Embedded Pilot Channel Estimation

• Embedded pilot channel estimation is a method that embeds known pilot symbols within data transmissions to acquire accurate channel state information across diverse wireless channels.
• It optimizes pilot placement and employs algorithms like ML, LS, compressed sensing, and neural networks to balance estimation accuracy, pilot overhead, and spectral efficiency.
• This technique is essential in modern communication systems, including OFDM, OTFS, and massive MIMO, ensuring robust performance in high-mobility and short blocklength scenarios.

Embedded pilot channel estimation techniques exploit the insertion of known pilot symbols within data transmissions to facilitate accurate channel state acquisition, enabling efficient coherent detection or robust noncoherent decoding. By tightly embedding pilots or pilot sequences within the frame structure—often tailored to underlying channel sparsity, delay-Doppler diversity, or coding methodology—these methods strike critical trade-offs between pilot overhead, estimation accuracy, spectral efficiency, and complexity. Embedded pilot channel estimation underpins ultra-reliable communications in short blocklength, high-mobility, massive MIMO, and next-generation waveforms including OTFS, OFDM, AFDM, and FBMC.

1. Channel Models and Embedded Pilot Placement

Embedded pilot channel estimation is designed to address various physical channel models, including SISO Rayleigh block-fading, massive MIMO with parametric Doppler-delay profiles, filter bank multicarrier channels, and doubly selective (delay and Doppler) channels:

• In block-fading SISO scenarios, pilot symbols are inserted in each coherence block to estimate the fading coefficients. The input-output relation is typically $y_i = h_i x_i + n_i$ per block, with $x_i$ including $n_p$ pilots and $n_n-n_p$ data entries (Coşkun et al., 2019).
• MIMO and wideband OFDM settings employ structured pilot matrices $\mathbf{P}$ learned via data-driven or analytic algorithms, often subject to pilot power constraints and placement strategies to optimize Fisher information or Cramér-Rao lower bound (CRB) (Magoarou et al., 2020, Mashhadi et al., 2020).
• Delay-Doppler channels, as in OTFS or AFDM, arrange pilots as impulses or sequences in dedicated DD bins, frequently surrounded by zero-guard regions to mitigate pilot-data interference. For higher spectral efficiency, pilot placement is increasingly intertwined with compressed sensing (CS) or low-PAPR pilot sequence design (Raviteja et al., 2018, Aghda et al., 2023, Li et al., 2024, Wang et al., 2 Dec 2025, Pachigolla et al., 2024, Yin et al., 2022, Zheng et al., 2024).

2. Channel Estimation Algorithms and Performance Bounds

Channel estimation from embedded pilots is typically executed via one or more of the following approaches:

• Maximum Likelihood (ML) and Least Squares (LS): Estimators employ pilot observations to solve $$\hat{h}_i = (x_i^p)^{\mathrm{H}} y_i^p / \|x_i^p\|^2$$ or analogous linear models (Coşkun et al., 2019).
• Parametric and Variational Optimization: Deterministic channel models parameterized by $\theta$ are estimated by solving for the minimal pilot dimensionality $N_m \geq \lceil N_p/2 \rceil$, optimizing the pilot matrix $M$ to achieve CRB-minimizing performance under a Frobenius-norm power budget (Magoarou et al., 2020).
• Threshold or Energy-Based Extraction: In DD and DAFT domains, pilot-window observations are thresholded to declare tap existence and estimate gains, often balancing missed detections and false alarms via tuning of the decision threshold $\mathcal{T}$ (Raviteja et al., 2018, Yin et al., 2022).
• Noncoherent and GLRT-Based Methods: For short code blocklengths, pilots produce rough channel estimates for list-construction. Afterwards, candidate codewords are ranked by noncoherent metrics such as $\sum_i |\langle y_i, x_i \rangle|^2$ (GLRT), operating only in-list for tractability (Coşkun et al., 2019).
• Compressed Sensing: CS solvers such as orthogonal matching pursuit (OMP) recover sparse channel impulse responses using pilots as sensing vectors, with improved scalability in multi-user and high-dimensional delay-Doppler grids (Aghda et al., 2023, Wang et al., 2 Dec 2025, Pachigolla et al., 2024).
• Neural Network Estimators: End-to-end deep learning frameworks jointly learn pilot matrices, expansion/reduction layers, and CNN-based estimators, occasionally integrating attention modules for capturing long-range spatial/frequency correlations and enabling pruning of pilots for overhead reduction (Mashhadi et al., 2020).

