---
title: Starlink Ku-Band OFDM Downlink
url: https://www.emergentmind.com/topics/starlink-ku-band-downlink-waveform
type: topic
---

# Starlink Ku-Band OFDM Downlink

The Starlink Ku-band downlink is a low-Earth-orbit satellite downlink in the 10.7–12.7 GHz band whose publicly characterized wideband form is an OFDM waveform with fixed frame timing, explicit synchronization sequences, and a large set of predictable symbols that can be exploited for passive positioning, navigation, and timing (PNT). Public reverse-engineering and receiver studies describe eight 240 MHz channels, 1024 subcarriers per channel, a 750 Hz frame rate, exact primary and secondary synchronization structures, dedicated edge pilots, and a broader deterministic frame template that supports matched-filter acquisition, Doppler tracking, and time-of-arrival (TOA) estimation [2210.11578][2602.02627]. Subsequent work formulates a “full beacon” model for recurring symbols and demonstrates Doppler-only PVT on captured Starlink signals, while separate work exploits a narrowband continuous-wave beacon tone for Doppler-rate aiding and explicitly does not disclose the wideband OFDM configuration [2602.06682][2605.20394].

## 1. Frequency plan, channelization, and basic physical layer

The blind-identification description of the downlink places the total allocated band at 10.700–12.700 GHz and resolves eight equally spaced OFDM channels, each 240 MHz wide and separated by 10 MHz guard bands. For channel index \(i=1,\dots,8\), the channel center frequencies are
\[
F_{c,i}
=
10.700\ \text{GHz}
\;+\; 0.25\,(i-1)\,\text{GHz}
\;+\; 0.1171875\ \text{MHz},
\]
so adjacent channel centers are 250 MHz apart and each channel is offset by half a subcarrier above the lower band edge. The OFDM subcarrier spacing is
\[
\Delta f=\frac{240\times10^6}{1024}=234{,}375\ \text{Hz}.
\]
These parameters define the baseline wideband Starlink Ku-band OFDM signal reported in the public signal-structure studies [2210.11578].

A later implementation-oriented waveform study adopts the same eight-channel, 240 MHz-wide view and uses \(s=1024\) subcarriers, again implying \(\Delta f \simeq 234.375\) kHz. In that formulation, each channel carries 302 OFDM symbols per frame plus an empty guard interval, and the symbol timing is consistent with the earlier blind-identification result [2602.02627].

One receiver study aimed at signal-of-opportunity navigation reports a different capture configuration rather than a different underlying standard: a nominal downlink carrier \(f_0 \simeq 11.325\) GHz for Starlink “channel 3” and an observed complex RF bandwidth of 100 MHz. In that experiment, the front end uses a feedhorn and an Othernet Bullseye TCXO LNB, with \(f_{\text{LO}}=9.75\) GHz and an IF near 1.575 GHz, followed by an Ettus USRP X300 sampling at \(F_s=100\) MS/s with 16-bit I/Q [2602.06682].

| Parameter | Value | Source context |
|---|---:|---|
| Total Ku-band downlink allocation | 10.700–12.700 GHz | Wideband OFDM characterization |
| Number of channels | 8 | Wideband OFDM characterization |
| Channel bandwidth | 240 MHz | Wideband OFDM characterization |
| Channel spacing | 250 MHz | Wideband OFDM characterization |
| Guard band | 10 MHz | Wideband OFDM characterization |
| Subcarriers per channel | 1024 | Wideband OFDM characterization |
| Subcarrier spacing | 234,375 Hz | Wideband OFDM characterization |
| Example capture carrier | \(f_0 \simeq 11.325\) GHz | 100 MHz capture setup |

The modulation observed on the wideband OFDM downlink is predominantly 4-QAM (QPSK), with 16-QAM reported for some payload symbols. With all subcarriers active and 4-QAM, the raw bit-rate expression reported in the blind-identification study is approximately 465.5 Mbps [2210.11578]. A plausible implication is that the publicly observed waveform has enough regularity to support both communications demodulation and opportunistic sensing, but the full proprietary higher-layer design remains undisclosed.

## 2. Frame timing and synchronization sequences

The publicly characterized OFDM waveform is frame-synchronous with frame period
\[
T_{\text{frame}}=\frac{1}{750}\ \text{s}\approx 1.3333\ \text{ms}.
\]
Each frame contains 302 consecutive nonzero OFDM-symbol intervals and one vacant frame-guard interval of approximately 4.444 \(\mu\)s, satisfying
\[
302\cdot 4.4\;\mu\text{s}+4.444\;\mu\text{s}=1.3333\;\text{ms}.
\]
The per-symbol structure uses \(N=1024\) subcarriers and \(N_{cp}=32\) cyclic-prefix samples, so the useful symbol interval is \(T=4.2667\ \mu\text{s}\), the cyclic-prefix interval is \(T_{\rm cp}=0.1333\ \mu\text{s}\), and the total OFDM-symbol interval is \(T_{\rm sym}=4.4000\ \mu\text{s}\) [2210.11578].

