PaperlayerView exact revision on arXiv

Source-linked AI summary

AFDM: Evolving OFDM Towards 6G+

Hyeon Seok Rou, Vincent Savaux, Zeping Sui, Giuseppe Thadeu Freitas de Abreu, Zilong Liu

arXiv:2602.08163v4eess.SP

TL;DR

6G+ needs waveforms that improve robustness in high-mobility and doubly dispersive channels without abandoning the OFDM infrastructure. This article develops a practical FDFD model and analyzes AFDM as an evolutionary waveform, finding extensive OFDM-chain reusability alongside tunable chirp-domain functionality. Open standardization questions remain around control-plane signaling and AFDM/OFDM coexistence.

  • Problem

    6G+ requires waveform capabilities for high mobility and ISAC while preserving backward compatibility and reuse of OFDM infrastructure.

  • Method

    The article develops a generalized pulse-aware FDFD channel model and analyzes AFDM compatibility, impairments, functionality, and reusability against OFDM.

  • Results

    AFDM retains extensive OFDM-chain reusability, with RF and FFT/IFFT components largely reusable while channel estimation and equalization require logic-level updates.

  • Takeaways & Limitations

    AFDM is presented as a modular, evolutionary candidate for high-fidelity 6G+ communications over doubly dispersive channels.

  • Takeaways & Limitations

    Control-plane signaling of chirp parameters, shared-resource-grid coexistence of AFDM and OFDM numerologies, and alignment with 3GPP study items remain unspecified.

Abstract

from arXiv · show

As sixth generation (6G) standardization accelerates, there is growing consensus in favor of evolutionary waveforms that add new capabilities while preserving compatibility with the orthogonal frequency division multiplexing (OFDM) core of 4G and 5G. This article positions affine frequency division multiplexing (AFDM) as such a candidate, providing structural robustness for high-mobility communications and integrated sensing and communication (ISAC) over doubly dispersive channels while remaining backward-compatible with the legacy OFDM air interface. We first develop a generalized fractional-delay-fractional-Doppler (FDFD) channel model that accounts for practical pulse-shaping filters and the resulting inter-sample coupling. Building on this model, we show that the AFDM transceiver reuses nearly the entire OFDM chain, adding only lightweight digital pre- and post-processing. We then analyze the impact of hardware impairments such as phase noise and carrier frequency offset, and examine the advanced functionalities enabled by the chirp-parameter domain, including index modulation and physical-layer security. Assessing reusability across the radio-frequency, physical, and higher layers, we conclude that AFDM offers an efficient path toward high-fidelity later versions of 6G and beyond (6G+) communications.

I. Introduction

6G+ waveforms must address high mobility, doubly dispersive channels, and emerging ISAC needs while preserving OFDM compatibility and hardware reuse. The article presents AFDM as an evolutionary candidate that combines chirp-based robustness and tunability with extensive OFDM-chain reuse, supported by a generalized practical FDFD model.

  • B. Why AFDM?: AFDM uses chirp-based subcarriers and tunable parameters to provide Doppler resilience, channel-diversity adaptation, index modulation, and physical-layer security.Its affine frequency-domain structure also enables secondary functionalities through tunable chirp parameters.
  • C. Contributions of this article: AFDM reuses nearly all conventional OFDM hardware and signal-processing blocks, adding only lightweight linear-complexity pre- and post-processing around FFT/IFFT operations.The reusable chain includes FFT/IFFT modules, pulse shaping, prefixing, and resource management.
  • C. Contributions of this article: AFDM is proposed as a 6G+ waveform candidate for extreme mobility, NTN, D2D, ISAC, and physical-layer security while preserving OFDM architectural compatibility.The article directly frames AFDM around these use cases and the OFDM legacy.
  • C. Contributions of this article: The article evaluates AFDM across compatibility, complexity, FDFD robustness, hardware impairments, advanced functionality, and reusability from RF through higher layers.Its stated scope includes direct comparison with OFDM across practical implementation dimensions.
  • II. System Model: The generalized FDFD model incorporates physical pulse-shaping kernels, capturing both fractional-Doppler spectral leakage and fractional-delay inter-sample interference.The model is intended to represent practical transceiver implementations rather than idealized grid-aligned channels.
  • B. Why AFDM?: Doppler shifts disrupt OFDM subcarrier orthogonality and introduce inter-carrier interference, motivating waveform designs that better accommodate doubly dispersive channels.The article identifies this limitation as especially relevant to high-mobility and ISAC scenarios.

