In-Band Full Duplex

Doubling Spectral Efficiency — The Full-Duplex Promise

All deployed cellular systems use half-duplex (HD) transmission: uplink and downlink are separated in either frequency (FDD) or time (TDD). This separation wastes spectrum — at best, the system can use each Hz of bandwidth in one direction at a time.

In-band full duplex (IBFD) transmits and receives simultaneously on the same frequency, potentially doubling the spectral efficiency. The fundamental obstacle is self-interference (SI): the transmitter's own signal, which may be 100 -- 130 dB stronger than the desired received signal, leaks into the receiver chain.

After decades of research, SI cancellation has matured to the point where IBFD is a serious candidate for 6G, particularly for:

  • Simultaneous uplink/downlink at base stations (infrastructure FD).
  • Integrated access and backhaul (IAB) nodes.
  • Full-duplex relay and repeater nodes.
  • Simultaneous communication and sensing (ISAC receivers).

This section analyses the three-stage SI cancellation architecture and quantifies the rate gains and residual interference floors.

Definition:

Three-Stage Self-Interference Cancellation

Self-interference cancellation in IBFD is performed in three cascaded stages, each operating in a different domain:

1. Propagation-domain cancellation (20 -- 40 dB): Physical isolation between Tx and Rx antennas through spatial separation, cross-polarisation, directional isolation (beamforming nulls toward the Rx), and absorptive shielding. Achievable isolation depends on antenna geometry: separate Tx/Rx arrays (∼\sim40 dB), shared array with circulators (∼\sim20 dB).

2. Analog-domain cancellation (30 -- 60 dB): A reference copy of the transmitted signal is passed through an analog cancellation circuit (tapped delay line with adjustable gains and phases) that models the SI channel's multipath profile. The resulting anti-SI signal is subtracted at RF before the LNA. This stage is critical to prevent receiver saturation and ADC clipping.

3. Digital-domain cancellation (30 -- 50 dB): After ADC conversion, the known transmitted digital signal is used to reconstruct and subtract the residual SI in baseband. Adaptive filters (e.g., widely linear MMSE) model both linear and nonlinear SI components (PA distortion, IQ imbalance, phase noise).

The total SI cancellation is the sum (in dB) of all three stages:

βSI[dB]=βprop+βanalog+βdigital\beta_{\text{SI}}[\text{dB}] = \beta_{\text{prop}} + \beta_{\text{analog}} + \beta_{\text{digital}}

State-of-the-art systems achieve βSI=100\beta_{\text{SI}} = 100--130130 dB in total, bringing residual SI close to the noise floor for transmit powers up to 20 -- 30 dBm.

The ordering of stages matters: analog cancellation must reduce SI below the ADC dynamic range (typically 60 -- 70 dB below the Tx power) to avoid quantisation noise floors that digital cancellation cannot remove. This is the key bottleneck — any SI power that saturates the ADC is irrecoverably lost.

Self-Interference Signal Model

The self-interference signal at the receiver input of a full-duplex node can be decomposed as:

ySI(t)=hSI∗x(t)⏟linear SI+fNL(x(t))⏟nonlinear SI+ntx(t)⏟Tx noisey_{\text{SI}}(t) = \underbrace{h_{\text{SI}} * x(t)}_{\text{linear SI}} + \underbrace{f_{\text{NL}}(x(t))}_{\text{nonlinear SI}} + \underbrace{n_{\text{tx}}(t)}_{\text{Tx noise}}

where:

  • hSIh_{\text{SI}} is the (multipath) SI channel impulse response,
  • x(t)x(t) is the transmitted signal,
  • fNL(⋅)f_{\text{NL}}(\cdot) captures PA nonlinear distortion (typically modelled by a parallel Hammerstein model with odd-order terms up to 7th or 9th order),
  • ntx(t)n_{\text{tx}}(t) includes transmitter noise (phase noise, DAC quantisation noise, oscillator leakage) that is uncorrelated with x(t)x(t) and therefore cannot be cancelled digitally.