3. Code and Frame Structure Integration

Modern embedded pilot schemes align pilot placement and channel estimation tightly with coding and decoding frameworks, enhancing reliability at low overhead:

• Short Blocklength List Decoding: Embedded pilots only yield rough channel estimates for initial decoding, permitting extremely low pilot overhead. Final decisions are made through list decoders coupled with noncoherent metrics (Coşkun et al., 2019).
• Interleaved and Split Pilot Structures: In delay-Doppler, splitting pilots or interleaving pilot positions (e.g., split-pilot in OTFS or interleaved pilots in FBMC) achieves lower guard requirements and enables interference cancellation over data regions, retaining effective channel estimation (Hosseiny et al., 2020, Li et al., 2024).
• Cyclic Shift Orthogonality and Pilot Sharing: In multi-user MIMO-OTFS, cyclic-shifted Zadoff-Chu sequences enable pilot sharing among users, leveraging orthogonality to drastically reduce pilot overhead, e.g., $>30\%$ savings, while spatial decomposition and CS localization yield accurate estimation (Wang et al., 2 Dec 2025).
• IM-based Pilot Index Modulation: In FTN signaling over HF channels, pilot sequence location itself embeds information bits, producing higher spectral efficiency with robust sequence identification via cross-correlation and auto-correlation metrics (Keykhosravi et al., 2024).

4. Overhead Minimization and Spectral Efficiency

Embedded pilot channel estimation drives aggressive pilot overhead minimization strategies across system paradigms:

• Guard Reduction and Superimposed Pilots: Methods such as split-pilot (Li et al., 2024) or superimposed pilot embedding in AFDM (Zheng et al., 2024) halve or eliminate traditional guard requirements, allowing overlapping of pilot regions and nearly 100% restoration of nominal spectral efficiency.
• Data-Driven Pilot Pruning: Neural network-based pilot pruning dynamically zeroes out subcarriers or antennas, focusing pilot energy where estimation accuracy benefits most, yielding graceful NMSE–overhead trade-offs (e.g., 25–30% pilot saving with <1 dB NMSE degradation) (Mashhadi et al., 2020).
• Spread Pilot and Symbol-based Arrangements: DVB and inter-vehicular OFDM deployments leverage Walsh–Hadamard spreading or pilot repositioning, reducing overhead and improving BER, notably under high Doppler mobility (0809.5189, Sassi et al., 2014).

5. Extensions to MIMO, Multi-User, and Future Directions

Embedded pilot channel estimation frameworks admit natural generalizations and ongoing innovations:

• MIMO and Multi-user Adaptation: Orthogonally shifted pilots, split pilot-pairs per TX antenna, and cyclic shift pilot overlays enable scaling to $N_t \times N_r$ systems and multiple simultaneous users without proportional increase in pilot consumption or collision, effectively supporting massive machine-type communications (Raviteja et al., 2018, Wang et al., 2 Dec 2025, Yin et al., 2022, Pachigolla et al., 2024).
• Sparse Recovery in Wideband and High Mobility: Joint spatial decomposition (e.g., AoA estimation, spatial projection using ESPRIT/MUSIC) and off-grid CS are critical for resolving fractional delay and Doppler paths in rich-multipath regimes (Wang et al., 2 Dec 2025).
• Low-PAPR and Complexity Considerations: Embedding pilots in zero-padded bins (ZP-OTFS), leveraging Zadoff-Chu sequences, and pilot spreading all contribute to low-PAPR and simpler hardware/algorithmic implementations at scale (Aghda et al., 2023, 0809.5189).
• Robustness and Security: IM-based pilot location modulation increases spectral efficiency “for free”, with robust sequence identification exploiting short, known channel tap structure (Keykhosravi et al., 2024). Phaseless pilot schemes (Walk et al., 2015) significantly depart from classical design, enabling pilot phase to carry user data or facilitate PAPR reduction.

