{"id":"437918f7-3751-4047-adf8-59889488fbb6","arxiv_id":"2502.09118","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":3.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"OTFS and its variants OSDM, V-OFDM, OTSM, ODDM, OCDM, and AFDM can be described by one shared linear model, with OSDM and V-OFDM being exact special cases of OTFS under rectangular pulses and Nyquist sampling.","lead":"This survey shows that several next-generation wireless modulation schemes, including OTFS, Vector OFDM, and OSDM, are mathematically the same transmission format when standard pulse shapes are used. It organizes the crowded OTFS-variant literature into one common signal model and compares performance in fast-moving wireless channels.","discovery_kind":"review","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The Section V circular-channel model is only approximate under the high-Doppler conditions the paper targets; detector comparisons and unified-matrix visualizations inherit that approximation.","rationale":"I read the paper as a survey whose central positive claim is the unification of OTFS variants under one linear model and the equivalence of V-OFDM and OSDM with rectangular-pulse OTFS. These algebraic identities are traceable to the cited primary literature and I do not dispute them. The place the claim becomes load-bearing is when the unified model is instantiated as the circular Hch of Eq. (45) and then used for BER simulations, effective-matrix visualizations, and detector comparisons. That instantiation is only approximate for time-varying channels, and the approximation error grows with Doppler and block duration; the reader's weakest assumption identifies the same issue. The missing OFDM baseline and unreleased simulation code are real but secondary, affecting the 'benefit' narrative rather than the unification itself. I therefore keep the reader's CONDITIONAL verdict unchanged, while sharpening the concrete test that would settle whether the circularity approximation actually changes the conclusions.","tokens_in":42280,"tokens_out":11422,"duration_ms":125481,"concrete_test":"Generate the true received vector for the Section IV-B setup by sample-by-sample linear time-varying convolution, with the same FIR transmit and receive filters, three off-grid paths, delays tau_i in [0, 4T/M], and Doppler nu_i = nu_max cos(theta_i), using a transmitted frame plus CP but without the modulo-MN wrapping of Eq. (42). Reconstruct the true linear effective map H_true by injecting basis vectors, and compute the normalized Frobenius error ||H_true - H_eq(45)||_F / ||H_true||_F. Then rerun the Fig. 12 and Fig. 20 BER comparisons using H_true with the OAMP detector; if the OTFS-REC/ODDM ranking or the detector ordering shifts by a non-negligible margin, the circular unified model is not an adequate basis for the comparative claims.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central unification rests on the discrete channel model in Section V-A. Equation (42) writes r[c] = sum_{p=0}^{P-1} h[c,p] s[[c-p]_{MN}] + n[c], and Eq. (45) presents Hch in circular shift form. With a single block-level CP, modulo-MN indexing is exact for an LTI channel whose delay spread fits in the CP. It is not exact for a time-varying channel: for c < P, the wrap replaces physically preceding transmitted samples (which experienced different Doppler phases) with the current block's tail. The time dependence in h[c,p] of Eq. (40) does not repair this; it only assigns per-sample gains. At the paper's own parameters (M=64, N=16, Delta f=15 kHz, block duration about 1.07 ms) and 300-500 km/h mobility, the Doppler phase rotation across the block is several radians, so the wrap error is not negligible. Sections IV-B and VI-D generate all BER curves, and Table V all effective-channel visualizations, from this circular model. If the approximation biases the structure of H, the detector comparisons and the claimed unified detector architecture are validated only for the approximate model, not for the physical high-mobility channel targeted by the survey. This is the weakest load-bearing point: the algebraic identities identifying V-OFDM and OSDM with rectangular-pulse OTFS are independent of the channel model and are well supported, but the unifying receive-model claim and all comparative conclusions depend on Eqs. (42)-(45).","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This manuscript is a survey/tutorial on OTFS and related waveforms. It presents OTFS as an ISFFT-precoded FBMC (Section II-C), shows that vector OFDM and OSDM are algebraically identical to rectangular-pulse OTFS (Sections III-A/B), describes OTSM and ODDM as related delay-domain schemes, and collects chirp-based OCDM and AFDM in an appendix. The paper then introduces a doubly-selective channel model with per-tap time-varying gains (Section V-A), expresses each modulation as y = Hx + w (Table IV), reviews linear, nonlinear, and deep-learning detectors under this model, and provides PSD, BER, and detector-comparison experiments. The central claim is that these variants share a common mathematical structure and can be studied and detected in a unified framework.","tokens_in":42548,"tokens_out":20158,"duration_ms":208683,"significance":"If