C. Generalized Matrix-form Input-Output Relationship

The generalized FDFD matrix formulation models pulse-shape-dependent inter-sample coupling while recovering legacy sparse channel structure under integer delays. Fractional delays produce dense coupling, whereas pulse selection controls the effective channel sparsity and detector complexity.

  • Generalized FDFD formulation: Fractional delays make the legacy separation into independent phase-offset and cyclic-shift matrices invalid because of pulse-kernel-induced inter-sample interference.The generalized formulation instead uses a chirp-periodically circulant fractional-delay matrix that combines fractional delay and prefix phase effects.
  • Integer-delay special case: For integer delays, the generalized formulation reconstructs the sparse, phase-rotated cyclic-permutation matrix used in legacy OFDM and AFDM.This establishes consistency between the generalized FDFD model and the IDID/IDFD special cases.
  • Fractional-delay effects: Fractional delays produce dense inter-sample coupling in which each received sample becomes a linear combination of all N transmitted samples.The effective interpolated delay matrix G(ℓp) is generally dense and Toeplitz before finite-kernel localization is considered.
  • Pulse-shape dependence: The transmit pulse shape governs coupling sparsity and diagonal-band thickness while leaving the main shifted-component position determined by the integer delay.Sinc interpolation yields denser smearing, rectangular pulses yield sparser coupling with spectral leakage and ICI, and raised cosine lies between them.
  • Unified representation: The explicit decomposition Ψ(ℓp)=G(ℓp)+Φ(ℓp) separates pulse-dependent interpolation from prefix-induced phase correction and unifies IDID, IDFD, and FDFD cases.The framework exposes pulse-dependent sparsity relevant to the complexity and performance of sparse detectors.

D. Extension to MIMO Channels

The SISO FDFD model extends to MIMO channels through a Kronecker-product construction that combines temporal channel structure with per-path spatial signatures. The resulting blockwise model preserves the same per-block interference characteristics as the SISO case.

  • MIMO construction: The MIMO FDFD model uses a standard Kronecker-product construction for Nt transmit and Nr receive antennas, with each path represented by a spatial signature.For uniform linear arrays, the spatial signature is formed from transmit and receive array response vectors.
  • Channel structure: MIMO blocks retain the SISO per-block interference characteristics, scaled by the spatial signatures of the propagation paths.Extensions to planar, non-uniform, and correlated arrays are treated as standard procedures outside the article’s scope.

III. On the Compatibility of AFDM over OFDM

AFDM preserves the OFDM processing foundation while replacing Fourier subcarriers with chirp-based affine-frequency processing to improve robustness under Doppler and fractional delay. Its effective channel remains more distinguishable than OFDM’s in doubly dispersive conditions, while selected parameters can recover conventional cyclic-prefix behavior.

  • Compatibility: AFDM reuses the OFDM hardware and processing chain, adding only lightweight baseband operations around the core transform processing.The compatibility analysis covers transform processing, pulse shaping, resource allocation, and prefixing.
  • Channel structure: Doppler shifts create off-diagonal effective-channel elements and inter-carrier interference in OFDM, with fractional delays or Dopplers making the channel even denser.The AFDM formulation addresses these effects by using chirp-based subcarriers in the DAFT basis.
  • Channel structure: Figure 3 shows that OFDM path matrices become indistinguishable under Doppler, whereas AFDM preserves a more distinguishable effective channel matrix and better orthogonality.The comparison uses a three-path doubly dispersive scenario with N = 64.
  • Prefixing: For specific AFDM parameter configurations, the chirp-periodic prefix reduces to a conventional cyclic prefix with no phase offsets.This occurs when 2Nλ1 is an even integer.

