are available.
- Clique topology: In a clique, transitions between AP ⇆M1 and AP ⇆M2 must pass through half-duplex modes.The first packet of a two-way full-duplex exchange is always half-duplex.
- Half-duplex transitions: Existing 802.11 contention allows all half-duplex modes to switch among one another.From M1 →AP, the directly available full-duplex transition is AP ⇆M1 through the two-way setup mechanism.
- Hidden-node topology: Hidden-node topology adds {M2 →AP, AP →M1} and {AP →M2, M1 →AP} to the two-way AP ⇄mobile modes.Each additional mode requires a corresponding half-duplex mode to initiate it.
- Transition mechanisms: FD-MAC combines shared random backoff and AP virtual contention to ensure required half-duplex starting modes can occur.These mechanisms also support transitions among the full set of modes in the protocol state diagram.
4.8 FD-MAC evaluations on WARP
The WARP evaluation implements FD-MAC on a real-time full-duplex system and compares two-way full-duplex throughput with half-duplex throughput.
- Experimental setup: FD-MAC was evaluated on two full-duplex WARP nodes exchanging packets with each other.Both nodes always had a head-of-line packet for the other, allowing the protocol to discover and establish full-duplex exchanges.
- Experimental setup: The evaluation compares full-duplex and half-duplex throughput at identical transmit power.Both implementations run on WARP, and the transmission modulation is QPSK.
- Result: 70% higher throughput was achieved by FD-MAC’s full-duplex two-way exchange than by half-duplex.This is the reported result for the WARP comparison under identical transmit power.
5. DISCUSSION AND CONCLUSIONS
The paper presents FD-PHY and FD-MAC as first real-time implementations of full-duplex physical and medium access layers, while identifying further system optimization as an open direction. Its real-time, fully operational network-stack results support practical use of full-duplex in deployed networks.
- FD-PHY and FD-MAC are presented as the first real-time implementations of full-duplex physical and medium access layers.
- Further optimization of the system remains possible through joint design of transmitted signals, cancellation mechanisms, and baseband processing.
- The demonstrated real-time implementations show fully operational network stacks and create a strong case for practical full-duplex use in deployed networks.