the unification is taken as an expository claim, it is largely sound and useful. The matrix identities in Sections II-III are standard and correctly presented, and the paper provides a convenient reference for the relationships among OTFS, V-OFDM, OSDM, OTSM, ODDM, OCDM, and AFDM, with a useful complexity comparison in Table VI. I also checked the circular-convolution formulation in Eqs. (42)-(45): with a CP longer than the maximum delay spread, the modulo-MN indexing is exact even for time-varying channels, because the CP contains copies of the same tail samples; the time variation is captured in h[c,p]. The concern that wrap-around under Doppler invalidates the model therefore does not land, provided the stated CP condition holds. The main limitations are scope and presentation: the numerical comparisons do not include an OFDM baseline, and several comparative tables use qualitative entries without definitions; these do not invalidate the central unification.","major_comments":[],"minor_comments":[{"comment":"The BER experiments compare only OTFS-REC, OTFS-SRRC, and ODDM, while the abstract and introduction motivate these schemes by their advantage over OFDM; please either add an OFDM baseline or explicitly state that the comparison is among OTFS variants and that the advantage over OFDM is taken from the cited literature.","section":"Section IV-B, Figs. 12-13"},{"comment":"The qualitative ratings (e.g., 'Sensitive', 'Robust', 'Strong', 'Good', 'Bad') have no definitions or supporting references; since the table presents comparative performance claims, please add a legend, explicit criteria, and citations for each attribute.","section":"Table III"},{"comment":"The continuous-time ODDM expression in Eq. (32) uses the phase e^{j2π nℓ/N}, while the correct expression in Eq. (35) contains the time-dependent phase e^{j2π n(t-mT/M)/(NT)}; please reconcile these two forms or explain that Eq. (32) is a simplified sampled form valid under the Nyquist pulse assumption.","section":"Section III-D, Eqs. (32) and (35)"},{"comment":"The symbol N is used both for the number of temporal slots (M=64, N=16 in the experiment) and for the number of data symbols in Eq. (46), which makes the complexity entries ambiguous; for example, the LMMSE complexity O(N^3) should be stated in terms of the data-vector dimension, e.g., O((MN)^3) or equivalent.","section":"Section VI-D and Table VI"},{"comment":"Before Eq. (42), the assumptions behind the modulo-MN indexing should be stated explicitly (CP length at least the maximum delay spread, no inter-block interference, and the receive filter response absorbed into h[c,p]); the current one-sentence justification may lead readers to believe the circular model itself is an approximation for Doppler channels.","section":"Section V-A, Eqs. (42)-(45)"},{"comment":"The index expression 'Pi,m+nM' appears garbled, and the definition Ns = M(N+11) is unexplained; please correct the notation and define Ns in terms of the FBMC prototype filter length.","section":"Section II-B, Eq. (13)"}],"recommendation":"minor_revision","confidential_remarks":"This is a competent survey with a sound but mostly expository central message. The equivalence of V-OFDM and OSDM with rectangular-pulse OTFS is not new, and the 'unified receiver architecture' is essentially the linear model y = Hx + w; the contribution lies in the consolidation and comparison, which is appropriate for a surveys journal. The main risk is overclaiming novelty, so I would ask the authors to temper the 'common receiver architecture' wording and to add the OFDM baseline or qualify the abstract accordingly."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Two things to know. First, this is a real service to the subfield: it explicitly writes down the modulation/demodulation equations for OFDM, FBMC, OTFS, OSDM, V-OFDM, OTSM, ODDM, and (in the appendix) OCDM/AFDM, and shows that OSDM and V-OFDM are exactly rectangular-pulse OTFS. That equivalence was in the primary sources, but having it in one place with a side-by-side I/O table and effective channel matrix pictures is genuinely useful. Second, the paper's own numerical evidence is weaker than its polished narrative: no OFDM baseline, no code/data, and a channel model that is only approximate in exactly the high-Doppler regime the survey targets.\n\nThe good parts first. The modulation algebra checks out; I sampled the key identities (Eqs. (19), (24), (26), (44)) and they are consistent with the cited literature. The detector survey (Section VI) is a fair, wide-ranging taxonomy, and the complexity table is honest about OAMP/VAMP's cost. The authors also do not overclaim the OTSM/ODDM relationship—they say those are not identical to OTFS, just same principle. Credit is given to Xia, Ebihara, Lin, Hadani, Raviteja, etc., so there is no circularity burden.\n\nNow the soft spots, in proportion.