6. Quantitative Performance and Trade-offs

Recent studies document concrete numerical improvements resulting from embedded pilot channel estimation:

Scheme / Paper	Pilot Overhead (%)	BER/BLER/SE Gain	Key Benefit
OSD+in-list GLRT (Coşkun et al., 2019)	Few pilots (e.g., $n_p=2$ per block)	$1.2$ dB BLER improvement @ $10^{-2}$ vs PAT	Near-RCUs finite blocklength channel estimation
NN + Pruning (Mashhadi et al., 2020)	Up to $30\%$ reduction	$<1$ dB NMSE loss	Frequency/antenna-aware pilot allocation
Cyclic-shift pilot OTFS (Wang et al., 2 Dec 2025)	$>30\%$ reduction	$3-5$ dB NMSE gain	Overlapping pilots via ZC sequence orthogonality
Split pilot OTFS (Li et al., 2024)	$\approx50\%$ reduction	Within $2$ dB of full-guard BER	Pilot self-interference cancellation & guard halving
Superimposed pilot AFDM (Zheng et al., 2024)	Guards eliminated	Full SE, iterative MSE/BER improvement	Iterative pilot-data interference suppression
IM–FTN pilot (Keykhosravi et al., 2024)	Location-encoded, minimal added overhead	$6$ dB MSE, $3.5$ dB BER gain vs baseline FTN	Free index-modulated SE increase, robust sequence identification
Spread pilot DVB (0809.5189)	$1/L$ (e.g., $L=16,32,64$)	+1–2.5 Mb/s useful bitrate increase over classical DVB	Block-wise precoding, flexible block design

Pilot overhead reduction can degrade estimation accuracy and error performance if not balanced against underlying channel sparsity, SNR, and coding structure. Techniques that jointly process pilots and moderate-complexity algorithms (e.g., iterative list decoding, message passing, CS, neural architectures) are effective at approaching theoretical bounds with moderate or low pilot load.

7. Future Research and Outstanding Problems

As physical layer systems evolve, embedded pilot channel estimation will increasingly intersect with sparse signal recovery, machine learning, noncoherent decoding, PAPR minimization, and security/SE optimization:

• Joint pilot-coding design: Tight integration with error-correcting codes, list decoders, and hybrid coherent/noncoherent schemes (Coşkun et al., 2019).
• Scalable multi-user and massive MIMO frameworks: Analytical pilot optimization under variation spaces, spatial frequency orthogonality, and parametric/sparse channel modeling (Magoarou et al., 2020, Wang et al., 2 Dec 2025).
• Spectral efficiency maximization: Superimposed pilot structures, split-pilot designs, and spread-pilot strategies in wideband, high-mobility, and low-latency settings.
• Low-complexity real-time implementation: Neural pilot reduction, auto-correlation and threshold extraction, practical zero-padded arrangements, and hardware-efficient spreading/interleaving.
• Robustness to channel uncertainty, asynchrony, and synchronization errors: Phaseless pilot schemes, dynamic pilot placement, and index-modulation in time-varying and uncertain environments.

Embedded pilot channel estimation remains a central mechanism for trading off pilot overhead, estimation fidelity, and complexity across a diverse range of modern communication architectures. Each incarnation, whether codebook-integrated, sparsity-exploiting, data-driven, or waveform-specific, advances the state-of-the-art in reliable, high-throughput wireless transmission.