Publications
You can also find my articles on my Google Scholar profile. N. Kim denotes myself.
Preprints
Received OFDM grids preserve both pilot and payload observations, yet wireless foundation models are rarely pre-trained on them. We introduce LWM-TF, a 9.58M-parameter factorized masked autoencoder pre-trained on complex 5G NR uplink grids taken after the FFT and before channel estimation. Each block attends first across OFDM symbols and then across resource blocks and receive antennas, and the mask follows the same two axes. Because the payload phase is unpredictable, the model reconstructs the normalized log-power of each masked token and predicts its scale separately; ablations show that the two terms are complementary and that this target outperforms complex-signal reconstruction. The encoder is pre-trained on 850,496 physical uplink shared channel (PUSCH) slots from a 10^6-slot dataset simulated with the NR transmit chain under various channel conditions. With the encoder frozen, linear probes classify modulation, predict CRC success, and detect UE mobility. On a separately generated test set, 16 labeled slots per class give 0.85 modulation accuracy, 0.82 CRC area under the ROC curve (AUROC), and 0.73 mobility AUROC, ahead of a supervised CNN on every paired draw. With 10,000 labels the probes reach 0.94, 0.96, and 0.92, well above engineered receiver statistics, and detect mobility where the CNN stays at chance. Without any impairment labels, distances in the frozen feature space flag unseen waveform impairments with 0.78 macro AUROC, against 0.68 for receiver statistics. A single pre-trained encoder thus provides receiver features that transfer across tasks, most strongly when labels are scarce.
In dense deployments, massive multi-user multiple-input multiple-output (MU-MIMO) base stations can acquire instantaneous channel state information (CSI) for only a limited subset of users per scheduling interval, restricting multiuser diversity. We therefore propose Digital Twin User pre-Screening (DiTUS), a digital-twin (DT)-aided framework that identifies promising users before instantaneous CSI acquisition. DiTUS forms spatial covariances from DT-inferred departure angles and path powers. Optional Gaussian-process (GP) calibration mitigates path-power bias, while the dominant rank-r eigenspace of the aggregate covariance yields common reference beams. It prescreens the pool using DiTUS-P, a low-complexity projection-energy rule, or DiTUS-L, a greedy log-determinant rule that promotes spatial compatibility. A two-level protocol collects scalar beam reports from shortlisted users and requests r-dimensional effective-channel vectors only from the scheduled set. The framework also supports proportional-fair scheduling. At 15 dB under DT imperfections, simulations with 128 candidates, a 64-user effective-CSI acquisition budget, and a 64-user shortlist show that DiTUS-L achieves 35.06 +/- 0.49 bps/Hz versus 30.28 +/- 0.60 bps/Hz for semi-orthogonal user selection (SUS) with full-dimensional CSI from 64 users, demonstrating that DT-based prescreening preserves substantial multiuser-diversity gains by identifying strong, spatially compatible users before acquiring effective-channel vectors.
2026
The received in-phase and quadrature (I/Q) baseband signals inherently encode physical-layer and channel characteristics of wireless links. Learning robust and transferable representations directly from such raw signals, however, remains challenging due to heterogeneous communication systems, diverse propagation environments, and limited labeled data. To address this, we present LWM-Spectro, a transformer-based foundation model pretrained on large-scale I/Q data represented as time-frequency spectrograms. The model leverages self-supervised masked modeling, contrastive learning, and a mixture-of-experts (MoE) architecture to learn general-purpose wireless representations. These representations transfer effectively to downstream tasks such as modulation classification and joint SNR/mobility recognition, even with minimal supervision. Across tasks, LWM-Spectro consistently outperforms state-of-the-art deep learning baselines in both few-shot and data-rich regimes, providing a unified foundation for wireless learning.