The frame indexing reported in the blind-identification study is highly specific. Symbol \(i=0\) is the Primary Synchronization Sequence (PSS), a time-domain sequence rather than an OFDM data symbol. Symbol \(i=1\) is the Secondary Synchronization Sequence (SSS), a single 4-QAM OFDM symbol. Symbols \(i=2\)–\(5\) are header or control symbols, symbols \(i=6\)–\(299\) are payload, symbol \(i=300\) is the coda-minus-one sync sequence (CM1SS), symbol \(i=301\) is the coda sync sequence (CSS), and \(i=302\) is the empty gap [2210.11578].

The PSS is defined with length \(N+N_{cp}=1056\) samples. Its 128-sample subsequence is differentially DPSK-encoded from a length-127 m-sequence generated by the primitive polynomial \(1 + D^3 + D^7\) with seed \((0,0,1,1,0,1,0)\), then repeated eight times; the first repetition and the 32-sample cyclic prefix are inverted in sign. The reported autocorrelation
\[
R_{pp}[\tau]
=
\sum_{n=0}^{1055} s_{\rm PSS}[n]\,s_{\rm PSS}^*[n-\tau]
\]
peaks at \(\tau=m(128+32)\), which encodes the repeated-subsequence structure [2210.11578]. In a later capture-oriented study, PSS autocorrelation is reported to exhibit clear periodic peaks at multiples of 8.8 \(\mu\)s, again confirming repeated subsequence structure in the observed signal [2602.06682].

The SSS is a full OFDM symbol with 4-QAM on all subcarriers except the four zeros at the channel center and edges, occupies symbol \(i=1\), and is identical for every satellite and frame. The CSS at \(i=301\) is reported as a full 4-QAM OFDM symbol rotated 90° relative to the SSS, while CM1SS at \(i=300\) fixes only a subset of QAM symbols [2210.11578]. Together, PSS, SSS, CM1SS, and CSS provide a rigid temporal scaffold on which later pilot and template-based processing builds.

## 3. Predictable symbols, edge pilots, and frame-wide determinism

A major extension beyond the initial synchronization-sequence disclosure is the characterization of predictable elements distributed throughout the frame. The most explicit of these are the “edge pilots,” which occupy subcarrier indices
\[
K=\{488,489,\dots,495;\ 528,529,\dots,535\}
\]
in every OFDM symbol slot \(i\in\{0,\dots,301\}\). These pilots are 4QAM and identical in every symbol, every frame, every beam, and every satellite. If \(\mathbf p=[p_0,\dots,p_{15}]^T\in\mathbb C^{16}\) is the pilot vector, then \(p_j=e^{j(\pi/2)(2q_j+1)/2}\) with \(q_j\in\{0,1,2,3\}\) fixed by a 150-hex-digit code listed in Appendix A of the paper, and
\[
\forall m,i,k\in K:\ X_{m,i,k}=p_k.
\]
This is a strong frame-invariant structure rather than a conventional sparse or beam-specific pilot pattern [2602.02627].

The same study identifies a frame-wide “reference template” \(T\in\mathbb C^{306\times301}\) over loaded non-pilot tones and symbol slots \(I=\{2,\dots,301\}\), excluding PSS and SSS. Its entries are the element-wise mode of hard-decoded QPSK symbols across “pure QPSK” frames:
\[
T_{i,k} = \arg\max_{c\in\{\pm1,\pm j\}} |\{m: \hat X_{m,i,k}=c\}|.
\]
This template is reported to be constant for all frames. After a short variable-length frame header, the demodulated QPSK symbols satisfy
\[
\hat X_{m,i,k}=D_{m,i,k}\,T_{i,k},
\]
where \(D_{m,i,k}\in\{\pm1\}\) is a BPSK “deviation.” The deviation field has a period 60 in subcarrier index and is circularly shifted by 16 tones per OFDM symbol increment:
\[
D_{m,i,k}=D_{m,i,k+60 \bmod |K|},\qquad
D_{m,i+1}=\operatorname{circshift}(D_{m,i},+16).
\]
The paper refers to these structured deviations as “T-codes” [2602.02627].