C. MIMO Extension of OFDM and AFDM Signal Models

The MIMO extension applies the OFDM and AFDM modulator-demodulator structures blockwise, preserving the same per-block interference behavior with antenna-dependent responses. Across practical pulse shaping and prefixing, AFDM retains the OFDM pipeline with lower structural integration burden than OTFS.

  • C. MIMO Extension of OFDM and AFDM Signal Models: AFDM’s MIMO model preserves the SISO per-block interference structure while adding the corresponding antenna response coefficients.The extension replaces the SISO channel component with the effective channel including modulator operations.
  • Compatibility: AFDM adds two element-wise phase-rotation blocks around the FFT/IFFT, while the remaining OFDM processing blocks can be reused directly.The rotations require N element-wise complex multiplications per block and are also used in established OFDM techniques.
  • E. Pulse-Shaping, Windowing, and Prefixing Blocks: AFDM supports the same pulse-shaping operation as OFDM without hardware changes because the DAFT basis and chirp-periodic prefix preserve the required structure.The AFDM prefix is prepended along a one-dimensional affine-frequency resource dimension.
  • Comparison with OTFS: Under practical pulse shaping, OTFS requires additional transforms, filtering, or two-dimensional processing, whereas AFDM and OFDM retain their orthogonality and convolutional forms with simple rectangular pulses.OTFS may require dedicated pulse shaping and complex equalization because finite support prevents ideal bi-orthogonality.
  • Compatibility: AFDM preserves the core OFDM hardware pipeline and one-dimensional framing using lightweight chirp rotations and a re-parameterized cyclic prefix.The article’s compatibility analysis evaluates this reuse across multiple transceiver layers.

A. Modulation Complexity and Scalability

AFDM adds only lightweight chirp-domain processing around reusable OFDM transforms, maintaining modest complexity and scalable implementation relative to OTFS.

  • AFDM complexity: AFDM adds two element-wise chirp rotations around the FFT/IFFT, while other OFDM processing blocks can be reused directly.The additional rotations have linear complexity in the number of subcarriers.
  • AFDM complexity: 30% at N = 256, 24% at N = 1024, and 20% at N = 4096 quantify AFDM's overhead over conventional OFDM.The overhead decreases logarithmically as the FFT size grows, while FFT/IFFT processing remains dominant.
  • Comparison with OTFS: AFDM maintains modest overhead over OFDM, whereas both practical OTFS implementations require significantly higher computational burdens from multistage transforms and 2D processing.The comparison includes pulse-shaping overheads summarized in Table 1.
  • Implementation scalability: AFDM's additional processing uses local element-wise multiplications without global reshaping, whereas practical OTFS can require interleaving, reshaping, and more complex data handling.This distinction affects memory movement, buffering, latency, and energy consumption in practical modems.
  • Implementation scalability: AFDM and OTFS complexity and memory requirements scale linearly with antenna count, so their relative overheads remain consistent in massive MIMO deployments.The analysis therefore applies directly across antenna counts.
  • Implementation scalability: OFDM and AFDM support arbitrary one-dimensional FFT sizes, whereas OTFS requires N = KL with integer K and L, limiting grid flexibility.The OTFS divisibility constraint can produce awkward FFT sizes for practical implementations.

B. Pilot Scheme and Channel Estimation

AFDM adapts OFDM-inspired pilot schemes to the affine-frequency domain, using isolated or multiplexed pilots and estimation procedures for integer and fractional channel parameters.

