PaperlayerView exact revision on arXiv

Source-linked AI summary

Experiment-driven Characterization of Full-Duplex Wireless Systems

Melissa Duarte, Chris Dick, Ashutosh Sabharwal

arXiv:1107.1276v2cs.IT

TL;DR

Full-duplex systems face strong self-interference, so the paper experimentally characterizes passive antenna separation and active analog and digital cancellation. It finds that cancellation and link rate depend on self-interference power and that digital cancellation should be selected based on measured analog suppression, while channel statistics change after cancellation.

  • Problem

    The paper addresses limited characterization of how increasing self-interference power affects active cancellation and full-duplex rate, and how digital cancellation depends on analog suppression.

  • Method

    The paper performs an experiment-based, data-driven characterization of passive antenna-separation suppression, active analog and digital cancellation, rate performance, and self-interference-channel distributions.

  • Results

    Active cancellation increases with self-interference power; under constant pre-cancellation SIR, full-duplex rate increases with self-interference power, while digital cancellation can increase noise after strong analog suppression.

  • Takeaways & Limitations

    Digital cancellation is most useful as a selectively applied safety net when analog cancellation provides poor suppression, and active cancellation changes the self-interference channel distribution.

Abstract

from arXiv · show

We present an experiment-based characterization of passive suppression and active self-interference cancellation mechanisms in full-duplex wireless communication systems. In particular, we consider passive suppression due to antenna separation at the same node, and active cancellation in analog and/or digital domain. First, we show that the average amount of cancellation increases for active cancellation techniques as the received self-interference power increases. Our characterization of the average cancellation as a function of the self-interference power allows us to show that for a constant signal-to-interference ratio at the receiver antenna (before any active cancellation is applied), the rate of a full-duplex link increases as the self-interference power increases. Second, we show that applying digital cancellation after analog cancellation can sometimes increase the self-interference, and thus digital cancellation is more effective when applied selectively based on measured suppression values. Third, we complete our study of the impact of self-interference cancellation mechanisms by characterizing the probability distribution of the self-interference channel before and after cancellation.

I. INTRODUCTION

The paper uses extensive experiments to characterize passive and active self-interference cancellation in full-duplex wireless systems, including how cancellation affects rate and channel statistics. It finds that cancellation depends on self-interference power and on how analog and digital cancellation are combined.

  • Large self-interference can occupy most of the receiver ADC’s dynamic range and increase quantization noise for the desired signal.
  • The study characterizes antenna-separation suppression, active analog cancellation, and active digital cancellation using extensive experimental data.Antenna separation provides passive path-loss suppression, while analog cancellation occurs before ADC conversion and digital cancellation occurs in baseband.
  • 74 dB of total average self-interference cancellation can be achieved in the characterized implementation.
  • For constant receiver-antenna SIR before active cancellation, full-duplex link rate increases as self-interference power increases.Average active cancellation also increases with self-interference power, which the paper attributes to lower channel-estimation error.
  • Digital cancellation after analog cancellation can increase system noise when analog suppression is large, so it is most useful selectively when analog suppression is poor.The paper reports that digital cancellation should be applied frame-by-frame based on measured suppression performance.
  • Before active cancellation, the self-interference magnitude is modeled as Ricean with a large K-factor; after cancellation, the reduced LOS component yields a smaller K-factor.The paper characterizes K-factor values before and after active cancellation and presents this as a previously unreported characterization for a full-duplex architecture.

II. CHANNEL MODEL FOR FULL-DUPLEX

The channel model represents a narrowband full-duplex node with transmitted symbols passing through a wireless self-interference channel after antenna separation. It models the channel as frame-constant and uses antenna separation as the passive suppression stage before active cancellation.

  • The considered narrowband full-duplex node includes UPS, MFD, DACs, ADCs, and transmit and receive radios.
  • The model includes self-interference channel stages with passive and active suppression before forming the two-way full-duplex system model.
  • A transmitted symbol is normalized through its constellation symbol and average symbol energy, with E[|s_i[n,f]|^2] = 1 and E[|x_i[n,f]|^2] = E_S,i.
  • The self-interference channel h_I,i[f] remains constant during frame f and changes between frames, with Ω_I,i = E[|h_I,i[f]|^2].
  • Antenna separation is the passive suppression mechanism, and its cancellation depends on propagation loss between the transmitting and receiving antennas.After separation, the received self-interference signal is h_I,i[f]x_RF,i(t).

B. With Analog Cancellation

Analog cancellation injects a wired cancelling signal whose coefficient estimates the self-interference channel, while imperfect channel estimates limit cancellation. Digital cancellation then subtracts a baseband estimate of the residual, but is not studied without analog cancellation.