\n\nThe Section V channel model is the one I'd want fixed before this becomes an authoritative reference. Equation (42) writes the received signal as a circular convolution with modulo-MN indexing. That is exact for an LTI channel whose delay spread fits in the CP, but for a time-varying channel the wrap substitutes the current block's tail for what were physically preceding samples that saw different Doppler phases. The paper's own parameters (M=64, N=16, 15 kHz spacing, 300 km/h) put the Doppler rotation across the block at several radians, so the error is not negligible. Every BER curve and every effective channel matrix in Table V inherits this approximation. The algebra that identifies V-OFDM/OSDM with OTFS does not depend on this model, so the core unification claim survives; but the 'unified receive model' and the detector comparisons are validated only for the approximate model, not for the physical channel the survey is about. The authors should state this caveat explicitly and, ideally, run one validation against a non-circular channel simulator.\n\nMissing OFDM baseline is a separate, smaller issue. The abstract promises to underscore the benefit against OFDM's Doppler sensitivity, but Figs. 12, 13, 20 compare only OTFS variants among themselves. That's fine for comparing pulses, but it doesn't support the superiority narrative.\n\nTable III has qualitative entries (e.g., anti-jamming) with no backing; that is a minor but careless detail for a survey that otherwise cares about precision.\n\nWho is this for? A graduate student or researcher entering OTFS who wants to know which waveform is which, and anyone designing a receiver who wants to know which detectors are worth trying. It deserves a serious referee; I would send it out with a request for a revision that addresses the channel-model caveat and adds an OFDM baseline to the BER runs. As is, it is a competent survey, not a definitive one.","headline":"A useful survey that correctly assembles the OTFS equivalence story, but the unified channel model and missing OFDM baselines keep it from being the authoritative comparison it aims to be.","tokens_in":43118,"tokens_out":3349,"would_cite":false,"duration_ms":34344,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper argues that OTFS is a precoded FBMC scheme, that V-OFDM and OSDM are special cases of it under rectangular pulses and Nyquist sampling, and that all major OTFS variants share one linear receiver model under doubly-selective…","keywords":["OTFS","delay-Doppler domain","inverse symplectic finite Fourier transform","doubly-selective fading","unified linear receiver model","high-mobility wireless","OSDM","vector OFDM"],"falsifier":"Run the same data block through V-OFDM, OSDM, and rectangular-pulse OTFS on one physical doubly-selective channel with Nyquist sampling and measure the effective channel matrices: the paper's equivalence predicts they coincide exactly, so any material discrepancy under sufficient CP length would falsify the unification, as would a demonstration that a CP shorter than the delay spread still yields identical outputs.","tokens_in":42045,"feed_emoji":"📡","tokens_out":7875,"duration_ms":77039,"temperature":0.7,"pith_summary":"This survey tries to dispel the confusion created by the many named variants of orthogonal time frequency space (OTFS) modulation by showing that they are largely the same idea expressed through different transforms. It presents OTFS as a precoded filter-bank multicarrier (FBMC) scheme whose precoder is the inverse symplectic finite Fourier transform (ISFFT). With rectangular transceiver pulses and Nyquist-rate sampling, vector OFDM (V-OFDM) and orthogonal signal-division multiplexing (OSDM) reduce to exactly the same transmit and receive expressions as OTFS. The paper then derives one effective channel model for doubly-selective fading and shows that OTFS, V-OFDM, OSDM, OTSM, ODDM, OCDM, and AFDM all fit the same linear input-output form $y = Hx + w$. If the unification is right, detectors and performance comparisons developed for one variant transfer to all the others.","feed_headline":"OTFS, V-OFDM, OSDM collapse into one modulation","feed_subtitle":"With rectangular pulses and Nyquist-rate sampling, all three share the same transmit and receive math.","key_machinery":"The load-bearing object is the ISFFT precoder together with the Heisenberg transform, whose discrete matrix form for OTFS is $s = (F_N^H \\otimes G)x$; with rectangular pulses the pulse matrix $G$ becomes the identity, so the expression collapses to the $(F_N^H \\otimes I_M)$ block transform shared by OSDM and V-OFDM. Around that transform pair the paper builds the unified channel machinery: the sampled doubly-selective channel is written as a time-dependent circular convolution $h[c,p]$ over modulo-$MN$ indices, producing an equivalent channel matrix $H_{\\mathrm{ch}}$, and every modulation's effective matrix has the same sandwich structure $H = T_{\\mathrm{rx}} H_{\\mathrm{ch}} T_{\\mathrm{tx}}$ with its own transmit and receive transforms. This gives all variants the common form $y = Hx + w$ and lets detector complexity be analyzed uniformly through the sparsity or bandedness of $H$.","core_discovery":"The central claim is that OTFS is not a separate waveform family but a precoding of FBMC: applying ISFFT to delay-Doppler symbols before the Heisenberg transform is exactly what distinguishes it from OFDM. Once the transmit and receive pulses are rectangular and sampling is at the Nyquist rate, the discrete OTFS signal becomes $s = (F_N^H \\otimes I_M)x$, which is the same expression as OSDM and