  • Pilot design: AFDM channel estimation adapts established pilot schemes, with pilot subcarriers isolated by DAFT-domain guard bands to capture path contributions without interference.The guard-band requirement depends on the channel's maximum normalized delay and Doppler shift.
  • Pilot design: A conventional single-pilot scheme places one pilot among null subcarriers, while remaining subcarriers carry data in the affine-frequency domain.Multiple pilots can improve estimation through diversity but reduce spectral efficiency because each pilot requires at least Q + 1 unavailable subcarriers.
  • Pilot design: Guard-interval-free and superimposed pilot schemes avoid or reduce the spectral-efficiency loss of null-subcarrier guards, but require receiver-side processing for channel estimation and detection.Superimposed pilots add pilot symbols to data symbols, while guard-interval-free schemes replace nulls with data.
  • Channel estimation: For integer delays and Dopplers, AFDM estimates the 3P path parameters separately rather than solving the intractable joint maximum-likelihood problem.The procedure estimates path delays and Dopplers before obtaining complex path gains.
  • Fractional parameters: AFDM extends estimation to fractional Doppler and fractional delay by decomposing fractional-delay channels into structured virtual paths.The resulting deterministic structure supports accurate delay and Doppler estimation beyond the Nyquist rate at Nyquist-rate sampling.

3) Special Case: OFDM-like Estimation Under Low Doppler

Under special AFDM parameter settings and low-Doppler conditions, frequency-domain pilot structures become regular enough to reuse conventional OFDM estimation methods while retaining affine-frequency flexibility.

  • Low-Doppler estimation: Under low Doppler, V_fp ≈ I_N and the AFDM channel matrix becomes diagonal, so prior OFDM frequency-domain estimation methods remain valid.These conditions correspond to channels commonly assumed in OFDM systems.
  • Special parameter setting: When 2Nλ1 ∈ Z and 1/(2λ1) ∈ Z, a single affine-frequency pilot produces a regular, constant-modulus, equispaced frequency-domain structure.The resulting pattern is illustrated in Fig. 6 and corresponds to pilot arrangements used in OFDM-based standards.
  • OFDM compatibility: AFDM demodulation can use the DFT matrix F_N instead of A in the special case, enabling conventional frequency-domain channel estimation.Data can also be multiplexed across pilot subcarriers in this OFDM-like mode.
  • OFDM compatibility: AFDM can switch between frequency-domain and affine-frequency-domain modulation, demodulation, and channel-estimation methods according to channel severity.The article describes this flexibility as preserving backward compatibility with OFDM while supporting doubly dispersive channels.

C. Signal Detection and Receiver Architectures

AFDM receivers exploit structured DAFT-domain channels through linear, iterative sparse, and turbo architectures, trading detection complexity against interference suppression and decoding refinement.

  • Turbo architectures: Turbo receivers exchange extrinsic LLRs between detector and decoder so decoder soft outputs help the equalizer suppress residual interference.This differs from one-shot receivers that perform detection without outer detector-decoder iterations.
  • Receiver categories: AFDM detection uses linear detectors or sparse-channel detectors, including ZF, LMMSE, MRC, MP, and EP algorithms.The sparse methods exploit the structured sparsity of the effective DAFT-domain channel matrix.
  • Low-complexity equalization: Banded LDL factorization reduces the memory and computational burden of MMSE equalization by exploiting the effective channel's sparse bandwidth.Equalization then uses two triangular solves and one diagonal solve.
  • Iterative detection: MRC iteratively updates symbol estimates and residuals for interference cancellation until convergence or a maximum iteration count is reached.These inner iterations are distinct from turbo iterations between detector and decoder.
  • Sparse detection: The effective channel matrix maps naturally to a sparse factor graph in which each received symbol connects to P transmitted symbols and each transmitted symbol connects to P observations.This structure supports iterative message passing and interference refinement.
  • Sparse detection: Direct MAP detection has exponential complexity in N, while message passing reduces complexity by approximating residual interference as Gaussian and scaling linearly with N.The MP detector exchanges Gaussian observation-to-variable messages and constellation PMFs between variable and observation nodes.

2) Turbo Receiver

Turbo receivers iteratively exchange soft information between detection and decoding, allowing AFDM equalization to refine estimates in dispersive channels while retaining the legacy receiver architecture around the detector.