  • Analog cancellation: Analog cancellation adds a cancelling signal at the RF adder to suppress the received self-interference signal.The implementation uses a DAC, transmitter radio, RF attenuator, and RF adder.
  • Analog cancellation: Imperfect estimates caused by additive noise and other system distortions prevent complete analog cancellation.The paper notes that perfect cancellation would require an exact channel estimate.
  • Analog cancellation: The noiseless analog cancellation coefficient is κAC,i[f] = hI,i[f]/hZ,i[f], while bκAC,i[f] denotes its noisy estimate.The coefficient compensates for the self-interference channel hI,i[f] and the cancellation-path channel hZ,i[f].
  • Digital cancellation: Digital cancellation adds −bκDC,i[f]xi[n, f] in baseband to the received signal after analog cancellation.Its ideal coefficient equals the residual channel hI,i[f]−hZ,i[f]bκAC,i[f].
  • Digital cancellation: The experiments exclude digital cancellation without analog cancellation because that configuration does not yield an interesting full-duplex system.Digital cancellation is therefore evaluated only after analog cancellation.

D. Notation Simplification and Summary of Self-Interference Model Parameters

The model summarizes cancellation through normalized post-cancellation channels and average cancellation factors. Active cancellation improves the pre-cancellation signal-to-interference ratio by the corresponding cancellation factor in a two-way full-duplex link.

  • Cancellation parameters: The normalized channels hAC,i[f] and hACDC,i[f] represent the self-interference channels after analog and combined cancellation, respectively.They provide compact parameters for rewriting the received-signal model.
  • Cancellation parameters: Average cancellation is defined as the ratio of self-interference energy before cancellation to energy after cancellation.Separate quantities are defined for analog cancellation and combined analog-plus-digital cancellation.
  • Received-signal model: With active mechanism Φ ∈ {AC, ACDC}, the post-cancellation self-interference depends on hΦ,i[f], ΩI,i, αΦ,i, and xi[n, f].AC denotes Analog Cancellation and ACDC denotes Analog and Digital Cancellation.
  • Received-signal model: Active cancellation improves the SIR by a factor of αΦ,1 relative to antenna separation before active cancellation.The rate analysis considers two-way full-duplex communication between two nodes operating simultaneously in the same frequency band.
  • Received-signal model: The received SINR at Node 2 follows by swapping the node indices in the Node 1 expression.The model includes desired signals, residual self-interference, and receiver noise at both nodes.

III. EXPERIMENT SETUP AND SCENARIOS CONSIDERED

The experiments use two WARP-based full-duplex nodes in a controlled, low-mobility 2.4 GHz laboratory setup. They vary antenna separation and transmit power while estimating analog cancellation from recent channel measurements.

  • Hardware and processing: The setup uses two WARP full-duplex nodes connected through an Ethernet switch to a host PC running MATLAB.Over-the-air reception uses WARP hardware, while sample processing is performed in MATLAB.
  • Scope: Characterizing coherence time and alternative channel estimators for optimal training is left for future work.The reported implementation therefore uses ten-frame averaging for analog coefficient computation.
  • Scenario parameters: The experiments fix node distance at D = 8.5 m and vary same-node antenna separation d ∈ {10 cm, 20 cm, 40 cm}.Both nodes are 2 m above the floor with line-of-sight between antennas.

IV. MEASUREMENT-BASED CHARACTERIZATION OF CHANNEL PARAMETERS

Measurements characterize average and frame-level cancellation as functions of received self-interference and analog suppression. Active cancellation improves with higher received interference, but digital cancellation can reduce total suppression when analog cancellation is already strong.

  • Average cancellation: 48 data points from 24 experiment scenarios quantify analog and combined analog-digital suppression across received self-interference powers.Each marker represents an 800-frame experiment, with scenarios varying antenna separation, transmit power, and node.
  • Average cancellation: As average received self-interference power PRI,i increases, average suppression from both AC and ACDC increases.Linear fits capture the dependence on received interference better than constant fits because channel-estimation error decreases at higher received power.
  • Analog and digital cancellation: As average analog suppression αAC,i increases, average digital suppression after analog cancellation αDC,i decreases.The linear fit again captures the dominant behavior better than the constant fit.
  • Frame-level cancellation: When frame-level analog suppression is lower, digital cancellation is more likely to increase total suppression.Digital cancellation increases total suppression when αDC,i[f] > 0.
  • Design implication: When average analog cancellation exceeds 30 dB, applying digital cancellation can increase average self-interference.This agrees with the frame-level finding that digital cancellation should be applied selectively based on measured suppression.

B. Passive Suppression and Total Cancellation

The experiments quantify passive suppression from antenna separation and total cancellation after combining passive, analog, and digital mechanisms. They also show that active cancellation reduces the self-interference channel’s K-factor, with greater suppression producing greater reduction.