The transmitter noise floor is the ultimate limit on SI cancellation. For a transmit power of 20 dBm and a transmitter EVM of −40-40 dB, the Tx noise power is 20−40=−2020 - 40 = -20 dBm. With a receiver noise floor of −90-90 dBm (in 20 MHz), this Tx noise is 70 dB above the thermal noise — setting a practical ceiling on the total achievable cancellation.

Theorem: Full-Duplex Rate Gain Over Half-Duplex

Consider a single full-duplex base station communicating with two half-duplex users (one UL, one DL) on the same time-frequency resource with bandwidth BB. Let PtP_t be the transmit power in each direction, ∣hDL∣2|h_{\text{DL}}|^2 and ∣hUL∣2|h_{\text{UL}}|^2 be the desired channel gains, and PSI,res=Pt/βSIP_{\text{SI,res}} = P_t / \beta_{\text{SI}} be the residual self-interference power after cancellation.

The full-duplex sum rate is:

RFD=Blog⁡2 ⁣(1+Pt∣hDL∣2σ2+PSI,res)+Blog⁡2 ⁣(1+Pt∣hUL∣2σ2)R_{\text{FD}} = B\log_2\!\left(1 + \frac{P_t |h_{\text{DL}}|^2}{\sigma^2 + P_{\text{SI,res}}}\right) + B\log_2\!\left(1 + \frac{P_t |h_{\text{UL}}|^2}{\sigma^2}\right)

while the half-duplex (TDD with equal UL/DL split) sum rate is:

RHD=B2log⁡2 ⁣(1+Pt∣hDL∣2σ2)+B2log⁡2 ⁣(1+Pt∣hUL∣2σ2)R_{\text{HD}} = \frac{B}{2}\log_2\!\left(1 + \frac{P_t |h_{\text{DL}}|^2}{\sigma^2}\right) + \frac{B}{2}\log_2\!\left(1 + \frac{P_t |h_{\text{UL}}|^2}{\sigma^2}\right)

The FD gain ρ=RFD/RHD\rho = R_{\text{FD}} / R_{\text{HD}} satisfies:

  • ρ→2\rho \to 2 as βSI→∞\beta_{\text{SI}} \to \infty (perfect cancellation) and high SNR.
  • ρ<1\rho < 1 (FD is worse than HD) when residual SI dominates: PSI,res≫Pt∣hDL∣2P_{\text{SI,res}} \gg P_t |h_{\text{DL}}|^2.
  • In practice, ρ∈[1.3,1.8]\rho \in [1.3, 1.8] for βSI=100\beta_{\text{SI}} = 100--120120 dB and typical urban channel conditions.

Full duplex uses each Hz in both directions simultaneously, so the factor-of-2 time loss of TDD disappears. However, the DL receiver sees residual SI as additional noise, reducing its SINR. The net gain depends on whether the SI cancellation is sufficient to make the DL SINR comparable to the SI-free HD SINR.

Proof
Derivation of the gain expression

Define γDL=Pt∣hDL∣2/σ2\gamma_{\text{DL}} = P_t|h_{\text{DL}}|^2/\sigma^2 and γUL=Pt∣hUL∣2/σ2\gamma_{\text{UL}} = P_t|h_{\text{UL}}|^2/\sigma^2 as the SI-free SNRs, and γSI=PSI,res/σ2=γDL⋅Pt/(βSI∣hDL∣2⋅something)\gamma_{\text{SI}} = P_{\text{SI,res}}/\sigma^2 = \gamma_{\text{DL}} \cdot P_t / (\beta_{\text{SI}} |h_{\text{DL}}|^2 \cdot \text{something}).