  • Turbo Receiver: Turbo receivers exchange soft reliability information between the detector and decoder so both modules progressively refine their estimates.The decoder’s soft outputs help the equalizer suppress residual interference.
  • Turbo Receiver: The detector’s extrinsic log-likelihood ratios are formed from estimated symbols, transmitted bit sequences, and constellation bit patterns.The symbol-to-bit converter produces the extrinsic information used in the iterative loop.
  • Turbo Receiver: Iterative feedback helps mitigate inter-carrier interference, reduce error propagation, and reshape the effective channel for subsequent decoding.This exploits coding and diversity gains simultaneously.
  • Turbo Receiver: AFDM turbo reception replaces specialized internal detection logic while preserving the surrounding high-level receiver infrastructure inherited from OFDM.AFDM may use LDL-based banded equalization or sparse factor-graph processing inside the detection block.

D. Impact of Fractional-Delay Model on Receiver Design

The generalized FDFD model materially affects AFDM receiver design by eliminating modeling errors caused by fractional delays and by determining sparse-detector complexity through pulse-shaping-induced coupling.

  • D. Impact of Fractional-Delay Model on Receiver Design: BER reaches an irreducible 10^-1 floor for the integer-delay receiver, whereas the proposed FDFD receiver removes the floor and recovers the expected waterfall.The comparison uses genie-aided MMSE receivers under a true FDFD channel with N = 64 and P = 3.
  • D. Impact of Fractional-Delay Model on Receiver Design: Ignoring fractional delay creates modeling error rather than estimation error, dominating performance at moderate-to-high SNR.The integer-delay model cannot represent fractional-delay inter-sample coupling.
  • D. Impact of Fractional-Delay Model on Receiver Design: Average taps per row capturing 99% of row energy are 2.6 for rectangular, 4.6 for raised-cosine α = 0.5, and 12.3 for sinc pulses.The pulse shape therefore controls the effective channel density used by sparse detectors.
  • D. Impact of Fractional-Delay Model on Receiver Design: Pulse shaping trades spectral containment against detector complexity: rectangular pulses minimize coupling, sinc maximizes it, and raised cosine offers a compromise.Message-passing and banded-MMSE costs scale with the number of nonzero taps per row.
  • D. Impact of Fractional-Delay Model on Receiver Design: MIMO-AFDM extends the SISO estimation and detection methods per antenna, while standard digital beamforming applies when Doppler is sufficiently small.Special parameter settings also permit frequency-domain channel estimation and beamforming.

F. Multiple Access and Coexistence with OFDM

AFDM supports multiple access and coexistence with OFDM through configurable frequency, affine-frequency, and spatial multiplexing schemes that preserve orthogonality or enable separable hybrid signals.

  • F. Multiple Access and Coexistence with OFDM: DAFT-spreading-based AFDMA spreads each user’s data across the full DAFT domain to exploit diversity and spreading gains in high-mobility environments.It is contrasted with per-block and per-subcarrier multiplexing in spectral efficiency, receiver complexity, and diversity exploitation.
  • F. Multiple Access and Coexistence with OFDM: Per-block AFDMA allocates subcarrier blocks to users and can replace an AFDM sub-block with an OFDM sub-block without inducing inter-carrier interference.The two paradigms are illustrated as per-block and per-subcarrier mappings.
  • F. Multiple Access and Coexistence with OFDM: Per-subcarrier AFDMA preserves orthogonality in both affine-frequency and frequency domains, enabling fine-grained AFDM/OFDM multiplexing within one block.This contrasts with per-block allocation and supports mixed symbols at subcarrier resolution.
  • F. Multiple Access and Coexistence with OFDM: The resulting hybrid signal is separable at the receiver, supporting flexible multi-user access under coexisting OFDM and AFDM signaling.This provides a direct receiver-side consequence of the coexistence design.
  • F. Multiple Access and Coexistence with OFDM: MIMO-AFDMA combines spatial and domain-specific multiplexing for simultaneous base-station communication with multiple user equipments.Uplink domain partitioning supports asynchronous high-mobility transmission while AFDM’s Doppler robustness limits mutual interference.
  • F. Multiple Access and Coexistence with OFDM: AFDM and OFDM signals can be spatially multiplexed without interference, demonstrating coexistence and backward compatibility particularly at low Doppler.The MIMO example uses four transmit antennas and two OFDM users.