  • Passive suppression: 34 dB at d = 10 cm, 41 dB at d = 20 cm, and 44 dB at d = 40 cm quantify passive suppression.Larger antenna separation yields larger passive suppression.
  • Total cancellation: 66 dB at d = 10 cm, 73 dB at d = 20 cm, and 74 dB at d = 40 cm are the measured average total cancellations.Total cancellation combines passive suppression with active analog and/or digital cancellation.
  • Total cancellation: 74 dB of total cancellation does not reach the approximately -90 dBm noise floor for transmit powers between 0 dBm and 15 dBm.Despite this, an average cancellation of 74 dB can produce full-duplex rates larger than half-duplex rates.
  • Channel distribution: Ricean distributions fit the measured channel-magnitude histograms better than Rayleigh distributions because their KL distances are lower.The Ricean model is used for both pre-cancellation self-interference and cancellation-related channel estimates.
  • Channel distribution: 25 dB to 40 dB describes the pre-cancellation K-factor for antenna separations between 10 cm and 40 cm.These large K-factors are attributed to the proximity of same-node antennas.
  • Channel distribution: Active cancellation reduces the self-interference K-factor, and the reduction increases as achieved suppression increases.The paper attributes this behavior to attenuation of the strong line-of-sight component.

A. Computation of Achievable Rates

Achievable rates are computed from measured per-frame error and SINR values, then averaged across frames for each transmission and summed across the two-way link. The half-duplex reference time-shares the link equally between directions.

  • Full-duplex rate computation: The full-duplex achievable sum rate averages the per-frame sum log2(1 + SINR_1[f]) + log2(1 + SINR_2[f]) across frames.Both directions contribute simultaneously to each full-duplex frame’s sum rate.
  • Full-duplex rate computation: The per-frame SINR at Node i is estimated as the reciprocal of AEVMS_i[f].The achievable rate for frame f is then log2(1 + SINR_i[f]).
  • Full-duplex rate computation: The achievable rate for transmission to Node i is the average of log2(1 + SINR_i[f]) across frames.This averages the measured frame-level rates rather than using a single aggregate SINR.
  • Half-duplex comparison: The half-duplex reference uses the same antenna and radio resources per node while allocating 50% of the time to each transmission direction.Its achievable sum rate weights each directional rate by one half.

B. Achievable Rates with Increasing Power

With approximately constant pre-cancellation signal-to-interference ratio, increasing received self-interference power increases achievable full-duplex rates because active cancellation and received signal power both increase. Equal transmit-power increases therefore raise rates in both directions and can produce full-duplex gains over half-duplex.

  • Rate behavior: At approximately constant SIRAS,i, the achievable rate for transmission to Node i increases as received self-interference power increases.The experiments varied both received signal and self-interference powers together by increasing transmit power at both nodes.
  • Rate-Power Increase: Increasing transmission power at both nodes by the same amount increases achievable rates in both directions and the full-duplex achievable sum rate.This is the paper’s Rate-Power Increase design rule for two-way full-duplex systems.
  • Experimental verification: The majority of FD-AC and FD-ACDC data points show that achievable sum rate increases as transmission power PT increases.The experiments include analog-only and analog-plus-digital full-duplex configurations.

C. Achievable Rates with Selective Digital Cancellation

Digital cancellation improves some frames but can worsen self-interference in others when applied after analog cancellation. Selective application based on measured suppression achieves the best average full-duplex performance and avoids harmful frames.

  • Compared systems: Digital cancellation is applied to all frames for FD-ACDC, selectively for FD-ACSDC, and not at all for FD-AC.The selective decision uses the measured per-frame digital-cancellation suppression.
  • Results: The FD-ACSDC system has the best average full-duplex rate among the three tested configurations.Average rates are represented by the thick bars in Fig. 11.
  • Results: Digital cancellation raises rates in frames where analog cancellation performs poorly by increasing total self-interference suppression.This makes digital cancellation a safety net for low-suppression frames.
  • Selective digital cancellation: Applying digital cancellation continuously can reduce achievable rates in some frames, whereas selective application avoids those decreases.The selection rule applies cancellation only when αDC,i[f] > 0.

VI. CONCLUSIONS

The paper characterizes full-duplex self-interference cancellation and its effects on rate and channel statistics using actual measurements. It shows that combined analog and digital cancellation must be evaluated in a complete implementation, while the reported results come from one implementation.

  • The study provides measurement-based characterizations of cancellation performance, rate, and self-interference-channel distributions in a full-duplex implementation.It examines the effects of increasing self-interference and transmission power and reports statistical characterization before and after active cancellation.
  • The paper characterizes the K-factor and distribution of self-interference before and after active cancellation using extensive measurements.This addresses a statistical characterization that the paper identifies as previously lacking in measurement-based full-duplex studies.
  • Digital cancellation after analog cancellation varies with the amount of analog cancellation, so independently measured cancellation values cannot determine combined-system benefits.The experiments show that digital cancellation becomes more effective for frames with lower analog cancellation, and that full-system implementation is needed to reveal the combined benefit.
  • The reported conclusions are based on a single implementation of one full-duplex system.The authors nevertheless present the measurements as analyzing fundamental characteristics of full-duplex systems.
Loading 1107.1276v2…