Pilot-Only Synchronizer

• Pilot-only synchronization is a method that leverages existing pilot signals to jointly estimate timing, carrier-frequency, and phase errors without extra overhead.
• It is applied across systems like distributed Massive MIMO OFDM, OTFS, PMCW radar-communication, URLLC, and optical CV-QKD to maintain efficiency under strict hardware and protocol constraints.
• The approach trades modest offline computational complexity for significant in-band overhead reduction and enhanced spectral efficiency in diverse communication settings.

A pilot-only synchronizer is a synchronization architecture in which known pilots, pilot blocks, or pilot tones are used to estimate timing, carrier-frequency, sampling-frequency, or phase impairments, often reusing structures already present for channel estimation or payload framing. In recent literature, the term spans distributed Massive MIMO and cell-free OFDM synchronization, CP-OTFS and OTFS timing/CFO recovery, PMCW-based joint radar-communication, pilot-assisted URLLC waveform channels, and optical pilot-tone local-oscillator synchronization in CV-QKD. Across these settings, the central objective is to extract synchronization information without introducing additional synchronization-specific overhead, while remaining compatible with practical waveform, hardware, and protocol constraints (Ryzhov et al., 2023, Bayat et al., 2023, Oliveira et al., 2023, Kislal et al., 2024, Sun et al., 2024, Sarmiento et al., 16 Sep 2025).

1. System-level motivation and operating regimes

Pilot-only synchronization arises when timing or frequency alignment must be recovered under stringent signaling budgets or implementation constraints. In distributed Massive MIMO or cell-free mesh architectures, many Access Points and User Equipments must share a common time reference so that joint multiuser transmissions produce the intended beamforming gains. The reported impairments include hardware-clock drift, residual Time Alignment Errors after 1588-based synchronization, and multipath- or two-way First Arrival Path-detection errors that produce overlapping OFDM symbols and Inter-Symbol Interference unless sufficiently large guard intervals are used. In that context, conventional point-to-point master-slave synchronization scales poorly because sequential exchanges consume one slot per node pair (Ryzhov et al., 2023).

In OTFS, pilot-only synchronization is motivated by the practical infeasibility of impulse pilots with large PAPR. One line of work uses a low-PAPR pilot structure with a cyclic prefix and exploits periodicity in both delay and time domains to estimate timing offset and CFO over a linear time-varying channel, using the same pilot already deployed for channel estimation. Another line embeds a single maximum-length sequence in the delay-time grid so that one-dimensional correlation and FFT processing can jointly recover the OTFS symbol boundary and sparse channel parameters (Bayat et al., 2023, Sun et al., 2024).

In PMCW-based joint radar-communication, pilot-only synchronization appears as a refinement stage after coarse preamble-based timing, CFO, and SFO acquisition. Regularly spaced pilot PRBS blocks are then used for residual timing and frequency tracking, while also supporting channel impulse-response estimation for equalization across repeated PRBS blocks (Oliveira et al., 2023).

In pilot-assisted URLLC, the same pilot symbols are used jointly for synchronization and channel estimation over a memoryless block-fading waveform channel. The critical distinction is whether block delays are fully dependent or independent across fading blocks: when the delays are common, pilot-only synchronization can be performed jointly across blocks; when they are independent, synchronization may dominate the error budget (Kislal et al., 2024).

In CV-QKD, the synchronization target is not packet timing alone but local-oscillator phase alignment. An optical or electrical pilot tone is extracted after coherent detection, and its instantaneous phase is used to compensate the quantum data. This formulation places pilot-only synchronization in direct contact with phase-noise estimation, excess-noise control, and hardware quantization limits (Sarmiento et al., 16 Sep 2025).

2. Pilot structures and sequence-design principles

The pilot structure is determined by the waveform constraints of each system. In distributed OFDM synchronization, the pilots are frequency-domain only: they are inserted in the allocated subcarriers of an OFDM symbol, with no direct access to time-domain shaping. The design requirements are a zero time-domain tail of length $T$, high peak-to-side-peak ratio in the ACF and MCF over a timing search window $[T_{\min},T_{\max}]$, minimal synchronization-slot count by multiplexing up to $N_{\rm pilots}$ pilots in one symbol, flexibility in $T_{\min}$, $T_{\max}$, and $T$, and an optional side constraint on PAPR. The zero-tail condition is enforced by constructing pilots in the null space of the bottom block of the IFFT matrix: if $W_{21}$ denotes the rows corresponding to the last $T$ time samples, then any pilot of the form $\mathbf y_{\rm FD}=V_0\mathbf x$ maps through the IFFT to a time-domain sequence whose last $T$ samples are exactly zero (Ryzhov et al., 2023).