More directly, for symmetric channels (γDL=γUL=γ\gamma_{\text{DL}} = \gamma_{\text{UL}} = \gamma):

ρ=log⁡2(1+γ/(1+γSI))+log⁡2(1+γ)log⁡2(1+γ)\rho = \frac{\log_2(1 + \gamma/(1 + \gamma_{\text{SI}})) + \log_2(1 + \gamma)}{\log_2(1 + \gamma)}

At high SNR (γ≫1\gamma \gg 1) and good cancellation (γSI≪γ\gamma_{\text{SI}} \ll \gamma):

ρ≈log⁡2(γ)+log⁡2(γ)log⁡2(γ)=2\rho \approx \frac{\log_2(\gamma) + \log_2(\gamma)}{\log_2(\gamma)} = 2

When γSI=γ\gamma_{\text{SI}} = \gamma (SI equals desired signal):

ρ=log⁡2(1+γ/(γ+1))+log⁡2(1+γ)log⁡2(1+γ)≈1+1log⁡2(γ)\rho = \frac{\log_2(1 + \gamma/(\gamma + 1)) + \log_2(1 + \gamma)}{\log_2(1 + \gamma)} \approx 1 + \frac{1}{\log_2(\gamma)}

which approaches 1 from above — the FD gain vanishes. ■\blacksquare

Full-Duplex SI Cancellation and Rate Gain

(Left) Waterfall diagram showing the SI power reduction through propagation, analog, and digital cancellation stages. (Right) Sum-rate comparison of full-duplex vs half-duplex as a function of SNR. Adjust the transmit power and cancellation depths to observe when FD provides a meaningful gain over HD.

Parameters
20
50
40

Full Duplex with Massive MIMO

Massive MIMO and full duplex are synergistic: the large number of spatial degrees of freedom at the base station can be used to simultaneously serve DL users, receive UL signals, and place spatial nulls toward the Rx array to suppress SI in the propagation domain.

Consider a base station with MM antennas split into MtM_t Tx and MrM_r Rx elements (or a shared array with circulators). The propagation-domain SI is an Mr×MtM_r \times M_t channel HSI\mathbf{H}_{\text{SI}}. By designing the Tx precoder F\mathbf{F} to include a null-space constraint:

HSIF≈0\mathbf{H}_{\text{SI}} \mathbf{F} \approx \mathbf{0}

additional 30 -- 50 dB of spatial SI suppression can be achieved without any analog hardware. This "beamforming-aided" SI cancellation is unique to multi-antenna FD systems and significantly relaxes the requirements on the analog canceller.

Recent results (e.g., Everett et al., 2016) demonstrate that a 64-antenna FD base station can achieve 90% of the theoretical 2×\times rate gain with total SI cancellation of only 80 -- 90 dB (vs 110+ dB for single-antenna FD), making massive MIMO FD a compelling 6G architecture.

Full-Duplex Applications in 6G

Beyond the raw rate gain, full duplex enables qualitatively new capabilities in 6G:

  1. Integrated access and backhaul (IAB): Small cells that simultaneously receive backhaul from the donor cell and serve access users, eliminating the half-duplex bottleneck at relay nodes.

  2. Simultaneous transmit and sense (STARS / ISAC): A radar receiver co-located with a communication transmitter can listen for echoes of its own signal while transmitting data — the foundation of monostatic ISAC.

  3. Latency reduction: In HD-TDD, a UE must wait for its allocated UL slot to transmit. FD allows immediate UL transmission during a DL slot, reducing round-trip latency by up to 50%.

  4. Spectrum efficiency in unlicensed bands: FD listen-before-talk (LBT) can sense the channel while transmitting, improving coexistence with Wi-Fi.

Self-Interference Cancellation Waterfall

Animated waterfall diagram showing how self-interference power is progressively reduced through propagation isolation, analog cancellation, and digital cancellation. The final residual SI is compared against the receiver noise floor.
Starting from 20 dBm Tx power, three cancellation stages reduce SI by 115 dB total, bringing residual SI to −95-95 dBm — just 6 dB above the noise floor.
⚠️Engineering Note

Full-Duplex SI Cancellation — State of the Art

State-of-the-art full-duplex demonstrators and their SI cancellation:

  • Rice University (2014): Single-antenna FD with 110 dB total cancellation (antenna isolation + analog + digital) at 20 MHz BW, enabling 1.87×\times throughput gain over HD in indoor tests.
  • Columbia/Stanford DARPA ACT (2020): 64-antenna massive MIMO FD with 95 dB total cancellation using spatial + analog + digital stages. Achieved 1.6×\times sum-rate gain over TDD.
  • Samsung 5G FD prototype (2023): Sub-6 GHz FD relay for IAB with 120 dB SI cancellation at 100 MHz BW.
  • Bandwidth scaling: SI cancellation degrades with bandwidth because the analog canceller must match the multipath SI channel across the entire band. Wideband cancellers (e.g., 400 MHz for 5G NR) require more tapped delay line stages, increasing complexity and power consumption.
  • Practical FD rate gain: Lab demos consistently show 1.4 -- 1.8×\times throughput gain; system-level simulations (accounting for inter-cell SI) project 1.2 -- 1.5×\times at network level.
Practical Constraints
  • •

    Best demonstrated: 130 dB SI cancellation at 20 MHz BW

  • •

    Bandwidth scaling: cancellation degrades ~3 dB per doubling of BW

  • •

    Network-level FD gain: 1.2-1.5x (less than point-to-point 1.4-1.8x)

,

Common Mistake: The "2x Rate Gain" Is a Best Case

Mistake:

Claiming that full duplex doubles the system throughput in all scenarios.

Correction:

The 2×2\times gain is an idealised upper bound assuming perfect SI cancellation, symmetric UL/DL traffic, no inter-node interference, and no cross-link interference (UL UE to DL UE). In practice: (1) residual SI reduces the DL SINR; (2) cross-link interference from UL UEs to nearby DL UEs can be severe; (3) traffic asymmetry means the idle direction wastes spectrum even in FD mode. System-level gains of 1.2×1.2\times -- 1.5×1.5\times are more realistic for network deployments.

Common Mistake: ADC Saturation Limits Digital Cancellation

Mistake:

Relying primarily on digital SI cancellation and under-designing the analog canceller.

Correction:

Digital cancellation operates on the quantised signal. If SI power at the ADC input exceeds its dynamic range, the quantisation noise floor rises irreversibly — no amount of digital processing can recover the lost information. The analog canceller must reduce SI to within the ADC's linear range (typically 60 -- 70 dB below the Tx power) before digitisation. This makes the analog canceller the most critical and most challenging component.

Key Takeaway

Full duplex achieves 1.4×1.4\times -- 1.8×1.8\times rate gain in practice through 100 -- 130 dB of cascaded SI cancellation. Massive MIMO spatial cancellation significantly relaxes the analog hardware requirements, making FD a compelling 6G architecture for infrastructure nodes (base stations, IAB relays, ISAC receivers).

In-Band Full Duplex (IBFD)

Simultaneous transmission and reception on the same frequency and time resource. Requires 100 -- 130 dB of self-interference cancellation through propagation isolation, analog circuits, and digital processing.

Related: Self-Interference Cancellation

Self-Interference Cancellation

The process of suppressing the transmitter's own signal at the co-located receiver in a full-duplex system. Performed in three cascaded domains: propagation (20 -- 40 dB), analog (30 -- 60 dB), and digital (30 -- 50 dB).

Related: In-Band Full Duplex (IBFD)

Quick Check

In a full-duplex transceiver with Pt=20P_t = 20 dBm, the analog cancellation stage achieves 50 dB of SI suppression and the digital stage achieves 40 dB. If propagation isolation provides 25 dB, what is the residual SI power?

−95-95 dBm

−75-75 dBm

−115-115 dBm

−50-50 dBm

Correction:
−95-95 dBm

Correct. Total cancellation: 25+50+40=11525 + 50 + 40 = 115 dB. Residual SI: 20−115=−9520 - 115 = -95 dBm. For a 20 MHz receiver with noise floor ≈−101\approx -101 dBm, the residual SI is 6 dB above the thermal noise — acceptable for moderate-SNR operation but it will degrade high-SNR performance.

Near-Field Communications and XL-MIMOOver-the-Air Computation