The practical consequence is a dense predictable set. Edge pilots, PSS, SSS, and T-coded QPSK symbols are reported to fill each frame sufficiently densely to yield up to approximately 69,500 known symbols per frame [2602.02627]. A separate lightweight pilot-estimation study expresses the same idea in a more compact receiver model: the deterministic part of each frame, repeated every \(T_{\rm fr}\) and common across satellites, is the “full beacon” \(b[n]\). It also states that, aside from PSS and SSS, there is one pilot every four subcarriers on approximately 62% of the OFDM resource elements [2602.06682]. This suggests that the predictable structure is not confined to a minimal synchronization layer but extends over a substantial fraction of the resource grid.

## 4. Receiver architectures and estimation methods

The matched-filter view begins with synchronization against known deterministic sequences. For the PSS, the blind-identification study gives the discrete correlator
\[
R[n] = \sum_{m=0}^{1055} y[n+m]\,s_{\rm PSS}^*[m],
\qquad
\hat n = \arg\max_n |R[n]|.
\]
Doppler is then acquired by coherent integration and FFT across the eight repeated PSS subsequences:
\[
C(\nu)=\sum_{m=0}^{1055} y[n+m]\,s_{\rm PSS}^*[m]\,e^{-j2\pi \nu t[m]},
\qquad
\hat\nu=\arg\max_\nu |C(\nu)|.
\]
Fractional-delay interpolation around \(\hat n\) refines the code phase to sub-sample resolution for meter-level pseudorange [2210.11578].

The pilot-centric acquisition framework developed later uses the edge pilots directly. Its coherent detection statistic is
\[
S_{\rm ep}(\tau,\beta)=\sum_{n=0}^{N_p-1} r[n+\tau]\,p[n \bmod 16]^*,
\]
with \((\hat\tau,\hat\beta)=\arg\max_{\tau,\beta}|S_{\rm ep}(\tau,\beta)|\). The full single-frame demodulation chain is reported as: coarse acquisition over Doppler/SFO hypotheses; frame extraction and Doppler/SFO compensation; segmentation into OFDM blocks with cyclic-prefix removal and FFT; channel equalization via the SSS to estimate \(H_k\) and frame gain \(Z_m\); per-symbol residual synchronization via maximum-likelihood phase and time estimation followed by a joint fit enforcing linear phase or time variation across symbol index; phase alignment; and constellation hard-decision on \(\hat X_{m,i,k}\) [2602.02627].

The lightweight SOP-navigation implementation uses a frame-by-frame beacon model after coarse time-frequency alignment:
\[
r_k[n] \approx b[n-d_k]\,e^{j\Theta_k[n]} + w_k[n],
\]
with
\[
\Theta_k[n]=\theta_k+\dot\theta_k\,nT_s+\tfrac12\,\ddot\theta_k\,(nT_s)^2.
\]
Here \(d_k\) is the code-phase offset, while \(\theta_k\), \(\dot\theta_k\), and \(\ddot\theta_k\) are the frame-start carrier phase and its derivatives. For a compensated frame \(r_k^c[n]\) and current beacon estimate \(\hat b[n]\), the correlation surface is
\[
Z(f,m)=\sum_{n=0}^{N_{\rm fr}-1}
\bigl(r_k^c[n]e^{-j2\pi f nT_s}\bigr)\,
\overline{\bigl(\widehat B(f)e^{j2\pi fn/N_{\rm fr}}\bigr)},
\]
and the residual frequency, phase, and delay are obtained as
\[
\Delta f_k,\Delta\Phi_k,\Delta d_k=\arg\max_{f,m}|Z(f,m)|.
\]
When the correlation magnitude \(c_k\) exceeds threshold, the beacon template is updated as
\[
\hat b_{k+1}[n]
=
\tfrac{k}{k+1}\,\bigl[\hat b_k[n-\Delta d_k]e^{j\Delta\Phi_k}\bigr]
+\tfrac{1}{k+1}\,r^c_k[n].
\]
The same study reports no additional resampling and performs baseband conversion, Doppler wipeoff, and FFT-based correlation at 100 MS/s; coherent phase tracking uses a Doppler Taylor expansion including first and second derivatives [2602.06682].

## 5. Navigation observables, processing gain, and measured performance

The disclosed waveform structures support both TOA-based and Doppler-based PNT observables. For TOA, the edge-pilot and full-frame predictable-symbol framework uses
\[
\hat\tau = \arg\max_\tau \left|\sum_{n=0}^{N-1} r[n]\,l[n-\tau]^*\right|
\]
as the correlation-peak estimator. The same work gives the high-SNR CRB for single-frame TOA as
\[
\sigma^2_{\rm CRB}=\frac{1}{8\pi^2\beta_2\cdot \text{SNR}},
\]
with \(\beta_2\) defined as the mean-square bandwidth, and notes that the Ziv–Zakai bound captures the low-SNR threshold regime [2602.02627].