In CP-OTFS, the pilot is a constant-amplitude Zadoff-Chu sequence of length $[T_{\min},T_{\max}]$0 placed on one Doppler index, with its last $[T_{\min},T_{\max}]$1 samples prepended as a cyclic prefix on neighboring delay bins in the same Doppler column. The design objective is practical low-PAPR operation while retaining the periodicity needed for timing-offset estimation in both delay and time dimensions (Bayat et al., 2023).

In the MLS-based OTFS JTSCE formulation, the pilot is a bipolar MLS of length $[T_{\min},T_{\max}]$2 scaled to $[T_{\min},T_{\max}]$3, followed by an appended zero at $[T_{\min},T_{\max}]$4. Its normalized aperiodic autocorrelation satisfies

$[T_{\min},T_{\max}]$5

and the corresponding $[T_{\min},T_{\max}]$6-point DFT is placed on one OTFS row, surrounded by a guard of at least $[T_{\min},T_{\max}]$7 zeros. The appended zero and the guard region are integral to the timing and delay-tap identification procedure (Sun et al., 2024).

In PMCW radar-communication, the relevant pilot objects are pseudorandom binary sequences generated by an $[T_{\min},T_{\max}]$8-stage LFSR with $[T_{\min},T_{\max}]$9 and feedback equation

$N_{\rm pilots}$0

After BPSK mapping, the resulting m-sequence has ideal two-level periodic autocorrelation,

$N_{\rm pilots}$1

which is exploited both in the preamble and in the pilot-only tracking blocks (Oliveira et al., 2023).

In optical pilot-tone synchronization for CV-QKD, the pilot can be generated electrically or optically through the IQ-modulator bias choice. The transmitted field is modeled as

$N_{\rm pilots}$2

where the pilot-tone power $N_{\rm pilots}$3 directly controls the pilot-extraction SNR, and therefore the synchronization-error variance (Sarmiento et al., 16 Sep 2025).

3. Estimation algorithms

The estimator architecture is as heterogeneous as the pilot design. In distributed Massive MIMO synchronization, the pilot set is synthesized offline by minimizing a worst-case cost over shift indices in $N_{\rm pilots}$4. The objective combines an autocorrelation term, a mutual-correlation term across distinct pilots, and optionally a PAPR term, in a minimax problem of the form

$N_{\rm pilots}$5

An iterative gradient-descent algorithm updates each pilot on the basis of the worst offending shift, while preserving the zero-tail constraint because all iterates remain in the column space of $N_{\rm pilots}$6. At the receiver, synchronization reduces to correlating the incoming OFDM symbol with each known pilot pattern and locating the maximum correlation peak within the search interval (Ryzhov et al., 2023).

In PCP-based CP-OTFS, timing-offset estimation is explicitly separated into a delay-domain part and a time-domain part. The delay metric is

$N_{\rm pilots}$7

and the time metric is

$N_{\rm pilots}$8

After TO compensation, CFO estimation is two-stage: a coarse angular estimate is obtained from periodic pilot correlations, and a refined estimate is produced by maximizing an ML criterion after approximating the time-varying channel with a generalized complex exponential basis expansion model. The low-complexity implementation rewrites the CFO metric using a structured coefficient sequence $N_{\rm pilots}$9, reducing the per-grid-point cost from $T_{\min}$0 to $T_{\min}$1 (Bayat et al., 2023).

In the pilot-assisted URLLC formulation, the pilot-only synchronizer is an ML estimator over the upsampled, matched-filtered pilot observations. For each block,

$T_{\min}$2

with

$T_{\min}$3

The optimization is reduced to maximizing

$T_{\min}$4

and in the synchronous-reception case the same $T_{\min}$5 is shared across all blocks. This produces a genuine joint synchronization-and-channel-estimation estimator, rather than a cascade of separate timing and channel stages (Kislal et al., 2024).