1) Characterization of PHN and CFO

AFDM’s chirp basis changes how phase noise and carrier frequency offset affect detection, while tunable chirp parameters support index modulation, physical-layer security, and reusable PAPR mitigation methods.

  • 1) Characterization of PHN and CFO: AFDM’s chirp slope partially aligns with CFO-induced phase rotation, making it more tolerant to moderate CFO than OFDM’s static subcarriers.OFDM impairments appear as common phase error and rapidly varying inter-carrier interference.
  • 1) Characterization of PHN and CFO: At BER 3 × 10^-3, OFDM loses approximately 2 dB from phase noise and 8 dB from CFO, while AFDM remains nearly ideal at θCFO = 0.1.At BER 10^-3 under phase noise, AFDM gains 10.5 dB over OFDM.
  • 1) Characterization of PHN and CFO: AFDM’s tunable chirp parameters λ1 and λ2 support index modulation, physical-layer security, and PAPR reduction without changing the fundamental framework.The same tunability preserves structural compatibility with the OFDM legacy.
  • 1) Characterization of PHN and CFO: Chirp-permutation index modulation maintains all active subcarriers while providing performance gains, and λ2 selection can increase aggregate bit rate without additional bandwidth.These mechanisms extend index modulation beyond conventional active-subcarrier selection.
  • 1) Characterization of PHN and CFO: AFDM physical-layer security uses chirp-sequence permutations and parameter hopping, with reported higher brute-force demodulation complexity than OTFS and OFDM.The security mechanisms exploit the tunable chirp dimensions.
  • 1) Characterization of PHN and CFO: AFDM has an OFDM-like PAPR profile, allowing legacy clipping, companding, and precoding methods to remain applicable.Dedicated methods include finite-set λ2 pre-chirp selection and block-diagonal DFT pre-chirp matrices.

I. Reusability of AFDM: A Layer-by-Layer Summary

AFDM preserves substantial OFDM infrastructure across RF, baseband, and physical-layer processing while adding chirp-aware functions for challenging channels. Its remaining barriers include synchronization, FDFD-aware channel estimation, and unspecified control-plane standardization details.

  • RF and baseband reuse: AFDM reuses OFDM radio-frequency components and wraps existing OFDM baseband blocks with limited additional processing.The article reports direct reuse of power amplifiers and transceivers, alongside FFT/IFFT-oriented processing reuse.
  • Waveform structure: AFDM retains one-dimensional time-frequency signaling, time-domain pulse shaping, and a prefix of the same length as OFDM’s cyclic prefix.Its chirp-periodic prefix replaces the cyclic prefix while preserving the prefix-length relationship.
  • Pilots and channel robustness: AFDM uses chirp-aware pilots with comparable overhead, whereas delay-Doppler alternatives require two-dimensional pilots and higher overhead.The supplied comparison also reports full delay-Doppler diversity for one alternative and lower Doppler robustness for OFDM.
  • Hardware impairments: AFDM improves phase-noise and carrier-frequency-offset resilience relative to OFDM while retaining OFDM-like peak-to-average power characteristics.The comparison reports phase-noise gains, reduced CFO-induced SNR loss, and statistical PAPR identity with OFDM.
  • Open challenges: AFDM’s compatibility is extensive but not complete: synchronization, FDFD-aware channel estimation, pilot overhead, and control-plane signaling remain open practical issues.The article assumes perfect channel state information for its structural analysis and identifies chirp-parameter signaling and AFDM/OFDM coexistence as standardization items.
  • Conclusion: The article concludes that AFDM offers an evolutionary route to high-fidelity 6G+ communications by combining doubly dispersive-channel capability with hardware reusability.This conclusion is framed as modular evolution rather than radical system redesign.
Loading 2602.08163v4…