The processing-gain argument is explicit. With
\[
L=\frac{\text{SNR}_{\rm post}}{\text{SNR}_{\rm pre}}=1+(N-1)|\mu|^2,
\]
fully known symbols satisfy \(|\mu|=1\), so \(L=N\). The theoretical maximum over a full frame is reported as \(N_{\max}\approx 302\cdot(1024+32)\approx 318{,}912\), corresponding to about 55 dB. Using only the exploitable predictable structure actually disclosed—PSS, SSS, edge pilots, and T-codes—gives \(\bar N\approx69{,}500\), so
\[
L\approx 1+(69{,}500-1)\approx69{,}500,
\qquad
G_p\simeq 10\log_{10}(69{,}500)\approx48.4\ \text{dB}.
\]
The same study reports that T-codes are recovered nearly error-free when SNR is at least \(-25\) dB, and that the processing gain extends acquisition to side beams whose pre-correlation SNR would be too low for PSS+SSS-only methods [2602.02627].

The practical SNR regime is correspondingly aggressive. A compact Ku-band feedhorn of diameter approximately 54 mm with gain about 15 dBi and beamwidth about 32° is reported together with an integrated LNB of noise figure about 1 dB. Pre-correlation SNR is stated as 1–8 dB for the assigned beam, with side-beam SNR reduced by 5–18 dB. PSS+SSS alone, with about 33 dB of gain, reportedly fails for side beams below about \(-17\) dB pre-correlation SNR, whereas pilots plus T-codes, with about 48 dB of gain, extend acquisition to at most approximately \(-25\) dB pre-correlation SNR [2602.02627].

For Doppler-only navigation, the lightweight beacon-estimation study defines the Doppler residual as
\[
\Delta f_k = f_{\rm estimated}-f_{\rm nominal},
\]
and embeds it into a Kalman-filter measurement step
\[
z_k = Hx_k + v,\qquad H=[0\ \ 1\ \ 0],\qquad z_k = 2\pi\,\Delta f_k.
\]
Using recurring symbols identified from Starlink satellites, it collects Doppler-shift measurements over a 600 s interval and computes a least-squares PVT solution. The reported performance is a positioning error of approximately 268 m after a post-fit refinement, with about 49% outlier rejection and initialization at the true site; without refinement or with only 300 s of data, the errors rise to much greater than 500 m. The same study notes that SNR, EVM, and BER are not explicitly reported, but beacon estimation was demonstrated at low SNR with a non-dish antenna [2602.06682].

## 6. Continuous-wave beacon studies, unresolved parameters, and scope boundaries

Not all Starlink PNT papers characterize the same waveform object. One recent navigation study focuses on “beacons” that are described not as wideband OFDM structure but as pure continuous-wave sine-wave tones in the Ku-band allocation 10.7–12.7 GHz. In that formulation, the beacon is a single continuous CW tone, transmitted without pulsing, with only one “central tone” per satellite exploited. The exact beacon-tone frequency offset, power level, and any real OFDM parameters are explicitly reported as not specified in the paper [2605.20394].

The observable in that CW-beacon framework is Doppler-rate. If \(r(t)=\|r_S(t)-r_U(t)\|\), then
\[
f_D(t)\approx-(f_c/c)\,\dot r(t),
\qquad
\alpha(t)\equiv \frac{d}{dt}f_D(t)\approx-(f_c/c)\,\ddot r(t),
\]
and the hardware extraction approximates
\[
\hat\alpha(t_i)=\frac{\hat f_D(t_{i+1})-\hat f_D(t_i)}{t_{i+1}-t_i}.
\]
The same paper defines an OFDM-derived range observable only as a simulation baseline and states that the number of subcarriers, subcarrier spacing, total bandwidth, symbol duration, cyclic prefix, guard bands, and modulation are all not specified for Starlink [2605.20394]. A recurring source of confusion is therefore the assumption that a successful Starlink PNT demonstration using a CW beacon tone has thereby identified the data-bearing OFDM waveform; the paper itself states the opposite.

Even within the wideband OFDM literature, some elements remain proprietary or only partially recovered. The blind-identification study did not recover an explicit public description of the FEC and states only that empirical data suggest strong coding such as LDPC or turbo with interleaving, while the exact rate and depth remain proprietary [2210.11578]. The lightweight beacon-estimation study likewise states that cyclic prefix and subcarrier spacing are not public in its formulation and that demodulation was achieved using known PSS, SSS, and OFDM parameters from prior work [2602.06682]. The public state of knowledge is therefore asymmetric: synchronization sequences, frame timing, edge pilots, and large deterministic symbol sets are characterized in detail, but the full proprietary communications stack is not.

Source: https://www.emergentmind.com/topics/starlink-ku-band-downlink-waveform