A new arXiv preprint reports that a compact DCT-based algorithm recovered nonlinear channel behavior in simulated OFDM signals at low signal-to-noise ratios. In its TWTA case, all channel components were recovered with negligible error at 30 dB, and accurate estimates continued to about 15 dB. When those estimates were used for compensation, the strongest reported result was nearly indistinguishable from an ideal linear OFDM system.
The study asks whether one ML-based estimator can jointly recover three parts of the distortion problem: multipath in the channel, nonlinear changes in amplitude, and nonlinear changes in phase. The model uses a compact DCT representation of those responses, which the paper describes as fast and well-conditioned in the tested settings. The recovered response can then be used for predistortion, meaning a pre-adjustment of the transmitted signal, or for iterative decoding.
The joint estimator alternates updates of channel taps, the coefficients representing the multipath channel, and DCT coefficients for amplitude and phase distortion. It applies stochastic-gradient rules over the training samples.
The test was deliberately compact
The main simulation used 1,024 OFDM subcarriers, 16-QAM, and a 16-sample cyclic prefix. The true channel length was three taps, while the estimator assumed six; the run used five iterations, a DCT resolution of 512, six amplitude coefficients and 12 phase coefficients. It transmitted 25,000 time samples, reported in the paper as 25.000, equivalent to 24 OFDM symbols or two TTIs. The comparison covered a Saleh-model TWTA and an SSPA scenario.
The amplifier changed the outcome
For the SSPA simulation, reliable estimation reached 0 dB, while errors at -10 dB remained small enough for excellent compensation. The reported training requirement in that setting was two OFDM symbols, or 2,048 time samples, compared with 24 symbols in the general setup.
In the TWTA bit-error-rate evaluation, predistortion was nearly indistinguishable from ideal linear OFDM, while iterative decoding showed a small degradation. The quality of the estimate still mattered: a channel estimate made at 0 dB produced an error floor at high signal-to-noise ratios, although performance stayed close to higher-quality estimates up to about 30 dB.
In the SSPA compensation test, both schemes had comparable performance, and channel estimation remained reliable even at -10 dB. The paper does not give exact tabulated bit-error-rate values or uncertainty intervals, so the size of the difference between the reported curves cannot be pinned down from the analysis.
The paper’s efficiency argument comes from the compact parameterization. It reports per-sample complexity of O(Q² + L) for channel adaptation, O(QL) for magnitude adaptation and O(L + Q) for phase adaptation, where Q denotes the DCT dimensions and L the channel length. It also describes the DCT-based methods as structurally more efficient than neural-network alternatives, which require substantially higher training and sample complexity. These are analytical complexity expressions, not measurements of runtime, memory, energy use or hardware-resource savings.
A promising result with a narrow boundary
Important limits sit inside the model. It uses an instantaneous nonlinear block followed by a linear FIR filter, leaving nonlinear memory effects and the Wiener-Hammerstein extension outside the study. The phase-domain reformulation also requires sufficiently high signal-to-noise ratio because phase extraction changes the noise structure; the paper places that requirement at about 10 dB.
The optimization is nonconvex: in the considered scenarios, the algorithm consistently reached a local minimum, a solution that may not be the best overall, but global convergence was not guaranteed. Underestimating the channel length impaired recovery, while overestimating it allowed redundant coefficients to be suppressed.
The supplied document is arXiv:2608.25847v1, an arXiv preprint dated 26 Aug 2026. The evidence is limited to the specified simulated OFDM, TWTA and SSPA configurations. Nonlinear channels with memory and measured real-time resource use remain outside the reported results.
Paper data and sources
Original title: Efficient DCT-Based Estimation and Compensation of Nonlinear Channels for OFDM Systems
Authors: Marc Martinez-Gost, Ana Pérez-Neira, Miguel Ángel Lagunas
Journal/Repository: arXiv
Status: Preprint, not yet peer-reviewed
First online: 2026-08-26
DOI: Not available
Original paper · Full text