2025
This paper presents a novel framework for low-latency frequency division duplex (FDD) multi-input multi-output (MIMO) transmission with Internet of Things (IoT) communications. Our key idea is eliminating feedback associated with downlink channel state information at the transmitter (CSIT) acquisition. Instead, we propose to reconstruct downlink CSIT from uplink reference signals by exploiting the frequency invariance property of channel parameters. Nonetheless, the frequency disparity between the uplink and downlink makes it impossible to get perfect downlink CSIT, resulting in substantial interference. To address this, we formulate a max-min fairness problem and propose a rate-splitting multiple access (RSMA)-aided efficient precoding method. In particular, to fully harness the potential benefits of RSMA, we propose a method that approximates the error covariance matrix and incorporates it into the precoder optimization process. This approach effectively accounts for the impact of imperfect CSIT, enabling the design of a robust precoder that efficiently handles CSIT inaccuracies. Simulation results demonstrate that our framework outperforms other baseline methods in terms of the minimum spectral efficiency when no direct CSI feedback is used. Moreover, we show that our framework significantly reduces communication latency compared to conventional CSI feedback-based methods, underscoring its effectiveness in enhancing latency performance for IoT communications.
In this paper, we propose a precoding framework for frequency division duplex (FDD) integrated sensing and communication (ISAC) systems with multiple-input multiple-output (MIMO). Specifically, we aim to maximize ergodic sum spectral efficiency (SE) while satisfying a sensing beam pattern constraint defined by the mean squared error (MSE). Our method reconstructs downlink (DL) channel state information (CSI) from uplink (UL) training signals using partial reciprocity, eliminating the need for CSI feedback. To obtain the error covariance matrix of the reconstructed DL CSI, we devise an observed Fisher information-based estimation technique. Leveraging this, to mitigate interference caused by imperfect DL CSI reconstruction and sensing operations, we propose a rate-splitting multiple access (RSMA) aided precoder optimization method. This method jointly updates the precoding vector and Lagrange multipliers by solving the nonlinear eigenvalue problem with eigenvector dependency to maximize SE. The numerical results show that the proposed design achieves precise beam pattern control, maximizes SE, and significantly improves the sensing-communication trade-off compared to the state-of-the-art methods in FDD ISAC scenarios.
The overhead of acquiring downlink (DL) channel state information at the transmitter (CSIT) poses a significant challenge for frequency division duplex (FDD) massive multiple-input multiple-output (MIMO) systems. To address this, we propose a framework that reconstructs DL CSIT solely from uplink (UL) pilots, leveraging partial frequency invariance of channel parameters. However, since perfect reconstruction is infeasible due to inherent discrepancies between the UL and DL bands, multi-user interference (MUI) arises. To address this, we formulate the ergodic sum spectral efficiency (SE) by incorporating the estimated channel and error covariance matrix (ECM). Additionally, we propose an ECM estimation technique using observed Fisher information and enhance system performance with a rate-splitting multiple access (RSMA) precoder optimization design. The simulation results validate the effectiveness of the proposed framework, demonstrating the critical role of ECM estimation in mitigating MUI and providing robust SE gains in practical FDD-based massive MIMO scenarios.
A critical hindrance in realizing frequency division duplex (FDD) massive multi-input multi-output (MIMO) systems is the overhead associated with the downlink (DL) channel state information at the transmitter (CSIT) acquisition. To address this, we propose a novel framework that eliminates the need for CSI feedback, while achieving robust sum spectral efficiency (SE). Specifically, by leveraging partial frequency invariance of channel parameters, we reconstruct the DL CSIT using uplink (UL) pilots with the 2D-Newtonized orthogonal matching pursuit (2D-NOMP) algorithm. Due to discrepancies between the two disjoint bands, however, perfect DL CSIT acquisition is infeasible; resulting in multi-user interference (MUI). To account for this, we reformulate the sum SE maximization problem using the reconstructed channel and its error covariance matrix (ECM). Then, we propose an ECM estimation method based on the observed Fisher information matrix and introduce a precoder optimization technique with rate-splitting multiple access (RSMA). Our simulation results verify the validity of the proposed framework in the practical FDD massive MIMO scenarios, highlighting the essential role of ECM estimation in mitigating MUI to attain RSMA gains.