In MLS-based OTFS JTSCE, the estimator performs a one-dimensional scan over trial indices $T_{\min}$6. For each trial, it extracts a row-wise sequence $T_{\min}$7, multiplies it by the local MLS, computes an $T_{\min}$8-point DFT, and forms the normalized peak metric

$T_{\min}$9

The first threshold crossing identifies the timing offset, after which additional crossings reveal integer-delay taps. Doppler and complex gains are then recovered in closed form from phase increments and pilot-compensated accumulation (Sun et al., 2024).

In PMCW radar-communication, the coarse stage is not pilot-only: it begins with Schmidl-Cox timing and CFO estimation over an SC preamble, using

$T_{\max}$0

After CFO/SFO correction, pilot-only refinement correlates the known PRBS with the received pilot block,

$T_{\max}$1

normalized by $T_{\max}$2, and declares frame alignment when $T_{\max}$3 (Oliveira et al., 2023).

In CV-QKD, phase synchronization is obtained directly from the pilot tone. The one-symbol phase-error variance is modeled as

$T_{\max}$4

and with an $T_{\max}$5-sample moving average,

$T_{\max}$6

The estimated phase is then used to rotate the quantum data prior to heterodyne demodulation (Sarmiento et al., 16 Sep 2025).

4. Reported performance and computational characteristics

The published results show that pilot-only synchronization can operate effectively across markedly different regimes, but the reported gains are system-specific.

Distributed Massive MIMO OFDM: Using $T_{\max}$7 kHz subcarrier spacing over a $T_{\max}$8 MHz bandwidth with $T_{\max}$9, $T$0, and a zero-tail length of $T$1s, pilot sets with $T$2 were synthesized. The minimax gradient descent typically converged in fewer than $T$3 iterations. With $T$4 pilots, the worst-case main-to-side-peak MCF was $T$5 dB and the ACF peak ratio was $T$6 dB; after a realistic urban multipath channel, the worst-case MCF fell to $T$7 dB. The reported synchronization-slot reductions were $T$8 for $T$9, $W_{21}$0 for $W_{21}$1, and $W_{21}$2 for $W_{21}$3, with one OFDM symbol synchronizing $W_{21}$4 pairs. With subcarrier spacing $W_{21}$5 kHz and FFT length $W_{21}$6, the sample period is approximately $W_{21}$7 ns, implying timing resolution well below $W_{21}$8s (Ryzhov et al., 2023).

PCP-based CP-OTFS: The per-block complexities are $W_{21}$9 for the delay metric, $T$0 for the time metric, $T$1 for coarse CFO, and approximately $T$2 for the fine CFO ML search over $T$3 grid points. Simulations over a 3GPP EVA channel with $T$4 taps, $T$5-QAM, $T$6 ns, and normalized Doppler up to $T$7 kHz showed lower TO bias and variance than the impulse-pilot method of the cited prior work, and an order-of-magnitude lower CFO-estimation MSE than the impulse-pilot coarse estimator, while retaining zero extra overhead because the same pilot supports both synchronization and channel estimation (Bayat et al., 2023).

PMCW radar-communication: On a $T$8 GS/s ZCU111 RFSoC proof-of-concept platform, coarse CFO estimates over $T$9 runs at SNR $\mathbf y_{\rm FD}=V_0\mathbf x$0 dB had mean $\mathbf y_{\rm FD}=V_0\mathbf x$1 of $\mathbf y_{\rm FD}=V_0\mathbf x$2 kHz for $\mathbf y_{\rm FD}=V_0\mathbf x$3, $\mathbf y_{\rm FD}=V_0\mathbf x$4 kHz for $\mathbf y_{\rm FD}=V_0\mathbf x$5, $\mathbf y_{\rm FD}=V_0\mathbf x$6 kHz for $\mathbf y_{\rm FD}=V_0\mathbf x$7, and $\mathbf y_{\rm FD}=V_0\mathbf x$8 kHz for $\mathbf y_{\rm FD}=V_0\mathbf x$9. After SC-based coarse correction and Tsai SFO compensation, pilot-only tracking yielded refined CFO estimates of $T$0 kHz, $T$1 kHz, $T$2 kHz, and $T$3 kHz, respectively, with the last case attributed to ambiguity due to pilot spacing below $T$4. Timing estimation error from the coarse stage was within $T$5 sample in $T$6 of runs for $T$7, and pilot-only $T$8-refinement was accurate to $T$9 sample (Oliveira et al., 2023).