equivalent to V-OFDM's block transform $S = XF_N^H$; hence the paper treats V-OFDM and OSDM as special cases of OTFS. The paper further posits a sampled doubly-selective channel model $h[c,p]$ in which the received block is a circular convolution with modulo-$MN$ indexing, and from it obtains each variant's effective channel matrix by sandwiching $H_{\\mathrm{ch}}$ between its modulation and demodulation transforms. The result is a single linear model $y = Hx + w$ covering all discussed variants, which is the basis for the receiver overview and complexity comparisons in the later sections.","pith_inferences":["If the equivalence is taken at face value, several published claims of 'new waveform' gains are likely reparameterizations: the actual levers are pulse shape, transform choice, and detector, so a fair comparison should hold those fixed across variants.","A testable consequence is that one software-defined radio chain with switchable unitary transforms (DFT, Walsh-Hadamard, Fresnel, affine Fourier) could implement all the variants in the unified model, lowering deployment cost in high-mobility systems.","The circular-convolution assumption is the brittle edge: in very wideband channels with frequency-dependent Doppler (Doppler squint), the unified model will need the delay-scale-space extension the paper lists as an open direction, so the unification is strongest for narrowband doubly-selective channels."],"forward_implications":["A detector built for OTFS with rectangular pulses applies unchanged to V-OFDM and OSDM, since their effective channel matrices coincide.","Apparent waveform advantages in PSD or BER are mostly pulse-shaping and filtering effects: the paper's simulations show OTFS-SRRC and ODDM nearly match each other, while OTFS-REC suffers from higher out-of-band emission and worse BER under narrowband filtering.","The OTSM, ODDM, OCDM, and AFDM variants can be served by the same receiver algorithms (linear, message-passing, memory AMP, and cross-domain iterative detectors) because all fit $y = Hx + w$ with structured channel matrices.","Detector complexity comparisons become meaningful on a common footing: for example, the paper's measurements show CD-MAMP provides BER close to OAMP at lower complexity.","The unified model gives a template for slotting future OTFS-related proposals into an existing analysis rather than treating each as a new modulation."],"supporting_citations":[{"why":"Defines OTFS and the ISFFT/SFFT transform pair that the paper reinterprets as FBMC precoding.","marker":"[15]"},{"why":"Supplies the OSDM modulation and demodulation expressions later shown identical to OTFS with rectangular pulses.","marker":"[36]"},{"why":"Supplies the V-OFDM block-wise IDFT/DFT structure identified as a special case of OTFS.","marker":"[39]"},{"why":"Defines OTSM, the delay-sequency variant, for the unified comparison.","marker":"[40]"},{"why":"Defines ODDM and its staggered pulse structure, the closest non-identical relative analyzed.","marker":"[41]"},{"why":"Introduces OCDM and the inverse discrete Fresnel transform used in the unified model.","marker":"[42]"},{"why":"Introduces AFDM and the discrete affine Fourier transform, the chirp-based member of the family.","marker":"[44]"},{"why":"Provides the discrete pulse-shaped OTFS transmit matrix form that makes the reduction to V-OFDM and OSDM explicit.","marker":"[58]"}],"fun_headline_variants":["OTFS, V-OFDM, OSDM share one linear model","OTFS is precoded FBMC, making V-OFDM and OSDM special cases","One math for OTFS, V-OFDM, OSDM in Doppler channels","OTFS, V-OFDM, OSDM: same math, different labels"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The unified model assumes the doubly-selective channel is a circular convolution with modulo-$MN$ indexing, which is only exact when the cyclic prefix exceeds the maximum delay spread and the per-sample gains fully track the time variation; in high-Doppler or very wideband conditions the circularity is approximate and every detector comparison built on it inherits that approximation.","fun_headline_variants_meta":{"raw":{"variants":["OTFS, V-OFDM, OSDM share one linear model","OTFS is precoded FBMC, making V-OFDM and OSDM special cases","One math for OTFS, V-OFDM, OSDM in Doppler channels","OTFS, V-OFDM, OSDM: same math, different labels"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001207,"raw_usage":{"total_tokens":4999,"prompt_tokens":1003,"completion_tokens":3996,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":619,"completion_tokens_details":{"reasoning_tokens":3905}},"tokens_in":619,"tokens_out":3996,"duration_ms":26501,"temperature":1.0,"reasoning_tokens":3905,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T22:34:55.557005+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the same data block through V-OFDM, OSDM, and rectangular-pulse OTFS on one physical doubly-selective channel with Nyquist sampling and measure the effective channel matrices: the paper's equivalence predicts they coincide exactly, so any material discrepancy under sufficient CP length would falsify the unification, as would a demonstration that a CP shorter than the delay spread still yields identical outputs.","supporting_citations":[],"review_version":1}