Pilot-assisted URLLC: Joint synchronization achieves the Cramér-Rao bound at moderate SNR, whereas per-block synchronization is several dB poorer. For packet-error probability below $[T_{\min},T_{\max}]$00, the required condition is $[T_{\min},T_{\max}]$01. Most of the gain from increasing the upsampling factor is achieved by $[T_{\min},T_{\max}]$02. For $[T_{\min},T_{\max}]$03 and target error probability $[T_{\min},T_{\max}]$04, joint synchronization operates within $[T_{\min},T_{\max}]$05 dB of perfect synchronization with pilot-aided channel estimation, whereas per-block synchronization incurs an extra approximately $[T_{\min},T_{\max}]$06 dB. The optimum pilot counts for joint and perfect synchronization coincide; examples include $[T_{\min},T_{\max}]$07 and $[T_{\min},T_{\max}]$08 (Kislal et al., 2024).

MLS-based OTFS JTSCE: With carrier $[T_{\min},T_{\max}]$09 GHz, subcarrier spacing $[T_{\min},T_{\max}]$10 kHz, $[T_{\min},T_{\max}]$11, $[T_{\min},T_{\max}]$12, $[T_{\min},T_{\max}]$13, and $[T_{\min},T_{\max}]$14-QAM, TO and delay detection accuracy exceeded $[T_{\min},T_{\max}]$15 once the MLS SNR reached $[T_{\min},T_{\max}]$16 dB, independently of data SNR. Off-grid Doppler and gain MSE dropped to below $[T_{\min},T_{\max}]$17 at MLS SNR $[T_{\min},T_{\max}]$18 dB. In BER, the JTSCE-aided receiver was within approximately $[T_{\min},T_{\max}]$19 dB of the perfect-sync-plus-CSI curve and outperformed the embedded-pilot benchmark by $[T_{\min},T_{\max}]$20-$[T_{\min},T_{\max}]$21 dB in the $[T_{\min},T_{\max}]$22 to $[T_{\min},T_{\max}]$23 region (Sun et al., 2024).

Optical pilot-tone CV-QKD: Comparative link-distance results were reported as a function of pilot-to-data ratio $[T_{\min},T_{\max}]$24, DAC resolution, and pilot-generation method. At $[T_{\min},T_{\max}]$25 dB and $[T_{\min},T_{\max}]$26 bits, both methods reached $[T_{\min},T_{\max}]$27 km; at $[T_{\min},T_{\max}]$28 dB and $[T_{\min},T_{\max}]$29 bits, both reached $[T_{\min},T_{\max}]$30 km, while a $[T_{\min},T_{\max}]$31-bit electrical pilot reached $[T_{\min},T_{\max}]$32 km and a $[T_{\min},T_{\max}]$33-bit optical pilot still reached $[T_{\min},T_{\max}]$34 km; at $[T_{\min},T_{\max}]$35 dB and $[T_{\min},T_{\max}]$36 bits, both reached $[T_{\min},T_{\max}]$37 km, while a $[T_{\min},T_{\max}]$38-bit optical pilot also reached $[T_{\min},T_{\max}]$39 km and the $[T_{\min},T_{\max}]$40-bit electrical pilot had $[T_{\min},T_{\max}]$41. These comparisons isolate the effect of DAC quantization on pilot extraction (Sarmiento et al., 16 Sep 2025).

5. Overhead reduction, protocol compatibility, and trade-offs

A defining advantage of pilot-only synchronization is its tendency to avoid dedicated synchronization overhead. In distributed OFDM synchronization, the pilots are carried in-band on allocated subcarriers exactly like existing 3GPP physical-channel reference signals, so no change is required to the 3GPP protocol stack. Because the synthesized pilots have exactly $[T_{\min},T_{\max}]$42 zeros at the end of the symbol, no additional zero-tail or extended cyclic prefix is required beyond the standard 3GPP CP, and the approach solves the guard-interval separation problem without reducing data-payload efficiency. The stated trade-off is modest offline computational effort to generate the pilot set in exchange for large online savings in synchronization-slot consumption and spectral efficiency (Ryzhov et al., 2023).

In PCP-based OTFS and in pilot-assisted URLLC, the same pilot used for channel estimation is leveraged for synchronization, so no extra pilot overhead is incurred. In URLLC this creates a pilot/data trade-off rather than a purely synchronization-specific overhead problem: adding pilots improves both timing and channel estimates but reduces the number of data symbols, so the optimal pilot count must be chosen jointly with the finite-blocklength error criterion (Bayat et al., 2023, Kislal et al., 2024).

PMCW radar-communication makes the overhead trade-off explicit. The SC+SFO preamble costs approximately $[T_{\min},T_{\max}]$43 ms, pilots cost $[T_{\min},T_{\max}]$44 symbols, and the remaining blocks carry data. Residual CFO directly affects MER and BER, with MER degraded by approximately $[T_{\min},T_{\max}]$45-$[T_{\min},T_{\max}]$46 dB if residual CFO exceeds $[T_{\min},T_{\max}]$47 Hz. This shows that pilot-only refinement is not merely a convenience layer: it materially influences communication quality after coarse synchronization has been acquired (Oliveira et al., 2023).

In optical pilot-tone synchronization, the trade-off is shifted to hardware dynamic range and phase-noise suppression. Optical pilot generation at quadrature bias does not burden the DAC’s dynamic range, whereas coarse DAC resolution corrupts electrical pilots through quantization replicas. The recommended operating regime is $[T_{\min},T_{\max}]$48 dB for positive SKR under realistic linewidths, $[T_{\min},T_{\max}]$49-$[T_{\min},T_{\max}]$50 dB for long links beyond $[T_{\min},T_{\max}]$51 km, and moving-average length around $[T_{\min},T_{\max}]$52 samples provided the averaging window does not exceed the pilot coherence time. This suggests that, in coherent optical systems, pilot-only synchronization is inseparable from front-end hardware design (Sarmiento et al., 16 Sep 2025).

6. Scope, limitations, and recurrent misconceptions

A recurrent misconception is that pilot-only synchronization necessarily means the absence of any dedicated synchronization structure. The PMCW case is an explicit counterexample: Schmidl-Cox and Tsai preambles are used for coarse timing, CFO, and SFO, and only afterward are pilot blocks used for refinement and drift tracking. In that setting, “pilot-only” refers to the refinement stage, not to the entire acquisition chain (Oliveira et al., 2023).

A second misconception is that pilot-only methods are always orthogonal or time-domain shaped. The distributed Massive MIMO construction is explicitly non-orthogonal, frequency-domain multiplexed, and constrained by the requirement that pilots be inserted only on allocated subcarriers. Its distinctiveness lies precisely in synthesizing distinguishable simultaneous pilots despite the lack of direct time-domain control (Ryzhov et al., 2023).

A third misconception is that pilot-only synchronization is universally sufficient. The URLLC analysis shows that sufficiency depends critically on the delay model. When fading blocks share one unknown delay, pilot-only joint ML synchronization reaches CRB performance at URLLC SNRs with no extra pilots. When each block has its own delay, synchronization becomes the bottleneck, and the reported remedies are increased pilot overhead, longer or more structured synchronization sequences such as Zadoff-Chu, or external timing references such as GPS or wired synchronization (Kislal et al., 2024).

A final practical limitation concerns feasibility rather than estimator form. In OTFS, the use of a low-PAPR PCP pilot is motivated by the impracticality of impulse pilots with large PAPR. In CV-QKD, electrical pilot generation becomes problematic at high pilot-to-data ratios and low DAC resolution, whereas optical pilot generation remains viable under the same conditions. A plausible implication is that future pilot-only synchronizers will continue to be evaluated less by estimator optimality in isolation than by how well their pilot structures coexist with waveform standards, analog front-end constraints, and channel-estimation requirements (Bayat et al., 2023, Sarmiento et al., 16 Sep 2025).